Efficient use of clinical EEG data for deep learning in epilepsy

• Augmenting datasets improves the performance of neural networks for interictal epileptiform discharge detection. Time shifting and different montages can reduce need annotated data. Deep learning may cause a fundamental shift in clinical EEG analysis. Automating detection Interictal Epileptiform Discharges (IEDs) electroencephalogram (EEG) recordings time spent on visual analysis diagnosis epilepsy. has shown potential this purpose, but scarceness expert data creates bottleneck process. We used EEGs from 50 patients with focal epilepsy, 49 generalized epilepsy (IEDs were visually labeled by experts) 67 controls. The was filtered, downsampled cut into two second epochs. increased number input samples containing IEDs through temporal using montages. A VGG C convolutional network trained to detect IEDs. Using dataset more samples, we reduced false positive rate 2.11 0.73 detections per minute at intersection sensitivity specificity. Sensitivity 63% 96% 99% model became less sensitive position IED epoch montage. Temporal use deep Dataset augmentation annotation, facilitating training networks, potentially leading