Analysis of EEG Signals for Predicting Neurological Recovery in Post-Cardiac Arrest Comatose Patients

Prognostic decisions made in the early days following cardiac arrest can significantly impact patient outcomes. This work leverages continuous electroencephalography (EEG) to predict neurological outcomes in patients who have become comatose after a cardiac arrest. We propose a two-stage stacked classifier (TSSC), combining an RNN model and an SVM model with a nonlinear kernel, with KNN serving as the meta-classifier. This approach involves an analysis of EEG features extracted using Welch's power spectral density, resulting in a reduced feature subset and a subsequent enhancement in accuracy. Experimental results demonstrate an impressive predictive accuracy of 95.1% with this approach. We also discuss the implications of our findings for clinical practice.