Machine learning model based on spontaneous brain activity and functional connectivity for identifying patients with internet gaming disorder.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, The Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, 646000, China.
- Department of Psychiatry, The Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, 646000, China. Electronic address: [email protected].
- Department of Radiology, The Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, 646000, China. Electronic address: [email protected].
Abstract
Internet gaming disorder (IGD) has been identified as a significant global mental health concern, emphasising the necessity for early diagnosis and intervention. This study aims to integrate multiple brain function metrics and machine learning algorithms to develop an advanced model for identifying IGD patients. The amplitude of low-frequency fluctuations (ALFF), fractional ALFF (fALFF), degree centrality (DC), regional homogeneity (ReHo) and voxel-mirrored homotopic connectivity (VMHC) from functional magnetic resonance imaging were extracted according to the anatomical automatic labelling (AAL) atlas from 244 IGD patients and 212 healthy controls (HCs). Combinations of 4 feature selection methods (analysis of variance (ANOVA), Kruskal‒Wallis (KW), Relief, and recursive feature elimination (REF)) and 10 classification algorithms were used to determine the most robust combinatorial model. Furthermore, a nomogram integrating functional metric features derived from the optimal model with independent clinical predictors was developed and evaluated via calibration curve and decision curve analyses (DCAs). The DC model derived from the bilateral olfactory cortex, bilateral insula and left orbitofrontal cortex (OFC), which combines the KW selector and Gaussian process (GP) classifier, and the ReHo model derived from the bilateral OFC, left olfactory cortex, left insula and left rectus, which combines the Relief selector and GP classifier, demonstrated superior performance in identifying IGD patients. The areas under the 0.971 and 0.989, respectively; corresponding AUCs for the ReHo model were 0.985 and 0.980. The internet addiction test (IAT) score was identified as an independent clinical predictor. The calibration curve and DCA demonstrated that the nomogram integrating DC with the IAT exhibited excellent reliability and significant net benefit (C index: 0.994). The nomogram based on DC and the IAT yields satisfactory classification performance and provides an effective tool for identifying IGD patients.