Related Experiment Video
Updated: Dec 12, 2025
A Standardized Pipeline for Examining Human Cerebellar Grey Matter Morphometry using Structural Magnetic Resonance Imaging
Published on: February 4, 2022
Direct cortical thickness estimation using deep learning-based anatomy segmentation and cortex parcellation
Michael Rebsamen1,2, Christian Rummel1, Mauricio Reyes3,4
1Support Center for Advanced Neuroimaging (SCAN), University Institute of Diagnostic and Interventional Neuroradiology, University of Bern, Inselspital, Bern University Hospital, Bern, Switzerland.
A new DL+DiReCT method improves cortical thickness measurement accuracy using deep learning and diffeomorphic registration-based cortical thickness (DiReCT). This approach enhances detection of neurological changes, outperforming existing techniques in correlation and sensitivity.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Biomarker Discovery
Background:
- Cortical thickness measurement is crucial for understanding neurodegenerative diseases.
- Existing methods like ANTs (using DiReCT) have limitations in accuracy and robustness.
- Surface-based methods like FreeSurfer are established but can be computationally intensive.
Purpose of the Study:
- To introduce DL+DiReCT, a novel method combining deep learning segmentation with DiReCT for enhanced cortical thickness analysis.
- To evaluate the accuracy, reliability, and sensitivity of DL+DiReCT against ANTs and FreeSurfer.
- To assess the performance of DL+DiReCT in detecting group differences in clinical populations.
Main Methods:
- Developed DL+DiReCT by integrating deep learning neuroanatomy segmentation with the DiReCT algorithm.
- Validated DL+DiReCT on two independent datasets.
- Compared DL+DiReCT results with FreeSurfer (surface-based) and ANTs (DiReCT-based) measures.
- Assessed scan-rescan robustness and sensitivity to group differences.
Main Results:
- DL+DiReCT demonstrated strong correlation with FreeSurfer (r = .887) for global mean cortical thickness, outperforming ANTs (r = .608).
- Both DiReCT-based methods showed higher sensitivity to cortical thickness changes than FreeSurfer.
- DL+DiReCT exhibited robustness comparable to FreeSurfer, unlike ANTs which showed low robustness.
- Deep learning segmentation led to the highest effect sizes for group differences between healthy controls and dementia patients.
Conclusions:
- DL+DiReCT offers accurate and reliable cortical thickness measurements efficiently.
- The method shows promise for detecting subtle neuroanatomical changes relevant to neurological disorders.
- DL+DiReCT represents a significant advancement by combining deep learning with established registration techniques.
