JCSM Rapid Communications (Jul 2021)

Sarcopenia diagnosis: comparison of automated with manual computed tomography segmentation in clinical routine

  • Louise Caudron,
  • Alexandre Bussy,
  • Svetlana Artemova,
  • Katia Charrière,
  • Salma El Lakkiss,
  • Alexandre Moreau‐Gaudry,
  • Jean‐Luc Bosson,
  • Gilbert R. Ferretti,
  • Eric Fontaine,
  • Cécile Bétry

DOI
https://doi.org/10.1002/rco2.37
Journal volume & issue
Vol. 4, no. 2
pp. 103 – 110

Abstract

Read online

Abstract Background Cross‐sectional muscle area (CSMA) at the mid third lumbar vertebra (L3) can be used for sarcopenia diagnosis. The measurement of CSMA is time‐consuming and thus restricted to clinical research. We aimed to compare the automatic module ABACS (Automatic Body composition Analyser using Computed tomography image Segmentation software) with manual segmentation for CSMA assessment into clinical routine. Methods The study population was screened retrospectively from a computed tomography‐scan (CT‐scan) database. All consecutive participants, hospitalized at the Grenoble University Hospital (CHU Grenoble Alpes) between January and May 2018, and with an abdominal CT‐scan including sagittal reconstruction were included. The software SliceOmatic complemented with the module ABACS (ABACS‐SliceOmatic) was compared with the software ImageJ. Their agreement was determined using Lin's concordance correlation coefficient and visualized in Bland–Altman plots for the CSMA measurement or with Cohen's kappa coefficient (κ) for sarcopenia status. Results Data from 680 participants were analysed (mean age 59 ± 19 years, %females: 45.7). The concordance correlation coefficient between both types of software was 0.93 (CI95%: 0.92 to 0.94). Mean CSMA was significantly higher with ABACS‐SliceOmatic (mean difference: 6.51 ± 10.50 cm2; P < 0.001). Kappa agreement for sarcopenia diagnosis was moderate: 0.68 (CI95%: 0.62–0.74) and 0.71 (CI95%: 0.65–0.76) for Prado's and Derstine's definitions, respectively. Conclusions ABACS‐SliceOmatic has moderate agreement with the manual software ImageJ in a routine clinical database. Our work suggests that ABACS‐SliceOmatic should be used with caution in clinical practice. To improve its reliability, we suggest to manually validate the automatic segmentation.

Keywords