A Radiomic "Warning Sign" of Progression on Brain MRI in Individuals with MS

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Multiple Sclerosis (MS) is a chronic, progressive demyelinating disorder.
• Diagnosis and monitoring of MS heavily rely on Magnetic Resonance (MR) imaging.
• New MR imaging lesions are critical early indicators of disease progression.

Purpose of the Study:

• To assess a machine learning model utilizing radiomics features for predicting MS progression on brain MR imaging.
• To identify potential imaging biomarkers for early detection of MS progression.

Main Methods:

• A retrospective cohort study with external validation was conducted.
• Longitudinal MR imaging data from MS patients were processed for machine learning.
• Radiomics features were extracted from prelesion areas and normal-appearing white matter.

Main Results:

• The best-performing model (XGBoost) achieved high accuracy (0.91) on the test set and good performance (0.74) on external validation.
• Key radiomics features were related to heterogeneity and gray-level emphasis in segmented regions.
• Probability maps were generated for potential clinical applications.

Conclusions:

• Machine learning models based on radiomics features can effectively distinguish prelesions from normal-appearing white matter.
• Radiomics features from normal-appearing white matter show promise as imaging biomarkers for MS progression.