AI Detects Diabetes on Abdominal CT Scan Performed for CTC Screening

Researchers at the University of Wisconsin School of Medicine and Public Health in Madison, WI, and the U.S. National Institutes of Health (NIH) Clinical Center in Bethesda, MD, have developed a segmentation method automated pancreatic for non-contrast CT images. They then tested the algorithm retrospectively on 9,000 consecutive patients who underwent CTC screening.

They found that the five most important CT biomarkers extracted using the deep learning-based segmentation model yielded an area under the curve (AUC) of 0.81 on a subset of patients whose CT exams were performed from zero to 2,550 days before they were diagnosed with type 2 diabetes.

“This study is a step toward the broader use of automated methods to address clinical challenges,” wrote the researchers led by first author Hima Tallam, a medical student and doctoral candidate at the University of Wisconsin. Dr. Perry Pickhardt of the University of Wisconsin and Dr. Ronald Summers, PhD, of the NIH Clinical Center served as co-lead authors.

A number of research studies have explored the usefulness of using CT image information to opportunistically screen for other conditions beyond the clinical indication of the study, including osteoporosis, cardiovascular events and metabolic syndrome. In the present study, researchers sought to investigate the utility of an automated deep learning approach to detect type 2 diabetes via CT biomarkers.

Researchers used data from 471 images – including contrast-enhanced CT images and contrast-free CT images – from a variety of public datasets to train a deep learning algorithm to segment the pancreas . Study participants had a diagnosis of diabetes that ranged from 5,055 days before diabetes diagnosis to 4,822 days after diagnosis.

In testing on a subset of 25 cases, deep learning-based pancreas segmentations were found to be accurate and reproducible, with a similar Dice coefficient of 0.69 equivalent to interobserver performance for two human readers.

Examples of segmentations of the pancreas on unenhanced abdominal axial computed tomography images in healthy participants and patients with type 2 diabetes mellitus. Images on the left are original CT images and images on the right show segmentations superimposed on the original CT images. (A) Images in a 61-year-old non-diabetic man with mean pancreatic CT attenuation of 35.50 HU ± 47.96 and pancreatic volume of 97.6 mL. (B) Images of a 59-year-old man with type 2 diabetes who was diagnosed 144 days before CT. The mean pancreatic CT attenuation was 20.66 HU ± 81.99 and the pancreatic volume was 77.10 mL. (C) Images in a 67-year-old man with type 2 diabetes who was diagnosed 595 days after CT. The mean pancreatic CT attenuation was 18.46 HU ± 48.30 and the pancreatic volume was 72.88 mL. The green area indicates complete segmentation and the yellow area indicates segmentation after erosion. Image and caption courtesy of RSNA.

Logistic regression analysis then showed that, on average, patients with type 2 diabetes had lower mean pancreas, muscle, and liver CT attenuation; higher standard deviation in pancreatic CT attenuation; and the fractal dimension of the lower pancreas. A progressive decrease in pancreatic attenuation was observed in patients with longer disease duration.

“These results indicate elevated amounts of peri-organ and intra-organ fat, which are related and consistent with previous work showing that diabetic patients tend to accumulate more visceral and intra-pancreatic fat than people without diabetes,” wrote the authors.

By using these CT biomarkers, the deep learning model was able to achieve a high level of accuracy in diagnosing type 2 diabetes.

Performance of Deep Learning Based CT Segmentation Model for Type 2 Diabetes Diagnosis
Detect diabetes from 0 to 2,550 days before diagnosis of diabetes AUC = 0.81
Detect diabetes from 0 to 2,550 days after diagnosis of diabetes AUC = 0.84

Adding clinical information to the model did not significantly improve performance.

The researchers said future work could focus on predicting type 2 diabetes in a prospective study. In addition, the current study may also inform future research on why the morphological characteristics of the pancreas change in patients with diabetes, they said.

“However, we ultimately hope that the CT biomarkers studied here may inform the diagnosis of the early stages of type 2 diabetes and allow patients to make lifestyle changes to alter the course,” the authors wrote.

Copyright © 2022

Comments are closed.