Epileptic Seizure Prediction Using Gan-Enhanced MRI Brain Imaging and Deep Learning

Baskar K1,*, Anjushree S2, Keerthika V3, Nisha P4, Mythili R5

1Associate Professor, Department of Biomedical Engineering, Vivekanandha College of Engineering for Women (Autonomous), Namakkal, Tamil Nadu, India.

Corresponding author: baskar@vcew.ac.in

2Assistant Professor, Department of Biomedical Engineering, Vivekanandha College of Engineering for Women (Autonomous), Namakkal, Tamil Nadu, India

3Department of Biomedical Engineering, Vivekanandha College of Engineering for Women (Autonomous), Namakkal, Tamil Nadu, India

4Department of Biomedical Engineering, Vivekanandha College of Engineering for Women (Autonomous), Namakkal, Tamil Nadu, India

5Department of Biomedical Engineering, Vivekanandha College of Engineering for Women (Autonomous), Namakkal, Tamil Nadu, India

Abstract

Epilepsy is a chronic neurological disorder characterized by recurrent and unpredictable seizures, posing significant risks to patient safety and quality of life. Accurate and early prediction of seizures remains a critical challenge in clinical neurology. This paper proposes a novel, non-invasive epileptic seizure prediction framework that integrates Magnetic Resonance Imaging (MRI) with advanced deep learning techniques, including Generative Adversarial Networks (GANs) and a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture. Initially, MRI scans undergo preprocessing steps such as skull stripping, normalization, and noise reduction to ensure data consistency. GANs are then employed to enhance image quality by removing artifacts and improving structural clarity, as well as to generate synthetic data for addressing dataset limitations. Subsequently, CNNs extract discriminative spatial features, while LSTM networks capture temporal dependencies associated with seizure progression. The proposed model classifies brain states into interictal, preictal, and ictal phases, enabling early seizure prediction. Experimental results demonstrate that the system achieves an accuracy of approximately 85–90%, with improved reliability and reduced false alarm rates compared to conventional methods. The proposed framework offers a robust and scalable solution for automated seizure prediction and has strong potential for integration into clinical decision-support systems and real-time patient monitoring applications.

Keywords: Epilepsy Prediction, MRI, Generative Adversarial Networks (GAN), Deep Learning, CNN, LSTM.