Predicting Patient Outcome in Multiple Sclerosis using a Quantitative Radiomic Approach

点击翻译按钮获取中文摘要

Project Summary: Multiple sclerosis (MS), a leading cause of disability in young and middle-aged adults, is a highly heterogeneous disease, with wide variations in clinical presentation, disease course and response to treatment. In order to personalize care in MS, it is important to harness its clinical heterogeneity and the diversity of its underlying pathology, and to develop models able to predict individual behavior of patients. Brain magnetic resonance images (MRI), acquired routinely in MS patients, contain information that reflects underlying pathophysiology, which may be brought into light through quantitative analyses. Radiomics, a technique well developed in oncology, converts routine medical images into mineable high-dimensional data that can be modeled to support clinical decision-making. The central hypothesis of the proposed project is that radiomic analysis, combined with careful feature selection and accurate modeling, can predict patient outcome and response to therapy using standard-of-care MRI in MS. Our hypothesis will be tested by leveraging existing 3-year imaging and clinical data from the CombiRx trial, a multi-center, phase-III investigation of combination therapy in 1008 relapsing-remitting MS (RRMS) patients. We will first determine the potential for non-invasive radiomic biomarkers of disease severity in RRMS. Towards this goal, we will extract radiomic features of MS lesions from FLAIR, pre and postcontrast T1-weighted MR images using an open-source radiomic pipeline. Through appropriate feature selection, we will identify an independent radiomic feature set able to characterize individual phenotype on MRI in a selection cohort. We will then evaluate the selected features cross-sectionally to determine their efficacy in characterizing disease severity, leveraging training and validation subsets. The performance of our radiomic approach will be compared to traditional models using clinical and standard imaging markers such as lesion volume. In the second stage of this proposal, we will explore the performance of radiomic-based models to predict long-term outcome and treatment response in MS. We will build models to predict disease activity free status (DAFS) at 3 years using selected baseline radiomic features, and identify treatment response phenotypes. We will investigate and compare the performance of various machine-learning models in an unbiased manner. Finally, we will assess the effect of each therapeutic regimen on radiomic features by comparing on-treatment changes across treatment arms. This may provide evidence for treatment-specific monitoring parameters.