Improving Features for Multiple Sclerosis Disability Progression Prediction through Temporal Alignment of Hospital Visits

Abstract

Predicting the disability progression in multiple sclerosis remains a significant challenge. Statistical features extracted from evoked potential signals are often discussed in literature as interesting biomarkers, yet their practical adoption remains limited due to their low effectiveness. Conversely, the use of deep neural network feature encoders, which allow for extracting more expressive biomarkers, is hindered by the limited availability of data. This study proposes a novel method of enhancing statistically extracted features by aligning hospital visits to embed the progression of the disease within the features. The results demonstrate a significant improvement in predicting disability progression using the statistically extracted features. Insight is provided through a study of correlation and importance of the enhanced features.