A diagnosis and prediction algorithm for juvenile myoclonic epilepsy based on clinical and quantitative EEG features.

Objective: To develop an objective ensemble machine learning model combining clinical features and quantitative EEG metrics (phase locking value [PLV] and multiscale sample entropy [MSE]) to support accurate diagnosis of juvenile myoclonic epilepsy (JME).

Methods: A total of 75 JME patients, 51 frontal lobe epilepsy (FLE) patients, and 30 normal controls were included. Eight clinical features, along with 684 PLV and 152 MSE features derived from EEG data, were extracted. Four models were constructed using ensemble XGBoost and GBDT classifiers, with performance evaluated through accuracy, precision, recall, F1-score, and AUC. The performance of these models was assessed using a five-fold cross-validation method. The Fisher Score method ranked the most influential features in the best-performing model.

Results: The combined model (clinical, PLV, and MSE features) achieved an accuracy of 85.26 % and an AUC of 0.97. Key features included specific PLV metrics and family history of epilepsy. Notably, the PLV of Fp2-O1 in the δ band (δ-PLV_Fp2-O1) significantly differed among JME, FLE, and normal controls.

Conclusion: The ensemble model effectively distinguished JME, and highlighted δ-PLV_Fp2-O1 as a potential distinguishing feature, paving the way for more objective diagnostic approaches.