Related Experiment Video
Updated: Sep 28, 2025
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Fully Automated Abdominal CT Biomarkers for Type 2 Diabetes Using Deep Learning
Hima Tallam1, Daniel C Elton1, Sungwon Lee1
1From the Department of Radiology and Imaging Sciences (H.T., D.C.E., S.L., R.M.S.) and Department of Biostatistics and Clinical Epidemiology Service (P.W.), Clinical Center, National Institutes of Health, 10 Center Dr, Bldg 10, Room 1C224D, MSC 1182, Bethesda, MD 20892-1182; and Department of Radiology, University of Wisconsin School of Medicine and Public Health, Madison, Wis (P.J.P.).
Deep learning analysis of abdominal CT scans identified key biomarkers, including pancreatic fat and visceral fat, for diagnosing type 2 diabetes mellitus. These CT-based markers showed strong predictive power, even outperforming models that included clinical data.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Abdominal CT biomarkers, both within and outside the pancreas, show potential for diagnosing type 2 diabetes mellitus.
- Previous research was limited by manual methods and small sample sizes.
Purpose of the Study:
- To investigate abdominal CT biomarkers for type 2 diabetes mellitus using a large dataset and fully automated deep learning.
- To assess the diagnostic performance of CT-derived pancreatic and extrapancreatic features.
Main Methods:
- Retrospective collection of noncontrast abdominal CT images from 8992 patients.
- Automated deep learning for pancreas segmentation and measurement of CT attenuation, volume, fat content, and fractal dimension.
- Assessment of visceral fat, atherosclerotic plaque, liver/muscle CT attenuation, and muscle volume; multivariable logistic regression analysis.
Main Results:
- Deep learning model achieved comparable pancreas segmentation to interobserver variability (Dice similarity coefficient 0.69).
- Patients with type 2 diabetes mellitus exhibited lower pancreatic CT attenuation and greater visceral fat volume (P < .0001).
- Pancreatic attenuation decreased progressively with longer diabetes duration. CT-based models showed high predictive performance (AUCs 0.81-0.85), with clinical data not significantly improving AUC.
Conclusions:
- Abdominal CT biomarkers, particularly pancreatic CT attenuation and visceral fat measures, are associated with type 2 diabetes mellitus diagnosis.
- Automated deep learning analysis of CT scans offers a robust method for identifying diabetes-related biomarkers.
- CT-based biomarkers, including intrapancreatic fat percentage and pancreatic fractal dimension, are significant predictors of type 2 diabetes mellitus.
