AI model able to ID early signs of type 2 diabetes on imaging results

Radiologists have used an advanced AI algorithm to identify early signs of type 2 diabetes on CT scans, sharing their findings in Radiology.[1] The group hope its efforts can help lead to improved diabetes diagnoses in the future.

“CT is a potentially useful modality for diagnosing type 2 diabetes,” wrote first author Hima Tallam, BSE, an MD/PhD student with the National Institutes of Health in Bethesda, Maryland, and colleagues. “CT is already widely used in clinical practice and can provide information on morphologic characteristics of the pancreas ... CT biomarkers outside the pancreas may also play an important role.”

The time it takes to manually evaluate low-dose, non-contrast CT images of the pancreas can be quite significant, Tallam et al. noted. Hoping to find a solution to this challenge, the group used a series of 471 CT images to develop a deep learning model for manually detecting signs of type 2 diabetes; this included 424 images to train the model, eight to validate it and another 39 to test its effectiveness.

The authors then externally validated their AI model with colorectal screening data from nearly 9,000 patients treated at a single U.S. facility. The mean patient age was 57 years old, and 55% of patients were women. While 6.3% of patients had type 2 diabetes, an additional 20.9% presented with dysglycemia.

Overall, the deep learning model achieved a performance that was comparable to human radiologists. It also helped confirm that patients with a lower pancreas density and higher levels of visceral fat were associated with a higher risk of diabetes.

Intrapancreatic fat percentage, pancreas fractal dimensions, plaque severity between the L1-L4 vertebra level, average liver CT attenuation and patient BMI were also listed as key predictors of type 2 diabetes among the study’s patient population.

“This study is a step toward the wider use of automated methods to address clinical challenges,” the authors wrote. “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. 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.”

Related Heart Health Content:

COVID-19 linked to a higher risk of diabetes for at least 1 year

Verapamil shows potential as a long-term oral treatment for type 1 diabetes

New data on the link between diabetes and high blood pressure

Metformin boosts outcomes for patients with HF and diabetes—sulfonylurea initiation makes things worse

Reference:

1. Hima Tallam, Daniel C. Elton, Sungwon Lee, et al. Fully Automated Abdominal CT Biomarkers for Type 2 Diabetes Using Deep Learning. Radiology 2022.

Around the web

Brian Ghoshhajra, MD, division chief, cardiovascular imaging, Massachusetts General Hospital, explains what specialized training is needed to perform coronary computed tomography angiography (CCTA) as interest rapidly rises in this field. 

Brian Ghoshhajra, MD, MBA, division chief, cardiovascular imaging, Massachusetts General Hospital and associate professor of radiology, Harvard Medical School, and a board member of the Society of cardiovascular Computed Tomography (SCCT), explains the rapidly expanding interest in cardiac computed tomography (CT) under the new chest pain evaluation guidelines.

A new study found Americans are more likely to have surgery during a pandemic if they are vaccinated, the hospital staff are vaccinated, the surgery is urgent and the surgery is an outpatient procedure.

Trimed Popup
Trimed Popup