Auditory paradigm-dependent EEG feature screening and representation for distinguishing consciousness states in male patients with disorders of consciousness.

Introduction: Accurate assessment of disorders of consciousness (DoC), including minimally conscious state (MCS) and unresponsive wakefulness syndrome (UWS/VS), remains a major challenge in clinical neurorehabilitation. EEG-based biomarkers offer a non-invasive approach for characterizing residual brain activity, yet their discriminative sensitivity may depend on auditory stimulus design. Methods: EEG recordings were obtained from 15 male patients with DoC during four auditory stimulation paradigms. Nonlinear, spectral power, and graph-theoretic features were extracted, yielding 136 candidate features. All individual features were independently evaluated using patient-level leave-one-out cross-validation (LOOCV) to characterize paradigm- and feature-domain-specific predictive performance. In a complementary exploratory optimization analysis, univariate statistical screening identified 26 features with nominal group differences ( Results: In the single-feature benchmark, the reversed-word sequence paradigm showed relatively higher averaged classification performance, while nonlinear features demonstrated comparatively stronger classification characteristics than spectral power and graph-theoretic features. In the optimization analysis, the PCA-derived representation achieved competitive classification performance, whereas the direct combination of all statistically screened features achieved the highest observed classification accuracy. Regression performance was generally limited. Discussion: These findings suggest that auditory stimulus design and EEG feature representation may influence the discrimination of consciousness states in DoC. In this small-sample exploratory study, reversed-word auditory stimulation and nonlinear EEG features showed promising discriminative characteristics, while direct feature combination provided additional predictive information for categorical classification. Further validation in larger and independent cohorts is required.