Volume 2 - Issue 2, March - April 2026

Routine interpretation of electroencephalograms (EEGs) is labor-intensive, subjective, and difficult to scale in clinical practice. This study proposes a custom convolutional recurrent neural network (CRNN) for automatic detection of abnormal EEGs from minimally processed multi-channel recordings. The architecture integrates 1-D dilated convolutions, lightweight bidirectional recurrent units, and channel-wise attention to model local waveform irregularities and long-range temporal dependencies. Evaluated on the TUH Abnormal EEG Corpus, the method attained 92.0% accuracy, 89.0% sensitivity, and 0.95 AUC-ROC, while sustaining inference below 100 ms per 10- s segment.

Malathi P, Jothika P, Vennila R, V.Vijayakumari,"Deep Learning Enabled Automatic Abnormal Eeg Identification Using Custom Recurrent Neural Network" International Journal of Advanced Multidisciplinary Research and Educational Development, V2(2): Page(1427-1432) Mar-Apr 2026. ISSN: 3107-6513. www.ijamred.com. Published by Scientific and Academic Research Publishing.