8th BigBrain Workshop - Challenges of Multimodal Data Integration

9 Sept 2024, 08:00 → 11 Sept 2024, 19:00 Europe/Rome

Palazzo della Salute

Katrin Amunts (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich), Alan C Evans (Montreal Neurological Institute McGill University Montreal), Paule-Joanne Toussaint (Montreal Neurological Institute McGill University Montreal), Susanne Wenzel (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich)

^{credit: Janet Lindenmuth, https://www.flickr.com/photos/j3net/4032850103}

---

You are cordially invited to attend the 8th BigBrain Workshop, taking place in the beautiful city of Padua, Italy, on September 10 and 11, 2024.

This workshop has established itself as the annual meeting place for the BigBrain community to come together and present their latest research, discuss prospects of the BigBrain associated data and tools, and brainstorm on how to better leverage high-performance computing and artificial intelligence to create multimodal, multiresolution tools for the high-resolution BigBrain and related datasets.

The BigBrain Workshop will be held in conjunction with a Training Day, taking place as a full-day event on September 9, on-site at the conference venue. 

Mark your calendars and plan to join us in Padua for this exciting event. We hope to see you there!

The event is free of charge but prior registration is required.  

Call for contribution

The workshop will be organized as a symposium, with both invited speakers and contributed talks as well as a poster session. We welcome short abstracts of current work and/or short proposals for future initiatives related to the BigBrain or similar data. Abstracts can be submitted until June 24 July 5th.

---

Keynote Speakers

Maurizio Corbetta HIBALL Lecture        in Brain Analytics and Learning

Michel Thiebaut de Schotten Sievers Lecture           in Computational Neuroscience

Michael Hawrylycz BigBrain Educational Lecture

---

Local Session: The Padova Neuroscience Center

Alessandro Salvalaggio

Alessandra Bertoldo

Lorenzo Pini

---

Important dates 

Registration open:	April 8, 2024
Call for contributions open:	April 8, 2024
Abstract submission due:	June 24, 2024 July 5, 2024
Acceptance notifications:	July 12, 2024
Registration deadline:	August 12, 2024
Training Day:	September 9, 2024
BigBrain Workshop:	September 10 and 11, 2024

 

---

Organizing Committee

Montreal Neurological Institute           McGill University Montreal           Alan C Evans           Paule-Joanne Toussaint	Institute of Neuroscience and Medicine (INM-1)           Forschungszentrum Jülich           Katrin Amunts           Susanne Wenzel​​
Local Organizing Committee:           University of Padova          Maurizio Corbetta          Alessandro Salvalaggio          Andreea Stefania Radu

Please contact the programme committee if you have any questions.  We will continuously update the information on this page and also share information via X / Twitter (@BigBrainProject) and e-mail. 

  

For some registered participants, mails from this platform seem to end up as spam. To make sure to receive all information, please check your junk folder.

Per default all times are given in CEST. To make sure to refer to your local time, please set your time zone accordingly. Click the menu at the top right of this page.

 

www.bigbrainproject.org

@BigBrainProject

 

 

Monday 9 September

08:30 Coffee and Registration

1 BigBrain Workshop Trainee Day - Welcome notes

Speakers: Prof. Alan C Evans (Montreal Neurological Institute McGill University Montreal), Krystle van Hoof (HBHL, McGill University)

BigBrain Project Educational Lecture: Michael Hawrylycz

2 The BRAIN Initiative Cell Atlas Network: Comprehensive multimodal atlases of the human brain

The BRAIN Initiative Cell Atlas Network (BICAN) is a collaborative effort between neuroscientists, computational biologists and software engineers to create a comprehensive atlas of the human brain. Supported by the U.S. BRAIN Initiative, the project is dedicated to advancing our knowledge of the brain by gathering and sharing new data that allows identifying a “parts list” of the brain, detailing its vast array of neurons and non-neuronal cells. The BICAN continues the work of the BRAIN Initiative Cell Census Network consortium which recently delivered a whole mouse brain transcriptomic atlas. In this Educational Lecture I will survey the open-access compendium of BICCN/BICAN data and tools that are available to help researchers better understand the structure and function of the brain and advance neuroscience research.

Speaker: Michael Hawrylycz (Modeling, Analysis, and Theory Group, Allen Institute for Brain Science)

10:15 Coffee and discussion break

Hands-on Session 1

3 Indirect network disconnection methods

Measuring behaviour and MRI signals to measure network-level dysfunction in patients complex, costly, and not easily implemented in clinical practice.
A possible solution to this problem has come in the last few years from the development of large databases of functional and diffusion MRI data, along with analytical pipelines for generating so-called ‘connectomes’, i.e. the ensemble of functional or structural connections among brain regions that are common in a large group of healthy subjects.

One method, known as ‘lesion network mapping’, computes whole-brain functional connectivity (FC) from the lesion by embedding the damaged brain region of a specific patient into the normative functional connectome, estimating which regions are connected to the site of damage. This functional map can serve as a proxy for network-level FC abnormalities caused by the lesion (functional disconnection, FDC). This method has been applied in several studies to investigate network dysfunction across a range of neurological and psychiatric conditions.
A similar method estimates structural disconnections from clinical structural MRI lesions. As for the functional disconnectivity maps, the lesion of patients is embedded into the normative structural connectome, that is a probability map of normal white matter tracts measured with advanced diffusion imaging protocols. In a structural disconnection (SDC) map, each voxel in the brain indicates the probability of structural disconnection caused by the lesion to healthy white matter tracts.

Similar approaches have been proposed for estimating indirect metrics of network damage requiring only clinical structural imaging in focal disorders (stroke, tumor), such as tract density index (TDI). This measure estimates the number of fibers damaged by a brain lesion.

In this session we will provide a general framework and a hands-on session to obtain structural and functional disconnection maps and tract density index from clinical MRI data.

• Theory: Applications, opportunities and limits of indirect disconnectivity approaches
• Hands on: How to obtain maps of structural and functional disconnection

Speakers: Alessandro Salvalaggio (Department of Neuroscience, University of Padova), Lorenzo Pini (Department of Neuroscience, University of Padova)

4 Introduction to the Allen Institute Single Cell Ecosystem

The Allen Institute for Brain Science (in collaboration with BICCN/BICAN) has recently created multiple large-scale transcriptomic taxonomies from single cell (sc)- and single nucleus (sn)-RNA sequencing, covering various species and disease states. To increase the usage and understanding of these taxonomies beyond their publications, the Institute created multiple web-based interactive tools. These tools allow neuroscience researchers to freely sort through, visualize, and analyze different taxonomy datasets directly from their browser. The first tool featured in this presentation is the Allen Brain Cell (ABC) Atlas, which is an interactive brain cell atlas featuring sc/snRNA-seq cell type taxonomies of: the whole mouse brain (Yao et al., 2023, Zhang et al., 2023), the whole human brain (Siletti et al., 2023), and two cortical regions (MTG and DLPFC) from brains with Alzheimer’s disease (Gabitto and Travaglini et al., 2024). Both the whole mouse brain and the MTG/DLPFC Alzheimer’s taxonomies feature spatial transcriptomic MERFISH data as well. The second tool featured is MapMyCells, which is a tool for mapping transcriptomic data onto the cell type taxonomies previously listed in the ABC Atlas. The third tool is the Cell Type Knowledge Explorer, which features: sc/snRNA-seq cell type taxonomies of the primary motor cortex of mouse, human, and marmoset (BICCN 2021); accompanying mouse patch-seq data showing electrophysiological and morphological characteristics (Scala et al., 2021), and an integrated mouse-human-marmoset taxonomy to identify cross-species cell type homologies (Bakken et al., 2021). All three tools will be demonstrated live with example use cases to introduce users how to use these tools for their own research questions and hypothesis testing. Users will need internet access, no prior software downloads are needed

Speaker: Rachel Hostetler (Allen Institute for Brain Science)

12:00 Lunch break

Hands-on Session 2

5 BigBrain data processing with CBRAIN

CBRAIN is a web portal that provides seamless access to high-performance computing clusters, and is a component of the NeuroHub ecosystem of neuroinformatics tools. This hands-on, interactive tutorial will cover the main functionalities of CBRAIN for the processing and management of data, illustrate them on BigBrain data, and demonstrate their interaction. Practical examples using scientific tools in CBRAIN will be covered along with how to access BigBrain-related datasets in various repositories.

**Expected learning outcomes: **

In this tutorial participants will learn how to access and use CBRAIN to access and process BigBrain data on HPC resources through an easy-to-use web portal. Specific topics covered will include:
* Working with data and files in CBRAIN
* Finding BigBrain datasets
* Processing BigBrain datasets with scientific tools
* Exploring and working with processing results
* Downloading and sharing the results

Speaker: Bryan Caron (NeuroHub, McGill University)

6 Sulcal morphometry for multiple subjects using BrainVisa / Morphologist

During this session, we will demonstrate how to perform a sulcal morphometry study for a set of T1 MRI images, using BrainVisa and especially its Morphologist toolbox.
The Morphologist pipeline will perform all preprocessings: brain alignment, brain extraction, tissues classification, cortical folds detection, structured representation of sulci, sulci identification, morphometric features extraction in a data table.
Users are then free tu use their favorite statistical or machine learning package to perform classification, regression, or correlation studies on populations.
The session will include manipulation of the main Morphologist pipeline, and some tools around, including 3D visualization, in order to show how users can operate more custom processing.
As it is a "hands-on" session, attending people are expected to come with a laptop with the BrainVISA distribution installed (https://brainvisa.info), and they may bring a small set of T1 MRI images to process. 3D Visualization demos will include data from the BigBrain image, preprocessed using Morphologist.

Speakers: Denis Rivière (Neurospin), Jean-Franois Mangin (Neurospin)

14:30 Coffee and discussion break

Hands-on Session 3

7 Making the multiscale organization of the human brain accessible to reproducible workflows using siibra-python

Understanding the human brain requires access to experimental data that capture relevant aspects of brain organization across a broad range of scales and modalities, and typically originate from a plethora of resources. To make multimodal and multidimensional measures of brain organization accessible, they need to be integrated into a common reference framework and exposed via suitable software interfaces. This tutorial will introduce participants to siibra toolsuite, which provides access to a multilevel atlas of the human brain built from “big data”. The atlas integrates brain reference templates at different spatial scales, complementary parcellation maps, and a wide range of multimodal data features. It links macroanatomical concepts and their inter-subject variability with measurements of the microstructural composition and intrinsic variance of brain regions, using cytoarchitectonic maps as a reference, and integrating the BigBrain model as microscopic reference template. The tool suite includes a web-based 3D viewer (siibra-explorer) and a Python library (siibra-python) to support a broad range of neuroscientific use cases. It makes use of EBRAINS as a data sharing platform and cloud infrastructure and implements interfaces to other neuroscience resources. The focus of this tutorial will be on building reproducible workflows with BigBrain data using the siibra-python library.

Speakers: Timo Dickscheid (Institute for Neuroscience and Medicine (INM-1), Forschungszentrum Jülich, Germany), Sebastian Bludau (Institute for Neuroscience and Medicine (INM-1), Forschungszentrum Jülich, Germany)

8 The BigMac Dataset: combining multi-contrast MRI and microscopy in the macaque brain

The BigMac dataset is an open access resource combining in vivo MRI, extensive postmortem MRI and multi-contrast microscopy for multimodal characterisation of a single whole macaque brain. The data is highly complementary to BigBrain facilitating complementary analyses across scales, datasets and species. Crucially, the BigMac MRI and microscopy data have been carefully co-registered together to facilitate quantitative multimodal analyses and data-fusion approaches which more heavily rely on multiple modalities being acquired in the same subject. This session will start with an introductory talk where I will outline the acquisition and curation of the BigMac data, alongside some example applications from the lab. The latter part of the session will feature hands-on interactive notebooks where participants can explore the BigMac data and curated outputs, as well as running microscopy-informed “hybrid” tractography and performing MRI-microscopy comparisons.

Speakers: Amy Howard (Wellcome Centre for Integrative Neuroimaging (FMRIB Centre), University of Oxford), Silei Zhu (Wellcome Centre for Integrative Neuroimaging (FMRIB Centre), University of Oxford)

16:15 Coffee and discussion break

Special Session NeuroAI

9 Approaches to brain control

Perturbing brain activity, e.g. through stimulation or brain-computer interfaces, can help us investigate brain mechanisms and restore normal activity patterns in subjects affected by neuropathologies. A major goal of applied neuroscience is to go beyond current perturbation approaches, that are largely empirical, and develop more advance, data- and theory-driven schemes to achieve controlled perturbations of brain activity. Any such scheme requires two key elements: (1) a computational model of brain activity, informed by observed neural activity data (2) a mathematical framework to design perturbations causing desired effects. In recent years, several authors argued that a linear approach, combining linear models of brain activity and linear control theory, could provide both. In this lecture, we will first describe the linear approach and its severe limitations. Next, we will discuss how advanced machine learning approaches might be used to overcome some of these limitations, possibly laying the ground for more advanced and effective neurostimulation experiments.

Michele Allegra is a physicist with a broad interest for neuroscience. His main research topic is the analysis and modeling of functional networks in the brain. Upon completing a Ph.D in quantum physics at the University of Turin and the ISI Foundation in Turin, he changed research field, moving into data analysis for neuroscience. He joined the Statistical Biophysics sector of the International School for Advanced Studies (SISSA), Trieste, where he worked in Prof. Alessandro Laio's group from 2015 to 2018. His research activity within Laio's group focused on advanced clustering techniques and their application to the study of dynamically changing brain networks. He deepened his focus on neuroscience during his stay at the Timone Institute for Neuroscience in Marseilles (2018-2021), where he joined the BraiNets group led by Andrea Brovelli. In Marseilles, he focused on the analysis of brain imaging data, with the goal of characterizing functional networks in the brain and their disruption in major diseases such as stroke. His current research at the Laboratory of Interdisciplinary Physics and the Padua Neuroscience Center, University of Padua, focuses on applying statistics, information theory, and complex system modelling to obtain new insights in neuroscience research.

micheleallegra.github.io/
pnc.unipd.it/allegra-michele/

Speaker: Michele Allegra (Padua Neuroscience Center, University of Padua.)

Panel Discussion: Integrating AI approaches for brain modelling

Rachel Hostetler, Amy Howard, Michele Allegra, Christian Schiffer
Moderator: Jane Roskams

18:30 Welcome Reception

Tuesday 10 September

08:30 Coffee and Registration

10 Welcome and Introduction

Sievers Lecture: Michel Thiebaut de Schotten

11 The emergent properties of the connected brain

Significant strides have been made in delineating the white matter architecture in the living human brain in the last two decades. These pathways have been identified as pivotal in supporting cognitive functions, with their variability closely associated with differences in cognitive performance, psychiatric conditions, and neurological manifestations. This underscores a hypothesis that brain functionality is not isolated within regions but emerges from the interaction facilitated by white matter connections. In our presentation, we will unveil cutting-edge methodologies developed recently in our lab – namely, the functionnectome and emuse – to explore these emergent properties. We will discuss their implications for understanding complex neuroscientific phenomena, such as consciousness and neuropsychological recovery post-stroke.

Speaker: Michel Thiebaut de Schotten (Neurofunctional Imaging Group, University of Bordeaux & Brain Connectivity and Behaviour Laboratory (BCBLab), Sorbonne Universities, Paris)

10:15 Coffee and discussion break

Contributed Talks - Multimodal data acquisition and processing (co-Chairs: Jordan DeKraker, Timo Dickscheid)

co-Chairs: Jordan DeKraker, Timo Dickscheid

12 Mapping neurotransmitter receptor distributions in the macaque cortex

Introduction: Quantitative maps of neurotransmitter receptor densities are important for characterizing the brain's molecular organization. We previously presented a 3D reconstruction pipeline for 2D autoradiographs to create 3D atlases at up to 50μm resolution [1]. Here, we use 3D reconstruction of autoradiographs from a macaque hemisphere to investigate patterns of receptor distribution and the balance of inhibitory, excitatory, and modulatory neurotransmitter receptors.
Methods: Four hemispheres (3 left, 1 right) from the brains of 3 adult male Macaca fascicularis were serially sectioned and visualized for 15 different neurotransmitter receptor binding sites using quantitative in vitro receptor autoradiography [2-3]. The 2D sections were reconstructed into 3D using our previously described BrainBuilder pipeline [1]. As no MRI was acquired for the macaque brains, the MEBRAINS template brain [4] was used as the reference volume to which the autoradiographs were reconstructed with our pipeline (Fig.1A). The 3D volumes were reconstructed at 0.5mm resolution for inhibitory (GABAA, GABAB), excitatory (AMPA, NMDA, Kainate), and modulatory (5-HT1A, 5HT2, M1, M2, M2, α1) receptors. Reconstructed volumes were flipped over the midline of the coronal plane and averaged together. Receptor densities were then projected onto the MEBRAINS surfaces and normalized by z-score. Vertex-wise gradients were calculated using 14 reconstructed receptor maps with principal component analysis and Pearson correlation. The ratios of excitatory glutamatergic to inhibitory GABA (E/I) receptors were calculated as well as the standard deviation and entropy of receptor densities. To identify unique receptor distribution patterns, each receptor was regressed onto the other receptor distributions with Elastic-Net, and the entropy of each receptor distribution was calculated over the cortex.
Results Fourteen 3D receptor maps were reconstructed (Fig.1B), and the first three principal components explained 49%, 14%, and 10% of the variance, respectively (Fig.1.C). The 1st gradient component highlights the visual cortex and segregates it from cortical areas associated with the default mode network and association cortices. This sensory-association axis was also reflected in the E/I ratio and measures of variance (standard deviation and entropy). The uniqueness of the receptor maps was characterized by calculating the linear dependence and entropy of the receptor maps, revealing a cluster of unique receptors: Muscarinic M2, 5-HT1A, α2, GABAA Benz., Adenosine 1 (Fig.1D).
Conclusion: We demonstrate gradients of receptor distribution across the macaque cortex with a particularly strong axis separating the visual cortex from the precuneus and posterior parietal cortex, both part of the default mode network [5]. The visual cortex presents a conspicuously low E/I ratio and high variability in receptor distributions. The 2D autoradiograph sections were reconstructed to 0.5mm resolution to provide gross anatomical information. The data supports reconstruction up to 50μm resolution. We will investigate microscale patterns of receptor distributions to elucidate the molecular architecture of the macaque brain at a previously inaccessible resolution.

[1] Funck, et al. 2022. biorxiv: https://doi.org/10.1101/2022.11.18.517039.
[2] Palomero-Gallagher & Zilles, 2018. Handb Clin Neurol 150: 355-387
[3] Rappan, et al. 2021. NeuroImage, 226:117574. https://doi.org/10.1016/j.neuroimage.2020.117574.
[4] Balan, et al. Imaging Neuroscience 2024;2:1–26. https://doi.org/10.1162/imag_a_00077
[5] Raichle. 2015 Annu Rev Neurosci. 2015 Jul 8;38:433-47: 10.1146/annurev-neuro-071013-014030

Speaker: Thomas Funck (Child Mind Institute)

BigBrain2024.png

13 Label-free biomolecular tissue analysis in the living brain via vibrational fiber photometry at arbitrary depth

Optical approaches for in vivo neural monitoring offer a precious window on brain functions and on the mechanisms of development, ageing or disease progression. Nonetheless, the existing methods still struggle to capture in situ the complex biomolecular alterations that accompany physiological and pathological dynamics. As a result, our grasp on the multifaceted components of brain activity is still limited.
To surpass this limitation, we propose a vibrational fiber photometry method based on spontaneous Raman scattering that allows monitoring the biomolecular content of arbitrarily deep brain volumes of the mouse brain in vivo without exogenous reporters. To do this, we employed a single, thin tapered optical fiber delivering and collecting optical signals to gather information on the local cytoarchitecture, to sense molecular alterations linked to circuit dysfunction caused by traumatic brain injury, and to detect diagnostic markers of brain metastasis with high accuracy.
In our view, vibrational fiber photometry offers an opportunity to capture a more comprehensive picture of neural activity in the biomolecular context of the local micro-environment. This capability, that can be employed alongside traditional fiber photometry or electrophysiological techniques, is particularly promising for empowering emerging research on brain-immune [1] and brain-cancer [2] bidirectional dynamics.

References
[1] Castellani, G. et al., Science (1979) 380, (2023).
[2] Mancusi, R. & Monje, M., Nature 618, 467–479 (2023).

Speaker: Filippo Pisano (Department of Physics and Astronomy, University of Padova, Padova, Italy & Padova Neuroscience Center, University of Padova, Padova, Italy)

14 Adaptation of FreeSurfer v7.4 pipeline for automated volumetric parcellation and cortical surface extraction of BigBrains 1 and 2

Introduction:

In 2013, we published BigBrain1 (BB1), a high-resolution (20µm^3) histological 3D-reconstructed model of the human brain (Amunts et al., 2013). Over the past several years, progress has also been made on BigBrain2 (BB2) (Mohlberg et al., 2022; Lepage et al., 2023), with preliminary reconstruction and segmented volumes [white matter (WM) and gray matter (GM)] now available (100µm^3).

Here, we utilize the segmented volumes as input to an MRI simulator, which can then be used as input to an adapted FreeSurfer (FS) v7.4 pipeline.

Outputs of this pipeline include:

(1) FS white and gray cortical surface extractions for BB1 and BB2, which may improve multimodal surface matching (MSM) surface registration (Robinson et al., 2018). Although MSM was developed and parameterized on FS surfaces, we (Lewis et al., 2023) have used it in the past to register (CIVET-extracted) BigBrain surfaces to structural MRI-derived population average surfaces that serve as important reference frames for multimodal data integration, e.g. FS’ fsaverage (Fischl 2012) and Human Connectome Project's (HCP) fs_LR (Van Essen et al., 2012). FS-extracted surfaces possess tessellation and medial cut methodologies expected to be more compatible with MSM surface registration.

(2) Initial automated FS volumetric parcellations (wmparc.mgz) of BB1 and BB2, which can provide regions of interest for higher-resolution analyses, such as FS’ hippocampal subfield segmentation (Iglesias et al., 2015).

Methods:

We created 2 versions each of segmented volumes (200µm^3) of BB1 and BB2. In both versions, we pre-masked the hippocampus and amygdala as GM. In version (A), we additionally masked all other subcortical gray regions as WM (to improve subsequent cortical surface extraction). This version was used as input to mrisim (Kwan et al., 1996; modified to allow high-resolution input and output) to create a simulated T1-weighted volume (200µm^3) which was then used as input to Freesurfer (FS) v7.4. In version (B), we did not pre-mask the subcortical gray regions. Version (B) was used as input to mri_synthseg (Billot et al., 2023), as described below.

The simulated T1 volumes were submitted to a modified version of the FS -hires (200µm^3) pipeline, with the following most notable changes. Instead of the default mri_ca_label, we used mri_synthseg (v1), which we found was superior for our subcortical masking needs (both in terms of quality and efficiency). For surface extraction, we additionally modified recon-all to produce high-resolution surfaces (~500k vertices per hemi).

While the white surfaces were of decent quality, we found that the pial surfaces were severely underexpanded for high-resolution surfaces (~500k vertices). Thus we had to additionally modify: mris_autodet_gwstats; mrisurf_mri.cpp; mris_place_surface.cpp.

Results:

Fig 1A shows examples of FS white and gray surface extractions for BB2.

Fig 1B shows automated FS volumetric parcellation (wmparc.mgz) for both BB1 and BB2 (200µm^3).

Fig 1C shows the FS hippocampal subfield segmentation output when the histological intensity volume (113µm^3) is input to mimic a T2-weighted volume.

Speaker: Lindsay Lewis (McGill University)

BB Abstract Figure(1).png

15 The Extremely Brilliant Brain: The Isotropic Micrometric Human Brain Dataset

Introduction
Brain atlases derived from MRI are a common tool for neuroscientists to understand the anatomy of the brain. However, as MRI has a limited resolution, these tools give a poor insight into fine structures [1]. This is why different groups have developed microscopy-based atlases which nevertheless require hours of sequential cutting and mapping [2,3]. Thus, we unveil a 7.72-micron brain dataset acquired with HiP-CT (Hierarchical Phase-Contrast Tomography), enabled by the Extremely Brilliant Source upgrade of European Synchrotron Radiation Facility (ESRF, Grenoble) and its beamline BM18, as a proof of concept for high-throughput anatomical studies.

Methods
The whole brain was obtained from LADAF (Grenoble). As part of our published protocol [4], it was fixed in formalin and prepared in a 70% EtOH / agar mix, followed by degassing. HiP-CT scanning was performed at the beamline BM18 of the ESRF with an isotropic voxel size of (7.72 μm)3; reconstruction and phase retrieval were performed with the PyHST2 toolbox. The dataset was aligned to the BigBrain space [2] using voluba (https://www.ebrains.eu/tools/voluba). Finally, structure tensor analysis [5] was conducted on the dataset.

Results & Discussion
Following the structure tensor analysis (cf. Figure 1), the fiber tracts can be studied in areas which were not resolved with MRI like the zona incerta [6]. Besides, the resolution enables the study of both white matter and blood vessels at the same time, along with the segmentation of smaller structures such as the choroid plexus. The strength of this dataset lies in the resolution, and in the isotropic and distortion-free imaging; thus, it should be used in a similar and complementary fashion to the BigBrain [2]. Alignment within the BigBrain space will enable the comparison of HiP-CT data with complementary microscopic modalities such as cytoarchitectonic maps and polarized light imaging.

Figure 1 (cf. attached file): Fractional-anisotropy map of a coronal slice of the Extremely Brilliant Brain, which reveals fibers in the striatum.

Conclusion
This unique dataset enables a label-free study of the brain at a micrometric scale, which bridges low-resolution in vivo techniques and high-resolution microscopy.

---

References
[1] K. H. Maier-Hein et al., Nature Communications 2017, 8, 1349.
[2] K. Amunts et al., Science 2013, 340, 1472.
[3] S. Ding et al., Journal of Comparative Neurology 2016, 524, 3127.
[4] C. L. Walsh et al., Nature Methods 2021, 18, 1532.
[5] N. Jeppesen et al., Composites Part A: Applied Science and Manufacturing 2021, 149, 106541.
[6] S. N. Haber et al., Biological Psychiatry 2023, 93, 1010.

Speaker: Matthieu Chourrout (University College London, Department of Mechanical Engineering)

Figure1_HIBALL2024.png

Panel Discussion (Moderator: Katrin Amunts): The benefit of making multimodal data interoperable

Moderator: Katrin Amunts

16 BigBrain as part of the Multimodal Atlas at EBRAINS

Speaker: Katrin Amunts (Forschungszentrum Jülich (INM-1) and Cécile and Oskar Vogt Institute for Brain Research, University Hospital Düsseldorf, Medical Faculty, Heinrich-Heine-University Düsseldorf)

17 The McGill multimodal data ecosystem

Speaker: Alan C Evans (Montreal Neurological Institute McGill University Montreal)

18 Moderated discussion

Michael Hawrylycz, Jean-François Mangin, Amy Howard, Timo Dickscheid, Alan Evans
Moderation: Katrin Amunts

13:00 Lunch

Contributed Talks - Segmentation and AI (co-Chairs: Claire Walsh, Jussi Tohka)

co-Chairs: Claire Walsh, Jussi Tohka

19 CytoNet: A Deep Neural Network for Whole-brain Characterization of Human Cytoarchitecture

The characterization of cytoarchitecture in the human brain provides an essential building block for the creation of a high-resolution multi-modal brain atlas. Cytoarchitecture is defined by the spatial organization of neuronal cells, including their shape, density, size, cell type, as well as their columnar and laminar arrangement, which differ between brain regions. High-throughput light-microscopic scanning of large, cell-body stained histological sections obtained by sectioning postmortem human brains enables detailed examination of cytoarchitectonic organizational principles across multiple brain samples, which is mandatory to capture the highly variable cytoarchitectonic organization. The limited scalability of existing methods to image and analyze datasets in the terabyte to petabyte range motivates current developments of AI methods for data-driven characterization and classification of human cytoarchitecture at large scale.

In this work, we present CytoNet, a deep neural network model that enables data-driven characterization of cytoarchitecture in the human brain. CytoNet is a convolutional neural network that is trained on 200 000 image patches (2048px@2μm/px) extracted from 4115 histological sections of 9 postmortem brains. The model is trained using a novel contrastive learning objective that derives the similarity relationship between image samples from their spatial distance in a common reference brain space. Using this loss, CytoNet is trained to map spatially close image samples, which likely show similar cytoarchitectonic structures, to similar feature representations.

We demonstrate that feature representations extracted by CytoNet allow classifying cytoarchitectonic areas, predicting spatial and morphological features, studying inter-individual variations, and enabling data-driven quantification and query-based exploration of microstructural principles at whole-brain level. Moreover, we show that the latent space learned by CytoNet exhibits an anatomically highly plausible structure that facilitates intuitive exploration of brain organization. CytoNet significantly extends existing methods for cytoarchitecture analysis and thus provides the foundation for novel analysis workflows that have the potential to facilitate studies relating the brain’s microstructure to connectivity and function.

Speaker: Christian Schiffer (Forschungszentrum Jülich)

20 Autoencoders for cluster analysis of rat brain histology

Introduction. Extracting quantitative information from the whole brain in histology is one big challenge in neuroscience [1]. Software programs [2] helped in histological quantification, although being time-consuming and relying on individual expertise. Further, machine learning allowed the automation of processes such as segmentation [3] and classification [4], but being, at best, semi-supervised. Here, we propose a novel method to automatically cluster rat brain histology using autoencoders.

Methods and materials. The method consists of first, using autoencoders (AE) to extract local tissue features from photomicrographs of myelin-stained rat brain sections; and second, to cluster the AE-derived features from patches of the sections using Gaussian mixture models (GMM). AE acquire a compressed representation of the tissue patterns from the myelinated axons. The compressed representation (latent space) can be used to create unsupervised maps using GMM, thus automatically separating tissue properties.

We used photomicrographs of myelin-stained sections from brains of three rats sham-operated and three rats after mild traumatic brain injury (mTBI). RGB images were converted into grayscale and differences in the staining intensity among brain sections were corrected by histogram matching, both between and within animals. After that, an AE was trained with 3 random million square patches of 64x64 pixels from the image sections, and the dimensionality of the latent space was set to 128. After training, all brain sections were passed through the AE, thus converting every 64x64 pixel patch into 128 values that are clustered using GMM. The optimal number of clusters was decided based on Bayesian Information Criterion (BIC) score.

Results. The resulting maps separate regions that vary in density and organization of myelinated axons. The figure shows two coronal sections from a sham (panel A) and mTBI brain (panel B), and their corresponding 10-cluster-maps (panels C and D). Clusters 7:9 represent highly myelinated areas, i.e., white matter areas such as the corpus callosum and optic tract. The cluster maps show the lamination in the sham cortex, whereas in the mTBI cortex the lamination changed at the lesion site. The sham external capsule is represented by clusters 8 and 9, and mTBI external capsule denotes less myelin content, with clusters 5:7. The granule cell layer of the sham is represented by cluster 2, while in the mTBI animal is represented by clusters 0 and 1. The overall presence of cluster 5 is higher in the mTBI animal compared to the sham-operated animal.

Discussion. The proposed method allows to classify and quantify tissue properties, thus allowing to identify local features of histological sections. Notably, the method is suitable for extracting information from massive histological samples in an automatic manner. Currently, we are working on training an AE with myelin-stained sections from more animals. Next steps will be to add Nissl-stained sections from the same animals, as well as to combine the information of both stainings. By co-registering and clustering consecutive sections stained for myelin and Nissl, we expect to distinguish histopathological changes in the whole brain after mTBI.

Speaker: Melina Estela Dalmau (University of Eastern Finland)

abstract_figure.tiff

21 Step by Step: Towards a gapless 1 micron BigBrain with Diffusion Models

Advances in microscopic imaging and high-performance computation have made it possible to analyze the complex cellular structure of the human brain in great detail. This progress has greatly aided in brain mapping and cell segmentation, leading to the development methods for automated analysis of tissue architecture and cell distribution in histological brain sections. However, histological image data can contain data gaps due to inevitable processing artifacts, which, despite careful precautions, may arise during histological lab work, such as missing sections, tissue tears, or inconsistent staining.

To address this issue, we present a convolutional neural network model that reconstructs missing or corrupted data from surrounding tissue, while preserving precise cellular distributions. Our approach is based on recent advancements in image generation and involves utilizing a denoising diffusion probabilistic model (DDPM) that is trained on light-microscopy scans of cell-body stained histological sections. We extend this model with the RePaint method to impute missing or replace corrupted image data.

To validate the model, we propose two new validation metrics based on two established deep learning models that were trained on the same type of data. In validation, we want to confirm a) the correct reproduction of cell statistics like cell size and count, and b) the generation of plausible cytoarchitectonic patterns, including brain area-specific laminar and columnar organization . We compare cell statistics using CPN, a cell segmentation model that provides precise cell statistics. Additionally, a model trained for cytoarchitecture classification is used to validate the structure of the inpainted regions by comparing how the inpainting process affects classification performance.

We find that images generated by the proposed DDPM exhibit realistic and anatomically highly plausible cell distributions, effectively filling in data gaps resulting from histological artifacts. The model achieves low errors in cell statistics of less than 10% and high accuracies in cytoarchitecture classification of above 85%, even with inpainted regions as large as 50% of the input patch. Our results demonstrate the potential of the proposed generative model to improve the accuracy and completeness of analysis workflows for histological brain imaging data and to provide the basis for the development of future whole-brain human brain atlases.

Speaker: Jan-Oliver Kropp (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich, Germany & Helmholtz AI, Forschungszentrum Jülich, Jülich, Germany)

22 BigBrain Image blind restoration and alignment with generative priors, U-Net and structural similarity

Human brain atlases play a crucial role in providing a spatial framework to organize information derived from diverse research on brains, integrating multimodal and multiresolution images. The preprocessing stage is crucial for enhancing the quality of the images, addressing issues related to degradation, and ensuring that subsequent analyses and interpretations are based on reliable and accurate data. By employing sophisticated preprocessing techniques, researchers can mitigate the impact of unknown degradations in the scanned BigBrain Images, resulting in improved image quality and, consequently, more robust findings in neuroscientific studies.

We present a pipeline that depends on two models to effectively address actual image restoration and alignment when the relationship between high resolution and low resolution images is unknown.

We propose a blind super-resolution model to address the resolution upscaling scenario when the function for mapping high- and low-resolution images is unknown. Our solution relies on three training modules with different learning objectives: 1. a degradation-aware network (U-Net)[1,4] to synthesize the high resolution image, given a low resolution image and the corresponding blur kernel; 2. a pre-trained generative adversarial network (GAN) to be used as prior, bridged to the U-Net by a latent code mapping and several channel-split spatial feature transforms (CS-SFTs); and 3. a rational polynomial image interpolation[2] into deep convolutional neural networks (CNNs) to retain details.

This pipeline considers the generic problem of dense alignment between two images, whether they be two frames of a video, two widely different views of a scene, two paintings depicting similar content, etc. Whereas each such task is typically addressed with a domain-specific solution, Near-Duplicates interpolation or alignment is an interesting new application, but large motion challenges existing methods. To address this issue, we adopt a feature extractor that shares weights across the scales and optimizes our network with the Gram matrix loss that measures the correlation difference between features. Then the fine alignment is learned unsupervised by a deep network that optimizes a standard structural similarity metric (SSIM) between the two images. The results on BigBrain images show the performance of the proposed approach.

References
1. Amunts K, Lepage C, Borgeat L, Mohlberg H, Dickscheid T, Rousseau MÉ, Bludau S, Bazin PL, Lewis LB, Oros-Peusquens AM, Shah NJ, Lippert T, Zilles K, Evans AC. BigBrain: an ultrahigh-resolution 3D human brain model. Science. 2013 Jun 21;340(6139):1472-5. doi: 10.1126/science.1235381.
2. Wang, X., Li, Y., Zhang, H., & Shan, Y. (2021). Towards real-world blind face restoration with generativefacial prior. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp.9168-9178).
3. Bian, S., Xu, X., Jiang, W., Shi, Y., & Sato, T. (2020, October). BUNET: blind medical image segmentation based on secure UNET. In International Conference on Medical Image Computing and Computer-AssistedIntervention (pp. 612-622). Springer, Cham.

Speakers: Mingli Zhang, Dr Paule-J Toussaint (Mcgill University)

fig2.png

23 Analyzing Regional Organization of the Human Hippocampus in 3D-PLI Using Contrastive Learning and Geometric Unfolding

Quantifiable and interpretable descriptors of nerve fiber architecture at microscopic resolution are an important basis for a deeper understanding of human brain architecture. 3D polarized light imaging (3D-PLI) provides detailed insights into the course and geometry of nerve fibers in whole postmortem brain sections, represented in large datasets. The large amounts of data, combined with complex textures in 3D-PLI images, however, make analysis challenging and limit access to data annotations. To this end, we propose using self-supervised contrastive learning to extract deep texture features for fiber architecture in 3D-PLI. We use the texture features to analyze the regional organization of the human hippocampus in combination with geometric unfolding to reduce the effects of its folded topology and project the features to a canonical reference space.

We analyze the fiber architecture of a human hippocampus of an 87-year-old male, measured with a polarizing microscope (PM) at 1.3 µm in-plane resolution on 60 µm thick brain sections. The volume comprises 545 brain sections, each 26757 × 22734 pixels in size. We apply contrastive learning to learn robust and descriptive representations by contrasting similar (positive) and dissimilar (negative) pairs of texture examples. Here, we leverage the volume reconstruction of individual brain sections in the learning objective to identify positive pairs based on a fixed distance between example image patches either in-plane (CL-2D) or across brain sections in 3D (CL-3D) (Fig. 1A). The objective is used to train a width-reduced ResNet-50 architecture on the full hippocampus, extracting 256 texture features for square patches of 128 pixels size (166 µm). After training, inference is performed using a sliding window approach to generate feature maps for whole brain sections (Fig. 1B). To analyze the folded architecture of the hippocampus, we apply HippUnfold and sample features from the feature maps at multiple depths of the pyramidal layer of the hippocampal Cornu ammonis (CA) region and the subicular complex (Fig. 1C). Subsequently, PCA is performed to reduce feature dimensionality for visualization and improve computational stability in further analysis (Fig. 1E).

To assess how well the deep texture features reflect the regional organization of the hippocampus, we perform k-means clustering for 6 clusters and compare the results with subfield labels. Clusters in CL-3D features show good visual agreement with hippocampal CA1 - CA4 regions and the subicular complex. In terms of mutual information (0.72), they align more clearly compared to clustering of baseline characterizations based on fractional anisotropy and mean transmittance (0.40), as well as CL-2D (0.61).

Without any supervisory signal, CL-3D features form a well-structured embedding space, following the general regional organization pattern of the hippocampus and additionally highlight an expected functional rostro-caudal heterogeneity. Projecting deep texture features to unfolded space using HippUnfold enables subsequent comparison with diverse modalities. This work thus lays the foundation for incorporating 3D-PLI texture information into a comprehensive multimodal mapping of the human hippocampus.

Speaker: Alexander Oberstrass

figure1.pdf

24 Towards Universal Instance Segmentation Models in Biomedical Imaging

Precise instance segmentation is a critical part of many fields of research in biomedical imaging. One key challenge is applying models to new data domains, typically involving pre-training a model on a larger corpus of data and fine-tuning it with new annotations for each specific domain. This process is labor- intensive and requires creating and maintaining multiple branched versions of the model. Working towards universal instance segmentation models in biomedical imaging, we propose to unify domain-adapted model branches into a single multi-expert model, following a foundation model paradigm. Our goal is to replace most existing fine-tuning scenarios with prompt-based user instructions, allowing the user to clearly state the task and object classes of interest. We hypothesize that such a combined approach improves generalization, as the base model can benefit from datasets that were previously only used for fine-tuning. A key challenge in the creation of such models is to resolve training conflicts and ambiguity in a pragmatic fashion when combining different segmentation tasks, datasets, and data domains. Such conflicts can occur if datasets focus on different classes in the same domain. For example, some datasets annotate all cells in microscopy images, while others focus on cells of a specific cell type. A naïve combination of such sets would create an ill-posed learning problem for most models, requiring them to infer their task from their input, which is undesirable in a universal setting. Models like SAM and MedSAM highlight the potential of prompting, but often require external detectors and fine-tuning. Here, we propose to leverage prompt-based task descriptions as a tool to manipulate general model behavior, such that user instructions yield domain expert models. We test our approach by training a Contour Proposal Network (CPN) on a multi-modal data collection, including the TissueNet dataset. Prompts, such as “cell segmentation” or simply “nuclei”, modulate underlying features, allowing the CPN to segment the respective object classes in TissueNet with a mean F1 score of 0.90 (0.88 for cells, 0.92 for nuclei), compared to 0.84 (0.81, 0.87) without prompting. Overall, the proposed approach introduces an interactive linguistic component that allows the conflict-free composition of various segmentation datasets, thus allowing to unify previously separated segmentation tasks. With that, we consider it an important step towards universal models.

Speaker: Mr Eric Upschulte (University Hospital Düsseldorf, Cécile & Oscar Vogt Institute for Brain Research)

segmentation-examples.png

16:00 Coffee and discussion break

LOC Session (Chair: Maurizio Corbetta)

Chair: Maurizio Corbetta

25 Brain connectivity and glioma: a new approach

The emerging field of “cancer neuroscience” reveals intricate functional interplays between glioblastoma and the brain’s normal cellular architecture encompassing neurons, glia, and vessels. Recent investigations underscore the role of structural and functional brain connections within within hierarchical networks, known as the connectome. These connections contribute significantly to glioblastoma’s location, spread, recurrence, and overall survival, revealing a complex interplay at the whole-brain level between the cancer and the nervous system. This mounting evidence prompts a paradigm shift, challenging the perception of glioblastomas as mere foreign bodies within the brain. Instead, these tumours are intricately woven into the structural and functional fabric of the brain. This radical change in thinking holds profound implications for the understanding and treatment of glioblastomas, which could unveil new prognostic factors and surgical strategies and optimise radiotherapy. Additionally, a connectivity approach suggests that non-invasive brain stimulation could disrupt pathological neuron-glioma interactions within specific networks.

Speaker: Alessandro Salvalaggio (Department of Neuroscience, University of Padova)

26 The Clinical Connectome: From Neurodegenerative to Focal Brain Diseases

At rest, our brain is never truly at rest. Even in the absence of external input, the brain continues to engage in a variety of intrinsic processes. This organization, termed the "functional connectome," consists of a hierarchical scaffold organized into polyfunctional neural networks, complemented by the "structural connectome," which includes distal and local structural connections between brain regions. Recent theories suggest that this functional and structural scaffold may form the foundation of cognitive abilities. Brain conditions that affect neural health impact the connectome, and the breakdown of connectivity can predict cognitive deficits across a broad range of neurological diseases. This reinforces the idea that the connectome is a fundamental characteristic of cognitive processes.
In this presentation, we will discuss the relationship between brain structural and functional connections and behavior in several neurological diseases, including proteinopathies, stroke, and brain tumors. By examining how the connectome and various pathophysiological mechanisms interact, we can gain valuable insights into the underlying processes that support cognitive abilities. Additionally, we will introduce a recently founded project, in which a work package is dedicated to the characterization of the clinical connectome, spanning from degenerative to focal brain diseases. This project will serve as a foundational repository for studying the impact of pathophysiological mechanisms on brain connectivity.

Speaker: Lorenzo Pini (Department of Neuroscience, University of Padova & Padova Neuroscience Center, University of Padova, Italy)

27 PET connectomics: Exploring brain network complexity from a molecular imaging perspective

tbd

Speaker: Alessandra Bertoldo (Padova Neuroscience Center, University of Padova)

Poster Session

28 Brain Signature for Emotional Burnout

Burnout syndrome is one of the forms of chronic occupation stress. There is no single view on the nature and structure of emotional burnout. Boyko's psychological construct of Emotional Burnout (EB) defines Syndrome as a mechanism of psychological defense in the form of complete or partial excluding of emotions in response to traumatic influences and includes three key stages: Anxiety Tension, Resistance, and Exhaustion. The neurophysiological mechanisms of emotional burnout remain insufficiently studied. Establishing burnout-specific changes in brain activity is necessary to understand the phenomenon of burnout and distinguish it from other emotional mental disorders. Defining the EEG markers of burnout was our aim. 752 volunteers, first-fifth year students from the Taras Shevchenko National University of Kyiv aged 18 to 26 years participated in this study. EEG was recorded during the resting state (3 min, closed eyes condition) monopolarly using EEG 23-channel system Neurocom. To establish EEG correlates of emotional burnout during rest state we used special software written in Python 3.6 to implement Power Spectral Density calculation, the interhemispheric and intrahemispheric average coherence and Detrended Fluctuation Analysis (DFA). We analyzed separate artifact-free EEG segments in all frequency bands from 0.2 to 45 Hz. Psychological testing was performed before the registration of EEG. To determine the formation of emotional burnout Boyko’s “Syndrome of emotional burnout” Inventory was used. The Exhaustion phase of emotional burnout was formed in 79 participants, and it was under development in 213 participants. In background EEG activity during the development of the Exhaustion phase of emotional burnout variations in EEG spatial synchronization were observed in low- and high-frequency EEG components and includes the formation of two separate networks of functional connections: interhemispheric prefrontal, anterior frontal, and frontal links (alpha and gamma low bands) and parietal-occipital links (alpha and gamma high bands). DFA describes the long-term temporal correlations in the cortex, which are involved in different aspects of brain functioning. We detected a high resting state DFA scaling exponent values (up to 0.90-0,95) under exhaustion development in the alpha 1 (left temporal, parietal area), alpha 2 (right frontal area), alpha3 (posterior regions). Obtained values of DFA exponent and average coherence suggest the exhaustion formation is accompanied by the changes in visual and verbal processing, emotional processes (discretion and analysis).
Keywords—Emotional burnout, Exhaustion, Detrended Fluctuation Analysis, Power Spectral Density, Functional connectivity

Speakers: Mr João Miguel Alves Ferreira (Universidade de Coimbra), Sergii Tukaiev (Taras Shevchenko National University of Kyiv)

29 Changes in connectivity between ‘higher order’ perceptual areas in apraxia after stroke

Introduction:
Limb apraxia, a disorder of skilled action not consequent on primary motor or sensory deficits, has traditionally been defined according to errors patients make on neuropsychological tasks. Recent lesion symptom mapping studies suggest extrastriate visual areas may be important in mediating them. This would suggest that perceptual deficits may account for some subtypes of apraxia. In this study we investigated the possibility of diaschisis affecting perceptual areas in the brain following left hemisphere stroke; and whether this related to patients’ apraxia deficits.
Methods:
We conducted a visual-perceptual localizer task involving 29 patients with left hemisphere stroke, comparing their performance to that of 17 age-matched healthy volunteers. Employing a standard block-design localizer task, participants were tasked with observing static colour photographs depicting familiar tools, headless bodies, non-tool objects, and scrambled versions of these stimuli (Valyear and Culham, 2010). Simultaneously, they engaged in a 1-back task. To pinpoint brain regions selectively engaged in tool-related visual processing, we conducted one-sample t-tests (p<0.05, FWE corrected) in each subject, seeking areas exhibiting heightened activation for tools in comparison to headless bodies, non-tool objects, and scrambled stimuli. Subsequently, we identified body-selective regions by contrasting activity for bodies against tools, non-tool objects, and scrambled stimuli. Psychophysical interactions were carried out to identify areas of diaschisis between these in patients, comparing them to healthy volunteers.
Results & Conclusion:
Our analyses consistently identified heightened activity for tools compared to other stimuli and bodies compared to other stimuli, respectively. These included pMTG (LOC) region and an anterior region along the IPS for tools and the extrastriate body area (EBA), known to selectively respond to human bodies and body parts when compared to objects and other control stimuli (Downing et al., 2001). The LOC region was consistently positioned laterally, ventrally, and anteriorly to EBA, in line with findings by Valyear and Culham (2010). There was no significant change in activation in any of these regions between healthy controls.
However, PPI analyses to unravel areas of functional disconnection (‘diaschisis’) identified that the left IPS showed increased connectivity with Left LOC in patients versus controls in the tools contrast; conversely, the left LOC and EBA regions appeared disconnected in the bodies’ task. The study did not involve many patients and therefore we could not identify a reliable effect of praxis errors on these specific functional disconnections.

Speaker: Elisabeth Rounis

Figure1_slide1.pdf

Figure2_PPIresults_Slide2.pdf

30 Cytoarchitectonic mapping of five new areas in the anterior lateral prefrontal cortex

The region from the vertical part of the intermediate frontal sulcus (infs-v) to the anterior inferior frontal gyrus (ifg) in the lateral prefrontal cortex is an extensive region that contains many functional subregions, which play essential roles in various human functions, including motor preparation (Vogt 2007), working memory (Mizuno 2008), empathy (Cui 2015), language control (Vingerhoets 2003, Abutalebi 2009), voluntary eye movement (Kleiser 2017) and music perception (Hyde 2011). However, their structural correlates are largely unknown because areal borders are not reliably associated with macroanatomical landmarks, especially in this region which exhibits a high variability in the sulcal pattern between individuals.
Five new areas were identified by analyzing the cytoarchitecture in serial sections of ten human post mortem brains, including the two BigBrains. Areal borders were detected by an observer-independent mapping approach (Schleicher 1999). The areas were named according to their anatomical localization: INFS1 (intermediate frontal sulcus area 1), IFMS1 (intermediate frontomarginal sulcus 1), MFG3 (middle frontal gyrus area 3), IFG1 and IFG2 (inferior frontal gyrus area 1-2), were arranged in a dorsal to ventral as well as rostral to caudal orientation. They were granular areas with well-developed layer IV. Area INFS1 had a detectable layer IV, without larger pyramidal cells in layer III and V. IFMS1 had larger cells in the deeper part of layer III and V than INFS1, and its layer IV was not as developed as INFS1. MFG3 was similar to IFMS1, while the cell size in the deep part of layer III and V was larger than IFMS1. IFG2 had a relatively diffuse layering with densely packed cells. Area IFG1 contained a thin layer IV with a blurry border with layer III and V, and few large pyramidal cells in the deeper layer III. A hierarchical cluster analysis was used to quantify cytoarchitectonic similarities of the newly mapped areas and neighboring cortices. INFS1 was more similar to IFMS1 and MFG3 than IFG1 and IFG2, while IFG1 was more distinct different from the other four areas. In addition, the newly defined five areas showed higher cytoarchitectonic similarity with the areas of the mfg (MFG1 and MFG2) and Broca’s areas 44 and 45 than to the frontal pole (Fp1), lateral orbitofrontal cortex (Fo5 and Fo6) (Amunts 2022).
The new areas have been 3D reconstructed and superimposed in reference spaces (MNI Colin27 and ICBM152casym). Additionally, the calculated 3D-probabilistic maps showed their interindividual variability. Furthermore, ultra-high resolution cytoarchitectonic maps of the areas were reconstructed in the BigBrain space (Schiffer 2021). The new maps will be implemented in the Julich Brain Atlas and provide a reference for localizing results from functional imaging studies and linking them to cytoarchitectonic areas.

Speaker: Zien Zhang

Figure_ZienZhang.pptx

31 Fast large-scale spiking neuronal network simulations with NEST GPU: memory and communications optimization

Efficiently simulating large-scale spiking neuronal networks is crucial for advancing neuroscientific research, where both high simulation speed and minimal memory consumption during network instantiation are essential. NEST GPU [1,2], a high-performance GPU-MPI library developed within the NEST Initiative [3], showcases remarkable simulation speeds for various network sizes, leveraging the computing power of GPU-based systems to accelerate research endeavours. To maximise the utilisation of existing and future computational resources of multi-GPU systems, it is of paramount importance to take into account the implementation of software structures that manage remote connections (i.e., connections between neurons located on different GPUs) and facilitate the efficient communication of spikes across these GPUs. Recent advancements in NEST GPU have been primarily concentrated on following this approach, enhancing its performance and scalability within multi-GPU environments.
Building upon previous work of dynamically constructing networks directly in GPU memory [4], the approach has been expanded to multiple GPUs operating in parallel. This has resulted in significant performance gains. To ensure accuracy, a validation pipeline is being established that automatically compares the spiking activity of neuroscientifically relevant models, such as the cortical microcircuit [5] and the multi-area model of macaque vision-related cortex [6], against their CPU counterparts as a reference. Additionally, an update is provided on the ongoing efforts to align the GPU and CPU components of NEST.

Acknowledgments:
European Union’s Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreement No. 945539 (Human Brain Project SGA3).
Initiative and Networking Fund of the Helmholtz Association in the framework of the Helmholtz Metadata Collaboration project call (ZT-I-PF-3-026).
Joint Lab “Supercomputing and Modeling for the Human Brain”.
Italian PNRR MUR project PE0000013-FAIR CUP I53C22001400006, funded by NextGenerationEU.
We are grateful for the use of Fenix Infrastructure resources, which are partially funded from the European Union’s Horizon 2020 research and innovation programme through the ICEI project under the Grant Agreement No. 800858. The authors further thank the INFN APE Parallel/Distributed Computing laboratory and the IAS-6.

[1] Golosio et al., Fast Simulations of Highly-Connected Spiking Cortical Models Using GPUs. Frontiers in Computational Neuroscience, 15, 627620, 2021.
[2] Tiddia et al., Fast Simulation of a Multi-Area Spiking Network Model of Macaque Cortex on an MPI-GPU Cluster. Frontiers in Neuroinformatics, 16, 883333, 2022.
[3] Gewaltig, Marc-Oliver, & Markus Diesmann. “NEST (neural simulation tool).” Scholarpedia 2.4 (2007):1430
[4] Golosio et al., Runtime Construction of Large-Scale Spiking Neuronal Network Models on GPU Devices. Applied Sciences. 13(17), 9598, 2023
[5] Potjans, T. C., & Diesmann, M. (2014). The cell-type specific cortical microcircuit: relating structure and activity in a full-scale spiking network model. Cerebral cortex, 24(3), 785-806.
[6] Schmidt et al., A multi-scale layer-resolved spiking network model of resting-state dynamics in macaque visual cortical areas. PLOS Computational Biology, 14(10), e1006359, 2018.

Speaker: Luca Pontisso (INFN - National Institute for Nuclear Physics)

32 Machine learning analysis of astrocytes in Sharpin mutation mouse model.

Astrocytes are a type of glial cell that play key roles in maintaining homeostasis of neuronal circuits, regulating neuronal activity, and modulating inflammation. Alzheimer's disease (AD), a common neurodegenerative disorder in the elderly, is characterized by amyloid beta plaques and tau tangles. The chronic inflammation observed in AD is mediated by glial cells including astrocytes, which not only exacerbate disease progression but also lead to a decline in their original glial cell functions. GWAS (Genome-wide association studies) on East Asian AD patients have most prominently observed an association between Sharpin mutations and brain shrinkage. Sharpin is a synaptic protein that binds to the Shank family and is also a component of LUBAC, which regulates NF-κB activity, suggesting its role in synaptic function and inflammation regulation. To investigate the impact of Sharpin mutation-induced changes in the inflammatory response on astrocyte activation, we observed the morphological changes in astrocytes in Sharpin mutant mice comparing to wild-type. Precise quantification of astrocyte activity is challenging and recent research has been attempting to automate microscopic image analysis and classification of them. In this study, we adapted one of those newly developed machine learning algorithms called “Morphious” and optimize it to analyze astrocytes in our mouse models. The optimized machine learning system learned the difference of astrocytes from wildtype and AD model mice. Afterward, Sharpin mutant mouse brains were tested to analyze the impact of the Sharpin mutation on astrocytes. We also measured the number of processes, process length, and stained area of astrocytes using automated batch image processing to characterize astrocyte activation regulated by Sharpin.

Keyword : Alzheimer’s disease, Sharpin, Morphious

“This research was supported by a grant of the Korea Dementia Research Project through the Korea Dementia Research Center(KDRC), funded by the Ministry of Health & Welfare and Ministry of Science and ICT, Republic of Korea (grant number: HU23C0199)”

Speaker: Bo-Ram Mun (Chonnam national university)

33 Mapping superficial white matter architecture on BigBrain

INTRODUCTION.
The superficial white matter (SWM) is the layer of white matter (WM) located immediately underneath the cortex. This SWM contains subcortical U-fibers interconnecting adjacent brain gyri, which remain incompletely myelinated until later in life [1]. U-fibers play a key role in brain plasticity and aging, and alterations in these fibers have been observed in conditions such as autism, epilepsy, and Alzheimer's disease [2,3,4]. Despite its importance, the SWM has been understudied, primarily due to technical difficulties and limitations [5]. Recent advances in ultra-high field 7 Tesla magnetic resonance imaging (MRI) technology have enabled precise imaging and mapping of brain microstructure, leading to reliable research on the SWM. However, the structural and functional role of the SWM is still unclear and histological data could improve the understanding of these relationship. Specifically, BigBrain histological data could unravel complex microstructural properties by characterizing changes in intensity related to the SWM oligodendroglia organization5. In this regard, this study assesses histological features of the SWM, and evaluate its association to macroscale motifs of brain organization.

METHODS.
We preprocessed BigBrain histology data by correcting the outliers in temporal poles and intensity changes along the y-axes. To examine the SWM, we solved the Laplace equation over the WM domain. This was achieved by initially computing a Laplace field across the WM and subsequently shifting an existing WM surface along that gradient. Stopping conditions were set by the geodesic distance traveled. SWM surfaces were sampled at fifty depths, each separated by 0.06 mm, beneath the gray and white matter interface. The microstructure intensity profiles, depicting the intensity values of BigBrain features, are presented in Fig. A. Additionally, we sample the gray matter intensities to correlate the gray matter profile with its SWM at each vertex. Finally, we applied diffusion map embedding (DM), a non-linear dimensionality reduction on the data. Vertex-wise intensity profiles were cross-correlated, resulting in microstructural profile covariance (MPC) matrices that represent vertex-specific similarity in histological feature across the SWM. The MPC matrix was converted to a normalized angle affinity matrix, and diffusion map embedding was applied. This procedure identified eigenvalues (gradients) describing the main spatial axes of variance in inter-regional similarity of microstructural profiles. Gradient analyses were conducted using BrainSpace [6].

RESULTS
The spearman correlation between the gray matter and SWM intensity changes of BigBrain shows a positive pattern for most areas of the brain except the medial insula. (Figure A, bottom). After we applied DM, we observe that the two first eigenvalues or “gradients” explained ~75% of the variance. The principal gradient differentiated highly myelinated areas, like motor and pre-motor cortes from association areas (e.g. parieto-temporal and middle and superior frontal lobe).

CONCLUSIONS
We generated precise and reproducible SWM surfaces and proposed a novel method to study the cytoarchitectural similarity of the SWM on BigBrain. We observed a continuum between the gray matter lamination and the SWM structure that distinguishes it from the deep white matter. With this study, we open a new window to extend further cytoarchitectural studies to include the SWM.

Speaker: Youngeun Hwang (McGill University)

34 Modelling re-entry excitation and interventions in a personalized neural field model of the cortex

Computational modeling and dynamical systems theory enhance our understanding of epileptic seizure dynamics and potentially provide new intervention strategies. Models like the Virtual Epileptic Patient (VEP) aim to make patient-specific predictions about the epileptogenic zone and subsequently identify targets for epilepsy surgery. The VEP is based on individual neuroimaging data, electrophysiological measurements of seizures, and a dynamical model of neural activity to construct a full brain network model of the patient. Currently, the VEP uses low-resolution neural mass models (NMM) to represent individual regions of the brain network. NMMs approximate neural activity within a single point, thus ignoring the spatial extent of the neural tissue and local propagation phenomena like the traveling waves observed in epileptic seizures. Therefore, we extended the model to high-resolution neural fields (NF) that represent the cortical sheet in its spatial extent.
We developed a high-resolution, personalized computational model for a patient with drug-resistant focal epilepsy in the left temporal lobe. Using T1-weighted and diffusion MRI combined with tractography, we reconstructed the cortical surface and estimated connections between surface points at a 1mm³ scale. Seizure dynamics were simulated using the two-dimensional Epileptor model in an excitable regime to test reentry effects, as suggested by empirical in-vivo and in-vitro studies. We applied the dynamical model to the cortical surface and explored the parameter space across coupling strengths and reentry frequencies. This approach revealed self-limiting excitations, spiral waves, and sustained reentry excitation. To terminate reentry, we tested two intervention strategies used in epilepsy treatment in our model: virtual thermocoagulation, which involved lesioning fiber tracts in the white matter to modify cortical connectivity, and virtual phase-dependent stimulation via virtually implanted electrodes.
Future research should focus on fine-tuning model parameters to match individual empirical data and optimize interventions.

Speaker: Jan paul TRIEBKORN (Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France)

Fig1_new.png

35 Novel insights into brain metabolism and functional coupling in healthy connectomes and their mismatch in pathology

Glucose stands as the primary fuel source driving the energy-demanding process of neuronal activity. Glucose metabolism can be explored in vivo in human by [18F]FDG-PET. DCM-based effective connectivity (EC) provides a linear mechanistic view of brain dynamics by conceiving it as a mixture of dissipative and solenoidal flow speaking to the concept of kinetic energy (Benozzo et al. 2023). This study aims to explore the association between kinetic energy and glucose metabolism by comparing EC-derived functional flow patterns with [18F]FDG-PET measurements in healthy controls (HCs), finally investigating the disruption of this coupling in gliomas.
RsfMRI data of 42 HCs are described in (Volpi et al. 2022), while rsfMRI data of 43 patients are in (Silvestri et al. 2022). EC matrices, estimated through sparse DCM (Prando et al. 2020), are decomposed into PCov (partial covariance of the neuronal states) and S (differential cross-covariance, temporal directionality of neural activation) as in (Benozzo et al. 2023). From dynamic PET data (60-min), individual Metabolic Connectivity (MC) matrices, expressing the relationships between the metabolic states of different brain regions, are estimated according to (Volpi et al. 2022). Standardized Uptake Value Ratio (SUVR) is derived from static PET.
Partial Least Square Correlation is applied to HC pairs: PCov-MC, S-MC (upper triangular matrices), SUVR-PCov and SUVR-S (nodal strengths for PCov/S, ROI values for SUVR). Generalizability of multivariate correlations were 7-fold cross-validated. For each generalizable effective-metabolic pair, the corresponding regression line was identified, defining a normality band at the 90-percentile of the HC distances from the line. Finally, patients’ metabolic and effective variables are projected onto the maximizing-covariance latent space identified in HCs. A tumor frequency map was obtained by combining tumor masks as weighted by the corresponding distance from the HC regression line (out-of-range patients only).
Cross-validation reveals good generalizability for PCov-MC and SUVR-S associations (r>0.76). The scatterplots between metabolic and effective scores for HCs and patient projections are in Fig. B. In both pairs, some patients notably diverge from the expected metabolic-effective coupling independently of tumor volume. As shown in Fig. C, SUVR-S pair is mostly altered for temporo-parietal tumors, while PCov-MC is mainly disrupted in patients with frontal lesions.
A dual metabolic-effective association emerges: one at local level (higher glucose uptake is related to sink nodes, i.e. nodes receiving inputs from the network) and one at network level (metabolic connectivity is related to undirected BOLD-signals covariance). Furthermore, our study unveils how these two distinct decoupling can differentiate patients based on the lesion location (and not its volume)—local disruption observed only for temporo-parietal lesions and network alterations for frontal lesions—providing novel insights into the physiopathology and the metabolic-kinetic link of glioma. While (Maleki Balajoo et al. 2022) already explored distinguishing pathologies (Alzheimer and Mild Cognitive Impairment) based on metabolism-function coupling, our study is a pioneer in gliomas and the introduction of EC measures allows both to link metabolism to the concept of kinetic energy, but also to decouple neuronal activity from the adverse effects of hemodynamic convolution, proposing metabolic-effective coupling as a new biomarker.

Speaker: Giulia Vallini (Department of Information Engineering, University of Padova)

figBigBrain.jpg

36 Switching patterns of cortical-subcortical interaction in the human brain

Aims. While human neuroscience has traditionally focused on the neocortex, recent literature highlights the key role of subcortical structures in brain dynamics and cognitive processes. We investigated cortico-subcortical interactions in the human brain at rest by analyzing dynamic functional connectivity at rest in a large cohort of healthy human
participants.

Methods. We considered 1078 human participants from the Human Connectome Project, for which resting-state fMRI scans were available. We computed sliding window functional connectivity (sw-FC) on windows of 60 s (Fig. 1A). sw-FC matrices were approximated by projecting onto the leading eigenspace, vectorized, and concatenated across windows and subjects. K-means clustering was used to identify a set of recurring sw-FC patterns or dynamic functional states (DFSs) (Fig. 1B).

Results. FC fluctuations were synchronized in cortex and subcortex. Cortical regions exhibited flexible connectivity with two core subcortical ‘clusters’ comprising, respectively, limbic regions (hippocampus and amygdala) and subcortical nuclei (thalamus and basal ganglia). We identified two alternating DFSs: in DFS1 the hippocampus coupled positively with the default mode network and negatively with the sensorimotor network, while the thalamus showed an opposite trend; in DFS2, this pattern of subcortical-cortical connectivity was reversed (Fig. 1C). Better cognitive health was associated with a stronger segregation of cortex and basal ganglia/thalamus in DFS1, and stronger integration in DFS2., i.e., with an alternation of states with higher and lower cortico-subcortical coupling.

Conclusions. Our findings hint at a general relevance of cortico-subcortical interactions in the generation of whole-brain spontaneous FC patterns in healthy subjects.

Speaker: Michele Allegra (Department of Physics and Astronomy, University of Padova)

illus_abs.png

37 The interaction between early sexual trauma and diagnosis of panic disorder on brain structural connectivity: A human connectome study

Background: Early traumatic experiences are associated with alterations in neural structures that increase risk factors for chronic psychopathology, including a 4- to 12-fold increase in the risk of developing psychiatric disorders (including symptoms of panic disorder (PD)) according to previous research. These early life adversities, such as early sexual trauma (EST), can have a lasting impact on an individual's brain development and functioning throughout adulthood. Although structural alterations in brain regions have been reported in previous studies, to date, no study has investigated connectivity in the brain networks of adult patients with PD experiencing EST through connectome-based analyses of diffusion tensor imaging (DTI) data. We investigated the interaction effects of EST on structural brain networks in PD patients and their relationships with clinical symptoms and trait factors.
Methods: A total of 82 participants with PD patients and 106 healthy controls (HCs) were enrolled in the study (94 men and 94 women; ages= 38.00 ± 12.69 years). Whole-brain structural networks were constructed using white matter tractography and network-based statistical methods were performed. In addition, we conducted exploratory correlation analyses between EST-associated degree in brain connectivity and clinical characteristics related to PD. We performed correlation analyses between the eigenvector centrality of nodes associated with EST and treatment outcomes in patients with PD. The Early Trauma Inventory Self Report-Short Form, Anxiety Sensitivity Inventory-Revised, State-Trait Anxiety Inventory, Neuroticism Scale, and Penn State Worry Questionnaire at baseline were administered. PD symptoms were measured at baseline and follow-up periods of 8 weeks, 6 months, and 1 year using the Panic Disorder Severity Scale.
Results: There were significant interaction effects between EST and the group on the middle temporal gyrus (MTG), parahippocampal gyrus (PhG) and inferior parietal lobule (IPL). Significantly hyperconnectivity in the MTG, PhG, and IPL was shown in patients with PD compared with HCs. Exploratory correlational analysis revealed a positive correlation between the global properties of the interaction networks and trait markers (e.g., neuroticism, anxiety sensitivity) and state markers (e.g., pathological worry) of anxiety levels in PD patients. Furthermore, the eigenvector centrality of the cingulate gyrus and inferior frontal gyrus was negatively correlated with short- and long-term pharmacological treatment responses in patients with PD (FDR passed, q=0.05).
Conclusions: Our findings revealed significant interaction effects between EST and the group on the MTG, PhG, IPL, suggesting that EST may differentially affect structural brain network connectivity in these regions, which are critically involved in memory performance, emotion processing, and regulation in PD patients. Consequently, patients with PD associated EST may have immature emotion processing due to hyperconnected brain regions, which could exacerbate anxiety symptoms and potentially influence treatment prognosis. Furthermore, the partial overlap observed between brain networks associated with EST and regions related to treatment response (e.g. cingulete gyrus, inferior frontal gyrus) suggests that these findings indicate the potential involvement of common neural circuitry in the pathophysiological mechanisms and therapeutic mechanisms of PD. These findings may provide a neuroscientific basis for understanding brain networks through EST.

Speaker: Dr Hyun-Ju Kim (CHA Bundang Medical Center)

Abstract_LSH_20240618.docx

38 3D reconstruction of BigBrain2: Progress report on semi-automated repairs of histological sections

The BigBrain2 is a second BigBrain data set supplementing and building on our expertise of the first BigBrain [1]. It will provide first insights into inter-subject variability at whole-brain, cytoarchitectonic level. Overall, BigBrain2 offers better quality staining, favourable to regional segmentation and registration, and contains fewer artefacts through sectioning and staining. In order to repair acquisition artefacts due to sectioning and histological preparation (tears, folds, missing tissue, excessive distortion etc.) [2], every fifth section was initially repaired, with comprehensive quality control (QC) [3], enabling an initial reconstruction at 100μm. The large variability with regard to the number, extent, and severity of the damages posed a particular challenge and therefore had to be specially addressed. In this work, we will report on new methods and approaches for repairing the remaining sections in a semi-automatic and cost-effective manner to complete the full reconstruction at 20μm. Based on the manual and semi-automatic repairs carried out so far, we present a new improved 3D reconstruction of BigBrain2.

The paraffin-embedded fixed brain of a 30-year-old male donor was sectioned coronally at 20μm thickness using a large-scale microtome. All 7676 sections were stained for cell bodies (Merker stain), then scanned at 10μm in-plane (flatbed scanner, 8bit grey level encoding) and subsequently at 1μm in-plane (Huron TissueScope scanner). The histological flatbed scanner sections were resampled at 20μm in-plane, to match the section thickness. Every fifth section was initially repaired, with comprehensive quality control (QC) and data provenance tracking of all repair operations [3] providing a means for assessing the extent of the repaired artefacts and for eventual reproducibility at the 1μm in-plane resolution.

A fully automated approach for repairing the remaining sections was deemed unviable given the nature and severity of the artefacts, therefore a semi-automated approach was developed with the objective to minimize human intervention. Sections with major artefacts were first identified and broken pieces were manually moved in place (about 10% of the sections). Subsequently, the section to repair was registered to the closest two previously repaired sections of the 5-series, 100μm apart, and a virtual reference image was interpolated from these two sections at the position of the section under review. Minor artefacts were corrected by interpolating good tissue from the reference section in place of missing tissue in manually identified regions. These masks ensure tracking of the extent of the repaired artefacts, with eventual reproducibility at 1μm.

Ongoing work includes the semi-automatic repairs of the remaining sections to obtain a complete volume at 20μm isotropic resolution onto which sections at the cellular resolution of 1μm can be progressively overlaid.

References:
[1] Amunts K. et al., BigBrain: An Ultrahigh-Resolution 3D Human Brain Model. Science, 2013.
[2] Mohlberg H. et al., 3D reconstruction of BigBrain2: Challenges, methods, and status of histological section repair – A progress report. BigBrain Workshop 2022
[3] Lepage C. et al., 3D reconstruction of BigBrain2: Progress report on updated processing pipeline and application to existing annotations and cortical surfaces. BigBrain Workshop 2023

Speaker: Hartmut Mohlberg (Institute for Neuroscience and Medicine (INM-1),Forschungszentrum Jülich, Germany)

39 Brain glial cell analysis using artificial intelligence: defining the role of Sharpin in AD.

Microglia and astrocytes are crucial glial cells in the central nervous system, with microglia primarily involved in regulating neuroinflammation and clearing cellular debris, while astrocytes support neuronal activity, modulate synaptic function, and help regulate the inflammatory response. In neurodegenerations, morphological and functional changes in microglia and astrocytes represent the pathological status and may contribute the disease progression. Alzheimer's disease is the most common neurodegenerative disorder and is characterized by the formation of amyloid plaques and neurofibrillary tangles resulting in the loss of memory and cognition. The chronic inflammation observed in Alzheimer's disease involves astrocytes and microglia, exacerbating disease progression and impairing glial cell function. Recent research suggests several methods to apply the artificial intelligence (AI) to the brain image processing and quantification. However, the AI coding and debugging knowledge required to adapt these tools for brain research could be a barrier for general neuroscience researchers to utilize them effectively. In this study, we tested Teachable Machine AI (Google), which is a new machine learning platform which doesn’t need significant programming, to analyze images of glial cells. Sharpin is a component of the LUBAC complex regulating NF-κB activity and resulting inflammation process, the mutations of which has association with Alzheimer's disease. However, the role of Sharpin in microglia and astrocyte activity regulating the inflammatory response is not well defined. We trained the machine learning tool using images of astrocytes and microglia from amyloid beta-treated and control groups. We then analyzed the images of astrocytes and microglia derived from Sharpin mutant mice to assess their similarity to the trained images. Our results suggest the potential usability of simple machine learning tools in brain imaging for researchers who are not proficient in AI coding.

Speaker: Ms Su Been Park (Chonnam national university)

40 Comprehensive Insights into Neural Activity and Metabolic Processes: Disentagling the Glycolytic Pathway to Examine the Resting-State [18F]FDG-PET/fMRI Coupling

Functional connectivity (FC) is used for investigating brain network organization. Given that the brain consumes 20% of the body’s glucose to support functions, understanding the relationship between glucose consumption and FC is essential for comprehending brain physiology (Raichle, 2006). Some studies have linked FC with the semiquantitative metabolic indices SUVR (Standardized Uptake Value Ratio) or the quantitative overall metabolic fractional uptake, Ki [ml/cm3/min], revealing a partial relationship. However, no studies have assessed how different components of the glycolytic process relate to network structure as expressed through FC.

fMRI and PET were collected for 42 healthy subjects (HCs) (Volpi et al., 2023). The fMRI data were band-pass filtered in the range 0.008-0.1 Hz (F1) to maximize the presence of spontaneous low-frequency oscillations, and in the range 0.008-0.21 Hz (F2) to include hemodynamic contributions. fMRI and PET signals were parcellated using a clustered version of the Yan functional atlas (Yan et al., 2023), plus 12 subcortical (AAL3). Individual FC matrices were obtained by calculating Pearson correlation. Static PET images were normalized to SUVR, while two-compartment kinetic modelling was applied to dynamic data to assess K1, k2, k3 rate constants and to calculate Ki using Variational Bayesian inference (Castellaro et al., 2017).
Partial Least Square Correlation was applied between K1 and FCSTR (nodal strength) and repeated between k3, Ki, SUVR and FCSTR (Figure 1a). The generalizability of significant multivariate correlations was tested via 7-fold cross-validation. For each generalizable pair, a regression line was established, defining a 90th percentile normality band of HC distances from the line.

The strongest relationship between the influx parameter K1 and FCSTR was in F2 (r=0.66). As the scatterplot between K1 and FCSTR scores shows (Figure 1b), brain regions with higher glucose transport often exhibit stronger functional connectivity.
While a positive correlation between SUVR and FCSTR was observed specifically in F1 (r=0.75), the scatterplot (Figure 1c) reveals a broader spread around the regression line, indicating higher variability across subjects.
The k3-FCSTR and Ki-FCSTR pairs showed a non-generalizability results and for this reason are not reported.

The stronger K1-FCSTR coupling in F2 was expected, as this band includes more hemodynamic contributions, which appear essential for linking the two metabolic and functional information sources (Amend et al., 2019). The modest yet reliable correlation between SUVR and FCSTR may be attributable to SUVR being a semi-quantitative measure compared to the quantitative FDG-PET analysis: SUVR cannot disentangle the different physiological processes describing brain glucose consumption kinetics (better described by K1), hindering a full understanding of its relationship with FC (Palombit et al., 2022; Volpi et al., 2024). The k3-FCSTR and Ki-FCSTR couplings do not surpass the level of statistical significance, highlighting that the major contribution to brain metabolic processes and the network, as described by FC and summarized in the strength graph measure, is sensitive only to blood flow and BBB permeability.

This study shows that the hemodynamic part of the BOLD signal is also crucial in studying its relationship with brain metabolism, as it is linked to glucose and oxygen supply.

Speaker: Claudia Tarricone (Department of Information Engineering, University of Padova, Padova, Italy; Padova Neuroscience Center, University of Padova, Padova, Italy)

Figure 1.jpg

41 Cortical Gyrification Patterns Associated with Neuroticism in Panic Disorder and Healthy Individuals

Background: Several studies have demonstrated an association between neuroticism and alterations in cortical folding. However, few studies have investigated the relation between neuroticism and gyrification in patients with panic disorder (PD) and healthy individuals. This study examined the relation between neuroticism and cortical gyrification patterns and their relation with symptomatology in patients with PD and healthy control (HC).
Methods: This study included 230 participants: 102 patients diagnosed with PD and 128 HC. Neuroticism, anxiety symptomatology, ways of coping strategies, and health-related quality of life was evaluated. Voxel-wise correlation analyses using FreeSurfer were conducted to determine the neural correlates of neuroticism related to cortical gyrification in patients with PD and HC.
Results: As neuroticism increased, cortical gyrification of the lingual gyrus decreased significantly in HC, whereas postcentral gyrus gyrification increased while lingual gyrus gyrification decreased in patients with PD. Although lingual gyrus gyrification in HC was significantly correlated with social phobia, interoception, and agoraphobia subscales, lingual gyrus gyrification in patients with PD revealed significant correlations with emotional coping, physical functioning, and the emotional role subscales. In addition, the local gyrification index in the postcentral gyrus was significantly correlated with excessive worry severity scale scores.
Conclusions: Our study suggests that increased neuroticism is associated with decreased cortical folding patterns in the lingual gyrus in both HC and patients with PD and increased postcentral gyrus gyrification only in patients with PD. These gyrification alterations may influence perceived quality of life as well as high levels of anxiety symptomatology in patients with PD.

Speaker: Prof. Hyun-Ju Kim (CHA Bundang Medical Center)

Abstract_20240701.docx

42 Cortical hierarchy relates to microstructural organization

The neocortex is composed of an extensive recurrent network in which distinct feedforward and feedback inputs from other brain regions are integrated to generate coherent percepts and execute voluntary actions 1. In recent decades, research has begun to shed light on how the cortex performs these functions and constructs the cortical hierarchy based on a principle of hierarchical distance 2. However, the spatial patterning of cortical organization and circuit mechanisms of the human cortex remain unclear. The recently proposed regression dynamic causal model (rDCM) offers a powerful and reliable tool for assessing the effective connectivity in large networks based on functional magnetic resonance imaging (fMRI) data 3. This study characterizes the hierarchy of input-output asymmetry between cortical regions and assesses its associations with cytoarchitecture and cortical types.

In this work, we analyzed 7 Tesla (7T) movie fMRI data from 10 unrelated healthy adults at the group level. Specifically, our work (i) utilized rDCM and high-resolution 7T MRI to construct effective functional connectomes (FC), (ii) calculated the input-output asymmetry index for each region (input minus output) and cross-correlated these indices for all regions using the Spearman correlation coefficient to generate the asymmetry index similarity (AIS) matrix, (iii) captured the principal component of AIS using diffusion map embedding, and (iv) assessed associations between AIS and cytoarchitecture as cortical types.

The group-level parcel-wise effective FC matrix was generated using the Glasser atlas 4, revealing asymmetry between the input and output streams of each region (Figure 1A). We generated the AIS by cross-correlating parcel-wise asymmetry indices, and then estimated its first component (AISG1). AISG1 clusters regions with closer input-output asymmetry patterns together. We observed an axis stretching between the dorsal attention network and the cingulo-opercular network and mesial frontal lobe (Figure 1B). Examining the distribution of AISG1 across cortical hierarchy levels 5, we found the highest values in the paralimbic system and the lowest in the unimodal system (Figure 1C), indicating the difference of the input-output asymmetry patterns between these two systems. In terms of cortical types, AISG1 was higher in the agranular area and lower in the parietal area (Figure 1D). To further investigate the relationship between AISG1 and cytoarchitecture, we sampled layer-specific intensity profiles between the pial and white matter surfaces using histological data 6. We assessed the histological skewness, which indexes the balance of cellular density in infra- vs supra-granular layers 7. Sensory areas exhibited lower skewness, while the paralimbic network showed higher skewness. Notably, we found a significant correlation between AISG1 and histological skewness (rho=0.54, pspin<0.001; Figure 1E).

Our findings point to a cortex-wide sensory-paralimbic differentiation of input-output asymmetry, which is associated with cytoarchitectural patterning. Our work provides new insights into macroscale interactions between cortical areas and how they are underpinned by cortical cytoarchitecture.

Speaker: Yezhou Wang (McGill university)

figure1_V1_with legend.png

43 Disconnectomic simulation reveals repetition pathways in a case of mixed transcortical aphasia

Introduction
Mixed transcortical aphasia (MTA) represents an uncommon aphasic syndrome, characterized by severe deficits in both comprehension and production across oral and written linguistic modalities, in contrast with retention of repetition abilities, often manifesting as echolalia. In the present single case study, we report the case of a woman who presented with MTA symptoms after a left hemisphere ischemic stroke involving perisylvian areas, with substantial preservation of reading abilities. The neural underpinning of her residual repetition and reading capacities is explored here employing recently developed lesion-based approaches for white matter disconnections’ probabilistic simulation.

Methods
Language abilities were evaluated with the ENPA battery (Capasso & Miceli, 2001) 6-days post-stroke.
We employed two state-of-the-art disconnectomic simulation toolkits called BCB Toolkit (Foulon et al., 2018) and Lesion Quantification Toolkit (Griffis et al., 2021) to reveal which white matter tracts are likely related to the reported pattern of language impairment. The first toolkit calculates the probability with which every tract is affected by the lesion, together with the proportion of lesioned volume of each tract. The second one, instead, calculates the percentage of streamlines that are crossed by the lesion, for each white matter bundle.

Results
Our patient presented with poor comprehension and reduced spontaneous production abilities, while repetition was preserved for words, short sentences, and complex number words up to three digits (e.g., three-hundred-forty-six). Notably, her reading abilities remained intact, in contrast with several previously reported MTA cases. An impairment of verbal and nonverbal short-term memory (verbal span – digit span = 2; spatial span – Corsi block-tapping test = 3) was found.
The results from the two disconnectomic simulation toolkits showed a major disconnection of the left arcuate fasciculus, left inferior fronto-occipital fasciculus, left inferior longitudinal fasciculus, left optic radiation, and anterior commissure. Specifically, regarding the arcuate fasciculus, she presented with disconnected long and posterior branches, while the anterior one was preserved.

Conclusions
Following recent results by Forkel and colleagues (2020), the observed impairments of the patient in the repetition of long sentences and of complex numbers exceeding three digits’ might be explained based on the disconnection of the arcuate fasciculus' long and posterior branches, related to impaired short-term memory abilities. Conversely, her preserved ability to repeat and read limited verbal material seems to be supported by the intact anterior branch of the left arcuate fasciculus.
In conclusion, this case report represents the second documented endeavor within the extant literature to provide a comprehensive delineation of white matter tracts’ involvement, utilizing state-of-the-art toolkits, in elucidating plausible mechanisms underlying brain disruption in an infrequent aphasia syndrome.

Speaker: Irene Bellin (University of Padua)

Bellin_abstract_methods.png

44 Glioma-Induced Alterations in Structural-Functional Connectivity Integration

Gliomas alter white matter (WM) integrity[1] and grey matter (GM) functions[2] through intra-tumoral changes, invasion and expansion[3], affecting structural (SC) and functional (FC) connectivity. Nevertheless, many researchers investigated functional/structural/microstructural abnormalities within specific WM[4,5]-GM[6] regions, lacking an integrated overview[7,8].

This study aims to examine SC-FC link at the individual level, evaluating if alterations in this relationship may influence tumours effects on key brain networks, especially in association with tumours topological characteristics.

41 glioma patients were enrolled in the study. Images were acquired with a 3T Siemens Biograph mMR PET/MR scanner: dMRI images (b-values:0/710/2855s/mm2, 100-directions, MRtrix3[9]-pre-processing/tractography[10,11]-10Mstreamlines-iFOD2[12]), rs-fMRI (15min-TR/TE-1260/30ms-3x3x3mm3, state-of-the-art pre-processing[13,14]). An expert neuroradiologist provided lesion masks.

Connectivity matrices were derived with the Yan-homotopic-functional atlas (100-parcels/17-networks)[15]. Number-of-streamlines matrix (SCnos) was computed. Eight dMRI microstructure maps were quantified with NODDI[16]-DTI[17]-DKI[18] models, to be fused[19] into a single-microstructure matrix (SNFmicro). FC matrix represented Pearson correlations[15].

Integration Matrix (IM-100x300), to define an individual SC/FC measure, was derived concatenating the three single modalities (FC, SCnos and SNFmicro).

To evaluate the interplay between an integrated approach and single modalities, connectivity abnormalities were derived by single and integrated connectivity matrices.

For each connectivity type, pseudo-reference matrix was computed as the median across all the patients. Tracts and parcels overlapping with the lesion were removed by the computation. Then, Spearman correlation was computed to measure the parcel’s connectivity similarity between each patient and its pseudo-reference group (row-by-row). Further, given the similarity distribution of all the subjects, a parcel was labelled as altered if the distance from the pseudo-healthy profile fell in the lower tail of the distribution (below 5%). Finally, the percentage of altered parcels within the same network represented the connectivity impairment of each network[20].

To investigate abnormalities-topology link, patients Network-Lesion distances were classified as Near/Far, according to the mean Euclidean distance between parcel-lesion centroids (threshold=76.17mm).

Panels A-B of Figure 1 illustrate, for networks Near/Far to the lesion, IM-alterations percentage across the subjects. Right-SalVentAttnB, Right-DefaultA and Right-LimbicB were the near networks more often altered among the patients (range 24%-29%). Left-DefaultA, Right-DefaultA and Right-SalVentAttnA were the networks far from the lesions more frequently altered across the patients (range 17%-27%). Just mentioned networks overlayed with higher frequency lesions position.

Panels A-B of Figure 1 also distinguish IM-alone (pink bar) and IM-in-overlap (grey bar) abnormalities. It is important to note that many networks highlighted both IM-alone and IM-in-overlap alterations. Networks near to the lesion were more frequently characterized by IM-alone rather than IM-in-overlap with single-modalities alterations (53%vs.32%).

These results suggest that, when performing studies on glioma, IM provides, especially for networks near to the lesion, a complementary view of connectivity changes. Hence, integration of brain functions with microstructure and structural integrity, could provide major insights about key networks alterations and highlights specifics topological-relevant tumours characteristics.

References:
1. https://doi.org/10.1093/brain/awac360
2. https://doi.org/10.1093/neuonc/noaa189
3. https://doi.org/10.1093/braincomms/fcaa216
4. https://doi.org/10.3390/cancers15143631
5. https://doi.org/10.3389/fonc.2022.998069
6. https://doi.org/10.3389/fonc.2020.00794
7. https://doi.org/10.1093/brain/aww194
8. https://doi.org/10.1155/2017/3530723
9. https://doi.org/10.1016/j.neuroimage.2019.116137
10. https://doi.org/10.1016/j.neuroimage.2012.06.005
11. https://doi.org/10.1016/j.neuroimage.2012.11.049
12. https://archive.ismrm.org/2010/1670.html
13. https://doi.org/10.1016/j.neuroimage.2004.07.051
14. https://doi.org/10.1016/J.NEUROIMAGE.2010.09.025
15. https://doi.org/10.1016/j.neuroimage.2023.120010
16. https://doi.org/10.1016/J.NEUROIMAGE.2012.03.072
17. https://doi.org/10.1016/S0006-3495(94)80775-1
18. https://doi.org/10.2214/AJR.13.11365
19. https://doi.org/10.1371/journal.pbio.3002314
20. https://doi.org/10.1152/jn.00338.2011

Speaker: Maria Colpo (Padova Neuroscience Center, University of Padova, Padova, Italy; Department of Information Engineering, University of Padova, Padova, Italy)

Figure1_MariaColpo.tiff

45 Modulation of Brain Resilience with Transcranial Magnetic Stimulation: a TMS-EEG Study

Background. Resilience reflects the ability of a complex system to sustain damage while still maintaining a proficient level of functioning. At the neural level, the brain's resilience to focal lesions or diffuse pathological processes develops as a function of favorable genetic predispositions and exposure to enriched environments.

Objective/Hypothesis. We tested if targeted, personalized exposure to other forms of stimulation, i.e. noninvasive brain stimulation, could momentarily alter one’s level of brain resilience, as tested through simulation of the induced effect of a random or targeted lesion on the brain connectome.

Methods. 23 subjects underwent a single-pulse transcranial magnetic stimulation (TMS) protocol during concurrent EEG recording targeting two regions belonging to two negatively correlated brain functional networks: the Dorsal Attention (DAN) and Default Mode (DMN) networks. We tested for induced changes in network topology using graph theory analysis, followed by the exploration of resilience through an in silico lesioning procedure, investigating the ability of the network to resist the progressive removal of its nodes and edges.

Results. For both networks tested, the delivery of a TMS pulse changed some resilience indexes accompanied by an immediate significant reduction in modularity. TMS targeting the DMN was also accompanied by a transient increase in clustering coefficient and local efficiency. Interestingly, all such TMS-induced changes in topology were significantly correlated with the increased resilience of the brain network in silico to the targeted removal of its edges.

Conclusion(s). Although exploratory, to our knowledge no previous studies have investigated the short-term impact of TMS pulses on one’s level of brain connectivity and resilience, as tested through simulation of the induced effect of a random or targeted lesion. In particular, the observed decrease in modularity following stimulation of both the DAN and the DMN suggests an overall strengthening of connectivity across the network’s nodes, resulting in a momentary increase between-modules connections with respect to those within. This appears in line with recent evidence highlighting the tendency of the induced TMS effect to spread across several different networks rather than remaining limited to the stimulated area. In line with such an increase in the processing capacities between the network’s nodes, we also observed a transient increase in the resilience of the brain network to the targeted removal of its edges. Indeed, more resilient systems benefit from distributed processing, as it allows the information transfer to be more equally distributed instead of relying on a few strong connections, which would cause a significant disruption in the system’s capacity if lesioned.

Speaker: Arianna Menardi (Precision Neuroscience and Neuromodulation Program, Gordon Center for Medical Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA & Department of Neuroscience & Padova Neuroscience Center, University of Padova, Padova, Italy)

BBWS24_Romanella_figure.png

46 The Semi-automated Quantification Approach of The Activated Microglia in a Mouse Model of Alzheimer's Disease

Alzheimer's disease (AD), the most common aging-related neurodegenerative memory disease, is characterized by amyloid beta plaques and tau tangles. Recently Sharpin mutations have been identified as Alzheimer's disease-related factors in Asian populations as a result of Genome-wide association studies (GWAS). SHARPIN is a component of the LUBAC complex which conjugates linear polyubiquitin chains and plays a key role in NF-kappa-B activation and regulation of inflammation. It is proposed that targeting those SHARPIN mutants may offer a new approach for treating Alzheimer's disease by modulating neuroinflammation and addressing synaptic dysfunction. Microglia are the main immune cells in the nervous system and are known to play a crucial role in neuronal damage and neurodegenerative diseases. Specifically, the activation of microglia has been observed in the cerebral cortex and hippocampus of Alzheimer's disease patients. This activation of microglia can be distinguished by their morphological changes or cell surface antigen expression, but analyzing these within tissue requires significant time and effort. Recently, many studies suggested automated image processing techniques for the massive quantification of microglia. In this study, we aimed to modify one of this machine learning based-method to improve the accuracy and quality of the automated analysis of microglia in the batch image processing and analysis stages. Then, we performed automated analysis of microglia in an AD mouse model to distinguish the AD-related modification of microglia compared to wildtype. Our revised machine learning system allowed us to analyze microglia in the AD mouse model in a more objective and efficient manner, and we applied it to the SHARPIN mutant mouse model to define the role of Sharpin in microglia regulation in AD.
Keyword : Alzheimer’s disease, SHARPIN, microglia
“This research was supported by a grant of the Korea Dementia Research Project through the Korea Dementia Research Center(KDRC), funded by the Ministry of Health & Welfare and Ministry of Science and ICT, Republic of Korea (grant number: HU23C0199)”

Speaker: Jin Ho Lee (Chonnam National University)

19:00 BigBrain Workshop Dinner

Wednesday 11 September

08:30 Coffee and Registration

HIBALL Lecture: Maurizio Corbetta

47 Architecture and function of spontaneous brain activity in health and pathology

I will argue that spontaneous brain activity plays a major role in brain function and behavior, and it does so in surprising and unexpected ways. While for more than hundred years we have focused on understanding cognition by studying the brain in action, I will argue that there are many lessons to learn by studying the brain at rest. Furthermore, the study of the brain at rest is fundamental to understand pathology since most neurological and psychiatric disorders are disorders of connectivity that we can study beautifully and conveniently by studying patters of activity at rest.

Speaker: Maurizio Corbetta (Department of Neuroscience, University of Padova)

10:00 Coffee and discussion break

48 Dimensional perspectives on multiscale cortical organization

The talk will overview conceptual and methodological advances to study spatial patterns in macroscale cortical organization. It will also present how open neuroinformatics approaches can be used to seamlessly integrate post-mortem information with high- and ultrahigh field neuroimaging datasets for targeted structure-function studies in the human brain.

Speaker: Boris Bernhardt (McGill University)

Contributed Talks - Mapping and Atlases (co-Chairs: Elizabeth Rounis, Thomas Funck)

co-Chairs: Elizabeth Rounis, Thomas Funck

49 HippoMaps: multiscale cartography of human hippocampal organization

The hippocampus has a unique microarchitecture, is situated at the nexus of multiple macroscale functional networks, contributes to numerous cognitive and affective processes, and is highly susceptible to brain pathology across common disorders. The hippocampus can be understood and modeled as a cortical (archicortical) structure with a 2D surface topology [1]. Taking inspiration from neocortical informatics tools like NeuroMaps [2], here, we introduce HippoMaps, an open access toolbox and data warehouse for the mapping and contextualization of hippocampal data on hippocampal surfaces in the human brain.

HippoMaps capitalizes on a novel hippocampal unfolding approach as well as shape intrinsic cross-subject and cross-modal registration capabilities [3]. We initialize this repository with data spanning 3D histology [4,5], structural MRI and resting-state functional MRI (rsfMRI) obtained at 3 and 7 Tesla [6,7], as well as intracranial encephalography (iEEG) recordings in epilepsy patients [8].

We present 30 novel, detailed maps of hippocampal structural and functional features. Structural measures derived from quantitative MRI and histology tend to show sharp subfield differentiation, whereas functional measures such as rsfMRI and iEEG band powers show gradual anterior-posterior differentiation. We show how such maps can be related to one another using a tailored approach for spatial map association that corrects for autocorrelation. This provides a method for contextualizing hippocampal data in future work. Code and tools are compliant with community standards, and are provided as comprehensive online tutorials that reproduce the figures shown here.

Bioinformatics data are not inherently useful unless context is given, for example, by their inter-relationships and their links to disease or cognitive processes. Here we provide a common space and toolbox for such comparisons in the hippocampus, spanning methodologies and modalities, spatial scales, as well as clinical and basic research contexts. Some maps have already been generated and uploaded to HippoMaps by members of the broader research community, and we further discourse in the spirit of open and iterative scientific resource development.

1. DeKraker J, et al. Automated hippocampal unfolding for morphometry and subfield segmentation with HippUnfold. Elife. 2022;11. doi:10.7554/eLife.77945
2. Markello RD, et al. neuromaps: structural and functional interpretation of brain maps. Nat Methods. 2022;19: 1472–1479.
3. DeKraker J, et al. Evaluation of surface-based hippocampal registration using ground-truth subfield definitions. Elife. 2023;12. doi:10.7554/eLife.88404
4. Amunts K, et al. BigBrain: an ultrahigh-resolution 3D human brain model. Science. 2013;340: 1472–1475.
5. Alkemade A, et al. A unified 3D map of microscopic architecture and MRI of the human brain. Sci Adv. 2022;8: eabj7892.
6. Royer J, et al. An Open MRI Dataset For Multiscale Neuroscience. Sci Data. 2022;9: 569.
7. Cabalo DG, et al. Multimodal precision neuroimaging of the individual human brain at ultra-high fields. bioRxiv. 2024. p. 2024.06.17.596303. doi:10.1101/2024.06.17.596303
8. Frauscher B, et al. Atlas of the normal intracranial electroencephalogram: neurophysiological awake activity in different cortical areas. Brain. 2018;141: 1130–1144.

Speaker: Jordan DeKraker (McGill University)

Fig1.png

50 Asymmetry of hippocampal subfield volumes in a heterogenous cohort of focal drug-resistant epilepsies

Introduction: Epilepsy affects 50-65 million individuals worldwide (World Health Organization). Drug-resistant epilepsy (DRE) is prevalent in 30-36% of clinic-based populations [1]. Surgical resection of epileptogenic zone is a common practice for treatment of focal DRE [2,3] with hippocampal sclerosis (HS), tumor-related malformations, and focal cortical dysplasia (FCD) the most common histopathological diagnosis in adults [1]. Studies on drug-resistant temporal lobe epilepsies (TLE) [4,5] and post-traumatic epilepsy [6] reveal associations of structural connectome reorganization in ipsilateral networks, hippocampal deformations, and variations in hippocampal signal intensity with memory disfunctions and risks of future epilepsies. We present an analysis of magnetic resonance imaging (MRI)-based hippocampal volumetry for patients diagnosed with HS and other epilepsy subtypes and age-matched healthy volunteers, supported by evaluations of hippocampal segmentation and post-operative histological diagnosis.
Methods: Seventeen patients (5 males, 12 females, 36.24 ± 14.21 years at the time of surgery) with focal DRE underwent resective surgery at Kuopio University Hospital. Pre-operative clinical MRI revealed structural etiological findings with presumption of temporal or frontal-lobe FCD (N=5), HS (N=3), dual FCD-HS (N=2), tumors (N=3), encephalocele (N=3), and structural without clear etiology (N=1). Additionally, 15 healthy volunteers (10 males, 5 females, aged 35.33 ± 7.81 years), were recruited. Patients and volunteers underwent MRI acquisition with a 3D MPRAGE T1-weighted sequence (3T Siemens, TR = 2.3 s, TE = 1.92 ms, 1.0x1.0x1.0 mm3 voxel size) before the surgery as part of a larger MRI study. Immunohistochemistry with NeuN and GFAP was conducted on resected tissues by a neuropathologist. T1w volumes were analyzed using HippUnfold v1.3.0 to segment the hippocampal subfields (Subicular complex (Sub.), Cornu Ammonis (CA) fields 1-4, dentate gyrus (DG), and stratum radiatum and lacunosum-moleculare (SRLM)) and compute their volumes [7,8]. This software uses a UNET deep convolutional neural network (CNN) and a template of ex vivo images to segment the tissue in and around the hippocampus and fits the hippocampal grey matter with a subject-specific and topologically constrained surface mesh [9].
Results: Brain tissues were resected during eight anterior temporal lobectomies, six extratemporal focal resections or limited lesionectomies, and three encephalocele disconnection procedures. Results of histological diagnosis are presented in Fig. 1A. Fig. 1B demonstrates the neuroradiologist’s evaluation of T1w-based hippocampal segmentation. Volumetric asymmetries for seven hippocampal subfields in volunteers and patients are presented in Fig. 1C with a summary of neuropathologist’s assessments of neuronal loss and gliosis in hippocampal subfields. While healthy volunteers exhibit minimal to mild volumetric asymmetries in CA2-4, patients with HS or FCD-HS demonstrate the largest volumetric asymmetries across all subfields, aligning with the laterality of their malformations. Two-sample, one-sided t-tests reveal patients with HS had greater absolute volumetric asymmetry in all subfields compared to other patients (pFDR < 0.05, Fig. 1D) as well as in all subfields except for CA2 compared to healthy volunteers (pFDR < 0.01).
Conclusion: We present an analysis of hippocampal volumetry between DRE patients and healthy volunteers. Results will be updated as data acquisition continues with the aim to present a multi-modal and multi-scale assessment of epileptogenesis in the human brain.

Speakers: Dr Mastaneh Torkamani-Azar (A.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland), Prof. Jussi Tohka (A.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland.)

Hippo_Fig1_20240701.jpg

51 High-resolution 3D Mapping of the Human Hypothalamus:Towards a Comprehensive Cytoarchitectonic Atlas

Introduction

The hypothalamus is crucial for maintaining homeostasis, regulating sleep-wake cycles, appetite, circadian rhythm, and thermal regulation (Nieuwenhuys et al., 2008). Despite its importance the structural organization, precise boundaries, and functional differentiation of its nuclei remain incompletely understood. Existing anatomical maps of the hypothalamus do not reflect interindividual variability in 3D space; they often lack the spatial resolution and morphological detail to provide a comprehensive understanding of this complex region and to inform neuroimaging studies about the brain’s microstructure. Therefore, we aimed to develop probabilistic cytoarchitectonic maps to address intersubject variability and provide a high-resolution 3D reference map for informing studies in the living human brain.

Methods

We delineated the hypothalamus and its nuclei on every 15th cell-body stained brain sections in 10 brains (5 female) including BigBrain (Amunts et al., 2013). For creation the high-resolution BigBrain model we used a deep-learning based tool (Schiffer et al., 2021) that delineated the remaining sections. Other brains were used to create probability maps that capture intersubject variability in space and location of areas. To do this, brains were 3D reconstructed and superimposed in standard reference space (Amunts et. al., 2020). Quantitative tools, including texture analysis (Devakuruparan, 2024) and object instance segmentation (Upschulte et al., 2021), were applied to analyse subdivisions in more detail.

Results

We generated high-resolution 3D map of 23 nuclei of the human hypothalamus (Figure 1), that show their shapes and neighbourhood relationships with high precision. These nuclei were categorized into four rostro-caudal zones:

Preoptic zone includes the anterior periventricular and median preoptic nuclei lining the third ventricle, with the uncinate and intermediate nuclei forming a cluster around the medial preoptic nucleus.

Anterior zone contains the paraventricular nucleus with dark magnocellular neurons ventrolaterally and less intense parvocellular neurons medially, the supraoptic nucleus with densely packed magnocellular neurons, and the retrochiasmatic, suprachiasmatic and periventricular nuclei.

Tuberal zone features the ventromedial nucleus with high peripheral cell density, the smaller posteromedial nucleus between the ventromedial nucleus and mammillary body, the dorsomedial nucleus with densely packed small neurons at its centre, and the arcuate nucleus within the tuber cinerium.

Mammillary zone includes the medial and lateral mammillary nuclei. The tuberomammillary and supramammillary nuclei contain large dark magnocellular neurons, and the lateral tuberal nucleus housing medium-sized neurons in the basolateral mammillary zone.

The mean hypothalamic volume was 1492 ± 264 mm³. The Lateral (514 ± 49 mm³) and Posterior hypothalamic areas (262 ± 33 mm³) showed the highest volumes, whereas the uncinate and lateral mammillary nuclei exhibited the lowest values (0.845 ± 0.15 mm³; 1.8 ± 0.3 mm³). Permutation tests found no significant effects of hemisphere, sex, or their interaction on the shrinkage-corrected volumes for each nucleus. Intersubject variability was reflected in the probabilistic maps that will be part of the Julich-Brain Atlas (Amunts, 2020) and available via EBRAINS and other platforms.

Conclusions

In sum, we provide a detailed microstructural map of the hypothalamus, serving as a profound anatomical basis for interpreting and comparing neuroimaging data helping to refine the functional organization of the hypothalamus.

Speaker: Mr Alexey Chervonnyy (Cécile & Oscar Vogt Institute for Brain Research)

3D reconstruction of the hypothalamus in the BigBrain.png

52 Mapping superficial white matter architecture on BigBrain

INTRODUCTION.
The superficial white matter (SWM) is the layer of white matter (WM) located immediately underneath the cortex. This SWM contains subcortical U-fibers interconnecting adjacent brain gyri, which remain incompletely myelinated until later in life [1]. U-fibers play a key role in brain plasticity and aging, and alterations in these fibers have been observed in conditions such as autism, epilepsy, and Alzheimer's disease [2,3,4]. Despite its importance, the SWM has been understudied, primarily due to technical difficulties and limitations [5]. Recent advances in ultra-high field 7 Tesla magnetic resonance imaging (MRI) technology have enabled precise imaging and mapping of brain microstructure, leading to reliable research on the SWM. However, the structural and functional role of the SWM is still unclear and histological data could improve the understanding of these relationship. Specifically, BigBrain histological data could unravel complex microstructural properties by characterizing changes in intensity related to the SWM oligodendroglia organization5. In this regard, this study assesses histological features of the SWM, and evaluate its association to macroscale motifs of brain organization.

METHODS.
We preprocessed BigBrain histology data by correcting the outliers in temporal poles and intensity changes along the y-axes. To examine the SWM, we solved the Laplace equation over the WM domain. This was achieved by initially computing a Laplace field across the WM and subsequently shifting an existing WM surface along that gradient. Stopping conditions were set by the geodesic distance traveled. SWM surfaces were sampled at fifty depths, each separated by 0.06 mm, beneath the gray and white matter interface. The microstructure intensity profiles, depicting the intensity values of BigBrain features, are presented in Fig. A. Additionally, we sample the gray matter intensities to correlate the gray matter profile with its SWM at each vertex. Finally, we applied diffusion map embedding (DM), a non-linear dimensionality reduction on the data. Vertex-wise intensity profiles were cross-correlated, resulting in microstructural profile covariance (MPC) matrices that represent vertex-specific similarity in histological feature across the SWM. The MPC matrix was converted to a normalized angle affinity matrix, and diffusion map embedding was applied. This procedure identified eigenvalues (gradients) describing the main spatial axes of variance in inter-regional similarity of microstructural profiles. Gradient analyses were conducted using BrainSpace [6].

RESULTS
The spearman correlation between the gray matter and SWM intensity changes of BigBrain shows a positive pattern for most areas of the brain except the medial insula. (Figure A, bottom). After we applied DM, we observe that the two first eigenvalues or “gradients” explained ~75% of the variance. The principal gradient differentiated highly myelinated areas, like motor and pre-motor cortes from association areas (e.g. parieto-temporal and middle and superior frontal lobe).

CONCLUSIONS
We generated precise and reproducible SWM surfaces and proposed a novel method to study the cytoarchitectural similarity of the SWM on BigBrain. We observed a continuum between the gray matter lamination and the SWM structure that distinguishes it from the deep white matter. With this study, we open a new window to extend further cytoarchitectural studies to include the SWM.

Speaker: Youngeun Hwang (McGill University)

SWM_abstract_hiball.png

53 Surface-based parcellation and vertex-wise analysis of ultra high-resolution ex vivo 7 tesla MRI in Alzheimer’s disease and related dementias

Introduction: MRI is the standard modality to understand human brain structure and function in vivo (antemortem). Decades of research in human neuroimaging has led to the widespread development of methods and tools to provide automated volume-based segmentations and surface-based parcellations which help localize brain functions to specialized anatomical regions. Recently ex vivo (postmortem) imaging of the brain has opened-up avenues to study brain structure at sub-millimeter ultra-high-resolution revealing details not possible to observe with in vivo MRI. Unfortunately, there has been limited methodological development in ex vivo MRI primarily due to lack of datasets and limited centers with such imaging resources. Therefore, in this work, we present one-of-its-kind dataset of 82 ex vivo T2w whole-brain hemispheres MRI at 0.3 mm^3 resolution spanning Alzheimer’s disease and related dementias. We developed a fast and easy-to-use automated surface-based pipeline to parcellate ultra-high-resolution ex vivo brain tissue at the native subject space resolution using the Desikan-Killiany-Tourville (DKT) brain atlas. This allows us to perform vertex-wise analysis in the template space and thereby link morphometry measures with pathology measurements derived from histology.

Methods: We perform structure-pathology association analysis in a large dataset of ultra-high-resolution 0.3 mm^3 T2w 7T MRI scans of whole brain hemispheres from 82 brain donors with ADRD diagnoses, a first study of such scale conducted at this resolution. We present a new computational pipeline that performs automated segmentation and whole-hemisphere FreeSurfer DKT atlas parcellation of the cortex in native subject space at sub-millimeter 0.3 mm^3 resolution, a first large-scale surface-based scheme for ex vivo whole-hemispheres analysis in diseased population. We achieve this by adapting the surface-based pipeline in FreeSurfer with an initial subject-space topology-corrected WM segmentation derived from a deep learning-based segmentation model as developed (Fig. 1). We evaluate the framework by correlating cortical thickness with neuropathological markers implicated in AD (measures of p-tau, neuronal loss; global amyloid-β, Braak staging, and CERAD) and perform vertex-wise generalized linear modeling.

Results: Fig. 2 shows the Spearman’s rho-value between the mean cortical thickness (mm) in each brain region (computed in subject space at native resolution) and five pathology measures: global ratings of amyloid-β, Braak staging, CERAD, and regional ratings of p-tau pathology and neuronal loss in the MTL, the region first implicated in AD. The analysis was covaried for age, sex and postmortem interval and corrected for multiple comparisons using Bonferroni method. Significant negative correlation was found in entorhinal, parahippocampal, medialorbitofrontal, temporal pole, inferior temporal and parietal lobes which are consistent with literature on progressive loss of cortical gray matter in AD.

Vertex-wise correlation between thickness and the neuropathology ratings was performed in MNI template-space by fitting a GLM at each vertex across the entire cohort with age, sex and PMI as covariates and corrected for multiple comparisons. Fig. 3 shows the t-statistics map on the pial and the inflated surfaces with the clusters outlined in white indicating regions where the significant strongest associations were observed surviving FDR correction. We observe that the strongest correlations were observed in MTL, the region associated with AD.

Speaker: Pulkit Khandelwal (University of Pennsylvania)

Khandelwal_figures.pdf

54 Exploring the links between the localization of cortical areas and the variability of folding patterns

Introduction
The variability of the cortical folding pattern is very complex. Hence, the links between folding and functional architecture have rarely been studied. To explore these links, we perform a stratification of the population according to the regional folding pattern before performing group analysis of functional maps [Sun et al., 2016]. To illustrate the potential of this strategy, we apply this approach to the study of the links between the shape of the precentral sulcus and fMRI-based proxies of areas 44 and 55b.

Material and methods
The 882 first HCP subjects (2015 release) were processed using the “Morphologist 2012” pipeline [http://brainvisa.info]. All brains were affinely normalized to the Talairach space. A group of seven sulci making up the Precentral Gyrus was formed for shape analysis. Pairwise shape similarity matrices were computed using rigid alignment. Isomap method was used to project the individual gyri into a low dimensional manifold. All gyri were also rigidly aligned to the most typical one, namely the gyrus with the minimum average distance to the set. Moving local average shapes were computed in this target space along each manifold dimension for visualization. The evolution of this moving average provides a detailed description of the shape-feature coded by the coordinates in the manifold. For each subject, a fMRI contrast used to define proxies of the language-associated areas 44 and 55b was transformed from the native space to the gyrus target space, in order to perform the same moving average process.

Results

Regarding the shape of the precentral sulcus, the first dimension of the manifold codes for the I-shape or T-shape of the inferior precentral sulcus. T-shapes are associated with larger proxies of the 44 area (see Fig.). The third dimension of the manifold codes for the length of the inferior precentral sulcus, and for the localization of the interruption usually separating the inferior precentral sulcus from the superior precentral sulcus. The 55b area is systematically located in this interruption, regardless of its localization (see Fig.).

Conclusion

Using manifold learning to represent the variability of the cortical folding patterns requires large datasets beyond reach for post mortem architectonic mapping. This is why we propose to use fMRI-based area proxies to establish a hypothesis, which we will test in a second phase with the brains used to establish the Jülich Brain. Fourteen of these have recently been analyzed with the Morphologist pipeline [Wang et al., 2024].

References
Sun Z. Y. et al., “Linking morphological and functional variability in hand movement and silent reading”. Brain Structure and Function 221(7):3361-71, 2016
Wang, X. et al. "A framework to improve the alignment of individual cytoarchitectonic maps of the Julich-Brain atlas using cortical folding landmarks." Cerebral Cortex 34.2 (2024): bhad538.

Speaker: Dr Zhong Yi Sun (Institut du Cerveau de Paris (ICM))

BigBrain.png

12:30 Lunch Break

Contributed Talks - Modelling and Connectivity (co-Chairs: Lorenzo Pini, Boris Bernhardt)

co-Chairs: Lorenzo Pini, Boris Bernhardt

55 Improving Brain Simulation Accuracy: Sensitivity Analysis and Realistic Time Delays in Neural Mass Models

The classical Jansen and Rit Neural Mass Model has been a cornerstone in computational neuroscience, offering valuable insights into brain activity and neural dynamics. However, traditional implementations of this model often fall short in capturing the complexities of real neural systems, particularly in terms of conduction delays and parameter sensitivity.
To address these limitations, we have reformulated the Jansen and Rit Neural Mass Model using Algebraic Random Differential Equations. The Local Linearization Method was employed as the numerical integrator due to its established advantages in efficiency and accuracy over traditional numerical methods. Furthermore, we introduced a more realistic modeling framework for distributed conduction delays by considering axonal properties such as length, diameter, myelination, and g-ratio. This innovative approach enables the simulation of thousands of interconnected cortical columns, achieving more accurate representations of neural dynamics.
A key aspect of our study involved conducting a comprehensive sensitivity analysis to evaluate the robustness and reliability of the model. To achieve this, we combined Machine Learning approaches using decision trees and Random Forests, focusing on characteristics that define healthy and epileptiform rhythms, such as amplitude, frequency, and the number of peaks per period. This analysis identified the most influential parameters affecting neural dynamics, providing critical insights into the factors essential for accurate simulations. By systematically varying each parameter and observing the resulting changes in the model's output, we identified both excitatory and inhibitory postsynaptic potentials (EPSP and IPSP, respectively), the inflection points of the sigmoid function (v0), and the axonal diameter (diam) as the most important. EPSPs are crucial for driving neural activity, while IPSPs are essential for regulation and stability. The parameter v0 can shift the entire sigmoid function along the membrane potential axis, changing the firing rate for a given membrane potential. Axonal diameter influences multiple aspects of neural dynamics; even small changes in this parameter can lead to significant alterations in model outputs, such as firing rates, oscillatory patterns, and synchronization. These findings not only validate our modeling approach but also highlight areas for further investigation and optimization.
These results highlight the potential of this new approach for advancing our understanding of neural mass models and improving the accuracy of large-scale brain simulations. This work provides a framework for multiscale analysis, allowing the incorporation of real information such as electrophysiology, neurotransmitters, and receptors. It offers a powerful tool for exploring brain activity and understanding neurological diseases, paving the way for new insights and therapeutic strategies.

Speaker: Anisleidy Gonzalez Mitjans (McGill Centre for Integrative Neuroscience, Montreal Neurological Institute, McGill University, Montreal, QC, Canada)

56 EEG Simulation through Integration of Structural Connectivity Data and High-Density Surface Meshes in Brain Region-Specific Network Models

This project presents a novel Region-Specific Brain Network Model (RSBNM) framework that integrates high-resolution multimodal data, exemplified through a case study focusing on the hippocampus. The core elements include leveraging the high-density surface mesh of the hippocampus from BigBrain [1, 2] to simulate EEGs within The Virtual Brain (TVB) environment, employing MRI-derived structural connectivity data [3] exported from EBRAINS, and generating regional mapping information.

A significant challenge in model development is balancing practical feasibility with biological authenticity. While high-resolution surfaces are ideal for accurately depicting intracortical connectivity, the computational demands are substantial. To address this challenge, we developed the BigBrain Network Model (BBNM) framework, designed to integrate the high-resolution brain atlas of BigBrain with TVB's simulation capabilities.

Structural connectivity maps from 200 HCP subjects were averaged. The HCP structural connectome dataset is derived from 20 state-of-the-art brain parcellations to reconstruct region-based empirical structural connectivity from diffusion-weighted MRI data. The following brain parcellation schemes, along with their respective number of parcels and associated publications, were utilized:

• MIST: 31, 56, 103, 167 parcels [Urchs et al., 2019]
• Craddock: 38, 56, 108, 160 parcels [Craddock et al., 2012]
• Shen 2013: 79, 156 parcels [Shen et al., 2013]
• Harvard-Oxford: 48, 96 parcels [Desikan et al., 2006; Frazier et al., 2005; Goldstein et al., 2007; Makris et al., 2006]
• Desikan-Killiany: 70 parcels [Desikan et al., 2006]
• von Economo-Koskinas: 86 parcels [Scholtens et al., 2018; von Economo & Koskinas, 1925]
• AAL (version 2): 92 parcels [Rolls et al., 2015; Tzourio-Mazoyer et al., 2002]
• Destrieux: 150 parcels [Destrieux et al., 2010]
• Brainnetome: 210 parcels [Fan et al., 2016]
• Schaefer: 100, 200 parcels [Schaefer et al., 2018]
• Julich-Brain (version 2.9)**: 294 parcels [Amunts et al., 2020; Amunts et al., 2021]

This comprehensive approach enables detailed simulations of EEG activity, providing valuable insights into the structural-functional relationships within the human brain and demonstrating the practical applicability of the region-specific BBNM framework.

[1] Katrin Amunts et al. BigBrain: An Ultrahigh-Resolution 3D Human Brain Model.Science340,1472-1475(2013).DOI:10.1126/science.1235381

[2] Jordan DeKraker, Roy AM Haast, et al. (2022) Automated hippocampal unfolding for morphometry and subfield segmentation with HippUnfold eLife 11:e77945

[3] Domhof, J. W. M., Jung, K., Eickhoff, S. B., & Popovych, O. V. (2022). Parcellation-based structural and resting-state functional brain connectomes of a healthy cohort (v1.1) [Data set]. EBRAINS. https://doi.org/10.25493/NVS8-XS5

Speaker: Alejandro Salinas Medina (School of Computer Science - McGill University)

BBW2024_Alejandro-abstract_rev.pdf

57 cuBNM: GPU-Accelerated Biophysical Network Modeling

Biophysical network modeling (BNM) of the brain is a promising technique for bridging macro- and microscale levels of investigation, enabling inferences about latent features of brain activity such as excitation-inhibition balance. This approach allows personalized models of the brain to be fitted to individual subjects' imaging data through parameter optimization. However, the process typically requires several thousand simulations per subject, making it computationally expensive and limiting its scalability to larger subject pools and more complex models.

To address this, we present cuBNM (https://cubnm.readthedocs.io), a toolbox designed for efficient simulation and optimization of BNMs using both GPUs and CPUs. The core of the toolbox operates highly parallelized simulations using C++/CUDA, while the user interface, written in Python, is intuitive and allows for user control over simulation configurations. The toolbox includes parameter optimization algorithms such as grid search and evolutionary optimizers, including the covariance matrix adaptation-evolution strategy (CMA-ES).

Currently, the toolbox supports two types of reduced Wong-Wang models, but its modular design allows for future inclusion of additional models. These models can incorporate global or regional free parameters that are fit to empirical data using the provided optimization algorithms. Regional parameters can be homogeneous or vary across nodes based on a parameterized combination of fixed maps or independent free parameters for each node or group of nodes. This flexibility enables the integration of data, such as the microstructural variability of the brain from the BigBrain project, to enhance the biological realism of large-scale brain simulations.

In this demonstration, we showcase the use of the cuBNM toolbox for running simulations and optimizing parameters in two scenarios: a homogeneous model and a heterogeneous model informed by BigBrain data. Through these examples, we demonstrate the potential of cuBNM to streamline brain simulation research, making it more accessible and scalable for larger studies and more complex modeling tasks.

Speaker: Amin Saberi (Institute of Neuroscience and Medicine (INM-7), Research Centre Jülich, Jülich, Germany; Otto Hahn Research Group for Cognitive Neurogenetics, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany; Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf, Düsseldorf, Germany)

58 Minimizing time-to-result: Cobrawap latest developments and applications

Abstract

Cobrawap (Collaborative Brain Wave Analysis Pipeline) [1,2] is an open-source, modular and customizable data analysis tool developed in the context of HBP/EBRAINS [3], with the aim of enabling standardized quantitative descriptions of cortical wave dynamics observed in heterogenous data sources, both experimental and simulated. The tool intercepts the increasing demand expressed by the Neuroscience community for reusability and reproducibility, offering a software framework suitable for collecting generalized implementations of established methods and algorithms, and for embracing innovative procedures. Inspired by FAIR principles [4] and leveraging the latest findings in software engineering, Cobrawap is structured as a collection of modular Python3 building blocks that can be flexibly arranged along sequential stages, implementing data processing steps and analysis methods, directed by Snakemake or CWL workflow manager [5,6].

Cobrawap faces a general drawback of complex analysis tools, namely the need for technical efforts required for the initial installation and configuration, and the non-negligible demand for computational resources for execution on local machines. Offering Cobrawap within the EBRAINS infrastructure allows for addressing a larger community, multiform in background, expertise and objectives, thus at the same time pushing developers to pay attention to increased usability and user facilitation for the software release and execution. Following this line, actions have been completed to successfully deploy Cobrawap on FENIX-ICEI federated HPC sites, executable through direct ssh login or from the EBRAINS Collab. Other actions under the same guiding light (i.e. minimize the time-to-result, with users focusing on the scientific side without caring about the technology behind the scenes) have been to wrap the source code and all the dependencies as a standalone installable Python package and as a Docker image. Finally, recent efforts have been addressed in optimizing and improving algorithms to pursue the benefits of parallel computing (e.g. through vectorization approaches and parallelization libraries).

Among the latest scientific developments, dedicated efforts have been focused in dealing with both high-resolution recordings from brain imaging, and simulations of neuronal dynamics in the human brain obtained from models implemented through heterogenous simulation engines (e.g. TVB [7]). In the first case, annotated images are passed through a recursive algorithm (“HOS”, Hierarchical Optimal Sampling) that dynamically tunes the resolution across the field of view, optimizing both the signal-to-noise ratio and the dataset size. The second case represents a crucial step along the challenging pathway to reliably analyze human brain data (e.g. from EEG), extending the quasi-planar approach assumed for murine cortical data [8] towards the more complex geometrical structure of the human cortex.

Acknowledgments

Research co-funded by: European Union’s Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreements No. 785907 (HBP-SGA2) and No. 945539 (HBP-SGA3); European Union’s Horizon Europe Programme under Specific Grant Agreement No. 101147319 (EBRAINS 2.0); European Commission NextGeneration EU (PNRR EBRAINS-Italy MUR-CUP-B51E22000150006).

References

[1] https://github.com/NeuralEnsemble/cobrawap, https://cobrawap.readthedocs.io
[2] Gutzen, et al. (2024) https://doi.org/10.1016/j.crmeth.2023.100681
[3] https://www.ebrains.eu
[4] Lamprecht, et al. (2019) https://doi.org/10.3233/DS-190026
[5] Mölder, et al. (2021) https://doi.org/10.12688/f1000research.29032.2
[6] Crusoe, et al. (2022) https://doi.org/10.1145/3486897
[7] Gaglioti, et al. (2024) https://doi.org/10.3390/app14020890
[8] Capone, et al. (2023) https://doi.org/10.1038/s42003-023-04580-0

Speaker: Cosimo Lupo (Istituto Nazionale di Fisica Nucleare (INFN), Rome)

Cobrawap in EBRAINS.png

59 Neuropil distributions in the human brain predict resting-state functional networks

The human brain is intrinsically organized as (anti-)correlated regions as shown repeatedly by resting-state fMRI (Biswal et al. 2010), where these regions are thought to represent activity fluctuations of neuronal populations across large cortical swathes (Chen et al. 2020). Yet elucidating properties of these functional networks based on neuroanatomy has remained elusive. Prior attempts to predict functional connectivity from structural connectivity, specifically DTI-measured long-range associations, have failed (Honey et al. 2009). We hypothesized if microscopic connectivity of neuropil in the human brain can recapitulate dominant resting-state fMRI networks.

The most popular functional network is the default mode network (DMN) which has been primarily observed in the cerebral cortex (Shulman et al. 1997, Buckner et al. 2008, Thatcher et al. 2014), but also in the subcortex (Li et al. 2021, Seoane et al. 2024). Other examples of functional networks (Lawrence et al. 2023, Mulders et al. 2015) include the language network (LN), salience network (SN), and central executive network (CEN). Hard-wired networks range from medial and basolateral limbic systems (LIMBm, LIMBb) to primary/secondary visual cortex (VN) and somatosensory/motor cortex (SMN).

Neuropil density is fundamental for in silico modeling of brain energy metabolism as it represents the infrastructure necessary for brain function (Hyder et al. 2013). The metabolic cost of electrical activity at the neuropil involves several processes, which collectively comprise an “energy budget” (Yu et al. 2018). Bottom-up energy budgets require a certain level of understanding of cellular (CellDen) and synaptic (SynDen) densities (Yu et al. 2023). Prior budgets utilized generalized neuropil density representing prototypical measures of CellDen and SynDen to model the brain’s energetic metabolism. To consider how these densities vary across regions, we created a machine learning algorithm that predicts neuropil density (CellDen, SynDen) from in vivo MRI scans, where ex vivo Merker staining (BigBrain) and in vivo synaptic vesicle glycoprotein 2A PET imaging (SV2A-PET) were reference standards for CellDen and SynDen, respectively (Akif et al. 2024). We used these neuropil data to examine if major resting-state fMRI networks are revealed in human brain.

In vivo MRI datasets from 64 healthy control subjects (Akif et al. 2024) were selected from the ADNI database (adni.loni.usc.edu) of whom 10 were randomly selected for training. Using 54 normal human brain neuropil data, we generated 54 datasets each for CellDen and SynDen. Using these data as subject series, we investigated the presence of networks. The neuropil data revealed both correlated and anti-correlated networks, which are comprised of limbic/sensory systems, but also functional networks (LN, SN, CEN, DMN) commonly identified by resting-state fMRI (Figure 1).

These findings address puzzling results of prior studies that report structural networks cannot fully predict functional networks. Furthermore, the intermediary of the metabolic network created by the energy budget, we demonstrate the promise of individualized anatomic/metabolic datasets to gain insights of microscopic effects underlying mesoscopic measures of functional networks on a per subject basis. Understanding mechanisms of how the human brain is inherently organized as functional networks has major consequences for neuroimaging markers of brain function.

Speaker: Brian Chang (Yale University)

Chang_BigBrain_Workshop_Figure1.pdf

60 Wrap Up and Closing

17:00 Visit University of Padua: Tour to Palazzo del Bo - Tour 1

18:00 Visit University of Padua: Tour to Palazzo del Bo - Tour 2