EEG-Based Seizure Onset Detection of Frontal and Temporal Lobe Epilepsies Using 1DCNN

Objective: The manual interpretation of electroencephalogram (EEG) signals for detecting epileptic seizures is time-consuming and labor-intensive, highlighting the critical importance of exploring automated seizure detection methods. Given this, this work concentrates on seizure detection using scalp EEG signals collected from people with frontal lobe epilepsy (FLE) and temporal lobe epilepsy (TLE). Method: 20 FLE patients and 20 TLE patients are utilized in our work, and a parallel one-dimensional convolutional neural network (1DCNN) model is built for classification. Our work explores two strategies: the patient-specific strategy and the patient-cross strategy, during seizure detection. Furthermore, the performances of our work are evaluated at both event- and segment-based levels simultaneously for a more comprehensive comparison. Results: In the patient-specific strategy, TLE patients achieve superior overall results of 100% sensitivity, 0.0/h false detection rate (FDR) and 16.4-sec latency (90.2% sensitivity, 0.0/h FDR and 14.9-sec latency for FLE patients) at the event-based level, and 70.3% sensitivity, 99.6% specificity, 99.4% accuracy and 0.849 area under curve (AUC) (58.0% sensitivity, 99.5% specificity, 99.4% accuracy and 0.788 AUC for FLE patients) at the segment-based level. In the patient-cross strategy, TLE patients also show superior overall performances of 98.0% sensitivity, 0.8/h FDR and 18.8-sec latency (87.8% sensitivity, 1.6/h FDR and 16.7-sec latency for FLE patients) at the event-based level, and 80.5% sensitivity, 95.2% specificity, 95.1% accuracy and 0.879 AUC (66.9% sensitivity, 88.3% specificity, 88.2% accuracy and 0.776 AUC for FLE patients) at the segment-based level. Conclusion: Our work can effectively detect seizures of FLE and TLE, and this may provide valuable reference for future research on seizure detection in FLE and TLE.