Longitudinal MS Lesion Analysis

Method Summary

White matter hyperintensity (WMH) lesions were automatically segmented on FLAIR MRI at both timepoints using a validated deep learning model. Baseline and follow-up images were co-registered using symmetric diffeomorphic registration (ANTs SyN) to a common midpoint space, ensuring unbiased spatial comparison.

Lesion changes were classified by comparing segmentation masks between timepoints: new (present at follow-up only), enlarging/persistent (present at both timepoints), and resolving (present at baseline only). Change categories use a 1 mm tolerance to reduce misregistration artifacts; volume measurements use original (undilated) masks. Volumes are reported in millilitres (mL), computed from voxel counts and native voxel dimensions.

Automated quality checks assess image quality, registration accuracy, and segmentation plausibility before results are reported. Cases failing quality thresholds are flagged and quantitative results are withheld.

Limitations & Intended Use

This software provides assistive quantification only and does not constitute a clinical diagnosis or replace radiological interpretation.
Final clinical interpretation is the responsibility of the treating physician.
All measurements depend on image quality, scan protocol consistency, and spatial comparability between timepoints.
Automated lesion detection may miss small or atypical lesions and may include false positives, particularly near the cortex and in the posterior fossa.
Volume changes below 0.1 mL may fall within measurement uncertainty and should not be over-interpreted.
Brain volume change estimates are approximate and are not a substitute for dedicated volumetric analysis.