Abstract
SIGNIFICANCE: The accurate assessment and classification of residual consciousness are crucial for optimizing therapeutic interventions in patients with disorders of consciousness (DOCs). However, there remains an absence of effective and definitive diagnostic methods for DOC in clinical practice. AIM: The primary objective was to investigate the feasibility of utilizing resting state functional near-infrared spectroscopy (rs-fNIRS) for evaluating residual consciousness. The secondary objective was to explore the distinguishing characteristics that are more effective in differentiating between the unresponsive wakefulness syndrome (UWS) and the minimally conscious state (MCS) and to identify the machine learning model that offers superior classification accuracy. APPROACH: We utilized rs-fNIRS to evaluate the residual consciousness in patients with DOC. Specifically, rs-fNIRS was used to construct functional brain networks, and graph theory analysis was conducted to quantify the topological differences within these brain networks between MCS and UWS. After that, two classifiers were used to distinguish MCS from UWS. RESULTS: The graph theory results showed that the MCS group ( n = 8 ) exhibited significantly higher global efficiency ( Eg ) and smaller characteristic path length ( Lp ) than the UWS group ( n = 10 ). The functional connectivity results showed that the correlation within the left occipital cortex (L_OC) was significantly lower in the MCS group than in the UWS group. By using the indicators with significant differences as features for further classification, the accuracy for K -nearest neighbors and linear discriminant analysis classifiers was improved by 0.89 and 0.83, respectively. CONCLUSIONS: The resting state functional connectivity and graph theory analysis based on fNIRS has the potential to enhance the classification accuracy, providing valuable insights into the diagnosis of patients with DOC.