Comprehensive Summary
This study, conducted by Mao et al., investigated how multi-dimensional electroencephalogram (EEG) features—relative power (RP), fuzzy entropy (FuzEn), and functional connectivity (FC)—combined with machine learning could support more objective and accurate diagnosis in pediatric schizophrenia. EEG data were obtained from 45 male patients with pediatric schizophrenia (ages 10 years and 8 months to 14 years) and 39 healthy, age-matched subjects (ages 11 years to 13 years and 9 months). EEG features were analyzed using Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), and Categorical Boosting (CatBoost), then selected with recursive feature elimination (RFE). Results showed that the CatBoost yielded the most optimal results with a 4-s window length with 99.60% accuracy. Pediatric schizophrenic patients exhibited abnormal patterns across frequency bands in the occipital and temporal regions, reduced fuzzy entropy patterns in alpha, beta, gamma, and theta bands, and widespread brain dysconnectivity across brain regions. These findings suggest brain dysconnectivity in pediatric schizophrenia is a core feature, underlying cognitive and perceptual disturbances. CatBoost was found to be the most accurate in diagnosing pediatric schizophrenia.
Outcomes and Implications
Pediatric schizophrenia is difficult to diagnose, but combining traditional clinical assessments with objective assessments such as CatBoost can improve diagnostic accuracy. These findings suggest that EEG biomarkers may aid in early diagnosis and treatment of schizophrenia. Further research may need to be conducted in order to generalize these findings, as this study was conducted with a relatively small, limited cohort (male only).