Thursday, December 5 | 8:30 a.m.-8:40 a.m. | R4-SSCH09-4 | Room E451A
In this Thursday morning session, researchers will describe how quantitative metrics extracted from chest CT scans using AI can be integrated with machine-learning algorithms to predict low bone mineral density (BMD).
Presenter Angelo Scanio of UT Southwestern Medical Center in Dallas and colleagues found that using what they call a "Light Gradient Boosting Machine Learning" (LightGBM) model with CT data showed significant efficacy for detecting low bone mineral density – outperforming the sensitivity of radiological reports.
Dual-energy x-ray absorptiometry (DEXA) is the gold standard for measuring BMD, but culling data on the condition opportunistically from CT exams shows promise as a mode of assessment. The researchers conducted a study that included 781 patients who underwent a noncontrast chest CT exam and a DEXA scan within a one-year period between September 2022 and December 2023. They used Hounsfield units of the trabecular bone from chest CT scans to train the LightGBM model to predict low BMD; they compared LightGBM with a logistic regression model (using only Hounsfield units at the most caudal thoracic spine vertebra for comparison).
The LightGBM model achieved a higher area under the receiver operating characteristics curve than the logistic regression model and showed higher accuracy and sensitivity for predicting low BMD compared with chest CT radiological reports, the group found.
"These findings advocate for a broader adoption of AI models in opportunistic screening for low BMD among patients undergoing chest CTs," the team concluded.
Stop by this Thursday morning session to get all of the details.