Deep feature fusion based childhood epilepsy syndrome classification from electroencephalogram

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate classification of pediatric epilepsy syndromes is crucial for effective diagnosis and treatment.
• Existing research predominantly focuses on seizure detection, with limited attention to syndrome classification.

Purpose of the Study:

• To develop and evaluate a novel feature fusion model for classifying two common childhood epilepsy syndromes: benign childhood epilepsy with centro-temporal spikes (BECT) and infantile spasms (WEST syndrome).
• To improve the diagnostic accuracy of pediatric epilepsy syndromes using electroencephalogram (EEG) data.

Main Methods:

• A novel feature fusion model integrating deep transfer learning with time-frequency representations of scalp EEG was developed.
• A fully connected network was employed for feature learning and syndrome classification.
• Experiments were conducted on a dataset from the Children's Hospital, Zhejiang University School of Medicine (CHZU).

Main Results:

• The proposed algorithm achieved an average classification accuracy of 92.35% for distinguishing BECT, WEST syndrome, and normal cases.
• The feature fusion model demonstrated effectiveness in characterizing epilepsy syndromes from EEG data.

Conclusions:

• The developed deep transfer learning-based feature fusion model shows significant promise for accurate classification of common childhood epilepsy syndromes.
• This approach offers a potential advancement in the computer-aided diagnosis of pediatric epilepsy.