Weak supervision as an efficient approach for automated seizure detection in electroencephalography

Abstract

Automated seizure detection from electroencephalography (EEG) would improve the quality of patient care while reducing medical costs, but achieving reliably high performance across patients has proven difficult. Convolutional Neural Networks (CNNs) show promise in addressing this problem, but they are limited by a lack of large labeled training datasets. We propose using imperfect but plentiful archived annotations to train CNNs for automated, real-time EEG seizure detection across patients. While these weak annotations indicate possible seizures with precision scores as low as 0.37, they are commonly produced in large volumes within existing clinical workflows by a mixed group of technicians, fellows, students, and board-certified epileptologists. We find that CNNs trained using such weak annotations achieve Area Under the Receiver Operating Characteristic curve (AUROC) values of 0.93 and 0.94 for pediatric and adult seizure onset detection, respectively. Compared to currently deployed clinical software, our model provides a 31% increase (18 points) in F1-score for pediatric patients and a 17% increase (11 points) for adult patients. These results demonstrate that weak annotations, which are sustainably collected via existing clinical workflows, can be leveraged to produce clinically useful seizure detection models.

Continental generalization of an AI system for clinical seizure recognition

“…Thus we did a prospective study on the RPAH dataset, trained our model on the TUH training dataset, and ran the inference on the RPAH dataset. First, we test our model on the TUH dataset and get a 0.84 AUROC score (show on Table II) using 12-s window, which is better than the performance achieved by Khaled et al [15]. Although the AUC score only improve by 0.06, we use a 12s window, which is 5 times shorter than the Khaled et al.…”

“…We test the deep neural network (DNN) on the TUH Development dataset and do a prospective study on the RPAH dataset. For the TUH dataset test, we compare our performance with the Khaled et al [15], where we only use 12-s input but still improve 6% average AUC. We also test AUC on the RPAH dataset, which reaches 0.82.…”

“…The need for clinical care has spurred the emergence of automated nonpatient specific seizure detection algorithms. Among them, deep learning methods provide more accurate and promising ideas for this problem [10], [11], [14], [15]. However, exist-ing techniques and solutions still cannot meet the minimum requirements for clinical usage.…”

“…As far as we know, there are a few commercialised prospective seizure documentation tools such as Persyst [21] and EpiScan [18], [22]. Our understanding of the two tools, as well as other research reports [15], suggest EpiScan achieves a low sensitivity of 72%, which achieves good control of the false-alarm rate. However, the sensitivity for epilepsy in adults is critical, as they statistically experience lesser seizure frequencies.…”

Automatic Electrophysiological Noise Reduction and Epileptic Seizure Detection for Stereoelectroencephalography

“…Second, as mentioned in the introduction, one weakness of a current machine learning-based method was the labor and time costs required for labeling. Saab et al [21] proved that semi-supervised learning is applicable to automatic seizure detection. Consequently, the current approach could be used as a collector to automatically include high-quality true SOZ into the seizure library, then experts could select most representative ones from the database to serve as the training set for the semi-supervised learning technique, which could be beneficial for further improve the reliability of the detector.…”