Project

Multi-Modal Analysis and Integration of MS Patient Data

Short Summary

We will develop computational models for predicting the state of Multiple Sclerosis in patients, the course of the disease, and will stratify treatment responses to support clinicians. Our models will integrate patient notes as part of clinical data through natural language processing, brain scans from MRI images, and behavioral data from wearable devices to generate reports for longitudinal comparisons of MS patients to support patient care.

Goals

In this project, we aim to create computational models for predicting the course of the disease and stratifying for treatment responses to support clinicians. Building on a data corpus collected with our partners at the University Hospital Zurich in a previous PHRT project as well as ongoing studies, we will adapt and develop novel machine learning methods to exploit the variety of data modalities focusing on disease state characterization and disease progression modeling. This includes natural language processing (NLP) for features extraction from clinical data, automatic feature detection in magnetic resonance images (MRI), models for longer-term temporal data, and integrative models that combine information from these three different data modalities and sources. Together, our models will allow generating reports for longitudinal comparisons of MS patients to support patient care.

Significance

We expect our novel developments to help discover predictive markers for accurate disease prognosis by integrating multi-modal data sources to extract longitudinal changes. Our methods will eventually also be impactful for ongoing patient care and assessment by circumventing the need for reannotation and model retraining upon new incoming data, allowing for uninterrupted use of our models and avoiding delays in the work of clinical experts given their busy work schedule.

Background

Multiple Sclerosis (MS) is highly heterogeneous in clinical and imaging presentation and key pathophysiological processes of inflammation and neurodegeneration. Hence, establishing an accurate clinical picture of disease activity and progression in individual patients is a core challenge requiring knowledge and integration of the key elements driving the disease and its temporal evolution. Essential elements are the spectrum of neurological manifestation, imaging markers of disease burden, patient age, disease duration, initial symptoms, previous treatment response, and to some extent, cerebrospinal fluid or blood biomarkers. The highly complex interplay of these many different features requires a high level of experience by the clinician.

Data-Intensive Research Project

Co-Investigators

Prof. Dr. Gunnar Rätsch, Biomedical Informatics Lab, Department of Computer Science, ETH Zürich, Switzerland
Prof. Dr. med. Roland Martin, Klinik für Neurologie, Universitätsspital Zürich, Switzerland
Dr. med. Andreas Lutterotti, Neurozentrum Bellevue, Bellevue Medical Group, Zürich, Switzerland

Consortium

Status

In Progress

1st Joint SDSC Call, Data-Intensive Research