Investigating EEG burst suppression for coma outcome prediction

Abstract

Every year, over 300,000 incidents of cardiac arrest occur in the United States. Of the people who are successfully resuscitated and brought to the hospital, approximately 80% remain unconscious for some amount of time Marion [2009]. Predicting whether or not a patient will wake up from coma, as well as the patient's neurological function after waking up, is an important task in guiding treatment decisions for physicians and family of the patient. This project seeks to improve this prediction process by analyzing features of the patients' EEG recordings during coma with the aim to determine quantitative metrics which are predictive of patients' outcome. Specifically, we focus on the analysis of the similarity of bursts during burst suppression, which has been hypothesized to be linked with poor outcome. Our work confirms that similarity of bursts is indeed linked with poor outcome, and we also find that dynamic time warping gives a viable alternative to the previously used method of cross-correlation as a measure of similarity of bursts, with good predictive power for patient outcome.