Epilepsy Classification using Multi-Sequence MRI’s: An Ensemble Approach

Epilepsy is a major threat, when it comes to human health. Classifying brain anomalies on MR images as a pathological clue of Epilepsy is crucial for diagnosis. Electroencephalography (EEG) and Magnetic resonance (MR) imaging are only two of the diagnostic tools that have been proposed for use in screening for epileptic seizures. The diagnosis can be made with the use of many MR images taken at different times by a neuroradiologist. The dataset under consideration is comprised of axial multi-sequences of MR images, and it features both epileptic and nonepileptic aberrant brain structures from a wide range of people. This article performs a comparative study among the models - CNN, MobileNet, InceptionV3, VGG16, VGG19, LSTM, GRU and also the Ensemble models - ‘CNN + MobileNet’, ‘VGG16 + LSTM’, ‘VGG16 + GRU’, ‘VGG19 + LSTM’ and ‘VGG19 + GRU’. The Epilepsy prediction for user upload images is achieved through an ensemble approach. We have demonstrated the effectiveness of these models in correctly detecting epileptic patients based on multi-sequence MRI data by achieving extremely accurate and robust classification findings through extended experimentation. The experiments demonstrated the classification best accuracy of 97.16% for ‘VGG19 + LSTM’, while the lowest score was 69.93 for InceptionV3 among all the twelve models.