Convolutional neural networks for cytoarchitectonic brain mapping at large scale.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Histology

Background:

• Human brain atlases are crucial for understanding brain organization across individuals.
• Cytoarchitecture, the study of neuronal cell arrangement, is key to brain connectivity and function.
• Automated and observer-independent methods are needed for reliable cytoarchitectonic area identification.

Purpose of the Study:

• To develop an efficient workflow for mapping cytoarchitectonic areas in large series of human brain histological sections.
• To leverage deep learning for accurate and rapid annotation of brain sections.
• To create a user-friendly interface for accessing advanced brain mapping techniques.

Main Methods:

• A Deep Convolutional Neural Network (CNN) was trained on annotated histological section images.
• The CNN model learned to automatically annotate un-annotated sections between training pairs.
• The workflow processes large datasets (Terabytes) efficiently without requiring 3D reconstruction and is robust to artifacts.

Main Results:

• The CNN-based workflow accurately maps cytoarchitectonic areas in large series of human brain sections.
• The new method is significantly faster than previous observer-independent mapping workflows.
• The workflow was integrated into a web interface for accessibility to researchers without deep learning expertise.

Conclusions:

• Deep neural networks, specifically CNNs, offer a powerful new approach for cytoarchitectonic mapping.
• This automated workflow enhances the creation of high-resolution human brain area models.
• The developed tool facilitates reproducible and efficient analysis of large-scale histological data.