An efficient multimodal electroencephalography and functional near-infrared spectroscopy approach using stroop task to evaluate internet gaming disorder with machine learning methods: A comprehensive study

Internet Gaming Disorder (IGD) has emerged globally in recent decades with increasing severity. Despite its growing risk to public health, the neurobiological mechanisms underlying IGD remain unclear. Thus, underpinning the diagnosis of IGD based on neurophysiological data using machine learning (ML) techniques is highly valuable. 102 male university students, including 51 IGD and 51 healthy controls aged 18–23, participated in the study. A classic Stroop task with congruent, incongruent, and neutral stimuli was presented during simultaneous electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS). Features extracted from EEG sub-bands included statistical, complexity, and frequency domain features, amplitude, and latency information from event related potential (ERP) components, as well as oxygenation changes from fNIRS. The features were selected based on statistical significance using a t-test; subsequent classification was performed over the well-known ML. EEG, ERP, and fNIRS features were compared between the groups and provided valuable insight into IGD, the best classification performance obtained with Support Vector Machines. The accuracy was 79.4 % for EEG and 87.25 % for a combination of EEG and fNIRS systems. The results indicate that combining EEG and fNIRS modalities provides better and more robust performance. Examining linear and nonlinear EEG features, ERP component features, and fNIRS hemodynamic activation offers comprehensive analysis and potential biomarkers. This study aimed to develop an ML based diagnosis of IGD in male university students using multimodal EEG and fNIRS data recorded during a Stroop task. This study investigated IGD for the first time using multimodal EEG-fNIRS responses elicited by the Stroop task.