Quantification of abdominal fat from computed tomography using deep learning and its association with electronic health records in an academic biobank

Objective The objective was to develop a fully automated algorithm for abdominal fat segmentation and deploy this method at scale and associated with diagnoses in an academic biobank. Materials and Methods We built a fully automated image curation and labeling technique using deep learning and distributive computing to identify subcutaneous and visceral abdominal fat compartments from 47,587 CT scans in 13,422 patients in the Penn Medicine Biobank (PMBB). A classification network identified the inferior and superior borders of the abdomen, and a segmentation network differentiated visceral and subcutaneous fat. Following technical evaluation of our method, we conducted studies to validate known relationships with adiposity. Results When compared with 100 manually annotated cases, the classification network was on average within one 5 mm slice for both the superior (0.3±0.6 slices) and inferior (0.7±0.6 slices) borders. The segmentation network also demonstrated excellent performance with interclass correlation coefficients of 0.99 (p<2e-16) for subcutaneous and 0.99 (p<2e-16) for visceral fat on 100 testing cases. We performed integrative analyses of abdominal fat with the phenome extracted from the electronic health record and found highly significant associations with diabetes mellitus, hypertension, renal failure, among other phenotypes. Conclusion This work presents a fully automated and highly accurate method for the quantification of abdominal fat that can be applied to routine clinical imaging studies to fuel translation scientific discovery.