7th BigBrain Workshop: Challenges of big data integration

4 Oct 2023, 09:00 → 6 Oct 2023, 20:00 GMT

Gróska Innovation and business growth center, Reykjavík, Iceland

Katrin Amunts (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich), Alan C Evans (Montreal Neurological Institute McGill University Montreal), Morris Riedel (School of Engineering and Natural Sciences, University of Iceland), Paule-Joanne Toussaint (Montreal Neurological Institute McGill University Montreal), Susanne Wenzel (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich)

You are cordially invited to attend the 7th BigBrain Workshop, taking place in the beautiful city of Reykjavík, Iceland, on October 5 and 6, 2023.

This workshop is an opportunity for the neuroscientific community to come together and present their cutting-edge research, discuss future prospects of the BigBrain associated data and tools, and explore how to better leverage high-performance computing and artificial intelligence to create multimodal, multiresolution tools for the high-resolution BigBrain and related datasets.

This year's BigBrain Workshop will be held in conjunction with an HBHL Training Day, taking place as a full day event on October 4, on-site at the conference venue. 

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

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

---

The programme is available here.

---

We are proud to announce an exciting list of keynote and invited speakers.

Maryann Martone BigBrain Project  Educational Lecture		Kári Stefánsson HIBALL Lecture  in Brain Analytics and Learning		Paul M. Thompson Sievers Lecture  in Computational Neuroscience
Viktor Jirsa Invited Talk			Pedro A. Valdes-Sosa Invited Talk

---

LOC Session: The Local Landscape of Neuroscience in Iceland

Magnús Örn  Úlfarsson  deCODE genetics contributions in imaging  genetics

Ragnhildur Thóra Káradóttir Neuronal activity modulates myelin plasticity and regeneration

Lotta María  Ellingsen Identifying Brain Imaging Biomarkers for Early Diagnosis of Rare Dementia through MRI Post-Processing

The full programme is available here.

---

Important dates 

• Registration open: April 25, 2023
• Call for contributions open: May 2, 2023
• Abstract submission due: July 17, 2023, July 31
• Acceptance notifications: August 8  
• Registration deadline: September 4, 2023
   

---

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: School of Engineering and Natural Sciences University of Iceland Morris Riedel Katrín Ólöf Egilsdóttir

Please contact the programme committee if you have any questions.  We will continuously update the information on this page and also share information via 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 GMT. 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

 

 

Wednesday 4 October

1 Welcome notes

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

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

2 BigBrain Project Educational Lecture

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Are we FAIR yet? Perspectives on the state of global neuroscience data sharing

Neuroscience has made tremendous strides in moving towards open neuroscience over the past decade. Spurring by large investments into neuroscience infrastructure and open datasets, we are close to taking our place alongside fields such as genomics and astronomy, the poster children for open science, in providing large pools of public data. Given the rapid pace of technological development, the multimodal nature of our data and the plethora of model systems, establishing these pools was no small feat. But are we FAIR yet? We’ve made progress in making data findable and accessible, but interoperable and reusable remain a challenge. We know that large prospective data sharing projects like HBP, SPARC, ABCD, Human Connectome, US BRAIN Initiative Cell Census Network and Cell Atlasing project, succeeded in moving large amounts of data to the public domain, but is creating routine and effective data sharing according to the FAIR data principles within the reach of individual scientists across all disciplines? In this presentation, I will consider the state of global neuroscience infrastructure and challenges and benefits of routine data sharing for both big and small science. I will argue that as a whole, the global neuroscience community has taken significant steps towards FAIR, even as work remains to be done.

Dr Maryann Martone is a Professor Emerita at University of California, San Diego, and Chair of the Governing Board for the International Neuroinformatics Coordinating Facility (INCF). She is past Chair of INCF’s Council on Training, Science and Infrastructure, and past President of FORCE11 where she promoted the advancement of scholarly communication and e-scholarship. Pr Martone is also the founder of SciCrunch, a tech startup based on a uniform resource description framework for neuroscience (first developed under her lead at the Neuroscience Information Framework - NIF) and featuring a portal for connecting researchers to resources (built under her leadership at the NIDDK Information Network - dkNET). She started her career as a neuroanatomist, specialising in light and electron microscopy, but her current research interests are focusing on neuroinformatics with her team at the FAIR Data Informatics Lab and consist in building ontologies for neuroscience data integration and supporting data sharing through several large consortia, including the Stimulating Peripheral Activity to Relieve Conditions (SPARC) project, the BRAIN Initiative Cell Atlas Network (BICAN) and Preclinical Interagency Research Resource for Traumatic Brain Injury (PRECISE).

Speaker: Prof. Maryann Martone (University of California)

10:15 Coffee break

Hands-on session 1A

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

3 Neuromaps - a toolbox for structural and functional interpretation of brain maps

The development of advanced neuroimaging techniques has made it possible to annotate the brain in increasingly rich detail. In parallel, the open science movement has given researchers from diverse disciplines access to an unprecedented number of human brain maps. Integrating multimodal, multiscale human brain maps is necessary for broadening our understanding of brain structure and function. Here I will introduce neuromaps, an open-access Python software toolbox for contextualizing human brain maps. Neuromaps currently features over 70 curated brain maps, including genomic, neuroreceptor, microstructural, electrophysiological, developmental, and functional ontologies. The toolbox implements functionalities for generating high-quality transformations between four standard neuroimaging coordinate systems (MNI152, fsaverage, fsLR, CIVET), and can parcellate vertex- and voxel-level data according to a specified brain atlas. Robust quantitative assessment of map-to-map similarity is enabled via a suite of spatial autocorrelation-preserving null models, including permutation-based and generative models. Neuromaps combines open-access data with transparent functionality for standardizing and comparing brain maps, providing a systematic workflow for comprehensive structural and functional annotation enrichment analysis of the human brain. The workshop will include a presentation as well as a live coding workshop. Attendees are encouraged to bring their laptops to follow along.

Speaker: Justine Hansen (Montreal Neurological Institute)

Hands-on session 1B

Zoom webinar:
https://zoom.us/j/93338792038?pwd=TFhHYzg1UGhBK1pnMy9CMC9nd01xQT09
Kenncode: 7404cor

4 Exchanging data & knowledge

Data should be FAIR, meaning they should be findable, accessible, interoperable, and reusable. Especially in complex scientific fields, such as neuroscience, graph databases, such as the EBRAINS Knowledge Graph, are particular suited to host FAIR data and exchange gathered knowledge. In this session you will learn how to prepare your data for sharing and how to represent those data and the knowledge they hold in a graph database. First, we will generally discuss good practices for data organizations, metadata annotations, and data descriptors. You will then learn how to represent your data in a graph database using the openMINDS metadata framework. Last we will explore the benefits of data shared through the EBRAINS Knowledge Graph.

Requirements: WiFi; Laptop; EBRAINS account (desired); Python basic knowledge (desired)

Speaker: Lyuba Zehl (Jülich Research Centre, INM-1)

12:00 Lunch break

Hands-on session 2A

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

5 siibra-explorer and siibra-python - How to use EBRAINS atlas services with the online Siibra Explorer and Siibra Python tools

siibra is a software tool suite that implements a multilevel atlas of the brain by providing streamlined access to reference templates at different spatial scales, complementary brain parcellations maps, and multimodal regional data from different sources which is linked to brain anatomy at different spatial scales. It addresses interactive exploration via an interactive 3D web viewer (siibra-explorer) and integration into data analysis and simulation workflows with a comprehensive Python library (siibra-python), supporting a broad range of workflows for anatomists, experimentalists and computational neuroscientists with varying experience levels, from beginners to those with a solid background in Python.

This session offers participants an immersive opportunity to explore the advanced tools and techniques for data analysis and visualization. We will briefly introduce the tool suite and highlight its features and benefits. Participants will learn to access 3D reference templates and maps, including anatomical, and connectivity atlases. We will interactively explore BigBrain cytoarchitectonic maps and cortical layer segmentation and extract region-specific information via the EBRAINS Knowledge Graph.

Moving beyond the graphical interface of siibra-explorer, the session will proceed with siibra-python. Participants will be guided through coding exercises demonstrating how to fetch brain region maps, access the BigBrain dataset, and extract multimodal regional features such as cortical thicknesses, cell and neurotransmitter densities, gene expressions, and connectivity data.
After completing the training, participants will have a first insight of the features of siibra-explorer and siibra-python to enhance their ability to explore brain atlases and perform advanced neuroimaging analyses.

Requirements: A laptop with an up-to-date web browser (Chrome or Firefox is recommended) is required for the hands-on examples. All examples will be run on pre-built Jupyter notebooks, which will be provided for downloading. Please register for an EBRAINS account in advance.

Speakers: Timo Dickscheid (Forschungszentrum Jülich), Sebastian Bludau (Forschungszentrum Jülich - INM1)

Hands-on session 2B

Zoom webinar:
https://zoom.us/j/93338792038?pwd=TFhHYzg1UGhBK1pnMy9CMC9nd01xQT09
Kenncode: 7404cor

6 An open access software ecosystem for multiscale neuroscience

Hands-on session: From BigBrains to BrainSpaces: open tools to integrate histology, imaging, and macroscale connectivity

The human brain is a complex system, which can be studied from multiple different angles: As microscopic neurons form intricate webs of connections, a structural backbone emerges to support communication across anatomically distant brain areas. As such, considering both brain anatomy and function across several scales of investigation promises to lead to a more complete understanding of this organ. By highlighting the need to study the brain in a more integrated manner, this perspective has required a significant shift in the way we collect and analyze brain data. This hands-on session will showcase open software and data resources facilitating such investigations, with a particular emphasis on multiscale methods bridging histology, neuroimaging, and macroscale brain networks in humans. Following a brief overview
of relevant tools, participants will be guided through the application of these methods via
microstructure-informed investigations of functional brain network organization and statistical analysis of multimodal imaging features.

Session requirements
This session will mainly rely on Google Colab, allowing participants to execute notebooks prepared for this session and by-passing any system-based or installation requirements. Although prior knowledge of Bash and Python is an asset, this is not a requirement. Participants will have the option of providing their own neuroimaging data mapped to the fs-LR 32k-vertex neocortical surface template for processing and analysis in the provided notebook, but processed data will be provided for all participants to use. Further information on how to set up the participants’ own environment will be provided.

Speakers: Raúl Rodriguez-Cruces (Montreal Neurological Institute, McGill University, Montreal, Canada), Jessica Royer (McGill University), Boris Bernhardt (Mcgill)

14:30 Coffee break

Hands-on session 3A

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

7 Atelier3D-MCIN Tutorial – A software suite for interactive visualization, annotation and mapping of the big brain

This tutorial will introduce the Atelier3D-MCIN software and utilities, and provide walk troughs of different annotations and mapping applications for the big brain. Atelier3D is a Windows and Linux-based software platform that provides powerful processing and visualization tools for the analysis of very large volume and surface datasets, and this version has been specifically adapted to support 3D annotation, segmentation and 3D surface extraction in the big brain in support of different use cases in McGill and Juelich. The tutorial will cover Installation and basic use of the software for local or remote visualization, and then focus on one or more specific scenarios of interest to the community, in collaboration with actual users of A3D. New and advanced features will also be discussed, including interoperability with other tools and video production.

Speaker: Mona Omidyeganeh (Digital Technologies at National Research Council Canada)

Hands-on session 3B

Zoom webinar:
https://zoom.us/j/93338792038?pwd=TFhHYzg1UGhBK1pnMy9CMC9nd01xQT09
Kenncode: 7404cor

8 BigBrain data processing with CBRAIN and DataLad

This tutorial will provide an introduction to two tools that can be used to process and manage BigBrain-related data: CBRAIN and DataLad.
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. DataLad is a data integration tool to keep track of distributed datasets. The tutorial will cover the main functionalities of CBRAIN and DataLad, illustrate them on BigBrain data, and demonstrate their interaction.

Expected learning outcomes:
In this tutorial participants will learn to use the following infrastructure tools and services for analyzing data in HIBALL:

• Part 1: CBRAIN to process BigBrain data on HPC clusters through a web portal, including:
• Finding BigBrain datasets in CBRAIN
• Processing BigBrain datasets in CBRAIN
• Exploring and working with the processing results
• Downloading and sharing the results
• Part 2: DataLad and Boutiques, to access and process BigBrain data through uniform command-line interfaces, including:
• Finding BigBrain datasets in DataLad
• Installing and downloading BigBrain datasets
• Processing BigBrain datasets with Boutiques
• Uploading derived data to DataLad

Preparations & Equipment:

• Laptop (Windows, Mac or Linux)
• NeuroHub or CBRAIN account
  Participants who do not already have a NeuroHub or CBRAIN account can obtain one in advance of the training session by registering at https://portal.neurohub.ca and selecting ‘Request Account’.

Speaker: Bryan Caron (NeuroHub, McGill University)

16:15 Coffee break

9 AI-driven brain modelling

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Artificial Neural Networks (ANNs) are the workhorse of modern AI, but they were originally invented as models of the brain. In recent years computational neuroscientists have used ANNs to model the brain to great effect, achieving state-of-the-art matches to the representations found in the brain. In this session we will outline two of the major uses for ANNs in neuroscience research: 1) as a tool to answer "why" questions and 2) as a tool for discovery. We will discuss how anatomical, connectomics, and functional data, including anatomical by the BigBrain, can facilitate progress in modeling the brain. We will end with a discussion session on best tips for using ANNs in neuroscience.

Speakers: Pouya Bashivan (McGill University; MILA), Blake Richards (McGill University; MILA)

10 Panel discussion: Launching your Lab and Steering your Career

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Moderation: Jane Roskams
Panel: Pouya Bashivan, Boris Bernhardt, Lotta María Ellingsen, Sofie Valk

18:00 Reception

Thursday 5 October

11 Welcome and Introduction

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Speakers: Prof. Katrin Amunts (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich), Prof. Alan C Evans (Montreal Neurological Institute McGill University Montreal), Prof. Jón Atli Benediktsson (Rector of University of Iceland)

12 HIBALL Lecture in Brain Analytics and Learning

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Research into human diversity in the multiomic era

Dr Kári Stefánsson is a founder and CEO of deCODE Genetics in Iceland. He is an International Member of the US National Academy of Sciences and a member of the European Molecular Biology Organization (EMBO). Presently Professor Emeritus at University of Iceland, he was previously a Professor of Neurology, Neuropathology and Neuroscience at Harvard University and Director of Neuropathology at Beth Israel Hospital in Boston, Massachussetts. Professor Stefánsson has received many accolades from the international genetics community, including the European Society of Human Genetics Award, and the prestigious William Allan Award from the American Society of Human Genetics for his lifetime achievements and substantial research contributions to human genetics. He is currently actively engaged in leading the work at deCODE genetics for the discovery of genetic variations associated with common diseases, pointing to potential drug targets.

Speaker: Prof. Kári Stefánsson (deCODE Genetics, Iceland)

13 Panel discussion: Challenges of big data integration

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

10:45 Coffee and discussion break

Contributed Talks 1

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Chair: Kathleen S. Rockland, Jordan DeKraker

14 Morphology-based correction of batch effects in histology and autoradiography

Introduction
Imaging of 2D post-mortem tissue sections presents challenges due to potential differences in digitized pixel intensities across batches of acquired tissue. This issue is particularly pronounced in human in vitro autoradiography, where chemical fixation is not possible and the brain's size prohibits uniform shock freezing. Instead, the donor brain must be sectioned into discrete slabs for independent processing under different experimental conditions. We propose a novel approach to correct batch effects across adjacent 2D sections derived from contiguous tissue slabs based on the cortical morphology of the donor brain.

Methods
Our method uses 2D sections that have been aligned to a 3D structural reference volume, i.e., the donor’s T1w MRI, using a reconstruction pipeline [1].

1. Corresponding pairs of points between adjacent caudal and rostral tissue slabs are identified on a cortical surface mesh extracted from the structural reference volume [2].
   These points minimize the geodesic distance along the cortical manifold between the adjacent tissue slabs (Fig.1)
2. The difference in ligand binding densities between the paired points is calculated and the mean of the differences is used to calculate a correction factor for the rostral tissue slab:
   $$𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑖𝑜𝑛 𝑓𝑎𝑐𝑡𝑜𝑟_{𝑠+1} = \frac{1}{𝑛} \sum_{𝑖∈𝑃} V_i^{caudal_s} - V_i^{rostral_{s+1}}$$ where $s$ is a tissue slab, $i$ is a pair of vertices in the set $P$ of $n$ vertices between the caudal tissue slab, $s$, and the rostral tissue slab in the next tissue slab $s+1$. For $s=0$ the correction factor is set to 0, hence all slabs are adjusted to the first tissue slab. 3. This procedure repeats until the correction factor has been calculated for all caudal-rostral pairs of tissue slabs. 4. The correction factors are used to shift the mean of each slab. To assess the efficacy of the morphology-based batch correction, we employed simulated data generated on a cortical surface mesh. Synthetic ground truth vertex values were assigned corresponding to each vertex's caudal-rostral position along the mesh. Synthetic data were then perturbed by introducing random batch effects. The performance of the batch correction algorithm was evaluated based on the $r^2$ correlation of the true and corrected vertex values. We also applied the batch correction algorithm to reconstructed autoradiographs that measure GABA$_B$ receptor density. **Results** The $r^2$ between the synthetic ground truth vertex values and the corrected values was 0.99 $(p<0.1)$. Visual inspection of the corrected GABA$_B$ receptor density volume revealed that the batch correction was able to remove the obvious batch effects seen in the uncorrected volume (Fig.2&3).

Discussion
We developed a batch correction method that successfully removes batch effects by minimizing the differences between nearby vertices in adjacent tissue slabs on the cortical surface. Future work will apply this method to real autoradiography data and compare corrected ligand binding densities to those observed using positron emission tomography.

[1] Funck, T, et al. "3D reconstruction of ultra-high resolution neurotransmitter receptor atlases in human and non-human primate brains." bioRxiv (2022): 2022-11.
[2] Lepage, C. et al. 2017. "Human MR Evaluation of Cortical Thickness Using CIVET v2.1", OHBM (poster 4166), Vancouver.

Speaker: Thomas Funck (Child Mind Institute, New York City, New York, USA)

BBWS2023_33_Fig1.png

BBWS2023_33_Fig2.png

BBWS2023_33_Fig3.png

15 3D reconstruction of BigBrain2: Progress report on updated processing pipeline and application to existing annotations and cortical surfaces

The development of BigBrain2 is a continuation of the first BigBrain [1] that will contribute new insight on inter-subject cytoarchitectonic variability. Overall, BigBrain2 offers better quality staining, favorable to regional segmentation and registration, and contains fewer artefacts through sectioning and staining. In this presentation, we will report about the initial 3D reconstruction of BigBrain2 at 100µm, which is suitable already for the extraction of cortical surfaces and the representation of annotations of some cortical and subcortical regions.
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, and manual and semi-automatic corrections were performed to repair acquisition artifacts 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), from which a first 3D reconstruction was obtained at an effective section spacing of 100µm. Data provenance tracking of all repair operations provides a means for assessing the extents of the repaired artifacts and for eventual reproducibility at the 1µm in-plane resolution. The repaired sections were aligned to the post-mortem MRI of the fixed brain (Siemens Sonata, 1.5T, MPRAGE, 0.5mm) in an iterative process by 3D registration of the stacked images to the MRI, followed by 2D registration of the individual images to the sliced MRI, while gradually increasing the degree of 2D and 3D registration from rigid-body to affine to non-linear across 10 global iterations. These extra global iterations helped resolve the lower-frequency alignment errors causing jaggies. Alignment to the MRI enables to correct for tissue compression caused by cutting and mounting of sections, and tissue shrinkage. Ultimately, section-to-section non-linear 2D alignment (without MRI) was performed to resolve high-frequency alignment errors. Optical-balancing was applied by normalizing image intensities to the MRI data to correct for staining imbalances across the brain. The reconstructed 3D volume is obtained at 100µm in the MRI ex-vivo space, which is suitable for the extraction of cortical surfaces. Finally, computed transformations are saved and can be applied to regions annotated on the original sections.
Ongoing work includes the semi-automatic repairs of the remaining sections (80%) 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

Speaker: Hartmut Mohlberg

BBWS2023_23_Fig.pdf

16 Advances in MSM surface registration to bridge data across the BigBrain and FS/HCP ecosystems

Introduction

Structural MRI-derived population average cortical surfaces such as FreeSurfer's (FS) fsaverage 'fsavg' (Fischl 2012) and Human Connectome Project's (HCP) fs_LR (Van Essen et al., 2012) serve as reference frameworks for multimodal data integration.

The cortical surface extracted from the high-resolution 3D-reconstructed histological BigBrain model (Amunts et al., 2013, Lewis et al., 2014, Wagstyl et al., 2018) is also a notable reference space for increasing amounts of data.

Accurate surface registration between BigBrain and MRI-derived average surfaces is critical, not only for a wide range of functional applications, but also for comparison and cross-validation of the pipelines, themselves.

However, the BigBrain surface presents several challenges to existing registration tools designed for use with MRI and validated within a particular software ecosystem .

Recently, we introduced the first surface registration pipeline linking these reference surfaces (Lewis et al., 2019, 2020) via HCP's Multimodal Surface Matching (MSM) tool (Robinson et al., 2014, 2018). In 2021, our registration was included in BigBrainWarp (Paquola et al., 2021).

In the present work, we incorporate new tool developments in both the MNI and FS ecosystems to optimize our pre-processing, leading to improved alignment of folding-based landmarks, as well as reductions in maximum distortion.

Methods

Preprocessing:

1. input: CIVET-tessellated BigBrain white surface (2021 update: 163k vertices per hemisphere, medial cut excluding hippocampus)

2. volumetric transform from histological to HCP ‘dedrift’ space

3. use FS tool mris_remesh (new in v7.1) to re-tessellate white surface to 800,000 vertices per hemisphere

4. inflate FS-tessellated white surface to sphere with modified FS inflation tools

5. MNI tool surf_surf_interpolate (new output map function) to obtain vertex correspondence between 2 versions of white surface (original CIVET-tessellated 163k and FS re-tessellated 800k)

6. ‘borrow’ FS sphere at 163k using obtained vertex correspondence

MSM BigBrain to fsavg (then concatenate to fs_LR via HCP registration fsavg-to-fs_LR):

• Stage 1: Affine rotation. 2 landmarks: single vertex near anterior and posterior border of medial cut

• Multiscale approach (discrete / non-linear processing). Stage 2: Sulcal depth (global scale); Stage 3: Curvature (finer features)

--trans flag used across all stages (accounts for distortions from all previous stages)

• sampling grid and datagrid were 1 level higher than default

Analysis:

18 landmarks were manually defined on both BigBrain and fsavg. Geodesic distance (error) between the resampled and original landmarks was used to assess accuracy of the registration process.

Results

Our modified MSM pipeline yields improvements in alignment of folding-based landmarks relative to standard MSM and our previous versions (reduction of mean and maximum error, and elimination of outliers; Fig 1), as well as reductions in maximum distortion values.

Conclusions

Future work aims to compare BigBrain with atlases in fsavg or fs_LR space. Such comparisons [e.g., BigBrain cytoarchitecturally-defined areas to multi-modally defined areas from the HCP-MMP1.0 (Glasser et al., 2016)] will be critical for validating that the modified folding-based registration shown here achieves accurate alignment of cortical areas (as HCP-data MSMSulc registration was optimized for brain areal alignment based only on cortical folds when multi-modal information is not available).

Speaker: Lindsay B. Lewis (McGill Centre for Integrative Neuroscience (http://mcin.ca), Montreal Neurological Institute, McGill University, Montreal, QC, Canada)

BBWS2023_05_Fig.png

17 Scaling for Big Data: An Enhanced Surface Reconstruction of the Hippocampus Leveraging Delaunay Triangulation for High-Resolution Mapping from Unfolded to Native Spaces

The ability to process and analyze large and complex neuroimaging datasets is crucial to develop computer simulations of the brain. A focus of significant interest is to model the hippocampus, a brain structure integral to memory formation and emotional regulation. This study introduces a state-of-the-art algorithmic pipeline for big data scaling and enhanced surface reconstruction of the hippocampus, utilizing Delaunay Triangulation for high-resolution surface generation.
Initially, the pipeline addresses the challenge of upscaling low-resolution surface models by leveraging Delaunay Triangulation. This algorithmic approach not only maintains but enhances anatomical detail, allowing for a high-resolution representation of the hippocampus' complex topology. This process is particularly advantageous for large datasets, making it scalable and big data-compatible.
Following the surface enhancement, the pipeline employs a specialized warp field algorithm to transform the high-resolution surface from an unfolded space to a native space. This ensures compatibility with existing volumetric datasets and enhances the granularity and precision of subsequent analyses. An extensive set of validation checks ensures the warp field's fidelity, safeguarding the integrity of the transformed model.
Finally, the warped high-resolution surface is integrated with BigBrain hippocampal volumetric data through a robust volume-to-surface mapping algorithm. This step harmonizes the dual approaches of surface-based and volume-based analyses, allowing for comprehensive, nuanced exploration of hippocampal structure and function.
The pipeline is implemented in widely-accepted neuroimaging formats like NIFTI and GIFTI, ensuring seamless integration with existing analytical tools and datasets. Preliminary results indicate a significant advancement in the accuracy and depth of hippocampal analyses. This scalable approach is versatile and holds promise for a myriad of applications, from basic neuroscience research to advanced investigations into neurodegenerative diseases and cognitive disorders. Overall, the pipeline sets a new precedent for high-resolution, big data-compatible computational analysis of complex brain structures.

Speaker: Alejandro Salinas-Medina (McGill University, Montreal)

18 Relationship between receptor covariance and functional connectivity in macaque somatosensory areas

Introduction:
Sensory information is processed in a hierarchical way across different areas of the somatosensory cortex. In primates, the sensory signal first arrives at the primary somatosensory cortex (SI), and then at the secondary somatosensory cortex (SII), finally at the association cortex, which consists of areas that are located on the superior (SPL), inferior parietal lobes (IPL), and intraparietal sulcus (ips). Neurotransmitter receptors are a key element of information processing. To understand how somatosensory information is processed between brain areas, we analyzed the multiple receptor covariance (RC) patterns of distinct somatosensory areas. Furthermore, we examined functional connectivity (FC) patterns of each area and explore the shared and specific characteristics between RC and FC.
Methods:
In the present study, we defined 118 areas throughout the macaque brain, 36 of which are somatosensory-related areas located within SI, SII, and parietal association areas. The densities of 14 different receptors in each of the 118 areas had been quantified by means of in vitro receptor autoradiography. To construct the RC of each somatosensory-related area with the remaining areas across the cortex, we calculated a representative feature vector consisting of 14 receptor densities for each area. Statistical similarity between two areas was measured by computing the Pearson correlation. Furthermore, we reconstructed the resting-state FC of the somatosensory cortex using fMRI data from the PRIME-DE dataset. A principal components analysis was performed on the BOLD activity time courses across all vertices within each area, where the first principal component was taken as the representative activity time course for this area. The representative time courses were subsequently correlated with the activity time courses for each vertex across the brain.
Results:
RC patterns are similar for areas that are 1) anatomically adjacent to each other; or 2) at the same level of hierarchical organization. All SI areas showed consistent correlations with caudal SII, rostral and ventral ips, rostral IPL and higher visual areas. SII areas displayed stronger correlations with rostroventral IPL and cingulate areas, but relatively weaker correlations with ips and visual areas. Regarding SPL, there is a clear segregation between areas located on the lateral surface and the areas located within the cingulate sulcus. Within IPL and ips, the RC patterns changed gradually from rostral to caudal. As in RC patterns, the strongest FC was found between neighbouring areas. Likewise, early and higher sensory areas could also be separated by their FC patterns. The FC and RC also have some differences. For example, FC patterns of SI show more consistent connections to the primary motor cortex instead of to higher visual areas.
Conclusions:
Our results show that areas belonging to SI, SII or the somatosensory association cortex have distinct connectivity patterns in both RC and FC. Furthermore, despite comparable features, there are also important differences between the RC and FC of the somatosensory cortex. More broadly, our findings provide a link between the chemoarchitectonic and functional organization of the macaque somatosensory cortex and thus a novel direction for a multiscale understanding of brain structure and function.

Speaker: Meiqi Niu (INM-1)

19 Integration of cytoarchitecture and brain-wide connectivity reveals topographic organization of macaque insula networks

One of the major challenges to carrying out in vivo neuroanatomical analyses is that there is significant regional cytoarchitectural variation that is often difficult to capture by MRI. As a result, parcellation of brain regions is often limited to a scale too coarse for the understanding of their functions. While this is a challenge for many regions of the brain, the insula is one such region that is known to be cytoarchitecturally heterogeneous in which cellular variation is believed to be related to variation in function. Different regions of the insula support specific functions that are critically important for understanding a broad array of psychological functions, and may ultimately be targets of interventions for neuropsychiatric diseases. In order to capture such heterogeneity among sub-insular regions, it is necessary to collect neuroanatomical data at different resolutions and with different modalities from the same subjects. We present our first Integrated Mic(ro) to Mac(ro) Macaque brain dataset, here called MicMac. MicMac is an extendable workflow, represented by a within-subject whole brain dataset that integrates aligned multi-parametric in vivo MRI, high resolution ex vivo MRI, and histology within a single, standardized space.

In establishing the study protocol, all multiparametric MRI and histology data were obtained from a 10.3 year old healthy female rhesus macaque. In vivo MRI scans were performed on a Siemens 3T equipped with an 8-channel monkey head coil. Ex vivo MRI were conducted on a Bruker 7T using a 72mm volume coil. Both in vivo and ex vivo imaging protocols were harmonized, and optimized for experimental factors such as tissue fixation. Multi-shell dMRI was acquired for structural connectivity analysis using fiber tractography. Complete 3D histological volumes were reconstructed from a stack of cell-body (Nissl) and myelin-stained (Gallyas) 2D microscopy sections (2x2 micron in plane, 40 micron slice thickness, 400 micron interslice spacing) with optical-balancing to account for histological staining artifacts. All processed data were spatially aligned in a common in vivo reference space using an adapted image registration framework that was previously established for the human BigBrain project (Amunts et al, 2013).

The 15 cytoarchitectonically-defined insula subregions (Evrard et al, 2014) were specified on the histological images, rendered in the surface space defined by CIVET-Macaque (Lepage et al, 2021), and used as seed-regions for diffusion MRI (dMRI) tractography to reconstruct the connections between these subregions with other cortical areas.

Results demonstrate excellent correspondence of insula connectivity that align with macaque histological tract tracing studies (Mesulam and Mufson, 1982) and previous human dMRI and resting-state MRI studies (Menon et al, 2020). The specificity of these projections, even for small subregions, was well-defined and occupied distinct patterns across the cortex with minimal overlap. Despite the focus on insula for this application, the workflow demonstrated here (Fig. 1) could be done for any other brain region, although it is currently unknown how well this would translate due to variations in cellular structure and network connectivity.

Fig 1. Cytoarchitectonically-defined fiber tracking was performed in the insula using the MicMac dataset.

Speaker: Erika Raven (NYU Langone Medical School)

BBWS2023_09_Fig.png

20 High resolution cyto-, myelo- and receptor architectonic atlas of the macaque monkey brain

One of the biggest challenges in understanding the brain is being able to discern how the molecular, cellular and systemic levels of organization relate to each other to enable cognitive functions and the control of behavior. The combined analysis of the cytoarchitectonic segregation of the cerebral cortex and of the regional and laminar distribution patterns of multiple neurotransmitter receptors, which constitute key molecules of signal processing in the brain, has revealed that their densities vary considerably between different cytoarchitectonically defined areas, thus revealing their borders and enabling their multimodal characterization. Furthermore, the specific balance in the expression levels of different receptors within a cytoarchitectonically defined area, i.e., the receptor fingerprint of that area, varies systematically depending on the participation of cortical areas in different functional networks, thus indicating the hierarchical aspects of systemic functional organization.
I here present a novel atlas of the macaque monkey brain encompassing an ultra high-resolution 3D histological volume as well as 3D cortical maps encoding the regional and laminar distribution patterns of 14 different neurotransmitter receptors and associated with the stereotaxic space created by the MEBRAINS template and the volumetric representation of the Yerkes 19 template. This resource which will enable for the first time a voxel-wise whole brain analysis of the regional and laminar distribution patterns of multiple receptor types in one and the same macaque monkey brain. This multimodal and multiscale atlas spanning multiple orders of magnitude is accompanied by a parcellation scheme of the frontal, parietal and occipital cortices based on a quantitative analysis of their cyto- and receptor architecture, as well as of the hippocampus and subcortical structures such as amygdalar nuclei, the striatum and the globus pallidus. This atlas will enable a comprehensive analysis of the molecular, cellular and systemic organization of the macaque monkey brain, which must be understood as being topographically specific and structurally/functionally segregated in order to recognize the organizational principles that make it an interconnected system of complex structural and functional units. This resource will enable systematic analyses of the hierarchical relationships between these units, thus providing crucial insights into the structural segregation underlying the brain’s functional organization.

Speaker: Prof. Nicola Palomero-Gallagher (Forschungszentrum Jülich, INM-1)

12:45 Lunch break

21 Leveraging the BigBrain for Electrophysiology

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Electrophysiology is an imaging technology with a high temporal resolution that is cost-effective and widely deployable in any economic setting. It is therefore of great interest to global public health. Electrophysiological Source Imaging (ESI) estimates the neural primary current densities that produce the observed EEG/MEG. While ESI is closer to the underlying neural processes of interest, there remain some critical underdeveloped research areas to fully interpret the data.
In this presentation, we facilitate the use of BigBrain project to address some of these issues with the following developments:

• Development of workflow to obtain ESI projected to the BigBrain atlas space.
• A toolbox, CITFITSTORM that produces BigBrain compatible ESI obtained with a diversity of availability of MRI data to compute the lead yield. These are HCP-compatible MRI collections, Legacy datasets, with only T1, as well as when there is no MRI available [1].
• We have developed methods and a toolbox to integrate external high-dimensional neural mass models with great accuracy and computational efficiency. A main feature of this approach is to allow a general model of disturbed conduction delays between neural masses via a Connectome Tensor formulation [2].

With this work, we aim to facilitate biophysical models for principled multimodal data fusion in the framework of the BigBrain Project.

References:
[1] Areces Gonzalez, Ariosky; Paz-Linares, Deirel; Riaz, Usama; Li, Min; Wang, Ying; Kpiebaareh, Michael Y.; et al. (2023). Multimodal pipeline for HCP-compatible processing and registration of legacy datasets (MRI, MEG, and EEG). TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.22276549.v1
[2] Mitjans, A. G., Linares, D. P., Naranjo, C. L., Gonzalez, A. A., Li, M., Wang, Y., ... & Valdes-Sosa, P. A. (2023). Accurate and Efficient Simulation of Very High-Dimensional Neural Mass Models with Distributed-Delay Connectome Tensors. NeuroImage, 274, 120137. https://doi.org/10.1016/j.neuroimage.2023.120137

Speaker: Prof. Pedro. A. Valdes-Sosa (Joint China Cuba Lab for Neuroscience and Neurotechnology, University of Electronic Science and Technology of China, Chengdu China)

Contributed Talks 2

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Chairs: Viktor Jirsa, Justine Hansen

22 A 3D coordinate system and volumetric flatmap of the whole human neocortex with applications to data visualization and modeling

The BigBrain dataset has provided a high-resolution view of the human neocortex, promoting a shift from its conceptualization as a surface, as is common in MRI and EEG studies, to a volumetric object. This opens the door for analyses and modeling approaches previously only possible with rodent atlases.

In this work, we apply methods originally implemented for rodent neocortex to create a 3D coordinate system adapted to the geometry of human neocortex. The principal axis in this coordinate system is locally orthogonal to layer boundaries and measures cortical depth (Z coordinate). The other two axes are parallel to layer boundaries and describe a flatmap of the cortical volume, a 3D to 2D mapping in which every voxel inside the cortical volume is projected onto a flat square mesh (XY coordinates). This transformation is area-preserving and reversible, allowing not only to map 3D data to 2D for visualization purposes, but also to decompose the 3D volume based on a parcellation described in 2D. Notably, the layer structure is preserved when going from 2D to 3D, which allows easy delineation of columnar subvolumes anywhere in the neocortex.

Taking as starting point the openly available BigBrain cortical layer maps and layer boundary surface meshes, we produced cortical flatmaps and auxiliary atlases of cortical thickness, depth and orientation for both hemispheres, at a resolution of 100 um in the histological BigBrain space. We show applications of the flatmap for data visualization and for parcellation of the cortical volume. We also discuss the potential application of the auxiliary atlases for creating detailed models of human cortical circuits. Finally, we identify some challenges and caveats of our methods as applied to this and future high-resolution human cortical datasets.

Speaker: Sirio Bolaños Puchet (EPFL - Blue Brain Project)

BBWS2023_03_Fig.png

23 HippoMaps: structural and functional mapping of the hippocampus follows a 2D organization

The hippocampus, a critical hub in both cognition and numerous neurological diseases, consists of a folded archicortex which can be digitally unfolded and aligned between individuals using the recently developed HippUnfold software. This enables the application of a wealth of bioinformatics methods to the hippocampus, including the use of spatial correlation between measures in a common space. For example, histology data pertaining to the prevalence of interneurons at the microscale can be directly compared to the temporal characteristics of fMRI or intracranial encephalography at the mesoscale.

Here, we present HippoMaps: a common-space and open source repository for comparing and sharing hippocampal data across scales, datasets, and labs. This work massively scales multimodal, multicontrast, and multiscale imaging of the hippocampus, such that thousands of individual histology slices from different sources are reconstructed in a unified space. Presently we include histology data spanning popular stains sensitive to cell bodies, neural fibers, interneuron cell types, and various others, as well as in-vivo and ex-vivo quantitative MRI pertaining to myelination, blood flow, water diffusivity, functional imaging characteristics, and functional connectivity from the hippocampus to the rest of the brain. Alongside careful considerations of interindividual alignment and statistical rigour, we feel this approach can usher in a new branch of methods for hippocampal research.

From the present data we show that much of hippocampal organization is well described by the intrinsic 2D topology of the hippocampus: structural features vary primarily across the proximal-distal (or subfield-related) axis of the hippocampus while connectivity to the rest of the brain varies primarily across the anterior-posterior (or long-axis) of the hippocampus. Though there is further nuance to explore within hippocampal structure and function, we feel that this is an important consideration for new labs given the growing literature and, in some cases, divergence in understanding about hippocampal structure across labs. In particular, issues like out-of-plane slicing or imperfect modeling of 3D folding of the hippocampus have led to misunderstanding of hippocampal structure in the field. Here, we show that a simple 2D topology provides a powerful explanation for several notable observations in recent literature, and should be considered prior to making novel claims about unique hippocampal features in the future.

To illustrate and facilitate the uses of HippoMaps in new labs, we provide open source tools and tutorials for fetching, visualizing, and statistically comparing hippocampal data. With this framework in place, hippocampal researchers can not only view normative, healthy spatial distributions of hippocampal tissue properties, but they can also download and statistically compare them to ongoing data collection and can even opt to share their results, expanding the availability of hippocampal data prospectively and maintaining open and reproducible research standards within the field.

Speaker: Jordan DeKraker (McGill University)

24 Connectome to graph: building neural networks with brain-like connections

Artificial neural networks (ANNs) have a long history of drawing inspiration from the brain, particularly the visual system. While the performance of ANNs on visual tasks continue to improve, even the most state-of-the-art visual architectures hit a brain similarity ceiling, making them unreliable tools for gathering insights into the brain. Our goal is to build a model of the visual system that breaks past this ceiling by incorporating brain-like connections, particularly top-down feedback.

To that end, we built a publicly available code base that converts a connectome file into a top-down recurrent model, where each anatomical node from the connectome is represented by a recurrent convolutional layer. The user can further customize the relative layer and receptive field size, the mechanism of top-down feedback, and the proportion of feedback and feedforward inputs for each layer to better match the brain area they wish to model.

We used our tool to compare multiple hypotheses on the mechanism of top-down feedback with more realistic connectomes. We found that all proposed mechanisms of top-down modulation can resolve ambiguous sensory input using auditory clues. However, simultaneous threshold and gain modulating feedback helped models perform better on more difficult tasks such as occluded image recognition and tasks that require completely ignoring sensory input. The increased performance of the dual threshold and gain-modulating mechanism is particularly interesting considering previous research which suggests the dual mechanism leads to more brain-like firing patterns for an artificial neuron.

We hope that our code base will enable users to easily compare different biological and non-biological connection schemes in silico, yielding similar insights on the effectiveness of different computations and connectivity schemes on functionality. Our demonstration of the utility of these top-down feedback architectures shows that this could lead to ANNs with more human-like capabilities.

Speaker: Mashbayar Tugsbayar

25 Personalized Human Brain Energy Atlas: Machine learning-based neuropil density from anatomical MRI

Neuropil represents the fundamental unit in which cellular signals are processed, and it consists of neuronal/glial cells whose finest filaments form synapses. Thus, neuropil density (NP) combines both cellular (i.e., neuronal/glial cells) and synaptic masses and is central to understanding the heterogeneity of brain metabolism. Current bottom-up energy atlases use NP-derived neuronal (NeuDen) and synaptic (SynDen) density maps. However, a great innovation would be to predict metabolism from personalized neuroanatomy.

We hypothesize that in vivo NeuDen and SynDen can be derived from longitudinal relaxation time weighted MRI (T1w) for gray/white matter distinction and diffusion MRI for tissue cellularity (apparent diffusion coefficient, ADC) and axon directionality (fractional anisotropy, FA). We present a proof-of-concept machine learning algorithm that successfully predicts NeuDen and SynDen from routine in vivo MRI scans, where ex vivo Merker stain (BigBrain) and in vivo SV2A-PET imaging were respective gold standards. Our machine learning algorithm used gaussian-smoothed T1w/ADC/FA on a voxel-by-voxel basis to predict NeuDen/SynDen maps. We trained and compared NP predictions to NP gold standards and used histogram/spatial correlations as a proxy for prediction efficacy.

Notably, training on group average MRI data, especially with high gaussian smoothing, was ineffective for NP predictions. We used different levels of isotropic gaussian smoothing to provide our neural network nondirectional neighborhood information. Providing only low levels of gaussian smoothed datasets resulted in grainy neuropil predictions, showing the importance of neighborhood information from highly smoothed input datasets. Neighborhood information also helps to mitigate SNR issues (by averaging out the noise) and minor misregistration errors. Additionally, smoothed datasets can also provide a regional baseline to calculate relative changes in T1w, ADC and FA by the neural network as needed for neuropil prediction. High subject-by-subject histogram correlations for SynDen (0.94) and NeuDen (0.88) demonstrated realistic estimates across all subjects, while lower spatial correlations illustrated individualized predictions for SynDen (0.85) and NeuDen (0.45).

In summary, NP represents the microscopic infrastructure that regulates function which can be measured at the mesoscopic level (i.e., millimeter sized voxels in the human brain) with various PET and MRI methods. Since decreases in NP are associated with disorders such as depression, Parkinson’s, Alzheimer’s and aging, this work paves the way for individualized mesoscopic energy atlas prediction, enabling microscopic interpretations of functional neuroimaging data across health and disease.

Speaker: Adil Akif (Yale University)

26 The status quo of automated cytoarchitecture analysis: Where are we, and where are we going?

Cytoarchitectonic brain maps provide a microstructural reference for multi-modal human brain atlases, representing important indicators for brain connectivity and function. Cytoarchitectonic areas are defined by characteristic microstructural cell distributions, including the size, shape, type, orientation, and density of neurons, as well as their distinct laminar and columnar arrangement. High-resolution microscopic scans of histological human brain sections enable identifying cytoarchitectonic brain areas. Modern high-throughput microscopic scanners enable large-scale image acquisition, resulting in petabyte-scale microscopic imaging datasets that provide the foundation for next-generation brain atlases. As established cytoarchitectonic brain mapping methods based on statistical image analysis do not scale to such large datasets, ongoing research aims to develop methods for automatic classification and characterization of cytoarchitecture based on large amounts of high-resolution images.

In this presentation, we will give an overview of the current state of automated cytoarchitecture analysis and provide an outlook on future developments in the field. We will discuss the roles, potentials, and challenges of supervised learning, self-supervised representation learning, and graph-based inference at whole-brain level in the context of cytoarchitecture analysis. Finally, we will comment on the potential impact of novel methods and technologies on the field, including zero-shot learning, data-driven cytoarchitectonic mapping, multi-modal latent space fusion, and exascale computing.

Speaker: Christian Schiffer (Forschungszentrum Jülich)

27 Reporting EEG Source Imaging clinical studies in the BigBrain space: an open challenge to compare results

While there are many studies using Electrophysiological Source Imaging (ESI) for clinical applications, there is a crisis in comparing results due to many factors. Some problems are related to experimental and clinical trial design or statistical methodology. These problems may be difficult to resolve in the short term. However, some problems, which are more feasible to solve, are related to the lack of standard datasets to compare methods. This problem is compounded by the diversity of ESI methods as well as the lack of an updated standardized atlas that allows interpretation of ESI against the background of other aiming modalities and neuroscience data.
As a contribution to facilitate solving the methodological problem mentioned we present two developments from our group:

• We announce an open-source dataset that can help future comparisons (https://osf.io/6jd7y/ ).
  This dataset contains clinical, cognitive, motor, and EEG data from a clinical trial of a pharmacological agent to improve cognitive function in patients with Parkinson’s disease [1]. We have shown that quantitative EEG (as a proxy for brain function) mediates the effect of the intervention on cognitive function.
• We show how the statistical results of this mediation study (and for that matter any other studies) can be easily morphed into the BigBrain space. We provide a simple workflow to carry out this step.

We offer the challenge to reanalyze this dataset (with any ESI methods of choice) and to produce standardized results. Data and programs will be shared via the CONP.

Reference:
[1] Bringas Vega, M. L., Pedroso, Ibáñez, I., Razzaq, F. A., Zhang, M., Morales Chacón, L., Ren, P., Galan Garcia, L., Gan, P., Virues Alba, T., Lopez Naranjo, C., Jahanshahi, M., Bosch-Bayard, J. F., & Valdes‐Sosa, P. (2022). The Effect of Neuroepo on Cognition in Parkinson ’ s Disease Patients Is Mediated by Electroencephalogram Source Activity. Frontiers in Neuroscience, 16(June), 1–11. https://doi.org/10.3389/fnins.2022.841428

Speaker: Prof. Maria L. Bringas Vega (Joint China Cuba Lab for Neuroscience and Neurotechnology, University of Electronic Science and Technology of China, Chengdu China; 2. International Center for Neurological Restoration (CIREN from Spanish) Havana, Cuba)

15:30 Coffee and discussion break

28 The Virtual Big Brain in clinical applications

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Over the past decade we have demonstrated that the fusion of subject-specific structural information of the human brain with mathematical dynamic models allows building biologically realistic brain network models, which have a predictive value, beyond the explanatory power of each approach independently. The network nodes hold neural population models, which are derived using mean field techniques from statistical physics expressing ensemble activity via collective variables. Our hybrid approach fuses data-driven with forward-modeling-based techniques and has been successfully applied to explain healthy brain function and clinical translation including aging, stroke and epilepsy. Here we illustrate the workflow along the example of epilepsy: we reconstruct personalized connectivity matrices of human epileptic patients using Diffusion Tensor weighted Imaging (DTI). Subsets of brain regions generating seizures in patients with refractory partial epilepsy are referred to as the epileptogenic zone (EZ). During a seizure, paroxysmal activity is not restricted to the EZ, but may recruit other healthy brain regions and propagate activity through large brain networks. The identification of the EZ is crucial for the success of neurosurgery and presents one of the historically difficult questions in clinical neuroscience. The application of latest techniques in Bayesian inference and model inversion, in particular Hamiltonian Monte Carlo, allows the estimation of the EZ, including estimates of confidence and diagnostics of performance of the inference. The example of epilepsy nicely underwrites the predictive value of personalized large-scale brain network models. The workflow of end-to-end modeling is an integral part of the European neuroinformatics platform EBRAINS and enables neuroscientists worldwide to build and estimate personalized virtual brains.

Originally trained in Theoretical Physics and Philosophy in the 1990s, Dr. Jirsa has made contributions to the understanding of how network structure constrains the emergence of functional dynamics using methods from nonlinear dynamic system theory and computational neuroscience. Dr. Jirsa has been awarded several international and national awards for his research including the Francois Erbsmann Prize in 2001, NASPSPA Early Career Distinguished Scholar Award in 2004, and Grand Prix de Recherche de Provence in 2018. He serves on various Editorial Boards and has published more than 150 scientific articles and book chapters, as well as co-edited several books including the Handbook of Brain Connectivity. Dr. Jirsa is one of the Lead Scientists in the Human Brain Project and The Virtual Brain.

Speaker: Dr Viktor Jirsa (Institut de Neurosciences des Systèmes, Aix-Marseille Université)

Poster Session

29 3D reconstruction of BigBrain2: Progress report on updated processing pipeline and application to existing annotations and cortical surfaces

The development of BigBrain2 is a continuation of the first BigBrain [1] that will contribute new insight on inter-subject cytoarchitectonic variability. Overall, BigBrain2 offers better quality staining, favorable to regional segmentation and registration, and contains fewer artefacts through sectioning and staining. In this presentation, we will report about the initial 3D reconstruction of BigBrain2 at 100µm, which is suitable already for the extraction of cortical surfaces and the representation of annotations of some cortical and subcortical regions.
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, and manual and semi-automatic corrections were performed to repair acquisition artifacts 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), from which a first 3D reconstruction was obtained at an effective section spacing of 100µm. Data provenance tracking of all repair operations provides a means for assessing the extents of the repaired artifacts and for eventual reproducibility at the 1µm in-plane resolution. The repaired sections were aligned to the post-mortem MRI of the fixed brain (Siemens Sonata, 1.5T, MPRAGE, 0.5mm) in an iterative process by 3D registration of the stacked images to the MRI, followed by 2D registration of the individual images to the sliced MRI, while gradually increasing the degree of 2D and 3D registration from rigid-body to affine to non-linear across 10 global iterations. These extra global iterations helped resolve the lower-frequency alignment errors causing jaggies. Alignment to the MRI enables to correct for tissue compression caused by cutting and mounting of sections, and tissue shrinkage. Ultimately, section-to-section non-linear 2D alignment (without MRI) was performed to resolve high-frequency alignment errors. Optical-balancing was applied by normalizing image intensities to the MRI data to correct for staining imbalances across the brain. The reconstructed 3D volume is obtained at 100µm in the MRI ex-vivo space, which is suitable for the extraction of cortical surfaces. Finally, computed transformations are saved and can be applied to regions annotated on the original sections.
Ongoing work includes the semi-automatic repairs of the remaining sections (80%) 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

Speaker: Hartmut Mohlberg

30 Cerebral chemoarchitecture shares organizational traits with brain structure and function

Chemoarchitecture, the heterogeneous distribution of neurotransmitter transporter and receptor molecules, is a relevant component of structure–function relationships in the human brain. Here, we studied the organization of the receptome, a measure of interareal chemoarchitectural similarity, derived from positron-emission tomography imaging studies of 19 different neurotransmitter transporters and receptors. Nonlinear dimensionality reduction revealed three main spatial gradients of cortical chemoarchitectural similarity – a centro-temporal gradient, an occipito-frontal gradient, and a temporo-occipital gradient. In subcortical nuclei, chemoarchitectural similarity distinguished functional communities and delineated a striato-thalamic axis. Overall, the cortical receptome shared key organizational traits with functional and structural brain anatomy, with node-level correspondence to functional, microstructural, and diffusion MRI-based measures decreasing along a primary-to-transmodal axis. Relative to primary and paralimbic regions, unimodal and heteromodal regions showed higher receptomic diversification, possibly supporting functional flexibility.

Speaker: Sofie Valk (INM-7)

31 High-resolution 3D mapping of the amygdala - piriform region in the BigBrain

The amygdala – piriform region plays a central role in olfaction (Buchanan et al., 2003; Gottfried, 2010). The allocortex (amygdala and piriform cortex) and related piriform periallocortex receive projections from the olfactory bulb (Sakamoto, 1999). Deep amygdala nuclei seem to be involved in emotional processing of olfactory stimuli (Anderson et al., 2003). The region is cytoarchitectonically heterogeneous and includes small areas with a complex geometry, which cannot be studied in detail using structural in vivo neuroimaging. Previously, based on 10 postmortem brains, we have introduced probabilistic maps of 10 areas and nuclei of the amygdala (Kedo et al., 2018). Here we build on this research, and generate high-resolution maps of four areas of the mesial piriform region and new subdivisions of the amygdala in the BigBrain template, to study the extent and topography of cortical and subcortical structures, as well as their neighborhood relationships.
Cytoarchitectonic mapping in serial histological sections and a Deep-Learning workflow were applied to 3D-reconstruct the structures in the BigBrain. Firstly, we identified and delineated 19 structures of the amygdala in 57 sections in the right hemisphere and in 59 sections in the left hemisphere (Kedo et al., 2022). Secondly, the piriform region was analyzed, resulting in reference delineations in 57 sections in the right hemisphere and 42 sections in the left hemisphere. All delineations were performed using the web-based annotation tool (MicroDraw) at 1-micron resolution in-plane, in each 3rd to 18th section. Convolutional Neural Networks (CNNs) were applied for image segmentation in the unmapped sections in-between (Schiffer et al., 2021), separately for each data set. The annotations of both regions (Fig. 1A-B) were non-linearly transformed to the sections of the 3D reconstructed BigBrain space at 20-micron isotropic resolution (Amunts et al., 2013). The amygdala – piriform region was visualized using the Neuroglancer (Figs. 1A-C).
We have identified allocortical areas PirTBd, PirTBv and periallocortical areas PirTit, PirTu. The PirTB areas are located ventromedially to the Claustrum (Figs. 1A,C). Their spatial relationship to the caudally adjacent amygdala is shown in Figs. 1C-D. PirTit and PirTu rostrally replace the amygdalopiriform transition area (APir) on the temporal brain surface. PirTBv lies dorsally to both APir and PirTit. The dorsal areas PirTBd and anterior amygdaloid area (AAA) show a bumpy surface. Both areas medially extend on the basal brain surface. The shape of PirTit is influenced by the macroanatomy of the temporal pole. Lateral nucleus (invisible) reveals recesses in places of contact with granular parts of the paralaminar nucleus.
These maps will be openly available on EBRAINS platform of the HBP and integrated with the BigBrain model (https://go.fzj.de/bigbrain/) to serve as a histological reference data.
Amunts el al. (2013), Science 340:1472-1475.
Anderson el al. (2003), Nat Neurosci 6:196-202.
Buchanan el al. (2003), Learn Mem 10:319-325.
Gottfried JA (2010), Nat Rev Neurosci 11:628-641.
Kedo et al. (2022), 6th BigBrain Workshop, 25.-27. Okt 2022 In. Zadar.
Kedo et al. (2018), Brain Struct Funct 223:1637-1666.
Sakamoto et al. (1999) In: Handbook of Chemical Neuroanatomy (Bloom et al., eds).
Schiffer et al. (2021), Neuroimage 240:118327.

Speaker: Olga Kedo (INM-1 Forschungszentrum Jülich)

BBWS2023_11_Fig.png

32 Multimodal gradients reveal cortical local-global organization

Functional specialization and integration of the human cortex are two basic principles in neuroscience. For decades, significant efforts have been made to explore the underlying mechanisms of these principles based on a broad range of neuroanatomical and neuroimaging features. However, the spatial patterning of cortical organization and the interrelationships between different regions are still unclear. Recent studies have derived low-dimensional, continuous representations of cortical organization, also referred to as gradients, using cortex-wide decompositions of functional, microstructural, and structural connectivity features. The current work characterizes functional, microstructural, and structural gradient features within the probabilistic atlas of cortical cytoarchitecture, and assessed the uniqueness and redundancy of gradient profiles across cytoarchitecturally-defined cortical areas.
In this work, we studied 7 Tesla (7T) T1-weighted Magnetic Resonance Imaging (MRI), resting-state functional MRI, myelin-sensitive quantitative T1, and diffusion MRI of 10 unrelated healthy adults. Specifically, our work (i) took advantage of high-resolution 7T MRI to construct vertex-wise structural, microstructural, and functional connectomes, (ii) captured gradients profiles of brain regions through the integration of multimodal MRI and probabilistic atlas of cortical cytoarchitecture, (iii) assessed inter-parcel heterogeneity and homogeneity of multimodal features to quantify the uniqueness and redundancy of gradient fingerprints, and (iv) assessed intra-parcel heterogeneity between cortical vertices. To verify these findings, we repeated the main analyses on an independent dataset with 50 healthy adults.
Vertex-wise gradients of multimodal MRI data showed different spatial patterns across the cortex, indicating diverse hierarchies between vary modalities. By estimating cosine similarity between cortical regions, we found higher homogeneity in paralimbic regions and lower homogeneity in idiotypic i.e., sensory and motor cortices. To examine inter-parcel heterogeneity, we computed cosine distance between parcel-wise gradient profiles. We observed the highest heterogeneity in primary sensorimotor cortices, and lowest heterogeneity in paralimbic network (p<0.05, FDR corrected). Finally, we found that for most cortical parcels, vertices within a parcel are homogenous and show similar patterns among multimodal gradients. However, higher intra-parcel heterogeneity was found in heteromodal system (p<0.05, FDR corrected), while lower heterogeneity was found in idiotypic and paralimbic systems (p<0.05, FDR corrected). We repeated the main analysis on the validation dataset and found similar results.
Our findings point to a sensory-paralimbic differentiation of cortex-wide gradient fingerprints, where sensory/motor regions being more heterogenous compared to less distinctive paralimbic cortices. By reconciling local and global cortical features, our work may provide new insights into the neuroanatomical basis of specialized and integrative cortical functions.

Speaker: Yezhou Wang (McGill university)

BBWS2023_17_Fig.jpg

33 neuromaps: a Python toolbox for the structural and functional interpretation of brain maps

Imaging technologies are increasingly used to generate high-resolution reference maps of brain structure and function. Comparing experimentally generated maps to these reference maps facilitates cross-disciplinary scientific discovery. Although recent data sharing initiatives increase the accessibility of brain maps, data are often shared in disparate coordinate systems, precluding systematic and accurate comparisons. Furthermore, no data sharing platforms integrate standardized analytic workflows. Here we introduce neuromaps, an open-access Python toolbox for accessing, transforming and analyzing structural and functional human brain maps.

Our toolbox implements functionalities for generating high-quality group-level transformations between four standard coordinate systems that are widely used in neuroimaging (MNI152, fsaverage, fsLR, CIVET), and integrates them via a set of accessible, uniform interfaces. We also curated more than 40 reference brain maps from the literature that have been published during the past decade to facilitate contextualization of brain maps with respect to the biological ontologies of the human brain, including molecular, microstructural, electrophysiological, developmental and functional ontologies. Finally, we implement spatial autocorrelation-preserving null models for statistical comparison between brain maps that will help researchers to perform standardized, reproducible analyses of brain maps.

Collectively, neuromaps represents a step towards creating systematized knowledge and rapid algorithmic decoding of the multimodal multiscale architecture of the brain.

Speaker: Vincent Bazinet (Montreal Neurological Institute)

34 On Uncertainty-aware Deep Learning for Cytoarchitecture Classification

High-resolution light-microscopic scans of histological brain sections allow identifying cytoarchitectonic areas. They are defined by the local characteristics of microstructural organization, which encompasses the size, type, shape, and distribution of neurons, as well as their distinct laminar and columnar organization. As established brain mapping methods relying on statistical image analysis are infeasible to handle the large size of high-resolution datasets acquired by high-throughput microscopic scanners, recent research focused on the development of automated cytoarchitecture classification methods based on deep learning. While the performance of these deep learning methods has steadily increased over the last years, they are unable to provide reliable estimates of prediction uncertainty. In particular, the softmax outputs of classification networks are generally not well suited to estimate a model's uncertainty. The lack of well-calibrated uncertainty estimates makes the interpretation of predictions challenging, in particular when dealing with out-of-distribution data.

To this end, we here studied the behavior of a state-of-the-art deep neural network for cytoarchitecture classification with respect to its uncertainty awareness. We compared it to two methods for uncertainty quantification: Dropout variational inference (DVI), which quantifies uncertainty based on the variance of multiple predictions acquired with inference-time dropout, and evidential deep learning (EDL), which is explicitly trained to output an informative uncertainty score. We apply both methods to in-distribution test data and out-of-distribution data from a brain not seen during training. We compare the models based on calibration metrics, uncertainty scores, and prediction entropy.

Our experiments revealed that the baseline model is generally overconfident, an often reported behavior of neural networks that manifests as high-prediction probability even for incorrectly classified samples. We observe similar behavior for out-of-distribution samples from a brain not included during training, where the model was unable to express its inability to make accurate predictions. In comparison to the baseline, both DVI and EDL resulted in considerably more plausible uncertainty measures. For example, we observed that the uncertainty scores obtained from models trained with EDL indicate high certainty in regions with highly distinct cytoarchitectonic properties, including the primary visual and motor cortex. While EDL outputs a single normalized uncertainty score per sample, DVI provides class-level uncertainty estimates based on per-class variance. This allows us to obtain localized uncertainty measures for specific brain regions. For example, we observed a low-certainty ribbon for the primary visual cortex at the transition between primary and secondary visual cortex, indicating cytoarchitectonic ambiguities at the boundary between the two regions.
These ambiguities could be linked to the complex border phenomena that are characteristic of this region, the so-called border tuft and fringe area.

Our study revealed that predictions of existing models for cytoarchitecture classification are not well calibrated and lack the ability to express uncertainty. The investigated methods address these issues, providing complementary methods to assess uncertainty and improve model calibration. Future research will focus on the refinement of the training strategy and the involved hyperparameters. Finally, we plan to exploit the obtained uncertainty measures to identify high-certainty predictions for self-training approaches, which we expect to improve classification performance.

Speaker: Christian Schiffer (Forschungszentrum Jülich)

BBWS2023_01_Fig.png

35 Personalized Human Brain Energy Atlas: Machine learning-based neuropil density from anatomical MRI

Neuropil represents the fundamental unit in which cellular signals are processed, and it consists of neuronal/glial cells whose finest filaments form synapses. Thus, neuropil density (NP) combines both cellular (i.e., neuronal/glial cells) and synaptic masses and is central to understanding the heterogeneity of brain metabolism. Current bottom-up energy atlases use NP-derived neuronal (NeuDen) and synaptic (SynDen) density maps. However, a great innovation would be to predict metabolism from personalized neuroanatomy.

We hypothesize that in vivo NeuDen and SynDen can be derived from longitudinal relaxation time weighted MRI (T1w) for gray/white matter distinction and diffusion MRI for tissue cellularity (apparent diffusion coefficient, ADC) and axon directionality (fractional anisotropy, FA). We present a proof-of-concept machine learning algorithm that successfully predicts NeuDen and SynDen from routine in vivo MRI scans, where ex vivo Merker stain (BigBrain) and in vivo SV2A-PET imaging were respective gold standards. Our machine learning algorithm used gaussian-smoothed T1w/ADC/FA on a voxel-by-voxel basis to predict NeuDen/SynDen maps. We trained and compared NP predictions to NP gold standards and used histogram/spatial correlations as a proxy for prediction efficacy.

Notably, training on group average MRI data, especially with high gaussian smoothing, was ineffective for NP predictions. We used different levels of isotropic gaussian smoothing to provide our neural network nondirectional neighborhood information. Providing only low levels of gaussian smoothed datasets resulted in grainy neuropil predictions, showing the importance of neighborhood information from highly smoothed input datasets. Neighborhood information also helps to mitigate SNR issues (by averaging out the noise) and minor misregistration errors. Additionally, smoothed datasets can also provide a regional baseline to calculate relative changes in T1w, ADC and FA by the neural network as needed for neuropil prediction. High subject-by-subject histogram correlations for SynDen (0.94) and NeuDen (0.88) demonstrated realistic estimates across all subjects, while lower spatial correlations illustrated individualized predictions for SynDen (0.85) and NeuDen (0.45).

In summary, NP represents the microscopic infrastructure that regulates function which can be measured at the mesoscopic level (i.e., millimeter sized voxels in the human brain) with various PET and MRI methods. Since decreases in NP are associated with disorders such as depression, Parkinson’s, Alzheimer’s and aging, this work paves the way for individualized mesoscopic energy atlas prediction, enabling microscopic interpretations of functional neuroimaging data across health and disease.

Speaker: Adil Akif (Yale University)

36 Towards an EBRAINS service for brain wave analysis: Cobrawap

The current variety of data from neuronal recordings paves the way to complementary approaches for understanding brain activity. At the same time, this poses the challenge to consistently compare data across experiments, species, and spatio-temporal scales, promoting the integration of multiple approaches from different neuroscience sub-domains. Also, experimental data are essential to benchmark and further develop theoretical models. These requirements can be fulfilled by defining a standardized, but still to some extent customizable, analysis workflow.
In the context of brain wave analysis, we developed Cobrawap (Collaborative Brain Wave Analysis Pipeline; Gutzen et al, 2022; https://cobrawap.readthedocs.io), a FAIR-compliant open-source software cooperatively developed as a HBP-EBRAINS UseCase. Written in Python3, it is structured as a collection of modular building blocks (that can be added, removed or replaced) arranged along sequential stages, implementing data processing steps and analysis methods, directed by a workflow manager. The final output of Cobrawap describes the cortical wave activity in a standardized manner through quantitative observables, also directly exploitable for model calibration and validation (Capone et al, 2023).
A general drawback of complex analysis tools is the need for some technical efforts for their initial installation and configuration, and the non-negligible demand for computational resources for their execution on local machines. Aiming at a wider community diffusion and user facilitation, recent efforts have been addressed to integrate Cobrawap as an EBRAINS component. In this regard, we succeeded in deploying Cobrawap on three FENIX-ICEI federated HPC sites (CSCS, JSC, CINECA). We made Cobrawap executable through ssh direct login and from the EBRAINS Collab through UNICORE; eventually, also the EBRAINS Workflows Dashboard will feature it. Data upload is ensured from local storages and from the HBP/EBRAINS Knowledge Graph, while single jobs are managed through SLURM.
A crucial step toward increased usability is represented by the development of dedicated CWL workflows, dynamically built at runtime through a “meta-approach” parsing configuration files delivered by users for their custom applications. Full back-compatibility with the native Snakemake workflow manager has been guaranteed. The goal is to minimize the time-to-result, letting the user focus on the scientific side without caring of the technology behind the scenes.
Among the latest Cobrawap scientific developments, a great effort has been dedicated to high-resolution recordings from brain imaging. Through suitable curation scripts, images are annotated and converted into standard formats (e.g. NEO). Then, we developed a recursive algorithm (“HOS”, Hierarchical Optimal Sampling) for optimizing the signal-to-noise ratio, so to dynamically tune the resolution across the field of view, decreasing it in the sub-regions where signal is less reliable (e.g. the boundaries).
Pushing further the portability and the user-friendliness of Cobrawap, we are going to deliver it as both a Docker image and a pip-installable Python package, regularly upgraded on both scientific and technological sides through a dedicated CI/CD pipeline.
This research is co-funded by the European Union's Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreements No. 785907 (HBP SGA2) and No. 945539 (HBP SGA3) and the European Commission - NextGeneration EU (EBRAINS-Italy MUR CUP B51E22000150006).

Speaker: Cosimo Lupo (INFN, Rome)

BBWS2023_20_Fig.png

37 BigBrain Analysis: Cellular-Level Precision at 1µm Resolution

Introduction
The BigBrain model [1] is a cornerstone for extracting quantitative measures of brain architecture at 20μm isotropic resolution. While this model has proven instrumental in extracting 3D histological features, there's a growing need for even higher spatial resolution to obtain measures at the level of individual cells. Building on previous work from 2022 [2], this project utilizes 2D 1μm sections to provide a more detailed characterization of cellular distributions in the human brain, and to further enhance the BigBrain model with accurate estimates of layer-wise cell densities across the entire cortex.

Methods
Expanding on our previous work [2], we investigated 78 additional areas of the Julich-Brain cytoarchitectonic atlas [3]. These patches were sampled by registering each section to BigBrain space, and then sampling cortical locations corresponding to a probability >60% of being the specific area. In each patch, cortical layer boundaries were annotated by experts and validated using a four-eye procedure. Automatic cell body detection state of the art Deep Learning model [4] was applied to all patches, enabling the extraction of laminar cell numbers and cell body sizes for all areas under investigation (Fig.1).

Results
The expanded dataset now encompasses 900 cortical patches, with a size of about 52GB, selected from well-defined cytoarchitectonic areas of the Julich-Brain Atlas. Each patch in the dataset includes the 1μm raw image, manual annotations of isocortical layers, and contours and spatial properties of the extracted cell body segmentations (Fig.2). This enhancement in resolution from the native 20μm BigBrain resolution to 1μm has unveiled significant differences in cell packing density across various laminae of the brain. A trend of decreasing cell densities from posterior to anteriorly located areas was observed across all lamina of the human cortex. This trend was especially pronounced in granular layers II and IV. Moreover, the new patches can be utilized to refine previously generated cortical laminae [5], which were based on a limited number of areas.

Conclusions
The shift from 20 to a 1µm resolution image data has enabled quantitative analysis of individual cell bodies. This approach gives precise cell counts from specific brain areas and integrates them with overall brain data, revealing both known and new brain architectural insights. The resulting dataset, rooted in the BigBrain framework, provides a well structured and accurate spatial representation. This dataset can potentially replace the century-old cell counts from von Economo and Koskinas [6]. Its strengths lie in its reproducibility, precise 3D anchoring in the BigBrain, and the availability of original images for each patch, allowing detailed verification down to individual cells.

[1] Amunts K, et al. (2013). BigBrain: An ultrahigh-resolution 3D human brain model. Science
[2] EBRAINS https://search.kg.ebrains.eu/instances/f06a2fd1-a9ca-42a3-b754-adaa025adb10
[3] Amunts K, et al. (2020). Julich-Brain: A 3D probabilistic atlas of the human brain’s cytoarchitecture. Science
[4] Upschulte E, et al. (2022). Contour Proposal Networks for Biomedical Instance Segmentation. Medical Image Analysis.
[5] Wagstyl K, et al. (2020). BigBrain 3D atlas of cortical layers. PLOS Biology
[6] von Economo C, Koskinas GN. (1925). Die Cytoarchitektonik der Hirnrinde des Erwachsenen Menschen. Springer

Speaker: Sebastian Bludau (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich)

BBWS2023_34_Fig1.png

BBWS2023_34_Fig2.png

38 BigBrain Image Alignment with Unet and structural similarity

This paper 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 the 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 optimize our network with the Gram matrix loss that measures the correlation difference between features. Then the fine alignment is learned in an unsupervised manner 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.

Speaker: Dr Paule Joanne Toussaint (McGill University)

BBWS2023_12_Fig1.gif

BBWS2023_12_Fig2.gif

39 From structure to function in the human

Background. The amygdala is a crucial structure for several aspects of cognitive and affective functioning (Pessoa, 2010). These functions are supported by complex connectivity patterns to other brain regions, stemming from distinct subnuclei with unique microstructural properties (Kedo et al., 2017). However, current investigations of structure-function coupling in this region are limited by a lack of datasets and tools for individualized and observer-independent delineation of amygdala subregions. Given the strong inter-individual variability of this region, more personalized approaches are needed to reliably study the microstructural determinants of amygdala functions and connectivity. Our overall goal is to study how the amygdala's complex microarchitecture gives rise to its function and dynamic connectivity to other brain regions. This goal is addressed in two main steps, harnessing postmortem histological data as well as ultra high-resolution myelin-sensitive and resting-state functional imaging data.
Step 1: Data-driven mapping of amygdala microstructure. Capitalizing on a combination of advanced image processing (radiomics; (van Griethuysen et al., 2017)) and dimensionality reduction methods (UMAP, (McInnes et al., 2018)), we designed a novel method to capture variations in regional microstructure within the amygdala (Fig 1B-C). This approach was applied to the BigBrain dataset (Amunts et al., 2013), a unique resource offering unprecedented spatial resolution to study the neuronal organization of the human brain (100m3 voxels) (Fig 1A). We identified two dimensions (U1 and U2) capturing large spatial variations in amygdala cytoarchitecture, which were consistent with gold standard labels built from detailed visual inspections of series of postmortem specimens (Amunts, Mohlberg, Bludau, & Zilles, 2020; Amunts, Mohlberg, Bludau, Caspers, et al., 2020, Fig 2A-B) . We also assess the generalizability of these results to in vivo data, specifically leveraging myelin-sensitive contrasts (quantitative T1 imaging) collected at a field strength of 7 Tesla (7T; 500m3 voxels) (Fig 3A-C). Subsequent analysis correlating postmortem and in vivo findings indicate promising replicability (n=10 healthy participants), specifically, the axis which could best describe spatial variation in histology (U1) was found to have very similar correlations to the coordinate axes of all 10 subjects (Fig 3D). This highlights the potential of this approach for personalized investigations of regional microstructure.
Step 2: Correspondence of amygdala microstructure and functional network organization. We then leverage data-driven and subject-specific representations of amygdala microstructure computed in aim 1 to map large-scale changes in this region’s functional connectivity to other brain regions. Specifically, by isolating then contrasting the highest and lowest 25% of U1 values for each participant in the amygdala mask brought to functional space (Fig 4A-C). Our 7T scanning protocol includes repeated resting-state functional acquisitions. Notably, all sequences were designed to enhance signal-to-noise (e.g., via the use of multi-echo acquisitions), an important consideration when studying mesio-temporal structures.
Conclusion. We propose a theoretically-grounded approach for data-driven explorations of subcortical cytoarchitecture applied to the human amygdala. This approach generalized to microstructurally-sensitive in vivo MRI data and could delineate distinct functional network embeddings. The present work thus offers a first step towards an integrated account of the amygdala’s microstructural composition and functional organization.

Speaker: Hans Auer

BBWS2023_16_Fig.pdf

40 High-Resolution 3D Mapping of the Human Hypothalamus: Towards a Comprehensive Cytoarchitectonic Atlas

The hypothalamus is a brain structure that plays a central role in maintaining homeostasis and regulating various physiological and behavioural processes. It encompasses distinct nuclei with diverse microstructure, connectivity, molecular structure and functions, including circadian rhythm regulation, sleep-wake cycles, appetite control, stress response, and thermoregulation. Dysfunctions of the hypothalamus have been reported in the context of cluster headaches, obesity, addictive behaviour, aggressive disorders, anxiety disorders, sleep disorders, eating disorders, hypertension, and epilepsy. Despite its importance, the structural organization and precise boundaries of the hypothalamus, as well as the functional differentiation of its nuclei, are still not fully understood. Currently, there are no maps available, which inform neuroimaging studies about the microstructural segregation of the hypothalamus in 3D space. Existing maps of the hypothalamus lack the necessary spatial resolution and morphological detail to provide a comprehensive understanding of this complex region. Therefore, in our project we aim to develop a high-resolution 3D map of the human hypothalamus in order to determine its microstructure and localization in the stereotaxic space.
To create a high-resolution 3D reconstruction of the hypothalamus in the BigBrain models, its nuclei were delineated on a subsample of the high-resolution digitized histological sections of the BigBrain datasets (Amunts et al., 2013). To make delineations on every remaining section a deep-learning based brain mapping tool was applied (Schiffer et al., 2021). The model was trained on manual expert annotations on every 15th section in the BigBrain 1 to predict the delineations of the hypothalamus on every remaining sections. The automatically generated maps was re-evaluated to exclude incorrectly delineated sections. The delineations will then be converted into a 3D reconstructed BigBrain space with the use of non-linear registration of the high-resolution digital section (Amunts et al., 2020). Therefore, the resulting map visualized the complex shape of the hypothalamus and its 19 nuclei with high anatomical details.
The hypothalamus is divided into different zones, both in the mediolateral and rostrocaudal directions. Mediolaterally, there are three zones: the periventricular zone next to the third ventricle, the medial zone, and the lateral zone mainly occupied by the lateral hypothalamic area. Rostrocaudally, there are four zones: the preoptic zone bordering the lamina terminalis, the anterior zone, the tuberal zone above the infundibulum, and the mammillary zone.
In the preoptic zone, we identified several nuclei, including the periventricular hypothalamic nucleus, medial preoptic nucleus, median preoptic nucleus, uncinate nucleus, and intermediate hypothalamic nucleus. In the anterior hypothalamic area we distinguished the suprachiasmatic nucleus, paraventricular hypothalamic nucleus, anterior part of the periventricular nucleus, and supraoptic nucleus. The tuberal hypothalamic region contains the ventromedial hypothalamic nucleus, dorsomedial hypothalamic nucleus, arcuate nucleus, and tuberal part of the periventricular nucleus. Lastly, in the mammillary region we identified the medial and lateral mammillary nuclei, supramammillary nucleus, tuberomammillary nucleus, and lateral tuberal nucleus.
These histology-based maps in 3D provide detailed anatomical information of a complex region of the hypothalamus and serve as a spatial and structural reference for diagnostic, prognostic and therapeutic neuroimaging studies of the healthy human brain and those of patients.

Speaker: Mr Alexey Chervonnyy (Cécile und Oskar Vogt-Institut für Hirnforschung)

41 Intracortical microstructural asymmetry in the human cortex: laminar differentiation, organization, heritability, and relevance to language function and psychopathological traits

The cerebral cortex shows subtle left-right morphological asymmetry supporting hemispheric specialization of functional processes including attention and language, which is also associated with neuropsychiatric conditions. However, previous studies consider gray matter as a morphological feature rather than a laminar structure to study asymmetry. Here, we leveraged intensity profiles from ultra-high resolution post-mortem histological data and in vivo magnetic resonance imaging to describe the layer-related intracortical microstructural asymmetry in the human cortex. We observed the left-right asymmetry wave of intracortical microstructure along the layers in post-mortem histological data. Extending our model to in vivo MRI, we observed that default mode, ventral multimodal, and somatomotor networks transfer the asymmetry directions along the intracortical depth. This pattern was observed to be more heritable in middle-depth surfaces, indicating cellular lateralization may share more genetic information to guide brain functions than molecular lateralization. Furthermore, in terms of the system-level laminar organization, inverted U-shape lateralization was observed along the sensory-fugal axis, which was rarely shaped by genetic factors. Last, using supervised machine learning, we observed microstructural asymmetry features in cingulate and prefrontal cortices are more often selected to predict language functions and psychopathological traits. In sum, using a multilevel model, we find laminar and organization differentiation along the sensory-fugal axis in the microstructural asymmetry in the human cortex and its potential application to psychiatric conditions.

Speaker: Bin Wan (INM-7, FZJ)

42 System-level cortical maturation links to adolescent resilience to adverse life events

Introduction: Adolescence is a period of ongoing brain reorganization that is essential to biological and psychosocial maturation, but also to mental health (Paus et al., 2008). Adolescent brain maturation as captured via neuroimaging follows two main modes: 1) conservative strengthening of initially strong inter-regional similarities, or 2) disruptive remodeling, i.e. strengthening of initially weak inter-regional similarities and vice versa (Váša et al., 2020). While adverse experiences and psychopathological processes can alter maturational trajectories (Stenson et al., 2021), adolescent reorganization may also hold potential for flexible adaptation to risk factors. Thus, normative maturation facilitating psychosocial skills may also aid well-being through resilience to adversity.

Methods: We analyzed age-related changes in myelin-sensitive Magnetic Transfer (MT) and microstructural profile covariance (MPC) in a longitudinal cohort of individuals aged 14-26 (n=295; 512 scans; 50.8% female). MPC reflects inter-regional similarity of intracortical MT profiles sampled at ten cortical depths. We first identified maturational modes by correlating the whole-brain MPC pattern of each region at age 14 with the age-related changes of this pattern (14-26y; computed via edge-wise linear mixed effect models). Positive correlations indicate conservative and negative correlations disruptive development. To disentangle this pattern further, we applied non-linear dimensionality reduction to MPC age effects and identified an underlying axis along which regions vary in similarity of their MPC maturational profile (Paquola et al., 2019). Next, we investigated whether adolescent resilience to adverse life events is associated with altered maturational patterns. We predicted general distress from several adversity measures and - conceptualizing resilience as better than expected wellbeing given the adversity faced - extracted residuals as resilience scores. We then contrasted the maturational index and the principal axis of MPC change in adaptive versus susceptible individuals.

Results: We observed conservative MPC development in dorsal/sensorimotor and ventral/temporal cortex, whereas disruptive MPC development was characteristic of a frontoparietal heteromodal midline. The MPC maturational index showed a U-shaped relationship with the MPC axis of age-related change, indicating re-organization of the heteromodal midline towards ventral and dorsal ‘anchors’. Resilient individuals showed more conservative / less disruptive development in confined regions in the frontal and parietal cortex, indicating that age-related change in MPC was more closely coupled to patterns already set up during early adolescence. Moreover, resilience was related to an expansion and bimodal distribution of loadings along the identified underlying axis, reflecting a clearer MPC differentiation of the heteromodal fronto-parietal midline with age.

Conclusion: In sum, our analyses underline a benefit of stability in microstructural maturation through differentiation that follows paths already set up during early adolescence more closely. Inter-individual differences in resilience were associated with altered microstructural development in association and paralimbic cortex. These regions are known to show protracted plasticity linked to both socio-cognitive refinement and psychopathological alterations (Sydnor et al., 2021) and may point to an inherent link between cognitive resources and resilience capacities. However, future studies shall further investigate to which degree adaptive brain development may be a resilience factor in itself, or a consequence of protective environmental factors.

Speaker: Sofie Valk (Research Center Jülich)

BBWS2023_06_Fig.png

43 Whole-brain dynamical modeling of the adolescent developing brain

Regulation of cortical microcircuits is crucial for optimal neural processing. Adolescence involves substantial macro- and microscale changes in the brain, including maturation of cortical microcircuits. Evidence from animal studies suggests a calibration of cortical microcircuits and excitation-to-inhibition (E-I) ratio during adolescence. However, in-vivo measurement of cortical microcircuits in the human developing brain is challenging, and therefore the supporting in-vivo evidence on maturation of E-I ratio in humans is limited. Whole-brain dynamical modeling is a promising approach that enables mechanistic inferences about hidden brain features, such as estimated properties of cortical microcircuits and E-I ratio. Here, we used whole-brain dynamical modeling to study age-related changes of whole-brain model parameters during adolescence.

We simulated cortical activity based on a mean-field model of excitatory and inhibitory neuronal ensembles in regions connected based on subject-specific or group-averaged structural connectomes. The fit of simulations to empirical resting-state functional images of each subject was evaluated based on comparison of simulated and empirical functional connectivity as well as functional connectivity dynamics matrices. We identified optimal model parameters for each subject using covariance matrix adaptation evolution strategy as well as GPU-accelerated grid search of the whole parameter space. Based on the simulations performed with the optimal parameters, we calculated the regional E-I ratios in the simulation as their time-averaged simulated excitatory firing rates. We observed region-specific changes of E-I ratio with age, which was decreased in parietal and frontal regions and increased in occipital regions. In addition, we observed association of grey-white matter contrast with E-I ratio in specifc regions. Following, we aim to increase regional specificity of the simulations by introducing heterogeneity in the model parameters based on biological maps of receptors as well as myelo- and cytoarchitecture.

Overall, we present a whole-brain modeling approach to estimate E-I ratio in developing adolescents which revealed region-specific changes of E-I ratio with age and its links to cortical microstructure.

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)

BBWS2023_19_Fig.png

Friday 6 October

44 Sievers Lecture in Computational Neuroscience

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

ENIGMA, AI & the Human Brain: Worldwide Neuroimaging and Genetics of 30 Brain Diseases in 100,000 Individuals from 45 Countries

Since 2009, the ENIGMA Consortium has published the largest worldwide neuroimaging studies of over 15 brain diseases and conditions, including Parkinson’s disease, epilepsy, ataxia and brain injury, PTSD, substance use disorder, bipolar disorder, and major depression, and neurodevelopmental conditions including OCD, ADHD and ASD. With over 2,000 participating scientists, ENIGMA has cooperatively analyzed data from 45 countries, and leads worldwide imaging genetics studies that discovered over 500 genomic loci that affect brain structure, disease risk, and brain synchrony and connectivity (Grasby et al., Science, 2020). In this lecture, we review ENIGMA’s major findings to date, including maps of disease effects on brain MRI, DTI, and functional MRI, as well as factors that influence them. A decade of genomic screens of worldwide multimodal brain images has discovered over 500 common and rare genomic variants influencing brain connectivity, brain function mapped using EEG, and rates of tissue loss in development and old age. We review the major factors influencing brain development and disease worldwide, highlighting novel work in populations of diverse ancestry and using geocoding to capture environmental drivers of disease. We also highlight new directions in AI for automatic diagnosis, disease subtyping, and prognosis based on worldwide brain data, and new efforts in ENIGMA that inform neuromodulation and interventional studies.

Dr. Paul M. Thompson is a Professor of Neurology, Psychiatry, Radiology, Pediatrics, and Engineering, at the University of Southern California (USC) where he directs the Imaging Genetics Center, and is Associate Director for the Stevens Neuroimaging & Informatics Institute. Professor Thompson is also Director of the ENIGMA Center for Worldwide Medicine, Imaging & Genomics, and is Principal Investigator and Co-founder of the ENIGMA Consortium. ENIGMA has cooperatively analyzed data from over 45 countries to publish the largest worldwide neuroimaging studies of over 15 brain diseases and conditions, including Parkinson’s disease, epilepsy, ataxia and brain injury, PTSD, substance use disorder, bipolar disorder, and major depression, and neurodevelopmental conditions including OCD, ADHD and ASD. In parallel, the ENIGMA Consortium has led worldwide imaging genetics studies that discovered over 500 common and rare genomic variants that affect brain structure, disease risk, and brain connectivity (Grasby et al., Science, 2020).

Speaker: Prof. Paul M. Thompson (University of Southern California)

LOC Session: The Local Landscape of Neuroscience in Iceland

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

45 deCODE genetics contributions in imaging genetics

deCODE genetics, a global leader in human genetics, has an extensive collection of DNA, multi-omic, imaging, and phenotypic data. Since 2003, it has identified numerous genetic variants linked to diverse diseases and traits. Noteworthy contributions include unraveling inherited risks for Alzheimer's disease, schizophrenia, some common cancer types, cardiovascular diseases, and traits and phenotypes such as hair and eye color and cognition. This presentation spotlights deCODE's achievements in imaging genetics, focusing on breakthroughs like decoding the genetic basis of intracranial volume, effects of sequence variants on neurodevelopment, brain structure and function, and cognition and brain age.

Speaker: Prof. Magnús Örn Úlfarsson (University of Iceland; deCODE Genetics, Iceland.)

46 Neuronal activity modulates myelin plasticity and regeneration

The CNS is responsive to an ever-changing environment. Until recently, studies of neural plasticity focused almost exclusively on functional and structural changes of neuronal synapses. In recent years, myelin plasticity has emerged as a potential modulator of neural networks. Myelination of previously unmyelinated axons, and changes in the structure on already-myelinated axons, can have large effects on network function. The heterogeneity of the extent of how axons in the CNS are myelinated offers diverse scope for dynamic myelin changes to fine-tune neural circuits. The traditionally held view of myelin as a passive insulator of axons is now changing to one of lifelong changes in myelin, modulated by neuronal activity and experience.

Myelin, produced by oligodendrocytes (OLs), is essential for normal brain function, as it provides fast signal transmission, promotes synchronization of neuronal signals and helps to maintain neuronal function. OLs differentiate from oligodendrocyte precursor cells (OPCs), which are distributed throughout the adult brain, and myelination continues into late adulthood. OPCs can sense neuronal activity as they receive synaptic inputs from neurons and express voltage-gated ion channels and neurotransmitter receptors, and differentiate into myelinating OLs in response to changes in neuronal activity.

This lecture will review myelin plasticity in adult animal, whether myelin changes occur in non-motor learning tasks, and questions whether myelin plasticity and myelin regeneration are two sides of the same coin.

Speaker: Prof. Ragnhildur Thora Káradóttir (Faculty of Medicine, University of Iceland; Cellular Neuroscience at the University of Cambridge, UK)

47 Identifying Brain Imaging Biomarkers for Early Diagnosis of Rare Dementia through MRI Post-Processing

Diagnosing dementia in its early stages is a significant challenge due to the phenotypic overlap of various types of dementia. Some diseases can go misdiagnosed for years before the correct diagnosis is reached. On the other hand, certain causes of dementia present with distinct structural changes in the brain; however, these changes can be extremely subtle in the early stages, making them hard to detect through visual inspection. This presentation introduces deep learning-based image processing strategies for accurate segmentation and labeling of anatomical structures linked to rare forms of dementia. Our ultimate goal is to facilitate fast and automated computation of novel imaging biomarkers that have the potential to help characterize the structural changes in the brain at earlier disease stages than what is currently possible.

Speaker: Prof. Lotta María Ellingsen (Faculty of Electrical and Computer Engineering, University of Iceland)

10:15 Coffee and discussion break

Demo Session

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Chair: Arthur W. Toga, Lyuba Zehl

48 MicroDraw multilayer: a web application to annotate multi-resolution multimodal data

Brain histology provides unique information on the cellular structure of the brain. However, high resolution datasets are challenging to visualise, analyse, register and segment. MicroDraw (https://microdraw.pasteur.fr) is our online tool for visualising and collaboratively annotating high resolution histology data such as the BigBrain. MicroDraw allows users to segment regions and annotate them based on common ontologies in a collaborative way. However, the ability to visualise different data modalities concurrently, or to superpose data and atlases was missing thus far.

We will demonstrate a new extension of MicroDraw which allows us to superpose different data layers, enabling users to bring together several high resolution data modalities into one smooth view, such as cell body stained slices with the underlying myeloarchitecture or receptor architecture, or a combination of histological data with registered anatomical or diffusion MRI. The visibility of each layer, including data and labels, is configured with a simple graphical interface. This combined view should facilitate high resolution anatomical labelling informed by multi-modal tissue sources, and help verify the registration across modalities. The combination of multiple modalities and the possibility to superpose atlases should significantly facilitate the manual segmentation of structures using MicroDraw’s vectorial annotation tools.

Speakers: Katja Heuer (Unit of applied and theoretical neuroanatomy, Institut Pasteur, Paris, France), Nicolas Traut (Unit of applied and theoretical neuroanatomy, Institut Pasteur, Paris, France), Roberto Toro (Unit of applied and theoretical neuroanatomy, Institut Pasteur, Paris, France)

49 Towards an EBRAINS service for brain wave analysis: Cobrawap

The current variety of data from neuronal recordings paves the way to complementary approaches for understanding brain activity. At the same time, this poses the challenge to consistently compare data across experiments, species, and spatio-temporal scales, promoting the integration of multiple approaches from different neuroscience sub-domains. Also, experimental data are essential to benchmark and further develop theoretical models. These requirements can be fulfilled by defining a standardized, but still to some extent customizable, analysis workflow.
In the context of brain wave analysis, we developed Cobrawap (Collaborative Brain Wave Analysis Pipeline; Gutzen et al, 2022; https://cobrawap.readthedocs.io), a FAIR-compliant open-source software cooperatively developed as a HBP-EBRAINS UseCase. Written in Python3, it is structured as a collection of modular building blocks (that can be added, removed or replaced) arranged along sequential stages, implementing data processing steps and analysis methods, directed by a workflow manager. The final output of Cobrawap describes the cortical wave activity in a standardized manner through quantitative observables, also directly exploitable for model calibration and validation (Capone et al, 2023).
A general drawback of complex analysis tools is the need for some technical efforts for their initial installation and configuration, and the non-negligible demand for computational resources for their execution on local machines. Aiming at a wider community diffusion and user facilitation, recent efforts have been addressed to integrate Cobrawap as an EBRAINS component. In this regard, we succeeded in deploying Cobrawap on three FENIX-ICEI federated HPC sites (CSCS, JSC, CINECA). We made Cobrawap executable through ssh direct login and from the EBRAINS Collab through UNICORE; eventually, also the EBRAINS Workflows Dashboard will feature it. Data upload is ensured from local storages and from the HBP/EBRAINS Knowledge Graph, while single jobs are managed through SLURM.
A crucial step toward increased usability is represented by the development of dedicated CWL workflows, dynamically built at runtime through a “meta-approach” parsing configuration files delivered by users for their custom applications. Full back-compatibility with the native Snakemake workflow manager has been guaranteed. The goal is to minimize the time-to-result, letting the user focus on the scientific side without caring of the technology behind the scenes.
Among the latest Cobrawap scientific developments, a great effort has been dedicated to high-resolution recordings from brain imaging. Through suitable curation scripts, images are annotated and converted into standard formats (e.g. NEO). Then, we developed a recursive algorithm (“HOS”, Hierarchical Optimal Sampling) for optimizing the signal-to-noise ratio, so to dynamically tune the resolution across the field of view, decreasing it in the sub-regions where signal is less reliable (e.g. the boundaries).
Pushing further the portability and the user-friendliness of Cobrawap, we are going to deliver it as both a Docker image and a pip-installable Python package, regularly upgraded on both scientific and technological sides through a dedicated CI/CD pipeline.
This research is co-funded by the European Union's Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreements No. 785907 (HBP SGA2) and No. 945539 (HBP SGA3) and the European Commission - NextGeneration EU (EBRAINS-Italy MUR CUP B51E22000150006).

Speaker: Cosimo Lupo (INFN, Rome)

BBWS2023_20_Fig.png

50 neuromaps: a Python toolbox for the structural and functional interpretation of brain maps

Imaging technologies are increasingly used to generate high-resolution reference maps of brain structure and function. Comparing experimentally generated maps to these reference maps facilitates cross-disciplinary scientific discovery. Although recent data sharing initiatives increase the accessibility of brain maps, data are often shared in disparate coordinate systems, precluding systematic and accurate comparisons. Furthermore, no data sharing platforms integrate standardized analytic workflows. Here we introduce neuromaps, an open-access Python toolbox for accessing, transforming and analyzing structural and functional human brain maps.

Our toolbox implements functionalities for generating high-quality group-level transformations between four standard coordinate systems that are widely used in neuroimaging (MNI152, fsaverage, fsLR, CIVET), and integrates them via a set of accessible, uniform interfaces. We also curated more than 40 reference brain maps from the literature that have been published during the past decade to facilitate contextualization of brain maps with respect to the biological ontologies of the human brain, including molecular, microstructural, electrophysiological, developmental and functional ontologies. Finally, we implement spatial autocorrelation-preserving null models for statistical comparison between brain maps that will help researchers to perform standardized, reproducible analyses of brain maps.

Collectively, neuromaps represents a step towards creating systematized knowledge and rapid algorithmic decoding of the multimodal multiscale architecture of the brain.

Speaker: Vincent Bazinet (Montreal Neurological Institute)

BBWS2023_22_Fig.png

Contributed Talks 3

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Chair: Arthur W. Toga, Lyuba Zehl

51 The brainstem perspective on cortical functional connectivity

The brainstem is a crucial yet understudied structure of the human brain. Due to the technical difficulties of imaging deep brain structures, the cerebral cortex has long been the focus of in-vivo human imaging studies. Functional magnetic resonance imaging (fMRI) in particular has deepened our understanding of the cortex, including the presence of functionally specialized brain regions, intrinsic functional networks, and our understanding of higher-order cognitive processes. The brainstem’s role in what is now considered primarily cortical organization and function remains an open question. Here we employ a recently acquired 7T fMRI brainstem dataset, which underwent rigorous physiological noise correction and was delineated according to 58 brainstem nuclei, to study how functional activity in the brainstem aligns with cortical function. We identify a set of brainstem hubs that are maximally connected to the cortex, including the periaqueductal grey, the dorsal raphe, and the laterodorsal tegmental nucleus. Likewise, cortical regions that are maximally connected with the brainstem appear in anterior regions. We demonstrate that these brainstem and cortical hubs reflect both slow (fMRI) and fast (MEG) dynamics, including lag-1 temporal autocorrelation, signal variability, and MEG alpha power. Next, we cluster brainstem regions with respect to how they connect to the cortex and identify modules of brainstem nuclei that subserve familiar cortical functional activation patterns related to memory, social cognition, movement and sensation, and emotion. Using PET-derived cortical profiles for 19 neurotransmitter receptors and transporters we show that neuromodulatory systems are likely mediating the relationship between brainstem and cortical functional activity. Finally, we demonstrate that unimodal and transmodal cortical regions have distinct patterns of connectivity to the brainstem. Altogether, this study extends our perspective of cortical function---including dynamics, cognitive function, and the functional hierarchy---to the brainstem, demonstrating the importance of brainstem activity to cortical function.

Speaker: Justine Hansen (Montreal Neurological Institute)

BBWS2023_04_Fig.png

52 Intracortical microstructural asymmetry in the human cortex: laminar differentiation, organization, heritability, and relevance to language function and psychopathological traits

The cerebral cortex shows subtle left-right morphological asymmetry supporting hemispheric specialization of functional processes including attention and language, which is also associated with neuropsychiatric conditions. However, previous studies consider gray matter as a morphological feature rather than a laminar structure to study asymmetry. Here, we leveraged intensity profiles from ultra-high resolution post-mortem histological data and in vivo magnetic resonance imaging to describe the layer-related intracortical microstructural asymmetry in the human cortex. We observed the left-right asymmetry wave of intracortical microstructure along the layers in post-mortem histological data. Extending our model to in vivo MRI, we observed that default mode, ventral multimodal, and somatomotor networks transfer the asymmetry directions along the intracortical depth. This pattern was observed to be more heritable in middle-depth surfaces, indicating cellular lateralization may share more genetic information to guide brain functions than molecular lateralization. Furthermore, in terms of the system-level laminar organization, inverted U-shape lateralization was observed along the sensory-fugal axis, which was rarely shaped by genetic factors. Last, using supervised machine learning, we observed microstructural asymmetry features in cingulate and prefrontal cortices are more often selected to predict language functions and psychopathological traits. In sum, using a multilevel model, we find laminar and organization differentiation along the sensory-fugal axis in the microstructural asymmetry in the human cortex and its potential application to psychiatric conditions.

Speaker: Bin Wan (INM-7, FZJ)

53 Whole-brain dynamical modeling of the adolescent developing brain

Regulation of cortical microcircuits is crucial for optimal neural processing. Adolescence involves substantial macro- and microscale changes in the brain, including maturation of cortical microcircuits. Evidence from animal studies suggests a calibration of cortical microcircuits and excitation-to-inhibition (E-I) ratio during adolescence. However, in-vivo measurement of cortical microcircuits in the human developing brain is challenging, and therefore the supporting in-vivo evidence on maturation of E-I ratio in humans is limited. Whole-brain dynamical modeling is a promising approach that enables mechanistic inferences about hidden brain features, such as estimated properties of cortical microcircuits and E-I ratio. Here, we used whole-brain dynamical modeling to study age-related changes of whole-brain model parameters during adolescence.

We simulated cortical activity based on a mean-field model of excitatory and inhibitory neuronal ensembles in regions connected based on subject-specific or group-averaged structural connectomes. The fit of simulations to empirical resting-state functional images of each subject was evaluated based on comparison of simulated and empirical functional connectivity as well as functional connectivity dynamics matrices. We identified optimal model parameters for each subject using covariance matrix adaptation evolution strategy as well as GPU-accelerated grid search of the whole parameter space. Based on the simulations performed with the optimal parameters, we calculated the regional E-I ratios in the simulation as their time-averaged simulated excitatory firing rates. We observed region-specific changes of E-I ratio with age, which was decreased in parietal and frontal regions and increased in occipital regions. In addition, we observed association of grey-white matter contrast with E-I ratio in specifc regions. Following, we aim to increase regional specificity of the simulations by introducing heterogeneity in the model parameters based on biological maps of receptors as well as myelo- and cytoarchitecture.

Overall, we present a whole-brain modeling approach to estimate E-I ratio in developing adolescents which revealed region-specific changes of E-I ratio with age and its links to cortical microstructure.

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)

BBWS2023_19_Fig.png

54 Align with the NMIND Consortium for better brain imaging

Robust, validated, and widely accessible tools are necessary for consistent scientific advancement and discovery. In neuroimaging, scientific advancement has outpaced the availability of well-implemented, trustworthy tools, forcing researchers to generate in-house software. Unfortunately, such in-house software rarely follows best practices of software engineering, documentation, testing, or validation, relying on lab- or individual-specific approaches instead.

Compounding the challenges at hand, descriptions of implementation, evaluation and/or benchmarking for neuroimaging software are typically sparse, reducing transparency for users and creating a risk of inappropriate usage. While the scale, complexity, and analytic acumen required for neuroimaging studies have increased exponentially over the last decade, there has not been a corresponding emphasis on software infrastructure to support such advances, and this has perpetuated a crisis of reproducibility through unaccounted heterogeneity across tools and algorithms.

We present the NMIND Consortium, a collaborative focused on standardizing scientific software development best practices, facilitating tool evaluation, and minimizing redundancy in the field.

NMIND has adopted three guiding principles:

• Alignment, which refers to development and adoption of standards for critical components of data processing, analysis pipelines, and associated software. These include, but are not limited to: 1) coding & infrastructure standards, 2) testing & benchmarking standards, and 3) documentation standards.

• Testing, which refers to accessible mechanisms for evaluating the compliance of tools with the NMIND standards. Tool contributors will be able to interact with these mechanisms through Web-based public interfaces and programmatic toolkits.

• Engagement, which refers to the widespread promotion and adoption of the NMIND collaborative standards and testing in the field, through the efforts of field researchers, educators, and resource generators.

Towards the above-mentioned objectives, there are several ways in which members of the community can currently participate in NMIND.

For tool evaluation, we invite use of the NMIND Coding Standards Checklist (available at https://nmind.org). Integration of the checklist into software development and release workflows can assist with publication, and improve documentation, infrastructure and testing surrounding tools. Importantly, we welcome feedback, and expect these checklists will evolve as living standards within the community.

You can join us monthly for our regular one-day Hackathons, oriented around establishing a community and a common time to work on improving NMIND toolkits and to interact with other community efforts (e.g., BIDS, ReproNim, HBCD). Outcomes of group-based tool reviews, a regular focus of these events, are shared under the Proceedings section of the NMIND website. Software coding skills are not required for participation.

The past decade of the neuroimaging community has witnessed landmark advances in data collection, processing and analysis. Though, arguably, the greatest advance has been the emergence of an open science culture, with open data, tools and knowledge bases serving as incubators and accelerators for collaboration. Looking forward, the NMIND collaborative model will be essential for the field to take the next major step in its evolution towards a reproducible science capable of delivering critically needed scientific and clinical deliverables.

Please see our full paper in Nature Human Behavior: https://rdcu.be/dfGtn.

Speaker: Gregory Kiar (Child Mind Institute)

BBWS2023_13_Fig.png

55 SynthEco: A digital system for analyzing the brain-to-society mechanisms of lifecourse human behavior in its multi-layered and dynamic environment and its mental health and well-being consequences

Biological, social, cultural and economics factors are important determinants of health/disease status and well-being throughout life. We are presenting the open-source ecosystem “SynthEco” which utilizes statistically representative synthetic populations derived from census data for a given geospatial granularity to create a complex digital ecosystem to analyze the brain-to-society pathways of lifecourse human behavior within their environment, zooming in on dopamine gene and brain systems as key biological bridge. In this context, we are introducing a way of parameterising and re-parameterising individuals through a combination of genomic, health as well as behavioural and social data, using a geospatially referenced synthetic population as merging layer. Through the geospatial layer the individuals are also anchored into context such as surrounding infrastructure and social environment, allowing for the analysis of individuals’ multidimensional behaviour and health outcomes over time while considering the context they are acting within, such as through the use of agent-based models or simulation studies.

Taking Montreal (Quebec, Canada) as development and Pittsburgh (Pennsylvania, USA) as replication side, we enrich a synthetic population through linkage and statistical extrapolation with longitudinal geo-referenced discovery (MAVAN; birth cohort) and population (CLSA; Canadian Longitudinal Study of Aging) cohorts. We trace molecular pathways through which dopamine-related genes in specific brain regions related to decision-making, emotional regulation and behavior may act in concert to shape mal/adaptive responses to the environment (expression-based polygenic risk scores based on the DRD4 co-expression gene network in the striatum (ePRS-DRD4)). The SynthEco platform views individuals as agents nested within modular systems of systems, which includes those operating at multiple scales in their biology and their psychology, as well as systems of a different type that form their multi-layered environment (e.g., birth weight, early child/family environment, area-level socio-economic status, access to affordable school, housing, transportation, food, as well as access to local social and health services in the form of enterprises and institutions).

The longitudinal cohorts and cross-sectional studies in SynthEco cover different aspects of human behavior, food security and financial, physical and mental wellbeing over the life course which are assessed at different age point along a lifecourse continuum (childhood, adolescence, young adult, aging adult). The time-sensitive re/parameterization of agent’s behavior made possible by stitching together information from different data bases in SynthEco opens new possibilities for allowing the consideration of the complex, multi-faceted behavior of these agents as they evolve in space and time. These combined data sources allow for the geospatial clustering of different dimensions of health and wellbeing in individuals and the identification of vulnerable groups sharing similar risk profile characteristics and environmental conditions, which could inform the development of targeted public health interventions and community planning, as well as better targeted public policy and investment.
Implications of this work are of interest to researchers modeling complex human behavior and agent-based models, as well as governments, policymakers, and NGOs interested in simulating interventions, analyzing individuals within a population, and identifying sub-populations or regions of specific concern to them.

Speaker: Antonia Gieschen (University of Edinburgh)

56 Wrap up and closing

Zoom webinar:
https://fz-juelich-de.zoom.us/j/67716299814?pwd=Q3dBRXZQa3NLL1Y2TVYvUlBqV0lQQT09
Kenncode: 749254

Speakers: Prof. Alan C Evans (Montreal Neurological Institute McGill University Montreal), Prof. Katrin Amunts (Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich)

13:00 [Optional] Excursion - Golden Circle Tour