JMIR Publications

ABSTRACT

Background:

Real-time monitoring of pediatric epileptic seizures is still a significant obstacle in clinical practice. Some investigations have delved into the potential of machine learning (ML) methods in the detection of epilepsy. However, there is limited systematic evidence to demonstrate its feasibility.

Objective:

As such, this study intends to elucidate the accuracy of ML in detecting epilepsy, providing evidence-based suggestions for the development of future real-time monitoring tools.

Methods:

PubMed, Cochrane, Embase, and Web of Science were thoroughly retrieved as of August 27, 2023. The risk of bias in the eligible studies was appraised utilizing the revised Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2). In the meta-analysis for ML, we pooled the results of both the training and validation sets and assessed the risk of overfitting, while for deep learning, we only summarized the data of the validation or test sets.

Results:

This systematic review included 28 original studies, with 15 studies on ML and 13 on deep learning. The pooled c-index, sensitivity (SEN), and specificity (SPE) of ML in the training set were 0.76 (95% CI: 0.69–0.82) 0.77 (95% CI: 0.73–0.80), and 0.74 (95% CI: 0.70–0.77), respectively. In the validation set, the pooled c-index, SEN, and SPE of ML were 0.73 (95% CI: 0.67–0.79), 0.88 (95% CI: 0.83–0.91), and 0.83 (95% CI: 0.71–0.90), respectively. Meanwhile, the pooled c-index of deep learning in the validation set was 0.91 (95% CI: 0.88–0.94), with SEN and SPE being 0.89 (95% CI: 0.85–0.91) and 0.91 (95% CI: 0.88–0.93), respectively.

Conclusions:

Our systematic review demonstrates promising accuracy of artificial intelligence methods in epilepsy detection. Deep learning appears to offer higher detection accuracy than traditional ML. These findings support the development of deep learning-based early-warning tools in future research.