Seizure prediction using very long-term minimally invasive subcutaneous EEG: generalizable between-patient models
This article was originally published here
Epilepsy. 2022 Apr 20. doi: 10.1111/epi.17265. Online ahead of print.
This study describes a generalized approach to inter-patient seizure prediction using recurrent neural networks with ultra-long-term subcutaneous EEG (sqEEG) recordings. Data from six patients diagnosed with refractory epilepsy and monitored with an sqEEG device were used to develop a generalized algorithm for seizure prediction using LSTM deep learning classifiers. Electrographic seizures were identified by a board-certified epileptologist. One-minute data segments were labeled as preictal or interictal based on their relationship to confirmed seizures. The data was separated into training and test datasets, and to compensate for the unbalanced data ratio during training, noise-added preictic data segment copies were generated to extend the dataset training. The mean and standard deviation of the training data were used to normalize all data, preserving the pseudo-prospective nature of the analysis. Different architecture classifiers were trained and tested using a patient-free cross-validation method, and the area under the receiver operating characteristic curve (AUC) was used to evaluate the performance classifiers. The significance of each input signal was assessed using a one-signal output method with repeated training and testing for each classifier. Cross-classifiers performed significantly better than chance in four of six patients and an overall mean AUC of 0.602 ± 0.126 (mean ± std). A mean time to warning of 37.386% ± 5.006% (mean ± std) and a sensitivity of 0.691 ± 0.068 (mean ± std) were observed in patients with results greater than chance. The analysis of the input channels showed a significant contribution (p