Deep-learning model detects diabetes on x-rays

Wednesday, November 30 | 12:45 p.m. - 1:15 p.m. | W5B-SPCH-9 | Learning Center - CH DPS
In this scientific poster session, researchers will share their experience with using a deep-learning model to detect diabetes from ambulatory frontal x-rays in a large clinical dataset.

A research team led by presenter Dr. Ayis Pyrros of Duly Health and Care in Chicago trained and tested a multitask deep-learning model on 303,604 frontal chest x-rays acquired between 2010 and 2021 at a single institution. The group then compared the discriminatory ability of the deep-learning model with the ground-truth label and with other predictors, such as body mass index (BMI) and social deprivation index (SDI).

The model's overall discriminatory ability for diabetes was an area under the curve (AUC) of 0.819 versus an AUC of 0.542 using SDI as a predictor, and an AUC of 0.686 using BMI, with all curves demonstrating comparative p-values < 0.001, according to the results.

"Chest radiography can be utilized opportunistically to aid in detecting patients with diabetes and prediabetes," Pyrros noted.

Check out this Wednesday session to get all the details.

Page 1 of 545
Next Page