Prediction of Epileptic Seizures from EEG signals Using Liquid Time-Constant Networks

Epileptic seizures significantly affect the health and quality of life of patients. Accurate prediction of seizures using EEG data can provide early warnings and improve intervention strategies. This paper explores the application of Liquid Time-Constant (LTC) networks, a recent innovation in recurrent neural network architectures, to classify epileptic events from EEG signals. The proposed methodology utilizes the Mendeley EEG dataset, and achieves a remarkable accuracy of $98 \%$ compared to conventional convolution neural networks (CNN), which signifies the diagnostic potential in a clinical setting and can improve patient safety by minimizing false negatives and ensures timely intervention by the physicians. Experimental results demonstrate the efficacy of LTC networks in seizure prediction tasks, notably in regard to the rapid convergence and flexibility in handling the complex temporal relationships present in EEG inputs.