Population-Scale CT-based Body Composition Analysis of a Large Outpatient Population Using Deep Learning to Derive Age-, Sex-, and Race-specific Reference Curves

Abstract

Background Although CT-based body composition (BC) metrics may inform disease risk and outcomes, obtaining these metrics has been too resource intensive for large-scale use. Thus, population-wide distributions of BC remain uncertain. Purpose To demonstrate the validity of fully automated, deep learning BC analysis from abdominal CT examinations, to define demographically adjusted BC reference curves, and to illustrate the advantage of use of these curves compared with standard methods, along with their biologic significance in predicting survival. Materials and Methods After external validation and equivalency testing with manual segmentation, a fully automated deep learning BC analysis pipeline was applied to a cross-sectional population cohort that included any outpatient without a cardiovascular disease or cancer who underwent abdominal CT examination at one of three hospitals in 2012. Demographically adjusted population reference curves were generated for each BC area. The z scores derived from these curves were compared with sex-specific thresholds for sarcopenia by using χ² tests and used to predict 2-year survival in multivariable Cox proportional hazards models that included weight and body mass index (BMI). Results External validation showed excellent correlation (R = 0.99) and equivalency (P < .001) of the fully automated deep learning BC analysis method with manual segmentation. With use of the fully automated BC data from 12 128 outpatients (mean age, 52 years; 6936 [57%] women), age-, race-, and sex-normalized BC reference curves were generated. All BC areas varied significantly with these variables (P < .001 except for subcutaneous fat area vs age [P = .003]). Sex-specific thresholds for sarcopenia demonstrated that age and race bias were not present if z scores derived from the reference curves were used (P < .001). Skeletal muscle area z scores were significantly predictive of 2-year survival (P = .04) in combined models that included BMI. Conclusion Fully automated body composition (BC) metrics vary significantly by age, race, and sex. The z scores derived from reference curves for BC parameters better capture the demographic distribution of BC compared with standard methods and can help predict survival. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Summers in this issue.

Graphical abstract

Figure 1:

Diagram shows pipeline for fully automated body composition analysis of abdominal CT examinations. PACS = picture archiving and communication system.

Figure 2:

Images demonstrate examples of model output for deep learning body composition pipeline. A, Examples of automated slice selection. The left CT slice was automatically selected by the model; the right CT slice was selected by a radiologist. Graph shows output of L3 regression model. Two vertical lines represent slices selected by radiologist (light blue) and by algorithm (orange). Dotted red line is unfiltered prediction; fluctuations in this curve correspond to adjacent vertebral levels. Zero crossing of filtered prediction is chosen as L3 level by algorithm (solid red line). B, Two examples of automated segmentation results, one per row. Left image shows automated segmentation by the model, the middle image is the input CT slice, and the right image is manually segmented ground truth produced by the radiologist. Color interpretation for segmentation masks are as follows: background or other, black; skeletal muscle, brown; subcutaneous fat, yellow; and visceral fat, white.

Figure 3:

Distribution of body composition parameters according to race and sex. Box-and-whisker plots show, A, skeletal muscle area, B, visceral fat area, C, subcutaneous fat area, D, skeletal muscle index, E, visceral fat index, and, F, subcutaneous fat index in outpatient population in White non-Hispanic women (red), Black women (orange), White non-Hispanic men (yellow), and Black men (green). Line within box represents the median, box represents interquartile range, and dashed lines represent maximum and minimum values in our population.

Figure 4:

Graphs show reference curves of body composition areas for, A, White non-Hispanic female patients (n = 4971), B, White non-Hispanic male patients (n = 3963), C, Black female patients (n = 629), and, D, Black male patients (n = 319). From inferior to superior, lines represent third (dark blue), fifth (light blue), 10th (pink), 25th (purple), 50th (black), 75th (green), 90th (yellow), 95th (orange) and 97th (red) percentiles.

Figure 5:

Bar charts show comparison proportion of patients classified with sarcopenia with varying age and race categories by standard method with sex-specific skeletal muscle index (SMI) threshold for calculating sarcopenia (35% of patients in total) versus skeletal muscle index z-score threshold of −0.40 corresponding to 35% of population across age, race, and sex (dotted line).

Figure 6:

Graphs show hazard ratios (solid lines) with 95% CIs (dotted lines) for risk of death (n = 9752 with 201 death events). A, Simple Cox proportional hazards models. B–D, Multivariable Cox proportional hazards models for, B, all areas (n = 12 128), C, all areas and weight (n = 7651), and, D, all areas and body mass index (BMI) (n = 7210). SFA = subcutaneous fat area, SMA = skeletal muscle area, VFA = visceral fat area.