DIAGNOSTIC ACCURACY AND RELIABILITY OF GEMINI AI FOR MENTAL HEALTH SCREENING

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Mental health problems among university students are increasing, while stigma and limited access to professional services often delay help-seeking. Artificial intelligence (AI)-based screening tools may provide an accessible approach for early mental health detection. This study aimed to evaluate the diagnostic accuracy and reliability of Gemini AI for mental health screening within Indonesia’s SATU SEHAT framework. A pilot cross-sectional diagnostic accuracy study was conducted among 63 undergraduate students. Participants completed mental health screening using both Gemini AI and the SATU SEHAT platform. Diagnostic performance was assessed using receiver operating characteristic (ROC) analysis, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), overall accuracy, and Cohen’s Kappa coefficient. Gemini AI demonstrated excellent discriminative ability with an area under the curve (AUC) of 0.957 (95% CI: 0.89–1.00). Sensitivity was 86.36%, specificity 97.56%, PPV 95.00%, NPV 93.02%, and overall accuracy 93.65%. Agreement with the SATU SEHAT reference standard was almost perfect (Cohen’s Kappa = 0.857, p < 0.001). Gemini AI showed high diagnostic accuracy and reliability for early mental health screening. These findings suggest that AI-assisted screening may complement existing digital mental health services by supporting early identification of individuals at risk and providing a foundation for larger validation studies.