A Multimodal Monitoring Approach to Predicting the Onset of Physiological Incidents Using Machine Learning

Traumatic Brain Injury (TBI) is a complex, heterogeneous disease affecting millions of people in the U.S. each year [1]. Multimodal monitoring (MMM) is a relatively new attempt to access and monitor the brain post injury by incorporating multiple sources of information recorded in the intensive care unit (ICU) [2]. As a result, MMM has led to the creation of exhaustive datasets, allowing for the integration of many different signals and modalities pertaining to the management of severe TBI [2]–[4]. One of the most pressing questions about a TBI patient is whether an incident, or an event, will occur in the near future [5]. Additionally, clinicians require sufficient time before the onset of an event to administer a drug to the patient to avoid the oncoming event. Therefore, the aim of this work is to detect whether physiological events will occur in the future using the multimodal physiological data as features. In this study, we leverage both the breath and resolution of physiological data to predict the onset of physiological incidents 30 minutes before their onset. Specifically, we focus on three physiological events, namely high intercranial pressure (ICP), out of range diastolic arterial blood pressure (ABP), and out of range systolic ABP, using seven relevant modalities.