Key result
Automated deep learning accurately quantifies CT abdominal fat with a 1.00 correlation.
Why the study?
The study was conducted to develop a fully automated algorithm for abdominal fat segmentation and deploy it at scale in an academic biobank.
Does a fully automated deep learning algorithm accurately quantify abdominal fat from CT scans and associate with clinical phenotypes in biobank patients?
Population
13,502 patients with 52,844 CT scans in the Penn Medicine Biobank
Comparison
Deep learning automated segmentation vs manual annotation
Design
Algorithm development and validation study
Authors
Loading...
May enable scalable fat quantification from routine CTs; leaves open clinical utility pending prospective validation.
Observational (n=13,502)
Does a fully automated deep learning algorithm accurately quantify abdominal fat from CT scans and associate with clinical phenotypes in biobank patients?
Effect estimate: ICC 1.00
p-value: p=<2 x 10^-16
A fully automated deep learning method accurately quantifies abdominal fat from routine clinical CT scans and demonstrates significant associations with cardiometabolic diseases.
MacLean et al. (2021) conducted an observational in General biobank population (n=13,502). Deep learning algorithm for abdominal fat segmentation vs. Manual annotation was evaluated on Intraclass correlation coefficient for subcutaneous and visceral fat segmentation (ICC 1.00, p=<2 x 10^-16). A fully automated deep learning algorithm accurately quantified abdominal fat from CT scans, achieving intraclass correlation coefficients of 1.00 for both subcutaneous and visceral fat.
Synapse has enriched 4 closely related papers on similar clinical questions. Consider them for comparative context: