Volumetric assessment from L2 to L4 improves the reliability and validity of body composition by CT

Body composition (skeletal muscle, adipose tissue, and bone) is a general measure of health that can be determined from abdominal computed tomography (CT) using deep learning (DL) algorithms. Use of a volumetric approach, rather than the more commonly used single axial slice, likely improves the reliability and validity of body composition biomarkers, but a standardized approach is needed.

A study was designed to test the reliability of volumetric body composition assessment in serial CTs. Residents of Southern Minnesota and Western Wisconsin who had two consecutive abdominal CT exams 1 to 14 days apart from 2010 to 2021 were identified. Previously-validated DL models were applied to both CT exams, and three methods of determining body composition biomarkers were compared: two-dimensional (2D) model on a slice at the midpoint of L3, three-dimensional (3D) model on a slice at the midpoint of L3, and 3D model on slices extending from L2 to L4. For each method, reliability was assessed by the intraclass correlation coefficient (ICC) separately in paired noncontrast and paired contrast-enhanced CT exams and validity was assessed by the strength of association of each body composition measure with age.

There were 3,376 persons with paired noncontrast and 9,103 persons with paired contrast-enhanced CT exams. The 3D body composition biomarkers from L2-L4 showed generally better reliability than the 2D or 3D biomarkers at the midpoint of L3 with higher ICCs observed with or without use of contrast. The 3D body composition more strongly correlated with age than 2D body composition for most biomarkers. The 3D bone area from L2-L4 more strongly correlated with age than 3D bone area at the midpoint of L3.

Volumetric L2-L4 body composition analysis shows better reliability and validity than single slice L3 analysis, and this may result in more accurate clinical associations with body composition.