Deep CNN ensemble framework for early detection of epileptic seizures using STFT-derived EEG features
TL;DR
The use of pre-trained models reduced the training time and resources required, and the unique application of the ensemble learning approach produces more robust and reliable results compared to individual deep learning models.
Abstract
Epilepsy is a major neurological illness that affects several million around the globe, and it has a major impact on their daily activities. The epileptic seizures occur any time and at any location and this unpredictable nature of seizures makes it challenging for epilepsy patients to lead a normal existence. This results in limitations on the day-to-day activities of an epilepsy patient. In this regard, a reliable seizure detection model can play a crucial role. Such a solution has been proposed in the present work. The proposed model encompasses two different pre-trained models, the MobileNet-V1 and Inception-V3. These two models are ensembled in the proposed model, which performs feature extraction as well as the classification task. The proposed model is validated against the Electroencephalogram (EEG) signals obtained from the Children’s Hospital Boston (CHB) and the Massachusetts Institute of Technology (MIT), known as the CHB-MIT dataset for seizure detection. The use of pre-trained models reduced the training time and resources required, and the unique application of the ensemble learning approach produces more robust and reliable results compared to individual deep learning models. The proposed seizure detection model achieved a reasonable performance, having an accuracy of 93.71%, a sensitivity of 93.60%, and a specificity of 92.61%.