Francesca Rigiroli1, Dylan Zhang2, Jeroen Molinger3, Yingqi Wang4, Andrew Chang5, Paul E Wischmeyer6, Brant A Inman7, Rajan T Gupta8. 1. Department of Radiology, Duke University Medical Center, Durham, NC, USA; Department of Radiology, Beth Israel Deaconess Medical Center, Boston, MA, USA. Electronic address: f.rigiroli@gmail.com. 2. Department of Radiology, Duke University Medical Center, Durham, NC, USA. Electronic address: Dylan.zhang@duke.edu. 3. Department of Anesthesiology, Duke University Medical Center, Durham, NC, USA. Electronic address: jeroen.molinger@duke.edu. 4. Division of Urology, Duke Cancer Institute, Durham, NC, USA. Electronic address: yingqi.wang@mohh.com.sg. 5. Division of Urology, Duke Cancer Institute, Durham, NC, USA. Electronic address: Andrew.Chang@moffitt.org. 6. Department of Anesthesiology, Duke University Medical Center, Durham, NC, USA. Electronic address: paul.wischmeyer@duke.edu. 7. Division of Urology, Duke Cancer Institute, Durham, NC, USA; Duke Cancer Institute Center for Prostate and Urologic Cancers, Durham, NC, USA. Electronic address: brant.inman@duke.edu. 8. Department of Radiology, Duke University Medical Center, Durham, NC, USA; Division of Urology, Duke Cancer Institute, Durham, NC, USA; Duke Cancer Institute Center for Prostate and Urologic Cancers, Durham, NC, USA. Electronic address: rajan.gupta@duke.edu.
Abstract
PURPOSE: Manual measurement of body composition on computed tomography (CT) is time-consuming, limiting its clinical use. We validate a software program, Automatic Body composition Analyzer using Computed tomography image Segmentation (ABACS), for the automated measurement of body composition by comparing its performance to manual segmentation in a cohort of patients with bladder cancer. METHOD: We performed a retrospective analysis of 285 patients treated for bladder cancer at the Duke University Health System from 1996 to 2017. Abdominal CT images were manually segmented at L3 using Slice-O-Matic. Automated segmentation was performed with ABACS on the same L3-level images. Measures of interest were skeletal muscle (SM) area, subcutaneous adipose tissue (SAT) area, and visceral adipose tissue (VAT) area. SM index, SAT index, and VAT index were calculated by dividing component areas by patient height2 (m2). Patients were dichotomized as sarcopenic, having excessive subcutaneous fat, or having excessive visceral fat using published cut-off values. Agreement between manual and automated segmentation was assessed using the Pearson product-moment correlation coefficient (PPMCC), the interclass correlation coefficient (ICC3), and the kappa statistic (κ). RESULTS: There was strong agreement between manual and automatic segmentation, with PPMCCs > 0.90 and ICC3s > 0.90 for SM, SAT, and VAT areas. Categorization of patients as sarcopenic (κ = 0.73), having excessive subcutaneous fat (κ = 0.88), or having excessive visceral fat (κ = 0.90) displayed high agreement between methods. CONCLUSIONS: Automated segmentation of body composition measures on CT using ABACS performs similarly to manual analysis and may expedite data collection in body composition research.
PURPOSE: Manual measurement of body composition on computed tomography (CT) is time-consuming, limiting its clinical use. We validate a software program, Automatic Body composition Analyzer using Computed tomography image Segmentation (ABACS), for the automated measurement of body composition by comparing its performance to manual segmentation in a cohort of patients with bladder cancer. METHOD: We performed a retrospective analysis of 285 patients treated for bladder cancer at the Duke University Health System from 1996 to 2017. Abdominal CT images were manually segmented at L3 using Slice-O-Matic. Automated segmentation was performed with ABACS on the same L3-level images. Measures of interest were skeletal muscle (SM) area, subcutaneous adipose tissue (SAT) area, and visceral adipose tissue (VAT) area. SM index, SAT index, and VAT index were calculated by dividing component areas by patient height2 (m2). Patients were dichotomized as sarcopenic, having excessive subcutaneous fat, or having excessive visceral fat using published cut-off values. Agreement between manual and automated segmentation was assessed using the Pearson product-moment correlation coefficient (PPMCC), the interclass correlation coefficient (ICC3), and the kappa statistic (κ). RESULTS: There was strong agreement between manual and automatic segmentation, with PPMCCs > 0.90 and ICC3s > 0.90 for SM, SAT, and VAT areas. Categorization of patients as sarcopenic (κ = 0.73), having excessive subcutaneous fat (κ = 0.88), or having excessive visceral fat (κ = 0.90) displayed high agreement between methods. CONCLUSIONS: Automated segmentation of body composition measures on CT using ABACS performs similarly to manual analysis and may expedite data collection in body composition research.
Authors: Carla M M Prado; Jessica R Lieffers; Linda J McCargar; Tony Reiman; Michael B Sawyer; Lisa Martin; Vickie E Baracos Journal: Lancet Oncol Date: 2008-06-06 Impact factor: 41.316
Authors: Yingqi Wang; Andrew Chang; Wei Phin Tan; Joseph J Fantony; Ajay Gopalakrishna; Gregory J Barton; Paul E Wischmeyer; Rajan T Gupta; Brant A Inman Journal: Eur Urol Oncol Date: 2019-05-25
Authors: Maryam Ebadi; Lisa Martin; Sunita Ghosh; Catherine J Field; Richard Lehner; Vickie E Baracos; Vera C Mazurak Journal: Br J Cancer Date: 2017-06-06 Impact factor: 7.640
Authors: Elizabeth M Cespedes Feliciano; Karteek Popuri; Dana Cobzas; Vickie E Baracos; Mirza Faisal Beg; Arafat Dad Khan; Cydney Ma; Vincent Chow; Carla M Prado; Jingjie Xiao; Vincent Liu; Wendy Y Chen; Jeffrey Meyerhardt; Kathleen B Albers; Bette J Caan Journal: J Cachexia Sarcopenia Muscle Date: 2020-04-20 Impact factor: 12.910