An Efficient EEG based Neurological Disorder Prediction and Classification using hybrid Modified Residual Attention U-net with ABMO Algorithm

Telemedicine is a medical service that involves the use of tools for exchanging information, which is part of the broader field of eHealth. A monitoring and classifying electroencephalography (EEG) signal of brain disorders has become a crucial aspect of neurological analysis in recent years. EEG signals provide valuable evidence pertaining to brain disorders and can be used to assess various neurological conditions in individuals. Nevertheless, conventional approaches frequently encounter difficulties in achieving precision as a result of the complex and diverse nature of EEG data, which poses challenges in extracting meaningful features and recognizing patterns. Consequently, this can result in inaccurate analysis. Deep learning approaches can overcome these shortcomings by automatically learning complex patterns and characteristics from raw EEG data, enhancing prediction accuracy and enabling more precise and trustworthy diagnosis of neurological illnesses. The goal of this work is to construct a telemedicine framework based on deep learning to forecast various neurological illnesses by analyzing EEG signals. The suggested detection method has three primary stages: The initial stage involves pre-processing the input signals using Finite Linear Haar wavelet-based Filtering (FLHF). The second phase involves extracting the features of each sub-band using the Daubechies Wavelet Transform (DWT). The features obtained from the third phase were inputted into the Modified Residual Attention U-net (MRAU-net) with the Adaptive Barnacle Mating Optimizer (ABMO) classifier. The proposed Adaptive Batch Multi-Objective optimization (ABMO) method enhances the performance of the MRAU-net model by efficiently picking features and maintaining a balance between exploration and exploitation during training. This results in improved detection and classification of intricate EEG patterns. This study utilizes two EEG databases, specifically the Children's Hospital Boston (CHB) MIT Dataset and the TUH-EEG Corpus dataset, to evaluate the effectiveness of the proposed method. The simulation results indicate that the proposed model has the potential to significantly enhance disease detection accuracy while reducing system complexity and time consumption compared to existing techniques such as Back Propagation Neural Network (BPNN), convolutional neural network (CNN), deep neural network (DNN), Gated Recurrent Unit (GRU), and Bidirectional deep long short-term memory (Bi-LSTM).The system attains an accuracy of 96.8%, precision of 89.2%, recall of 98.6%, F-Score of 94.5%, and a processing time of 0.09 seconds using the CHB-MIT Dataset. Additionally, it attains an accuracy rate of 95.8%, a precision rate of 98%, a recall rate of 99%, an F-Score rate of 96%, and a processing time of 0.08 seconds when utilizing the TUH-EEG Corpus dataset.