• AI
  • Molecular Imaging
  • CT
  • X-Ray
  • Ultrasound
  • MRI
  • Facility Management
  • Mammography

Could a Deep Learning System Facilitate the Diagnosis of Type 2 Diabetes on Abdominal CT?

Article

Employing a deep learning system for pancreatic segmentation, researchers found that intrapancreatic fat percentage and pancreatic fractal dimension were among the key predictors for type 2 diabetes mellitus in a multivariable analysis.

In a newly published retrospective study that incorporated deep learning technology and involved nearly 9,000 patients, researchers found that abdominal computed tomography (CT) biomarkers, such as pancreatic CT attenuation and visceral fat, were associated with the diagnosis of type 2 diabetes mellitus.

In the study of 8,992 patients (including 572 patients with type 2 diabetes), the study authors utilized a deep learning system to segment the pancreas and provide measurements of pancreas fractal dimension, CT attenuation and fat content. Other assessed biomarkers included visceral fat, liver CT attenuation and atherosclerotic plaque, according to the study, which was published in Radiology.

The researchers found that patients with diabetes had lower average pancreatic CT attenuation (mean of 18.74 HU vs. 29.99 HU) and greater visceral fat volume (mean of 235 mL vs. 130.9 mL) in comparison to those without diabetes. The study authors also noted that greater duration of diabetes also corresponded with a progressive decrease in pancreatic attenuation.

Pointing to the findings from the multivariable analysis, the study authors said the CT-derived factors had “considerable predictive power.” Five of the six optimal factors for predicting type 2 diabetes were CT measures of total and eroded volumes of intrapancreatic fat percentage, pancreas fractal dimension, average liver attenuation and plaque severity between the L1 and L4 vertebra levels.

“This proves the final model’s ability to discern patients with type 2 diabetes before and after diagnosis from participants without diabetes,” wrote Ronald M. Summers, MD, a senior investigator and director of the Imaging Biomarkers and Computer-Aided Diagnosis Laboratory with the National Institutes of Health (NIH) Clinical Center in Bethesda, Md., and colleagues.

The study authors also noted that machine-to-person variability for the deep learning technology was similar to interobserver variability.

In regard to the limitations of the retrospective study, the authors acknowledged that the timing of the CT scans varied considerably, ranging between 5,055 days prior to diabetes diagnosis to 4,822 days after a patient had been diagnosed. They also noted that disease stage could not be determined and that the final study model did not include factors such as family history, hypertension, and blood glucose levels.

While noting that further improvement is necessary when it comes to the clinical use of automated pancreas segmentation, Dr. Summers and colleagues said that CT biomarkers may have the potential to facilitate earlier diagnosis of type 2 diabetes.

Recent Videos
Does Initial CCTA Provide the Best Assessment of Stable Chest Pain?
Nina Kottler, MD, MS
Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients
Related Content
© 2024 MJH Life Sciences

All rights reserved.