Electroencephalography (EEG) based epilepsy diagnosis via multiple feature space fusion using shared hidden...

Area of Science:

• Neurology
• Artificial Intelligence
• Machine Learning

Background:

• Epilepsy is a common neurological disorder diagnosed using electroencephalography (EEG).
• Multi-view learning (MVL) aids automatic epilepsy recognition from complex EEG features.
• Existing MVL methods struggle with inter-view sample differences.

Purpose of the Study:

• To propose a novel shared hidden space-driven multi-view learning algorithm for epilepsy recognition.
• To address challenges in current MVL approaches for EEG analysis.
• To improve the accuracy and robustness of automatic epilepsy diagnosis.

Main Methods:

• Developed a shared hidden space-driven multi-view learning algorithm.
• Utilized kernel density estimation to construct a shared hidden space.
• Combined original and shared spaces to create an expanded learning space.

Main Results:

• The proposed algorithm demonstrated promising performance on a University of Bonn epilepsy dataset.
• Achieved an average classification accuracy of 0.9787.
• Outperformed single-view methods by at least 4% in classification accuracy.

Conclusions:

• The shared hidden space-driven MVL algorithm effectively enhances EEG-based epilepsy recognition.
• The method fully utilizes relevant information within and across different data views.
• This approach offers a significant advancement in automatic epilepsy diagnosis.