Deep learning method for epilepsy detection using embedded zero tree wavelet transform and FRCNN-AIV3 classifier

The protracted diagnostic process for Epilepsy, a neurological illness that is now incurable, begins in childhood, when the symptoms first appear. Early diagnosis is crucial in reducing adverse outcomes, as it allows for the immediate initiation of compensatory instruction. Timely action is crucial in resolving these situations due to the seriousness of the situation and the potential for severe consequences. Deep Learning (DL), a novel method of soft computing, has proven beneficial in various fields, including pattern recognition and healthcare diagnostics. The use of deep learning techniques in the treatment of neurological and neuropsychiatric diseases, particularly Epilepsy, is the subject of this study's in-depth investigation. This research delves into the inner workings of DL algorithms and how they diagnose various neurological illnesses in humans. Epilepsy problems can be detected using electroencephalogram (EEG) data. To remove extraneous data and analyze the same data in both temporal and frequency domains, the EEG dataset is segmented and filtered. The next step is to decompose the EEG signals into their component bands and extract characteristics from them using the Embedded Zero tree Wavelet (EZW) Transform. Afterwards, five statistical methods are used to extract information from these EEG sub-bands: mean, variance, entropy, skewness, and kurtosis. Several classifiers, such as the Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), Convolutional Neural Network (CNN) with Visual Geometry Group (VGG) net (CNN-VGG Net), and Support Vector Machine (SVM) enhanced with Feed Forward (FF) Neural Network (SVM-FF), are fed these extracted features. For feature categorisation according to their respective classes, a sophisticated deep learning approach is applied, specifically the CNN with optimised Adam optimiser-optimised Inception V3 architecture (CNN-AVI3). Notably, the suggested model obtains an outstanding accuracy rate of 97.28% by combining EWZ and CNN-AVI3. In addition, it surpasses current state-of-the-art techniques with a specificity of 97.86%, an F-score of 93.74%, a recall of 95.14%, and an accuracy of 92.39%.