Topographic differences in EEG microstates: distinguishing juvenile myoclonic epilepsy from frontal lobe epilepsy

This study aims to develop an exploratory classification model for Juvenile Myoclonic Epilepsy (JME) based on electroencephalogram (EEG) microstate features to assist clinical diagnosis and reduce misdiagnosis rates. A total of 123 participants were included in this study, consisting of 74 patients diagnosed with JME and 49 patients with Frontal Lobe Epilepsy (FLE). Resting-state EEG data were retrospectively collected from all participants. After preprocessing, microstate analysis was performed, and 24 microstate features (including duration, occurrence rate, coverage, and transition probability) were extracted and analyzed. Finally, the extracted microstate parameters were used to train six machine learning classifiers to distinguish between the two types of epilepsy. The performance of these models was assessed by calculating accuracy, precision, recall, F1 score, and area under the curve (AUC). The study found that all parameters of microstate A showed high consistency between the two groups. However, the JME group exhibited lower occurrence and smaller coverage of microstate B compared to the FLE group, while showing longer durations for microstate C. Additionally, the transition probabilities from microstate B to C and D were lower in the JME group, while the transition probability from C to D was significantly higher. When EEG microstate features were integrated into the six machine learning classifiers, the linear discriminant analysis (LDA) algorithm achieved the best classification performance (accuracy of 76.4%, precision of 79.5%, and AUC of 0.817). This study found significant differences in EEG microstate characteristics between JME and FLE. Based on 24 microstate features, a classification model was successfully developed and validated. These findings underscore the potential of EEG microstates as neurophysiological biomarkers for distinguishing between these two epilepsy types.

The online version contains supplementary material available at 10.1007/s11571-025-10255-9.