Type 2 diabetes (
T2D) is a global public health concern due to its increasing prevalence. Risk assessment and early detection of
T2D are vital in improving individualsÂ’ health, reducing the burden on national health insurance, and enhancing well-being. This study leverages artificial intelligence, specifically eXtreme Gradient Boosting (XGBoost), to develop predictive models for
T2D based on genetic and medical imaging data. The models aim to establish a prediction model and identify high-risk subgroups for
T2D within a cohort of 68,769 Taiwan Biobank participants. The approach integrates the Polygenic Risk Score (PRS) and Multi-image Risk Score (MRS) with demographic factors and environmental exposures to assess
T2D risk. The model's performance is evaluated using the Area Under the Receiver Operating Curve (AUC). Results demonstrate that genetic information alone is insufficient for accurate
T2D prediction (AUC = 0.73), whereas medical imaging data, including abdominal ultrasonography, vertebral artery ultrasonography, bone density scan, and electrocardiography, significantly improves prediction accuracy (AUC = 0.89). The best-performing model integrates genetic, medical imaging, and demographic variables (AUC = 0.94), successfully identifying subgroups at high risk of developing
T2D. The study also presents an online risk assessment website for
T2D. In summary, this research represents the first integration of whole-genome and medical imaging data for
T2D risk assessment. The genetic-only model outperforms previous genetic prediction studies, and integrating genetic and medical imaging information significantly enhances AUC. By utilizing artificial intelligence to analyze genetic, medical imaging, and demographic factors, this study contributes to the early detection and precision health of
T2D.
Yi-Jia Huang1, Chun-houh Chen2, and Hsin-Chou Yang1,2,*
1Institute of Public Health, National Yang Ming Chiao Tung University, Taipei, Taiwan
2Institute of Statistical Science, Academia Sinica, Taipei, Taiwan
*Corresponding author
Hybrid: Meeting Room and Zoom Link to Follow
8 November 2023
9:00am