Deep Learning Framework With Integrated Feature Extraction for Detection of Epilepsy and Basal Ganglia Diseases

The prevalence of epilepsy is estimated to be 1 percent of the world, with recurring fits of epilepsy because of excessive electrical activity within the brain, and basal ganglia diseases, which include Parkinson disease, are mostly related to motor and cognitive aspects of the human body. The current diagnostic algorithms typically use manual feature extraction and a substantial human input, which is not scalable and lowers diagnostic accuracy. To overcome these shortcomings, this research will suggest a hybrid deep learning model in the form of one-dimensional convolutional neural network-long short-term memory (1D-CNN-LSTM) where 1D-CNN layers will be used to extract automatic features in representations of EEG signals and LSTM layers will be used to capture the temporal relationships in sequential brain activity data. It is tested on binary epilepsy classification problems, and contrasted with the already known deep learning baselines, such as conventional CNN, deep neural network (DNN), and standard CNN-LSTM model, all operated under the same experimental conditions. The proposed model is found to have an accuracy of 98.15% on testing epilepsy compared with CNN (97.6%), DNN (96.6%), and standard CNN-LSTM (97.7%), and a lower testing loss (4.48%), which indicates a better generalization. In the case of basal ganglia disorder, 700 epochs of training gives the optimal classification accuracy of 91.0%, which is better than the performance of baseline CNN and CNN-LSTM setups used in this research. These findings objectively indicate that the proposed 1D-CNN+LSTM model has better and stronger performance in terms of detection of neurological disorders over the state of art deep learning models.