Purpose: The purpose of this study was to assess the accuracy of liver tumor volumetry by manual contouring on computed tomography (CT) compared to pathological tumor volume determined from surgical specimen that served as a reference method.
Patients and method: Thirty-eight patients with planned liver surgery and a total of 41 liver tumors were included. There were 24 men and 14 women (mean age: 57 years; range: 32-85 years). Two readers calculated tumor volume by manual contouring on axial CT images. The reference tumor volume was calculated by manual contouring with dedicated software applied to the liver specimen slice. CT and pathology volumes were compared and the percentage of error (PE%) was calculated. Intraobserver and interobserver variabilities were calculated using Bland and Altman plots, and intraclass correlation coefficients (ICC).
Results: A strong correlation was found between CT tumor volumes and pathology tumor volumes (r=0.994; P<0.001 for both readers). The mean (±SD) and median (range) PE% were 19%±12% and 16% (1%, +42%) and 19%±15% and 17% (0%, +55%) for readers 1 and 2, respectively. Readers 1 and 2 significantly overestimated tumor volume (i.e., PE%>40%) in 3 (7%) and 2 (5%) tumors on CT, respectively. Tumor volume was not significantly underestimated in any of the patients (i.e. PE%>33%). Tumor size, CT attenuation, time between imaging and surgery, contours and margin definition did not influence the results of PE% values (all P values>0.05 for both readers). The bias and limits of agreement between the two readers were +4.6% and (-24%, +33%) with an ICC of 0.997.
Conclusion: There was a strong correlation between tumor volume measured on CT and that assessed with surgical specimen. Tumor size, visibility of contours and tumor margins and the time between CT and surgery did not influence the results.
Keywords: Liver; Liver volumetry; Semiautomatic method; Tissue segmentation.
Copyright © 2017 Editions françaises de radiologie. Published by Elsevier Masson SAS. All rights reserved.