Artificial Intelligence-Based Body Composition Analysis Using Computed Tomography Images Predicts Both Prevalence and Incidence of Diabetes-A Cross-Sectional and Longitudinal Analysis of 10-Year Retrospective Cohort in Korea

Background: We aimed to assess how the ageing-related changes in body composition contribute to the prevalence and incidence of diabetes using artificial intelligence (AI)-based analysis of abdominal computed tomography (CT) images. Methods: In this retrospective cohort study, we identified 15330 subjects age ≥18 years with abdominal CT scans at baseline, who underwent medical checkup at Seoul National University Hospital Healthcare System Gangnam Center from January 1, 2011 to September 30, 2012. Of these, 11693 subjects with available follow-up data were included in the longitudinal analysis. The volume of each body segment involved in abdominal CT images was measured by using an AI-based image analysis software. Findings: The ratio of visceral fat to subcutaneous fat (VF/SF ratio) increased with ageing. The optimal cutoffs of VF/SF ratio to predict the prevalence of diabetes were 1.2 and 0.5 in men and women, respectively. A VF/SF ratio over the cutoff was associated with a higher prevalence of diabetes (age and BMI adjusted OR 2.1, [95% CI 1.8-2.4] in men; 3.1, [2.4-3.9] in women). The same cutoffs of VF/SF ratio were used to predict incident diabetes in ten years of follow-up. Subjects with normal glucose tolerance at baseline who had higher VF/SF ratio had increased risk of progression to prediabetes or diabetes (age and BMI adjusted HR 1.2, [95% CI 1.1-1.4] in men; 1.4, [1.2-1.6] in women). Subjects with prediabetes at baseline who had higher VF/SF ratio also more frequently progressed to diabetes (age and BMI adjusted HR 1.4, [95% CI 1.2-1.6] in men; 1.8, [1.5-2.3] in women). Interpretation: VF/SF ratio in the abdomen change with ageing and are associated with the prevalence and future incidence of diabetes. AI-based analysis of abdominal CT images may help easily obtain body composition data to clinically assess the risk of incident diabetes. Disclosure Y.Kim: None. H.Son: None. J.Yoon: None. H.Choe: None. T.Oh: None. Y.Cho: Consultant; LG Chem. Funding Ministry of Health & Welfare, Republic of Korea