1615-P: Adrenal Gland Volume and Development of Type 2 Diabetes Using Deep Learning Techniques

Objectives: Current insights into the dynamic between adrenal gland volume (AGV) and changes in blood glucose levels are insufficiently explored. This research aims to establish an association between AGV, evaluated through a deep learning-based method, and the risk of developing type 2 diabetes (T2D). Research Design and Methods: Our study included an analysis of 9,708 adults who underwent abdominopelvic CT scans during health examinations between 2011 and 2012. Eligibility for the study was contingent on the absence of adrenal nodules in participants. We obtained AGV measurements from CT scans using a three-dimensional nnU-Net deep learning framework. The research approach to understanding the AGV-T2D relationship was designed to be both cross-sectional and longitudinal. Results: The AI model demonstrated notable precision in segmenting the adrenal gland, with a correlation coefficient of 0.71 (standard deviation [SD], 0.11) and an average volume discrepancy of 0.6 mL (SD 0.9) in an external dataset. At the outset, individuals with T2D showed higher AGV compared to those without T2D (7.3 vs. 6.7 cm³ in males and 6.3 vs. 5.5 cm³ in females, all P <0.05). Furthermore, AGV was positively correlated with various metabolic factors, including waist size, area of abdominal visceral fat, triglyceride-glucose index, and levels of glycated hemoglobin. We determined optimal AGV cutoffs for T2D prediction to be 7.2 cm³ for males and 5.5 cm³ for females. Following a median 7-year observation period, 938 subjects were diagnosed with T2D. A significant observation was that individuals with higher AGV faced a greater cumulative risk of developing T2D, this finding remained significant after controlling for other known risk factors (hazard ratio [95% confidence interval], 1.268 [1.105-1.455]). Conclusions: In summary, AGV evaluated using deep learning techniques demonstrated a significant association with current diabetes status and can be considered a reliable predictor for T2D. Disclosure E. Ku: None. S. Yoon: Stock/Shareholder; MEDICAL IP. S. Park: None. J. Kim: None. J. Yoon: None. Funding This study was supported by the National Research Foundation of the Ministry of Science and ICT of Korea (Project No. NRF-2020R1C1C1010723) and a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute funded by the Ministry of Health and Welfare of the Republic of Korea (Project No. HI21C0032 and HI22C0049).