Deep Learning-Based 3D CT Body Composition Is Associated With Mortality in Patients Undergoing Transcatheter Aortic Valve Replacement.
Authors
Affiliations (19)
Affiliations (19)
- Cardiovascular Imaging Research Center (CIRC), Department of Radiology (I.H., J.M.B., M.J., L.H.C., S.E., I.L.L., M.C.L., S.A., V.K.R., B.G., S.H., A.T., M.T.L., B.F.), Massachusetts General Hospital, Harvard Medical School, Boston.
- Radiology and Nuclear Medicine, CARIM & GROW, Maastricht University, the Netherlands (I.H., H.J.W.L.A.).
- Department of Radiology, University of Tuebingen, Germany (J.M.B., K.N.).
- Center for Preventive Medicine and Digital Health, University of Heidelberg, Mannheim, Germany (T.M.).
- Department of Diagnostic and Interventional Radiology, University Medical Center Freiburg, Faculty of Medicine, University of Freiburg, Germany (M.J.).
- Metabolism Unit (M.K.), Massachusetts General Hospital, Harvard Medical School, Boston.
- Gottsegen National Cardiovascular Center, Hungary (M.K.).
- Physiological Controls Research Center, University Research and Innovation Center, Óbuda University, Hungary (M.K.).
- Heart and Vascular Institute (E.Y., N.P.), Massachusetts General Hospital, Harvard Medical School, Boston.
- Department of Nuclear Medicine, Cardiac Imaging (S.E.), University Hospital Zurich, Switzerland.
- Department of Endocrinology and Clinical Nutrition (S.E.), University Hospital Zurich, Switzerland.
- Department of Cardiology, University Heart Center, University Hospital Zurich, Switzerland (S.E.).
- Medical Imaging Centre, Semmelweis University, Hungary (V.J.).
- Clinic of Diagnostic and Interventional Radiology, University Hospital Augsburg, Germany (I.L.L., M.C.L.).
- Cardiology Division (S.A., A.T.), Massachusetts General Hospital, Harvard Medical School, Boston.
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA (V.K.R., H.J.W.L.A., B.F.).
- Duke Clinical Research Institute (P.S.D.), Duke University School of Medicine, Durham, NC.
- Division of Cardiology, Department of Medicine (P.S.D.), Duke University School of Medicine, Durham, NC.
- Department of Radiation Oncology, Brigham and Women's Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA (H.J.W.L.A.).
Abstract
Recent advances in artificial intelligence enable fully automated 3-dimensional computed tomography (CT)-based body composition quantification. We investigated whether automated CT-derived thoraco-abdominopelvic body composition measures are associated with mortality after transcatheter aortic valve replacement. In this retrospective multicenter study, consecutive patients undergoing transcatheter aortic valve replacement with preprocedural CT imaging were included between 2014 and 2023. A deep learning-based pipeline performed automated 3-dimensional segmentation of thoraco-abdominopelvic noncontrast CT scans to quantify volumes (L/m<sup>2</sup>) and attenuations (Hounsfield units) of subcutaneous adipose tissue, visceral adipose tissue, muscular adipose tissue, and lean skeletal muscle, indexed to body surface area. Associations with mortality were evaluated using Cox models and Harrell C statistics, adjusting for demographics (age, sex, body mass index) and cardiometabolic risk factors (hypertension, diabetes, chronic renal failure). Among 915 patients (median age, 84 years [interquartile range, 77-90]; 58% men), 160 (17.5%) died over a median follow-up of 22 months (interquartile range, 14-36). Restricted cubic spline analyses demonstrated that lower subcutaneous adipose tissue volume and higher adipose tissue attenuation were associated with increased mortality, whereas higher skeletal muscle volume and attenuation were associated with improved survival. Addition of body composition measures provided incremental discriminatory value beyond demographics and cardiometabolic risk factors (Harrell C, 0.66 versus 0.71; <i>P</i><0.001). Automated 3-dimensional CT-derived body composition measures provide independent prognostic information and improve risk stratification for mortality after transcatheter aortic valve replacement. Low subcutaneous adipose tissue volume and higher adipose tissue attenuation identify patients at increased risk, whereas higher skeletal muscle volume and attenuation are associated with lower mortality.