Diagnosis of Neurological Dysfunction from EEG Signals Using DuelQ-SeizureNet and Meta Black Ant Optimization for Epileptic Seizure Detection.
TL;DR
A new hybrid deep learning model based on one-dimensional Convolutional Neural Network, Bidirectional Long Short-Term Memory, and Dueling Q-Learning is proposed to accurately classify epileptic seizures from EEG signals in an adaptive manner to introduce a novel reinforcement learning assisted optimization approach in deep seizure detection architecture.
Abstract
Background
Early diagnosis of neurological dysfunctions, particularly epilepsy, is vital for early intervention and improvement of patients' quality of life. However, traditional seizure detection techniques suffer from low detection accuracy, high false positive rate, and high computational complexity, making it difficult to effectively capture the complex spatiotemporal characteristics of electroencephalography (EEG) signals. Despite the significant improvements in seizure detection accuracy brought by deep learning techniques, the current models suffer from inaccuracies, limited adaptability, and inability to operate in real time. NEW
Method
To address these challenges, a new hybrid deep learning model based on one-dimensional Convolutional Neural Network (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM) network and Dueling Q-Learning is proposed to accurately classify epileptic seizures from EEG signals in an adaptive manner. In addition, a novel approach is proposed called Metaheuristic Based Adaptive Optimization (MBAO) to adaptively select an optimal temporal window size for the effective extraction of features, while minimizing the required information loss and computation burden.
Results
and Comparison with existing methods: The proposed model has tested in various experiments conducted in a large number of benchmark datasets like CHB-MIT, Kaggle EEG Epileptic datasets etc. which justifies the effectiveness of the proposed model. DuelQ-SeizureNet has an accuracy of 99%, a precision of 96%, a recall (sensitivity) of 98%, a specificity of 99%, and an F1 score of 99% with a low execution time of 50ms in seizure prediction.
Conclusions
This proposed framework introduces a novel reinforcement learning assisted optimization approach in deep seizure detection architecture. It can operate with lower false detection rates (1.8%), higher area under the ROC curve (AUC) (0.995), and lower computational speed than the existing scheme, ensuring reliable real-time implementation.