Automatic Detection of Abnormal EEG Signals Using WaveNet and LSTM.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Neurological disorders impact over a billion people globally, causing significant mortality.
• Manual electroencephalogram (EEG) analysis is time-consuming, requires expertise, and faces shortages of trained neurologists.
• Automated diagnostic processes are essential for timely and accessible neurological disorder identification.

Purpose of the Study:

• To develop and validate a novel deep learning model for automatic detection of abnormal EEG data.
• To address the limitations of manual EEG interpretation and the scarcity of neurological expertise.

Main Methods:

• A novel deep learning model integrating WaveNet-Long Short-Term Memory (LSTM) and LSTM architectures was proposed.
• The model was trained and evaluated on the TUH abnormal EEG Corpus V.2.0.0 (TUAB).
• Model generalization was assessed on an independent TUEP dataset without hyperparameter tuning.

Main Results:

• The proposed model achieved 88.76% classification accuracy on the TUAB dataset, surpassing state-of-the-art.
• On the independent TUEP dataset, the model demonstrated high robustness with 97.45% accuracy in classifying normal versus abnormal EEG recordings.
• Ablation experiments confirmed the significance of each component within the proposed deep learning architecture.

Conclusions:

• The developed deep learning model offers an effective and robust solution for automated abnormal EEG detection.
• This approach holds promise for improving the early diagnosis and management of neurological disorders, especially in resource-limited settings.
• The model's strong performance and generalization capabilities highlight its potential for clinical application.