Grant: Project Grant

Multiple sclerosis (MS) is a common neurological disease that disables young Canadians. There are 14 approved therapies now for many who have a relapsing remitting form of MS (RRMS). Yet within 10-15 years of disease onset, >50% of RRMS patients develop secondary progressive MS (SPMS), which remains untreatable. While many factors may cause this poor outcome, the inability to optimize therapy based on an individual's risk of worsening is crucial. There is no reliable way to identify the risks but if there is, early aggressive treatment would be possible for those with a greater risk of worsening, which is critical to save the tissue before irreversible damage accumulates. This project will establish a novel deep machine learning method to predict the 5-year risk of disability score worsening in RRMS patients. We will take a unique approach by simultaneously evaluating patient outcomes from brain pathology, clinical measures, and brain magnetic resonance imaging (MRI). Deep learning has shown enormous progress recently in disease prediction, especially when combing pathology-directed outcomes, and multiple types of patient data. Using standard brain MRI, our pilot deep learning data is 83% accurate in predicting RRMS and SPMS. By assessing the pixel pattern (texture) of MRI, our pioneer work also discover that the scale and direction of MRI texture detects specific elements of MS pathology. We have now built new deep learning methods using 3 types of outcomes from MRI texture pathology, clinical data, and brain MRI, and cutting-edge techniques to unveil how deep learning works. Built on these solid data and large cohorts of MS patients, this project will establish a novel objective method to identify the high and low risks of disability worsening for individual RRMS patients so that each can receive the best treatment regimen to improve prognosis. This personalized medicine approach will directly benefit MS patients and families, the health care system, and society.