MRI-Based Radiomic Machine-Learning Model May Accurately Distinguish between Subjects with Internet Gaming Disorder and Healthy Controls

Abstract

Purpose To identify cerebral radiomic features related to the diagnosis of Internet gaming disorder (IGD) and construct a radiomics-based machine-learning model for IGD diagnosis. Methods A total of 59 treatment-naïve subjects with IGD and 69 age- and sex-matched healthy controls (HCs) were recruited and underwent anatomic and diffusion-tensor magnetic resonance imaging (MRI). The features of the morphometric properties of gray matter and diffusion properties of white matter were extracted for each participant. After excluding the noise feature with single-factor analysis of variance, the remaining 179 features were included in an all-relevant feature selection procedure within cross-validation loops to identify features with significant discriminative power. Random forest classifiers were constructed and evaluated based on the identified features. Results No overall differences in the total brain volume (1,555,295.64 ± 152,316.31 mm3 vs. 154,491.19 ± 151,241.11 mm3), total gray (709,119.83 ± 59,534.46 mm3 vs. 751,018.21 ± 58,611.32 mm3) and white (465,054.49 ± 51,862.65 mm3 vs. 470,600.22 ± 47,006.67 mm3) matter volumes, and subcortical region volume (63,882.71 ± 5110.42 mm3 vs. 64,764.36 ± 4332.33 mm3) between the IGD and HC groups were observed. The mean classification accuracy was 73%. An altered cortical shape in the bilateral fusiform, left rostral middle frontal (rMFG), left cuneus, left parsopercularis (IFG), and regions around the right uncinate fasciculus (UF) and left internal capsule (IC) contributed significantly to group discrimination. Conclusions: Our study found the brain morphology alterations between IGD subjects and HCs through a radiomics-based machine-learning method, which may help revealing underlying IGD-related neurobiology mechanisms.