Comparison of liver volumetry on contrast‐enhanced CT images: one semiautomatic and two automatic approaches

Abstract

This study was to evaluate the accuracy, consistency, and efficiency of three liver volumetry methods— one interactive method, an in‐house‐developed 3D medical Image Analysis (3DMIA) system, one automatic active shape model (ASM)‐based segmentation, and one automatic probabilistic atlas (PA)‐guided segmentation method on clinical contrast‐enhanced CT images. Forty‐two datasets, including 27 normal liver and 15 space‐occupying liver lesion patients, were retrospectively included in this study. The three methods — one semiautomatic 3DMIA, one automatic ASM‐based, and one automatic PA‐based liver volumetry — achieved an accuracy with VD (volume difference) of −1.69%,−2.75%, and 3.06% in the normal group, respectively, and with VD of −3.20%,−3.35%, and 4.14% in the space‐occupying lesion group, respectively. However, the three methods achieved an efficiency of 27.63 mins, 1.26 mins, 1.18 mins on average, respectively, compared with the manual volumetry, which took 43.98 mins. The high intraclass correlation coefficient between the three methods and the manual method indicated an excellent agreement on liver volumetry. Significant differences in segmentation time were observed between the three methods (3DMIA, ASM, and PA) and the manual volumetry (p<0.001), as well as between the automatic volumetries (ASM and PA) and the semiautomatic volumetry (3DMIA) (p<0.001). The semiautomatic interactive 3DMIA, automatic ASM‐based, and automatic PA‐based liver volumetry agreed well with manual gold standard in both the normal liver group and the space‐occupying lesion group. The ASM‐ and PA‐based automatic segmentation have better efficiency in clinical use.PACS number(s): 87.55.‐x

Accuracy of actual resected liver volume in anatomical liver resections guided by 3‐dimensional parenchymal staining using fusion indocyanine green fluorescence imaging

“…Meanwhile, multiple evidence reveal the accuracy of PRLV assessment by these software. Wei Cai et al reported that MI‐3DVS provides virtual surgical planning and is accurate in assessing PRLV . Therefore, it can be considered that the liver volume that virtually resected using MI‐3DVS for surgical planning is equal to the liver volume of our targeted resection.…”

“…Accurate PRLV was calculated by virtual surgery in MI‐3DVS, as proposed by Cai et al The ARLV was calculated by drainage, as a standard method for measuring the liver volume. To assess the accuracy of ARLV obtained using fusion IGFI, the agreement between ARLV and PRLV must be evaluated.…”

“…planning and is accurate in assessing PRLV. 18,19 Therefore, it can be considered that the liver volume that virtually resected using MI-3DVS for surgical planning is equal to the liver volume of our targeted resection.…”

Fully automated whole-liver volume quantification on CT-image data: Comparison with manual volumetry using enhanced and unenhanced images as well as two different radiation dose levels and two reconstruction kernels

“…Volumetry of the liver is by now performed using different techniques which generally rely on different segmentation techniques, contour recognition, etc. [ 7 – 9 ]. Manual contour tracing for organ volumetry is knowingly limited as it is cumbersome, time-consuming and underlies inter- and intra-observer variability [ 10 ].…”

Atlas-based liver segmentation for nonhuman primate research

“…On non-contrast CT scans, the liver is isodense compared to nearby tissues, such as muscle and kidney (Fig. 1), making margin decisions difficult [6]. Automated methods to segment the liver have recently been shown to be effective [7][8][9], however, most of the work involves segmentation of human liver.…”