Enhancing Epileptic Seizure Detection Using Convolutional Neural Networks and Data Augmentation Techniques

Epilepsy stands as one of the most prevalent neurological disorders worldwide. Diagnosis is typically conducted by examining electroencephalogram (EEG) recordings in clinical settings. In this research paper, a novel deep-learning model is put forward to detect epileptic seizures. Due to the scarcity of seizure-related epileptic samples and the need for an adequate dataset to train convolutional neural networks (CNNs), the study employs a data augmentation technique called Cubic Spline Interpolation (CSI). After applying the low-pass filter, Cubic Spline Interpolation successfully augments the training data, addressing the challenges associated with limited samples. The Following training and testing sets undergo short-time Fourier transform (STFT) to enable time-frequency analysis. Subsequently, the transformed data is fed into a convolutional neural network (CNN) model, integrated with dropout layers, to facilitate categorizing three distinct states, including healthy, pre-ictal, and inter-ictal. The findings of this study reveal compelling results, with an impressive accuracy rate of 99.3%. These outcomes underscore the effectiveness of the proposed approach, indicating its superiority in achieving higher accuracy when contrasted with conventional methods.