Related Experiment Video
Updated: Dec 26, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.3K
Automatic labeling of cortical sulci using patch- or CNN-based segmentation techniques combined with bottom-up
Léonie Borne1, Denis Rivière1, Martial Mancip2
1Université Paris-Saclay, CEA, CNRS, Neurospin, Baobab, Gif-sur-Yvette, 91191, France.
Medical Image Analysis
|March 13, 2020
Summary
This study introduces novel methods for automatically identifying human cortical sulci, overcoming challenges posed by their complex folding patterns. A convolutional neural network (CNN) approach significantly improves accuracy and speed in sulcus recognition.
Area of Science:
- Neuroscience
- Medical Image Analysis
- Computational Anatomy
Background:
- Human cortical sulci exhibit extreme folding variability, making automated and manual recognition challenging.
- Existing methods struggle with accurate and comprehensive sulcus identification, limiting large-scale machine learning dataset creation.
- The Morphologist toolbox, while useful, has limitations in recognizing the full complexity of cortical sulci.
Purpose of the Study:
- To enhance the accuracy and efficiency of human cortical sulcus recognition.
- To improve upon the existing Morphologist toolbox model for sulcus recognition.
- To develop robust methods that account for inter-subject variability in cortical folding patterns.
Main Methods:
- Implemented patch-based multi-atlas segmentation (MAS) techniques.
- Developed convolutional neural network (CNN)-based approaches, specifically a U-Net architecture.
- Integrated domain-specific geometric constraints and a top-down perspective for fold cleavage to refine segmentation.
Main Results:
- Both MAS and CNN approaches outperformed the current Morphologist model.
- The CNN U-Net-based approach demonstrated superior performance.
- The refined methods effectively incorporated geometric and topological properties for improved sulcus morphometry.
Conclusions:
- Novel MAS and CNN-based methods significantly advance cortical sulcus recognition.
- CNN U-Net models offer a highly efficient and accurate solution for automated sulcus identification.
- The integration of geometric constraints enhances the reliability of automated neuroanatomical segmentation.
