A template-based customization approach for compensating missing brain MRI sequences in people with multiple sclerosis

Motivation: Clinical brain MRI is critical for care delivery but not always consistent, limiting advanced secondary analyses. Goal(s): To develop a new method for synthesizing unavailable brain MRI using data with multiple sclerosis. Approach: We first create templates for the common T1, T2, and FLAIR MRI based on available scans of different cohorts. Then we use a tissue-wise z-score approach to derive person-specific contrasts including lesions using the other sequences of the scan. Results: 45 individuals appear to be sufficient for constructing robust templates of the sequences. Further customization allows to create personalized sequences equivalent to the ground truth. Impact: Robust synthetization of unavailable brain MRI sequences is essential in various secondary analytic tasks including machine learning for advanced disease characterization and monitoring, which can benefit people with different diseases beyond multiple sclerosis.