Epileptic seizure detection using transfer learning

Abstract

The most notable advancement in the 21st Century has been in artificial intelligence (AI). Despite how far AI has progressed, how it applies to healthcare remains a significant challenge for brilliant minds all over the world. A neurological condition known as epilepsy can strike a person at any time in their life. An individual with epilepsy therefore experiences frequent to infrequent seizures, which can occasionally result in death. Electroencephalogram (EEG) signals aid in the diagnosis of this condition. However, lengthy EEG signals frequently take a day or longer to detect this disorder, even for trained neurologists, and may even cause human error. Therefore, it is essential to create a reliable and computationally efficient system. This study aims to classify seizures by creating Convolutional Neural Network (CNN) Inception ResNet V2 and short-time Fourier transform (STFT) to extract the time-frequency plane from time domain signals. This study helped to better classify health and seizures by achieving up to 100% the highest classification accuracy.