Agreement of Skeletal Muscle Area and Muscle Quality Between Two Different Image Analysis Software Tools

ABSTRACT

Background: Both clinicians and researchers have a growing interest in assessment of muscle mass utilizing diagnostic abdominal computed tomography (CT) scans. Different imaging analysis software tools exist for the assessment of muscle mass;however, minimal information is available to describe the agreement between tools. The objective of this project was to determine the agreement, reliability, and strength of the relationship between skeletal muscle cross-sectional area (CSA) and muscle quality at the third lumbar region (L3) between two different image analysis software tools (3D Slicer vs SliceOmatic) using a convenient sample of individuals who have undergone diagnostic abdominal CT scan imaging. Method(s) A retrospective sample of individuals who had undergone a diagnostic abdominal CT scan was utilized. For both image analysis software tools, L3 skeletal muscle CSA was identified using a Hounsfield Unit (HU) range of -30 to +150 and muscle quality was defined as the mean HU. For each patient, L3 skeletal muscle CSA (cm2) and mean HU was calculated using 3D Slicer (version 5.0.3) and SliceOmatic (version 4.3, TomoVision, Quebec, Canada). Lin's correlation coefficient (LCC), intraclass correlation coefficient (ICC), and Spearman correlation coefficient (SCC) were used to examine the agreement, reliability, and strength of the relationship with both L3 skeletal muscle CSA and muscle quality using3D Slicer versus SliceOmatic. Bland Altman plots were created to depict the agreement of L3 CSA and muscle quality between the two tools. Result(s) A total of 504 patients were included;the sample included 128 healthy adults and 376 patients who had the following diagnoses breast cancer (n = 175), colorectal cancer (n = 127), sepsis (n = 37) and COVID-19 (n = 37). The mean L3 skeletal muscle CSA measured using SliceOmatic was 140.6 +/- 36.0 cm2 and using 3D Slicer was 137.6 +/- 36.1 cm2. When examining the agreement, reliability, and strength of the relationship of L3 skeletal muscle CSA between SliceOmatic and 3D Slicer, LCC was 0.934 (p < 0.001), ICC was 0.968 (p < 0.001), and SCC was 0.930 (p < 0.001). The mean muscle quality measured using SliceOmatic was 35.1 +/- 10.8 HU and using 3D Slicer was 34.6 +/- 11.0 HU;LCC was 0.928 (p < 0.001), ICC was 0.964 (p < 0.001), and SCC was 0.957 (p < 0.001). Both the Bland Altman plots for L3 skeletal muscle CSA and muscle quality using SliceOmatic and 3D Slicer displayed overall strong agreement (Figures 1 and 2). However, 27 outliers were identified when visualizing the agreement L3 skeletal muscle CSA;further investigation of these outliers revealed that most of these measurements were conducted in critically ill patients (sepsis and COVID-19). Examining L3 skeletal muscle CSA between SliceOmatic and 3D Slicer among a subgroup of patients with sepsis and COVID revealed lower overall agreement (LCC 0.679, p < 0.0001), reliability (ICC 0.811, p < 0.001), and strength (SCC 0.642, p < 0.001). Similar findings were observed with muscle quality between SliceOmatic and 3D Slicer among a subgroup of patients with sepsis and COVID (LCC 0.585, p < 0.0001;ICC 0.741, p < 0.001;SCC 0.592, p < 0.001). Conclusion(s) Overall, both the SliceOmatic and 3D Slicer imaging analysis software tools had strong agreement, reliability, and strength when examining muscle mass and muscle quality. However, the agreement, reliability, and strength between muscle mass and muscle quality was lower between the two tools among critically ill patients compared to healthy controls and patients with cancer. Further research is needed to describe the etiology of this lower agreement in critically ill patients. (Table Presented).