AI Analysis of Pancreatic Scans Could Improve Type 2 Diabetes Diagnoses

Man using blood sugar measurement device to monitor type 2 diabetes
Credit: Elva Etienne/Getty Images

Research from the National Institutes of Health (NIH) in Bethesda shows artificial intelligence (AI)-driven analysis of computed tomography (CT) scans of the pancreas has the potential to improve diagnosis of type 2 diabetes.

Type 2 diabetes causes a huge health burden on both patients and healthcare providers. Around 13% of all U.S. adults have type 2 diabetes, with another 35% with ‘prediabetes’ at high risk for developing the condition.

Due to the slow onset of symptoms and potential of lifestyle changes as a treatment it is important to diagnose patients as soon as possible so treatment can be started and the development of symptoms slowed.

The use of different types of AI, such as deep learning, to aid medical diagnostics has really taken off in the last few years, thanks in large part to advances in image recognition technology. However, it has mostly been applied to diagnose and classify conditions such as cancer.

But CT imaging has the potential to provide a lot of information about the abdomen and pancreas. “Studies have shown that the pancreas of a patient with diabetes has a smaller volume than that of a person without diabetes, with an increased amount of intrapancreatic fat and hence lower CT attenuation,” explain the authors.

In this study, published in the journal Radiology, the researchers used abdominal CT images collected from 8992 patients at routine colorectal cancer screens between 2004 and 2016 to train a deep learning algorithm to predict type 2 diabetes. Of these, 572 had been diagnosed with type 2 diabetes and 1880 with dysglycemia, or blood sugar levels that go too low or too high.

The researchers used a deep learning method to segment the pancreas so measurements of interest can be collected. These included: CT attenuation, volume, fat content, and pancreas fractal dimension. Visceral fat, atherosclerotic plaque, liver and muscle CT attenuation, and muscle volume were also recorded by the team.

The results from the deep learning model did not significantly differ from manual analysis of the images. The findings showed that patients with diabetes had lower pancreatic density and higher visceral fat than those without the condition. These differences became more pronounced as the condition progressed.

“We found that diabetes was associated with the amount of fat within the pancreas and inside the patients’ abdomens,” said lead author Ronald Summers, senior investigator and staff radiologist at the NIH. “The more fat in those two locations, the more likely the patients were to have diabetes for a longer period of time.”

In the final model, the team found that intrapancreatic fat percentage, pancreas fractal dimension, plaque severity between the L1-L4 vertebra level, average liver CT attenuation, and body mass index were all predictors of type 2 diabetes.

“Future work may be focused on predicting type 2 diabetes in a prospective study. The current study may also inform future research on the reasons for the changes that occur in morphologic characteristics of the pancreas in patients with diabetes,” write the authors.

“However, we ultimately hope that the CT biomarkers investigated herein might inform diagnosis of early stages of type 2 diabetes and allow patients to make lifestyle changes to alter the course.”

This site uses Akismet to reduce spam. Learn how your comment data is processed.