JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Assessing the Reliability of Pancreatic CT Imaging Biomarkers for Diabetes Prediction: A Dual Center Retrospective Study

  • 0David Geffen School of Medicine at UCLA, Los Angeles, California (A.S.); Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Radiology and Imaging Sciences, National Institutes of Health Clinical Center, Building 10, Room 1C224D MSC 1182, Bethesda, MD 20892-1182 (A.S., P.M., R.M.S.).
Academic Radiology +

|

March 22, 2025

|

March 22, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

Pancreatic imaging biomarkers from CT scans are reliable for predicting diabetes, even with variations in segmentation quality and contrast. These robust biomarkers show consistent predictive ability across different clinical settings.

Area Of Science

  • Radiology and Medical Imaging
  • Endocrinology and Diabetes Research
  • Computational Pathology

Background

  • Pancreatic imaging biomarkers on computed tomography (CT) scans are linked to diabetes.
  • Previous studies have not assessed the resilience of these biomarkers to variations in segmentation quality and contrast status.
  • Understanding this robustness is crucial for reliable diabetes prediction.

Purpose Of The Study

  • To evaluate the robustness of pancreatic imaging biomarkers to variations in segmentation quality and contrast status.
  • To determine how these factors affect the predictive ability of CT-derived imaging biomarkers for diabetes.
  • To assess the consistency of predictive performance across different segmentation algorithms and clinical scenarios.

Main Methods

  • Retrospective analysis of CT scans and HbA1c tests from 9772 patients.
  • Classification of patients into diabetic/incident and nondiabetic groups based on HbA1c.
  • Measurement of pancreatic imaging biomarkers (attenuation, fat fraction, fractal dimension) using three segmentation algorithms (TotalSegmentator, nnU-Net, DM-UNet).
  • Assessment of algorithm agreement and predictive ability for diabetes using generalized additive models.

Main Results

  • Attenuation-based imaging biomarkers demonstrated high algorithm agreement (ICC ≥0.93).
  • Models using these biomarkers showed good predictive performance for diabetes (AUC 0.84-0.91 overall).
  • Predictive ability remained consistent across contrast and noncontrast scans, with high positive (0.79-0.84) and negative (0.89-0.94) predictive values.

Conclusions

  • Attenuation-based CT imaging biomarkers are robust to segmentation algorithm variations.
  • These biomarkers maintain consistent predictive ability for diabetes across diverse clinical conditions.
  • CT-derived pancreatic imaging biomarkers show promise as a reliable tool for multi-institutional diabetes screening.

Keywords: