EEG microstate analysis in children with prolonged disorders of consciousness

Prolonged disorders of consciousness (pDoC) in children lack objective and effective diagnostic methods to assess consciousness states, hindering targeted treatment selection and delaying recovery. It remains unclear whether EEG microstate analysis, a method capturing spatial and temporal dynamics of brain activity, can serve as a reliable tool to differentiate consciousness states in children with pDoC, warranting further investigation. Resting-state EEG data (32 channels) were collected over 10 minutes from 45 children, divided into three groups (n=15 each): Vegetative State/Unresponsive Wakefulness Syndrome (VS/UWS; 7 females, 5.9±1.2 years), Minimally Conscious State (MCS; 7 females, 5.7±1.0 years), and healthy controls (HC; 7 females, 5.8±1.3 years). Spatial and temporal properties of EEG microstates were compared across groups. Correlation analysis examined relationships between microstate parameters and Coma Recovery Scale-Revised (CRS-R) scores in children with pDoC. Support vector machine (SVM) models were trained using combined temporal and spatial microstate features, optimized via grid search and random search algorithms. Model performance was evaluated using standard metrics, and features were ranked by permutation importance. CRS-R scores differed significantly between VS/UWS and MCS (p < 0.001). The microstate (MS) C in VS/UWS showed only 1.1% topographic similarity to standard templates. HC showed stronger connectivity for MS A, C, and D, while MCS exhibited stronger functional connectivity for MS C and D compared to VS/UWS. VS/UWS had longer MS B duration, shorter MS C duration, and lower MS C coverage than MCS and HC (p < 0.05). CRS-R scores showed moderate correlations with MS B duration (r=-0.504, P=0.005), MS C coverage (r=0.679, P<0.001), and occurrence of MS C (r=0.744, P<0.001) and D (r=-0.709, P<0.001), but weak correlations with MS C duration (r=0.488, P=0.006) and MS B coverage (r=-0.376, P=0.041). Particle Swarm Optimization Support Vector Machine (PSO-SVM) classification outperformed Grid Search SVM (GS-SVM) and Quantum PSO-SVM (QPSO-SVM) (area under the curve [AUC] = 0.722), with MS C occurrence ranked as the top feature . This study demonstrates that EEG microstate analysis is an objective, user-friendly tool for differentiating consciousness states in children with pDoC. Machine learning algorithms, specifically support vector machines, revealed that MS C occurrence is a potential neurophysiological biomarker.