EEG microstate-based classification using machine learning in depressed adolescents with and without non-suicidal self-injury

Abstract

Non-suicidal self-injury (NSSI) is prevalent and concerning behavior among adolescents, particularly those diagnosed with major depressive disorder (MDD). The identification of the neural correlates associated with NSSI in this population is crucial for early detection and intervention. This study investigated the distinct electroencephalography (EEG) microstate patterns in adolescents with depression and NSSI. A total of 134 participants, including 44 MDD adolescents with NSSI, 41 MDD adolescents without NSSI, and 49 healthy controls (HCs) were enrolled, and their EEGs were recorded. Microstate analyses were conducted on all three groups. We performed a correlation analysis between the microstate parameters and psychological assessments. Machine learning algorithms were employed to classify the groups to identify patterns specific to the NSSI group. As for the global explained variance, duration, occurrence, and coverage, microstate B had significantly higher values in HCs than in MDD adolescents with and without NSSI. MDD adolescents with NSSI showed a significantly lower transition probability from class D to B than MDD adolescents without NSSI and HCs. Microstate parameters significantly correlated with psychological assessments. The best classification performances showed 77.42 % accuracy (95 % confidence interval 0.69-0.86, p < 0.001) between MDD adolescents with NSSI and HCs, 71.11 % accuracy (95 % confidence interval 0.61-0.80, p < 0.001) between MDD adolescents without NSSI and HCs, and 63.53 % (95 % confidence interval 0.53-0.74, p = 0.009) accuracy between MDD adolescents with NSSI and MDD adolescents without NSSI. This study investigated the utility of EEG microstate features as biomarkers of NSSI in adolescents with MDD. These findings help elucidate the neural mechanisms underlying NSSI and offer a potential biomarker for MDD adolescents with NSSI.