Computed tomography derived cardiometabolic biomarkers in high-risk breast cancer using fully automated deep learning algorithms.
Authors
Affiliations (6)
Affiliations (6)
- Johns Hopkins University School of Medicine, Baltimore, MD, USA.
- Department of Anesthesiology, University of California, San Francisco, CA, USA.
- Department of Oncology, Johns Hopkins University School of Medicine, Baltimore, MD, USA.
- Department of Statistics, University of Florida, Gainesville, FL, USA.
- Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Radiology and Imaging Sciences, National Institutes for Health Clinical Center, Bethesda, MD, USA.
- Department of Oncology, Johns Hopkins University School of Medicine, Baltimore, MD, USA. [email protected].
Abstract
Cardiometabolic disease and obesity are important factors when considering breast cancer risk and survivorship. We aimed to utilize a fully automated deep learning algorithm to evaluate CT-derived cardiometabolic biomarkers in patients with breast cancer. We applied fully automated deep learning algorithms to derive cardiometabolic markers from baseline computed tomography (CT) abdomen scans of females with breast cancer diagnosed between 2016 - 2018 at Johns Hopkins. Among 127 females (64.6% postmenopausal; 57.5% White and 26.8% Black race), BMI (body mass index) was positively correlated with CT-derived adipose tissue area and muscle area, and negatively correlated with liver intensity, but not associated with presence of aortic calcification. However, myosteatosis and menopausal status outperformed BMI as clinical markers of cardiometabolic health in breast cancer, as they were associated with both adiposity (especially visceral adiposity) and the presence of aortic calcification, with myosteatosis also being associated with liver steatosis. Artificial intelligence algorithms can feasibly and directly quantify body composition, liver fat, and abdominal aortic plaque, and be a tool in evaluating metabolic disease risk in patients with breast cancer.