Automated and Expert-Level Identification of Interictal Epileptiform Discharges with AI-Powered methods

Focal epilepsy manifests as localized abnormal neuronal activity, producing interictal epileptiform discharges that clinicians traditionally identify manually. To enhance detection accuracy and efficiency, we propose a deep learning-based approach combining a convolutional autoencoder (CAE) for feature extraction and a Kolmogorov-Arnold Network (KAN) for classification. Our method integrates candidate selection to balance datasets, latent space compression for feature optimization, and feature pruning for interpretability. The model achieved 100% precision, 71% sensitivity, and 83.78% accuracy, with feature pruning reducing the latent space to 23 inputs while maintaining 87% accuracy. Compared to conventional deep learning models, this novel approach demonstrates higher classification efficiency with fewer parameters, making it suitable for real-time EEG analysis. These findings suggest that KAN-based classification could improve automated EEG monitoring and seizure detection, providing a scalable and clinically viable solution for epilepsy care.Clinical Relevance- Accurate interictal epileptiform discharges detection is essential for epilepsy diagnosis and care, yet manual identification remains time-consuming and highly variable across clinicians. This study offers a deep learning-based approach that improves IED classification precision while maintaining computational efficiency. By integrating feature pruning and an interpretable classification model, this method supports the development of real-time EEG monitoring systems and wearable seizure detection devices, potentially improving early diagnosis, treatment planning, and patient outcomes.