NeuroHero: A QEEG and HRV Based Neural Network for Explainable Post-Anoxic Coma Prognosis

Annually, 4.8 million people enter a post-anoxic coma following cardiac arrest, necessitating early recovery prognosis using Electroencephalograms (EEG) and Electrocardiograms (ECG). Challenges with continuous EEG monitoring and interpretation limit its real-time application. The best machine learning models for this issue use raw EEG data, lacking in explainability. This paper proposes a Quantitative EEG (QEEG) and Heart Rate Variability (HRV) based Dense Neural Network to enhance interpretability. Using the I-CARE database, signals were preprocessed through resampling, normalizing, and filtering. QEEG and HRV features from time, frequency, and other domains were extracted for model training. Notably the model achieved an AUC-ROC of 0.81 and sensitivity of $\mathbf{6 6 \%}$ at $\mathbf{1 0 0 \%}$ specificity f or p oor outcomes, o utperforming s ome state-of-theart models by up to 106%. This research reveals the significance of QEEG and HRV features in post-anoxic coma prognosis to prioritize model explainability.