Deep-Learning-Assisted Prediction of Neurological Recovery from Coma After Cardiac Arrest

We develop a deep-learning-based algorithm to predict the probability of recovery of a comatose patient who has suffered a heart attack by analyzing electroencephalogram (EEG) and electrocardiogram (ECG) data. These have been provided to participants in the George Moody Physionet Challenge (2023); our team name is RPG@IISC. Given EEGs and ECGs, we extract, from hour-long traces for each patient, the burst-suppression (BS) rate, interchannel EEG correlations, time intervals between successive peaks of ECG, and associated heart variability rate (HVR) metrics. We also use other information provided, e.g., patient age, sex, return of spontaneous circulation (ROSC), in-hospital or out-of-hospital cardiac arrest, presence of a shockable rhythm, and targeted temperature management. With these features, we then use combinations of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to make predictions of (a) the probability of recovery
$\mathcal{P}$
and (b) the cerebral performance category (CPC), at hourly scales; we then combine these hourly results to predict final values for
$\mathcal{P}$
and CPC. In the official phase, when evaluated at 72 hours after ROSC, the score obtained by our algorithm on the hidden-validation data and hidden-test data is 0.63, and 0.43(ranked 24
^th
), respectively.