Multiview Feature Fusion Representation for Interictal Epileptiform Spikes Detection

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Interictal epileptiform spikes (IES) in electroencephalogram (EEG) signals are crucial for identifying the epileptogenic zone.
• Current visual inspection for IES detection is subjective, time-consuming, and error-prone.
• Existing computer-aided methods struggle with the complex, nonlinear, and nonstationary nature of EEG signals.

Purpose of the Study:

• To develop a novel multiview feature fusion representation (MVFFR) method for accurate IES detection in EEG signals.
• To improve upon the performance of existing computer-aided detection methods for epilepsy diagnosis.
• To address the limitations of subjective and time-consuming visual inspection of EEG data.

Main Methods:

• Developed a multiview feature fusion representation (MVFFR) method.
• Integrated temporal, frequency, temporal-frequency, spatial, and nonlinear features.
• Employed an unsupervised infinite feature-selection method for optimal feature representation.
• Combined MVFFR with a robustness classifier for IES detection.

Main Results:

• MVFFR achieved optimal detection performance: 89.27% accuracy, 89.01% sensitivity, 89.54% specificity, and 89.82% precision on a balanced dataset.
• Demonstrated superior performance compared to other feature ranking methods.
• Maintained excellent generalization capacity with a low false detection rate (0.15/min) on an independent, unbalanced dataset.

Conclusions:

• The MVFFR method offers a significant advancement in automated IES detection from EEG signals.
• This approach effectively captures latent information in EEG, enhancing diagnostic reliability for epilepsy.
• MVFFR provides a robust and accurate tool, complementing existing diagnostic procedures.