Tuesday, November 28 | 1:40 p.m.-1:50 p.m. | T6-SSCH06-2 | Room E350
Using deep learning with chest CT to assess muscle composition can help predict whether patients with pneumonia and known chronic obstructive pulmonary disease (COPD) will require hospitalization, research to be presented Tuesday afternoon has found.
Hamza Abad, MD, PhD, and colleagues analyzed data from a study called the Multi-Ethnic Study of Atherosclerosis, which included CT exams conducted between 2010 and 2012; 3,031 were used to develop a deep-learning algorithm for chest muscle 2D segmentation and 2,595 exams used to validate it. The group assessed any associations between chest muscle composition and risk of hospitalization for pneumonia using Cox Proportional Hazards Models adjusted for age, sex, physical activity, race, smoking history, hypertension, diabetes, and cholesterol levels.
Abad and colleagues reported that in the overall study cohort, chest muscle measurements didn't predict incident pneumonia, but among 507 individuals with COPD, deep learning-derived CT measures -- such as extramyocellular fat index -- predicted incident pneumonia (p = 0.02).
"Deep-learning algorithms may opportunistically measure pectoralis muscle composition to predict downstream adverse health outcomes," they concluded.