Related Experiment Video
Updated: Jun 8, 2026
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
Bayesian classification of multiple sclerosis lesions in longitudinal MRI using subtraction images
Colm Elliott1, Simon J Francis, Douglas L Arnold
1Centre for Intelligent Machines, McGill University, Canada.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|October 1, 2010
Summary
This study introduces a new Bayesian framework for automatically detecting changes in multiple sclerosis (MS) lesions using MRI scans. The method precisely identifies new, enlarging, and resolving lesions, aiding disease monitoring.
Area of Science:
- Medical Imaging
- Neurology
- Biostatistics
Background:
- Accurate monitoring of multiple sclerosis (MS) lesions in longitudinal MRI is crucial for tracking disease progression and evaluating treatment efficacy.
- Existing methods may be limited by registration artifacts and sensitivity to subtle lesion changes.
Purpose of the Study:
- To present a novel probabilistic framework for the automated detection of new, enlarging, and resolving MS lesions in longitudinal MRI scans.
- To overcome limitations of previous methods by explicitly modeling variability and incorporating prior information.
Main Methods:
- Development of a Bayesian framework utilizing multimodal subtraction magnetic resonance (MR) images.
- Explicit modeling of difference image variability, tissue transitions, and neighborhood classes as likelihoods.
- Inclusion of reference scan classification as a prior to mitigate registration artifacts.
Main Results:
- The framework was validated on a scan-rescan dataset (3 MS patients) and a multicenter clinical dataset (89 RRMS patients, 212 scans).
- Demonstrated high precision in identifying MS lesions in longitudinal MRI.
- Showed sensitivity to active lesion changes, including new, enlarging, and resolving lesions.
Conclusions:
- The proposed probabilistic framework accurately and precisely identifies MS lesion activity in longitudinal MRI.
- This method offers a robust tool for monitoring disease progression and treatment response in multiple sclerosis patients.
