Predicting Quantitative Myelin Images from Clinical Scans and DTI using Linear and Neural Network Models

Motivation: Quantitative myelin imaging methods are usually not included in clinical MRI scans for persons with multiple sclerosis. Goal(s): Predict macromolecular pool-size-ratio maps from standard clinical scans. Approach: Compare PSR map prediction results from linear model and neural network model. Results: Most of the relationship between clinical scans plus DTI and PSR maps is captured by a simple linear model and the more complex nonlinear model offers minimal improvements. Impact: Leveraging typical clinical scans to predict PSR maps using a simple linear model can improve diagnostic capabilities for multiple.