MIMoSA: An Automated Method for Intermodal Segmentation Analysis of Multiple Sclerosis Brain Lesions

Area of Science:

• Neuroimaging
• Medical image analysis
• Multiple sclerosis research

Background:

• Magnetic resonance imaging (MRI) is essential for detecting white matter lesions (WMLs) in multiple sclerosis (MS).
• Automated segmentation of WMLs using MRI remains a significant challenge despite decades of research.
• Multimodal MRI techniques offer complementary tissue property information, enhancing WML identification.

Purpose of the Study:

• To introduce the Method for Inter-Modal Segmentation Analysis (MIMoSA), a novel fully automatic algorithm for WML segmentation.
• To leverage intermodal coupling regression and mean structure for improved lesion probability modeling within each voxel.
• To validate MIMoSA's performance against expert manual segmentation and established automated methods.

Main Methods:

• MIMoSA utilizes covariance features from intermodal coupling regression and mean structure.
• Validation involved comparison with expert manual segmentation and two automated methods (OASIS, LesionTOADS) on a dataset of 98 subjects.
• A secondary validation was performed using a publicly available dataset from a segmentation challenge.

Main Results:

• MIMoSA achieved an average Sørensen-Dice coefficient (DSC) of 0.57 and partial AUC of 0.68 (at 1% false positive rate) in the Johns Hopkins dataset.
• Performance of MIMoSA was superior to OASIS and LesionTOADS in the primary validation.
• MIMoSA demonstrated competitive performance in the segmentation challenge dataset.

Conclusions:

• MIMoSA provides statistically significant improvements in WML segmentation accuracy compared to LesionTOADS and OASIS.
• The algorithm shows robust and competitive performance across different validation datasets.
• MIMoSA represents an advancement in automated WML segmentation for multiple sclerosis research.