An Automatic Epilepsy Detection Method Based on Improved Inductive Transfer Learning.

Area of Science:

• Neurology and Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Epilepsy is a chronic neurological disorder characterized by recurrent seizures due to abnormal brain neuron discharge, impacting patient health and cognition.
• Electroencephalogram (EEG) is crucial for diagnosing, assessing, and localizing epilepsy, providing essential data for understanding its pathogenesis.
• Machine learning (ML) shows promise in classifying epilepsy EEG signals, but performance is often limited by insufficient training data in real-world scenarios.

Purpose of the Study:

• To address the challenge of limited EEG data in epilepsy detection by developing a novel transfer learning method.
• To enhance the adaptability and accuracy of epilepsy EEG signal recognition using a discriminate least squares regression (DLSR)-based inductive transfer learning approach.

Main Methods:

• Introduced a novel inductive transfer learning method based on discriminate least squares regression (DLSR).
• Leveraged DLSR's ability to expand inter-class intervals for improved classification.
• Simultaneously utilized target domain data and source domain knowledge to boost model performance.

Main Results:

• The proposed DLSR-based transfer learning method demonstrated superior performance in epilepsy EEG signal recognition.
• The method showed improved adaptability and accuracy compared to other representative techniques.
• Effectiveness was particularly noted in scenarios with limited available EEG data for model training.

Conclusions:

• The developed DLSR-based inductive transfer learning method effectively mitigates the impact of insufficient data on epilepsy detection.
• This approach offers a robust solution for enhancing the accuracy and reliability of ML-based epilepsy diagnosis using EEG signals.
• The findings support the utility of transfer learning in improving clinical applications of AI in neurology.