DilANN based Detection of Abnormal Brain Activity in Critically Ill Patients

Identifying the occurrence of seizures and abnormal brain activities like lateralized, generalized periodic discharges and rhythmic delta activities is crucial for critically ill patients for prompt diagnosis. The conventional method of EEG analysis is time-consuming and requires expert interpretation, which is often unavailable, particularly in developing countries. Most studies have focused only on automating seizure detection; other seizure-like patterns have been largely overlooked due to inadequate and poorly labelled data. Using the HMS dataset, this study proposes a unique method of computing bipolar montages and classifying EEG signals using the Dilated Adaptive Neural Net (DilANN), which processes signals from four brain regions to classify abnormal patterns. With an average area under the Receiver Operating Characteristic curve (AUROC) of 9 1. 5 % and a Precision-Recall curve (AUPRC) of 73.8 %, this approach enhanced abnormal brain activity detection, improving overall AUROC by 3.9 % and AUPRC by 16 % as compared to the state-of-the-art EEGNet, enabling more timely medical interventions.