Machine learning model based on spontaneous brain activity and functional connectivity for identifying patients with...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Internet Gaming Disorder (IGD) is a growing global mental health issue.
• Early diagnosis and intervention are crucial for managing IGD.
• Advanced computational models are needed for accurate IGD identification.

Purpose of the Study:

• To develop an advanced model for identifying IGD patients.
• To integrate multiple brain function metrics and machine learning algorithms.
• To create a reliable diagnostic tool for IGD.

Main Methods:

• Extracted brain function metrics (ALFF, fALFF, DC, ReHo, VMHC) from fMRI data of 244 IGD patients and 212 healthy controls.
• Employed feature selection methods (ANOVA, KW, Relief, REF) and classification algorithms (10 types).
• Developed and validated a nomogram integrating functional metrics and clinical predictors (IAT) using calibration curves and DCAs.

Main Results:

• Degree Centrality (DC) and Regional Homogeneity (ReHo) models showed superior performance in identifying IGD patients (AUCs up to 0.989).
• The Internet Addiction Test (IAT) score was a significant independent clinical predictor.
• A nomogram combining DC and IAT demonstrated excellent reliability and net benefit (C-index: 0.994).

Conclusions:

• The nomogram integrating DC and IAT provides a highly effective tool for classifying IGD patients.
• This approach offers satisfactory diagnostic performance for early identification and intervention.
• The study highlights the potential of combining neuroimaging and clinical data for IGD diagnosis.