Predicting stroke outcomes based on multi-modal analysis of physiological signals

Abstract

Stroke is a leading cause of death and disability. Early prediction of stroke patients' functional outcomes is helpful for treatments. However, current diagnosis machines, such as computed tomography (CT) and magnetic resonance imaging (MRI), are expensive, not portable, and may cause side effects. Additionally, current diagnosis scales, such as National Institutes of Health Stroke Scale (NIHSS), should be evaluated by professional medical staff, and thus cannot be conducted continuously. In this paper, we propose a multi-modal analysis methodology to predict a stroke patient's functional outcome based on physiological signals, including EKG, ABP, and PPG. By applying the multi-modal framework to analyze the stroke patients' physiological signals in intensive care unit (ICU), we find that the accuracy of stroke outcome predictions achieves 82.7%, which performs better than a single-modal built by any single phase. In addition, the joint EKG-ABP-PPG analysis achieves performance comparable to NIHSS, implying that the multi-modal analysis framework has potential for predicting functional outcomes of stroke.