Related Experiment Video
Updated: Jul 15, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
8.9K
Learning joint surface reconstruction and segmentation, from brain images to cortical surface parcellation
Karthik Gopinath1, Christian Desrosiers1, Herve Lombaert1
1ETS Montreal, Canada.
Medical Image Analysis
|September 29, 2023
Summary
SegRecon is a novel deep learning method that reconstructs and segments brain cortical surfaces from MRI in one step. This integrated approach significantly speeds up analysis and improves accuracy compared to traditional multi-step methods.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Medical Image Analysis
Background:
- Cortical surface reconstruction and segmentation from MRI are crucial for brain analysis.
- Existing multi-step methods are computationally intensive and time-consuming.
Purpose of the Study:
- To introduce SegRecon, an integrated end-to-end deep learning method for joint cortical surface reconstruction and segmentation.
- To overcome the computational limitations of traditional multi-step approaches.
Main Methods:
- A volume-based neural network predicts signed distances to nested cortical surfaces and their atlas space representations.
- The method jointly reconstructs and segments the white-to-gray-matter interface and the pial surface.
- Experiments were conducted on the MindBoggle, ABIDE, and OASIS datasets.
Main Results:
- SegRecon achieved reconstruction errors below 0.52 mm and 0.97 mm (average Hausdorff distance) compared to FreeSurfer.
- Parcellation results showed over 4% improvement in average Dice score versus FreeSurfer.
- Computation time was reduced from hours to seconds.
Conclusions:
- SegRecon offers a significantly faster and more accurate solution for cortical surface reconstruction and segmentation.
- The integrated deep learning approach demonstrates potential for advancing large-scale neuroimaging studies.
