Evaluation of Various Methods of Liver Measurement in Comparison to Volumetric Segmentation Based on Computed Tomography

Background: A reliable assessment of liver volume, necessary before transplantation, remains a challenge. Our work aimed to assess the differences in the evaluation and measurements of the liver between independent observers and compare different formulas calculating its volume in relation to volumetric segmentation. Methods: Eight researchers measured standard liver dimensions based on 105 abdominal computed tomography (CT) scans. Based on the results obtained, the volume of the liver was calculated using twelve different methods. An independent observer performed a volumetric segmentation of the livers based on the same CT examinations. Results: Significant differences were found between the formulas and in relation to volumetric segmentation, with the closest results obtained for the Heinemann et al. method. The measurements of individual observers differed significantly from one another. The observers also rated different numbers of livers as enlarged. Conclusions: Due to significant differences, despite its time-consuming nature, the use of volumetric liver segmentation in the daily assessment of liver volume seems to be the most accurate method.


Introduction
The assessment of liver size based on physical examination is disadvantageous, with a significant error possible due to considerable dependence on the experience of the examiner [1].Therefore, a more reliable quantitative assessment of liver enlargement is possible in imaging studies, especially computed tomography (CT) and magnetic resonance (MRI).Volume estimation based on CT measurements is superior to that based on ultrasound, but its margin of error is still evident [2].Three-dimensional imaging methods seem to be more accurate in this regard than two-dimensional ones.Currently, the need to move from two-dimensional to three-dimensional analysis is discussed more and more often, especially in the context of liver surgery.Although 3D segmentation technologies have been implemented for over a decade, traditional methods are still widely used in clinical practice.Our study aims to compare these older methods with modern techniques to better understand their limitations and provide insights into their comparison with current advanced techniques [3].Currently, work has already begun on the use of the convolutional neural network to automate liver segmentation across different imaging modalities.While 3D liver segmentation is considered highly accurate, it is not without errors.Future studies could include post-surgical liver weighing and volume measurement as a more reliable reference standard [4,5].The repeatability and reproducibility of liver measurements is a significant problem, especially in the case of manual measurements [6].As liver volume estimation is mandatory prior to a living donor transplantation or partial resection, the mismatch between imaging and intraoperative findings is still a reported issue [7,8].The tumor volume is a known confounding factor.The future remnant to total liver size assessment is faster when the total volume of the liver can be calculated from biometric data.The estimated total liver volume based on biometrics has been proposed particularly in order to size liver grafts in transplants [9].Over the years, the number of formulas used to calculate the volume of the liver has increased, which has made it difficult to choose the best one.The use of such formulas may lead to over-or underestimation, which again may lead to an error in the assessment of potential liver failure after resection or transplantation.The number of publications comparing the proposed methods with volumetry remains scarce.
Our work aimed to assess (1) if the measurement of one of the dimensions of the liver allows for a reliable estimation of its volume, (2) if we can reliably estimate the liver volume based on linear liver measurements and the patient's physiological data, (3) if hepatomegaly can be reliably qualified without taking measurements, and (4) if liver measurements based on simple instructions vary significantly between observers.In search of answers to the above questions, we relied on the method of computed tomography.

Materials and Methods
A total of 105 CT examinations (performed between October 2021 and March 2022) of the abdominal cavity have been selected and anonymized by an independent observer with five years of CT evaluation experience from the archives of the Department of Radiodiagnostics and Interventional Radiology, University Clinical Center prof.K. Gibi ński of the Medical University of Silesia in Katowice, Poland.The CT images that were included in the liver had to have visible hepatic veins and no focal lesions.All analyzed liver computed tomography examinations were performed on a SIEMENS SOMATOM Definition Edge device using the three-phase liver protocol or the classic abdominal protocol.The inclusion criteria included an examination performed in at least two phases after administering intravenous contrast agent, no focal lesions, no signs of significant fibrosis, no previous surgeries on the liver, and no signs of cholestasis.All observers measured the liver using all methods.To increase accuracy, it would be beneficial to have more than one radiologist perform 3D liver segmentation.The sex, age, weight, and height of participants were noted, and their BMI and BSA were calculated.The selected group consisted of 27 men and 78 women aged 57.13 ± 16.44 years (56.74 ± 16.66 years and 57.27 ± 16.47 years, respectively) with a BMI of 25.40 ± 5.51 kg/m 2 (24.88 ± 4.40 kg/m 2 and 25.59 ± 5.87 kg/m 2 , respectively).
In the next step, a group of 8 observers, composed of 2 radiology specialists, 5 radiology residents, and 1 medical physicist with 4.13 ± 1.46 years of experience in CT evaluation, was tasked with assessing the following parameters using all methods: whether, in their opinion, the liver is enlarged; the maximum anteroposterior (AP), craniocaudal (CC), and left-right (SD) dimensions of the liver; and the AP and CC dimensions in the midclavicular line.The liver volume was calculated according to Muggli The Mosteller method has been used for body surface area calculations [21]: The observers rated the studies independently of each other using diagnostic software.Measurements were made in millimeters, with an accuracy of one decimal place.
At the same time, another observer, a radiology specialist with seven years of experience in CT analysis, performed 3D liver segmentations for the above examinations.The segmentation was performed with the liver analysis module of the syngo.via(Siemens Healthineers: Erlangen, Germany.syngo.via,version VB30A), according to the manufacturer's instructions.The segmentation was conducted according to the manufacturer's instructions; the liver outline was automatically detected, followed by a semi-automatic segmentation of hepatic veins and the portal vein after setting seed points.Axial images of the portal venous phase were used for obtaining the liver volume.The liver outline was segmented automatically, based on a hierarchical, learning-based approach.After the detection of the liver outline, hepatic veins and the portal vein were segmented semiautomatically.After setting seed points to the hepato-caval confluence and the main stem of the portal vein, the system automatically segments the hepatic veins and portal vein.The volumes of the intrahepatic vessels were included in the CT volume.Surrounding extrahepatic vessels, extrahepatic bile ducts, and the gallbladder were excluded.The liver volume obtained as a result of segmentation was adopted in the study as the gold standard against which other parameters and calculated volumes were compared [22][23][24].For a better understanding, the workflow has been presented in Figure 1 and the data of the test group has been presented in Table 1.
At the same time, another observer, a radiology specialist with seven years of ence in CT analysis, performed 3D liver segmentations for the above examinatio segmentation was performed with the liver analysis module of the syngo.via(S Healthineers: Erlangen, Germany.syngo.via,version VB30A), according to the m turer's instructions.The segmentation was conducted according to the manufactur structions; the liver outline was automatically detected, followed by a semi-automa mentation of hepatic veins and the portal vein after setting seed points.Axial im the portal venous phase were used for obtaining the liver volume.The liver outli segmented automatically, based on a hierarchical, learning-based approach.After tection of the liver outline, hepatic veins and the portal vein were segmented sem matically.After setting seed points to the hepato-caval confluence and the main s the portal vein, the system automatically segments the hepatic veins and portal ve volumes of the intrahepatic vessels were included in the CT volume.Surrounding hepatic vessels, extrahepatic bile ducts, and the gallbladder were excluded.The liv ume obtained as a result of segmentation was adopted in the study as the gold st against which other parameters and calculated volumes were compared [22][23][24] better understanding, the workflow has been presented in Figure 1 and the data of group has been presented in Table 1.The results were gathered and analyzed using Statistica 13.3 (TIBCO Software Inc.: Palo Alto, CA, USA (2017)).Statistica (data analysis software system) version 13.The normality of distribution was checked with a Kolmogorov-Smirnov test.Due to the abnormal distribution of most of the analyzed quantitative parameters, Mann-Whitney and Kruskal-Wallis tests were implemented.A Spearman test was used for the correlation strength analysis.The qualitative variables were analyzed with the Pearson test.Using stepwise regression, an attempt was made to build a model calculating the liver volume based on the collected data.As the value of statistical significance, a p of 0.05 was adopted.

Liver Volume in Relation to Other Variables
The analysis of single measurement and parameter correlations with the segmented liver volume showed that there was a significant correlation (p < 0.001) for all of them; their correlation coefficients (r) and determination coefficients (r 2 ) are presented in Table 2.The difference between the calculated and segmented liver volumes is significant (p < 0.001) and presented in Figure 2. The results of multiple two-sided comparisons are attached as Supplementary Material Table S1 The difference between the calculated and segmented liver volumes is significant (p < 0.001) and presented in Figure 2. The results of multiple two-sided comparisons are attached as Supplementary Material Table S1.An attempt to build an optimized model based on the acquired data was performed with stepwise regression (both stepwise and progressive).Considering only the measurements made by observers, a viable model was calculated with a corrected r 2 = 0.749 based on maximum AP, CC, and SD measurements and the CC in the midclavicular line.With the addition of the rest of the acquired quantitative variables, a viable model with a corrected r 2 of 0.824 and p < 0.001 was obtained, based on the measurements above, with the addition of age and weight.An attempt to build an optimized model based on the acquired data was performed with stepwise regression (both stepwise and progressive).Considering only the measurements made by observers, a viable model was calculated with a corrected r 2 = 0.749 based on maximum AP, CC, and SD measurements and the CC in the midclavicular line.With the addition of the rest of the acquired quantitative variables, a viable model with a corrected r 2 of 0.824 and p < 0.001 was obtained, based on the measurements above, with the addition of age and weight.

Differences between the Observers
There was a significant difference observed for all of the measurements made by the observers, but the calculated volumes differed only for the Muggli, David et al. method.The p values are shown in Table 3. 1.0 Fu-Gui et al. [13] 1.0 Urata et al. [14] 1.0 Hashimoto et al. [15] 1.0 Yuan et al. [16] 1.0 Poovathumkadavil [17] 1.0 Vaughtney et al. [18]-BSA-based method 1.0 Vaughtney et al. [18]-weight-based method 1.0 Lin [19] 1.0 Yu et al. [20] 1.0 Heinemann et al. [21] 1.0 The difference between the observers in terms of the number of livers marked as enlarged was also significant, p < 0.001; the results are presented in Table 4.

Discussion
As can be seen in our results, the repeatability of liver measurements between independent observers is a factor that can significantly affect the assessment of a liver's volume.The obtained results confirm the previous reports that it is impossible to reliably estimate liver volume based on a single measurement.To minimize inter-operator variability, future studies should include multiple operators for 3D measurements.In our study, we used a single operator to ensure consistency; however, we recommend considering inter-operator variability in further analyses [7].Liver-volume-calculating formulas based on biological data, such as patients' weight and height, seem not to be significantly affected by this problem, as their results were comparable to each other.The observers' years of experience in CT evaluation was not a significant factor affecting the measurements.As reported before, only minimal training is needed to reliably make these measurements [25].
Of all analyzed methods, only Heinemann et al. offered results comparable to volumetric segmentation.The obtained results are consistent with the reports to date on the significant differences between the analyzed formulas used for assessing the liver volume [10,26].
The evaluation of the presence of hepatomegaly without measurements is not reliable due to the significant differences in this assessment.
So far, many authors have presented various solutions to the problem of liver volume measurement, but each of the proposed methods is burdened with some error [10,26].Currently, there is no consensus on which calculation method based on patient's physiological data is better [27].Even automated liver segmentation, used as the gold standard in this work, is affected by factors such as scan phase and slice thickness [8,28].The CT and MRI methods seem to correlate strongly; thus, the use of only magnetic resonance imaging to assess liver volume does not appear to be economically viable [29].The implementation of automatic methods of liver segmentation allowed for a significant reduction in the process time compared to semi-automatic and manual methods, which should significantly facilitate this type of measurement in everyday practice [30].
To average out possible uncertainties due to individual biases and/or experience levels, the measurements were performed by a group of specialists with a wide range of experience years.Another limitation of the method is the measurement of examinations obtained from the same CT scanner.The last limitation of the method was the small number of studies on which the analysis was based.

Conclusions
Our study concluded that a single liver measurement does not allow for reliable volume estimation.Most methods based on linear measurements and patient physiological data do not provide accurate liver volume estimates.Hepatomegaly cannot be reliably assessed without measurements.Liver measurements based on simple instructions vary significantly between observers.The use of volumetric liver segmentation in daily volume assessments appears to be the most accurate and time-optimized method.Implementing 3D segmentation techniques in clinical practice can significantly improve the accuracy of preoperative assessments and patient outcomes.This study highlights the significant discrepancies between various liver volume estimation methods and underscores the importance of transitioning to more reliable and accurate 3D segmentation techniques in clinical practice.Our results confirm previous reports of significant differences between the analyzed formulas in liver volume assessments.The significant differences in measurements between observers highlight the need for more standardized and precise methods, such as 3D segmentation.Future research should focus on further improving automatic segmentation methods and their implementation in daily clinical practice to minimize errors and improve preoperative outcomes.By adopting volumetric liver segmentation, clinicians can achieve more precise assessments, potentially improving preoperative planning and patient outcomes.

Supplementary Materials:
The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/jcm13133634/s1,Table S1.The results of two-sided multiple comparisons as result of Kruskal-Wallis post-hoc tests, presented as p value.Significant differences are marked in red.

Figure 1 .
Figure 1.Diagram presenting the workflow at the stage of data collection and preparation.

Figure 1 .
Figure 1.Diagram presenting the workflow at the stage of data collection and preparation.
Author Contributions: Conceptualization, M.C.; methodology, M.C.; software, M.C.; validation, M.C., J.P.-K.and K.G.; formal analysis, M.C.; investigation, all of the authors; resources, S.M.; data curation, A.B.; writing-original draft preparation, M.C.; writing-review and editing, M.C. and J.K.; visualization, M.C.; supervision, K.G.; project administration, J.P.-K.and K.G.; funding acquisition, J.K. All authors have read and agreed to the published version of the manuscript.Funding: This research received no external funding.Institutional Review Board Statement: This study's ethical review and approval were waived due to the retrospective nature of the study and patient data anonymization.Informed Consent Statement: Patient consent was waived due to the retrospective nature of the study and patient data anonymization.

Table 1 .
Data of the analyzed group.

Table 1 .
Data of the analyzed group.

Table 2 .
Correlation and determination coefficients of single measurements and parameters with the segmented liver volume. .

Table 3 .
Differences between observers in terms of liver measurements and calculated liver volumes.

Table 4 .
Difference between the observers in terms of the number of livers marked as enlarged.