Population-based and Personalized Reference Intervals for Liver and Spleen Volumes in Healthy Individuals and Those with Viral Hepatitis
Abstract
Background
Reference intervals guiding volumetric assessment of the liver and spleen have yet to be established.
Purpose
To establish population-based and personalized reference intervals for liver volume, spleen volume, and liver-to-spleen volume ratio (LSVR).
Materials and Methods
This retrospective study consecutively included healthy adult liver donors from 2001 to 2013 (reference group) and from 2014 to 2016 (healthy validation group) and patients with viral hepatitis from 2007 to 2017. Liver volume, spleen volume, and LSVR were measured with CT by using a deep learning algorithm. In the reference group, the reference intervals for the volume indexes were determined by using the population-based (ranges encompassing the central 95% of donors) and personalized (quantile regression modeling of the 2.5th and 97.5th percentiles as a function of age, sex, height, and weight) approaches. The validity of the reference intervals was evaluated in the healthy validation group and the viral hepatitis group.
Results
The reference and healthy validation groups had 2989 donors (mean age ± standard deviation, 30 years ± 9; 1828 men) and 472 donors (mean age, 30 years ± 9; 334 men), respectively. The viral hepatitis group had 158 patients (mean age, 48 years ± 12; 95 men). The population-based reference intervals were 824.5–1700.0 cm3 for liver volume, 81.1–322.0 cm3 for spleen volume, and 3.96–13.78 for LSVR. Formulae and a web calculator (https://i-pacs.com/calculators) were presented to calculate the personalized reference intervals. In the healthy validation group, both the population-based and personalized reference intervals were used to classify the volume indexes of 94%–96% of the donors as falling within the reference interval. In the viral hepatitis group, when compared with the population-based reference intervals, the personalized reference intervals helped identify more patients with volume indexes outside the reference interval (liver volume, 21.5% [34 of 158] vs 13.3% [21 of 158], P = .01; spleen volume, 29.1% [46 of 158] vs 22.2% [35 of 158], P = .01; LSVR, 35.4% [56 of 158] vs 26.6% [42 of 158], P < .001).
Conclusion
Reference intervals derived from a deep learning approach in healthy adults may enable evidence-based assessments of liver and spleen volume in clinical practice.
© RSNA, 2021
Online supplemental material is available for this article.
See also the editorial by Ringl in this issue.
Summary
Reference intervals derived from a deep learning approach enable clinical implementation of evidence-based assessment of liver and spleen volumes.
Key Results
■ This retrospective study established population-based and personalized reference intervals for liver and spleen volumes, encompassing the central 95% of healthy individuals in a reference group (2989 donors), and validated them by using a healthy validation group (472 donors).
■ In the viral hepatitis group (158 patients), when compared with population-based reference intervals, personalized reference intervals helped identify more liver volumes (21.5% [34 of 158] vs 13.3% [21 of 158], P = .01), spleen volumes (29.1% [46 of 158] vs 22.2% [35 of 158], P = .01), and liver-to-spleen volume ratios (35.4% [56 of 158] vs 26.6% [42 of 158], P < .001) outside the reference interval.
Introduction
Assessment of liver and spleen volume has clinical implications because various disease conditions are associated with abnormal liver or spleen volume. Liver volume may increase in association with acute hepatitis, veno-occlusive disease, hepatic deposition diseases, and hematologic malignancies, and it may decrease in the context of advanced liver fibrosis or cirrhosis (1–4). Splenomegaly can be caused by portal hypertension (5,6) or by infections or hematologic diseases (7). Liver-to-spleen volume ratio has been suggested as a prognostic index to predict clinically important portal hypertension, hepatic decompensation, and survival among patients with cirrhosis (2,5,8). Furthermore, liver volume measurement is an essential component of the work-up before liver resection; remnant liver volume is a major predictor of postoperative hepatic dysfunction and morbidity (9–11).
Although CT volumetry is the most reliable noninvasive method with which to measure liver and spleen volume (12), the time-consuming segmentation process has limited its clinical use. During interpretation of CT results, liver or spleen volumes are usually evaluated via visual estimations or measurements of organ length, despite limited accuracy (13,14). Recent advances in deep learning algorithms have enabled highly accurate automated CT-based liver and spleen volume measurements (12,15–17). These may allow for hepatic and splenic volumetric assessments via CT in routine clinical practice. In this regard, reference intervals for liver and spleen volumes are needed to assess whether measured volumes are normal or abnormal. Hepatic and splenic volumetric reference intervals have not yet been well established. One published study attempted to determine normal ranges for spleen volume, but it was limited by indirect volume estimations based on US measurements of spleen length (18).
Unlike decision limits, which are cutoff points for detecting specific diseases, reference intervals represent the parameters that define a healthy state (19). Thus, a reference interval is commonly defined as the range of values encompassing the central 95% of healthy individuals (19,20). However, this population-based approach may not be sufficiently robust in determining the reference intervals for liver and spleen volumes because liver and spleen volumes depend on personal factors, such as age, sex, and body size (2,13,18,21–24). Quantile regression is a statistical method for modeling specific percentile distributions of an outcome (25,26) and has been used to estimate covariate-dependent reference intervals (26–28). This method may also be useful for determining personalized reference intervals of liver and spleen volume indexes accounting for multiple variables.
We aimed to establish reference intervals for liver volume, spleen volume, and liver-to-spleen volume ratio by using population-based and personalized approaches in a large sample of healthy adult liver donors.
Materials and Methods
Our institutional review board approved this study and waived the requirement for informed consent because of its retrospective design.
Study Sample
Healthy liver donor candidates who underwent US-guided percutaneous liver biopsy at a single tertiary institution in Korea (Asan Medical Center) between April 2001 and October 2016 were consecutively registered. Eligible candidates were those who underwent CT, clinical, and laboratory examinations within 1 month before or after liver biopsy. Exclusion criteria were (a) a history of excessive alcohol consumption (ie, 20 g of ethanol per day); (b) clinically relevant hepatic steatosis, defined as macrovesicular steatosis of 5% or greater at liver biopsy; (c) liver diseases (ie, nonalcoholic steatohepatitis, viral or autoimmune hepatitis, and liver cancer) incidentally detected at biopsy or serologic tests; (d) abnormal laboratory findings (liver enzymes more than three times the upper normal limit or abnormal complete blood count or coagulation test result); (e) conditions precluding accurate organ volume measurements, including CT section thickness greater than 5 mm, motion artifact, or previous splenectomy; and (f) technical failure of deep learning in organ segmentation due to situs inversus. Finally, healthy liver donors made up the study sample, which was divided into the reference group (from April 2001 to December 2013) and the healthy validation group (from January 2014 to October 2016). As a disease control, we included patients with viral hepatitis B or C infection (viral hepatitis group) who underwent US-guided percutaneous liver biopsy at our institution (between 2007 and 2016) or at two tertiary hospitals (Haeundae Paik Hospital, Inje University College of Medicine, Busan, Korea; Hanyang University Medical Center, Hanyang University School of Medicine, Seoul, Korea) (between 2014 and 2017). The healthy liver donors were part of the study population in a previous study (29) that evaluated the criteria for the diagnosis of hepatic steatosis at nonenhanced CT. The viral hepatitis group was part of the study population in a previous study (2) that used the volumetric data to assess severity of liver fibrosis.
CT Examination
Contrast-enhanced CT was performed by using four- to 128-channel multidetector machines, with the following parameters: tube voltage, 100 or 120 kVp; tube current, 200 mA or 200–440 mA with an automatic exposure control; and section thickness, 2.5–5.0 mm with no gaps. Portal venous phase images were obtained 70–80 seconds after administration of contrast material. CT techniques are summarized in Table E1 (online).
Volume Measurements Using a Deep Learning Algorithm
Portal venous phase CT images were processed by using a convolutional neural network for automated liver and spleen segmentation, which was implemented in a web-based digital imaging and communications in medicine viewer (GoCDSS; Smart-Careworks). The deep learning algorithm is described by Ahn et al (12), and its source code is provided at https://github.com/seungsoolee0007/liver_spleen_segmentation. Briefly, the algorithm was developed by using the labeled CT data of 813 individuals and performed liver and spleen volumetry with measurement errors less than 5% of the measured volume; computation time was 33 seconds for a typical CT examination containing 70 sections (12). After the CT data were uploaded, the software automatically performed liver and spleen segmentation and measured the liver and spleen volumes by summing the consecutive areas of the liver and spleen multiplied by the section interval (21). The liver-to-spleen volume ratio was calculated as the liver volume divided by the spleen volume. After completion of the automated image analysis using the deep learning algorithm, one of three board-certified radiologists (D.W.K., J.H., S.S.L.; 9, 6, and 23 years of experience, respectively) reviewed the deep learning–generated segmentation results and corrected any segmentation errors by using the software. The CT review typically took less than 1 minute, and the time required for correcting segmentation errors was recorded.
Clinical and Pathologic Data
Clinical data, including age, sex, height, weight, and body mass index, were obtained within 3 months (median, 1 day; range, 0–87 days) of CT examinations. US-guided percutaneous liver biopsy was performed by using an 18-gauge needle (Stericut 18G coaxial; TSK Laboratory), with at least two biopsy specimens (approximately 1.5 cm long) obtained from different sites in the right hepatic lobe. Liver fibrosis was evaluated by using the Meta-analysis of Histologic Data in Viral Hepatitis fibrosis staging system (30): F0, no fibrosis; F1, portal fibrosis; F2, periportal fibrosis; F3, septal fibrosis; and F4, cirrhosis.
Statistical Analysis
Agreement between the volume indexes automatically measured with deep learning and those measured after the radiologists’ corrections was evaluated using 95% Bland-Altman limits of agreement. The reference intervals of the volume indexes were determined in the reference group by using population-based and personalized approaches. For this study, we chose the central 95th percentiles to define the reference intervals. The population-based reference intervals were determined by the intervals between the 2.5th and 97.5th percentiles of the distributions of the volume indexes in the reference group (20) and were obtained for the entire reference group as well as for multiple subgroups stratified by sex, age (by 10-year increments), height (by 10-cm increments), and weight (by 10-kg increments). To account for variations in the volume indexes depending on demographic and anthropometric factors (2,18), personalized reference intervals were generated by constructing separate quantile regression models for the 2.5th and 97.5th percentiles of the volume indexes as a function of age, sex, height, and weight (25,26). Separate models were also constructed to calculate personalized reference intervals when height or weight data were missing.
The reference intervals were validated in the healthy validation and viral hepatitis groups. The volume indexes of each individual were classified as within or outside the population-based (ie, those obtained for the entire reference group) or personalized reference intervals. In the healthy validation group, the reference interval validity was assessed on the basis of whether approximately 95% of healthy donors fell within the reference interval. The proportions of individuals falling within the reference interval were compared between the healthy validation and viral hepatitis groups by using the χ2 test. The differences in volume index classifications (ie, within or outside the reference interval) between the population-based and personalized reference intervals were evaluated with the McNemar test.
Quantile regression models for the 50th percentile (median) of the volume indexes were obtained to assess the standard volume index representing an expected volume index in a healthy individual with a given body size (ie, body surface area, height, or weight). The standard liver volume has been widely used to estimate the optimal liver volume for the metabolic demands of an individual patient before liver surgery (9–11). In the healthy validation cohort, we evaluated the validity of the estimated median liver volume as the standard liver volume. We also assessed a potential effect of ethnicity on liver volume by comparing our model with the three previously reported models for standard liver volume that were derived from Japanese (21), Middle Eastern (31), and Western (23) populations (Table E2 [online]). For each individual, the standard liver volumes were estimated with the quantile regression model for median liver volume and the three previous models (21,23,31). Then, the agreement of the estimated standard liver volumes with the observed liver volumes were evaluated by using the concordance correlation coefficient (CCC). All statistical analyses were performed by using SPSS software, version 21.0 (IBM) and R software, version 3.6.0 (R Foundation for Statistical Computing). P values less than .05 were considered to indicate significant differences.
Results
Study Sample Characteristics
The characteristics of the study sample are outlined in Table 1. Of the 5469 eligible donors, 2008 were excluded according to the exclusion criteria as shown in Figure 1. A total of 3461 healthy donors (2162 men and 1299 women; mean age, 30 years ± 9 [standard deviation]; age range, 18–62 years) were divided into the reference group (n = 2989; 1828 men and 1161 women; mean age, 30 years ± 9) and the healthy validation group (n = 472; 334 men and 138 women; mean age, 30 years ± 9). The viral hepatitis group included 158 patients (95 men and 63 women; mean age, 48 years ± 12), and 53 patients (33.5%) had liver cirrhosis (F4). The causes of viral hepatitis were hepatitis B virus (n = 78, 49.4%), hepatitis C virus (n = 72, 45.6%), or both (n = 8, 5.1%).
 |
Figure 1: Study enrollment flow diagram.
Deep Learning Analysis of Volume Indexes
The radiologists’ reviews of the deep learning–generated automated segmentation found minor segmentation errors in 182 (5.0%) of 3619 individuals overall. The segmentation errors required a short correction time (median time for correction, 35 seconds; range, 4–241 seconds) and were associated with small changes in the volume indexes (95% limits of agreement: −4.1% to 3.3% of measure volume indexes for liver volume, −7.5% to 8.3% for spleen volume, and −10.3% to 8.8% for liver-to-spleen volume ratio) (Table E3 [online]).
Population-based Reference Intervals
Figure 2 shows the distributions of the volume indexes. The population-based reference intervals were 824.5–1700.0 cm3 for individual liver volume, 81.1–322.0 cm3 for spleen volume, and 3.96–13.78 for liver-to-spleen volume ratio. The reference intervals according to age, height, and weight strata among men and are provided in Table E4 (online).
Figure 2: Distribution of (A) liver volume, (B) spleen volume, and (C) liver-to-spleen volume ratio in the reference group. Red vertical lines indicate lower (2.5th percentile) and upper (97.5th percentile) reference limits for the population-based reference intervals.
Personalized Reference Intervals
The quantile regression models for the 2.5th, 50th, and 97.5th percentiles of the volume indexes are summarized in Table 2. Personalized reference intervals of the volume indexes are determined as the fitted value from the quantile regression models of the 2.5th and 97.5th percentiles of the volume indexes, which are calculated by the sum of variable values multiplied by the corresponding coefficient values shown in Table 2 (also see the formulae in Table E5 [online]). A web-based calculator for the personalized reference intervals and a table of the population-based intervals are provided at https://i-pacs.com/calculators. The formulae to calculate the personalized reference intervals if height or weight is missing are also presented in Table E6 [online].
 |
Validation of Reference Intervals for Volume Indexes
Table 3 summarizes the validation results for the population-based and personalized reference intervals of the volume indexes in the healthy validation and viral hepatitis groups. In the healthy validation group, the volume indexes of approximately 95% (ie, 93.6%–96.0%) of individuals fell within the population-based and personalized reference intervals, supporting the validity of the reference intervals (Fig 3).
 |
Figure 3: Representative examples of liver and spleen volume measurements in (A) a healthy individual and (B) a patient with viral hepatitis. Liver (pink) and spleen (green) masks generated by the deep learning were overlaid on axial poral venous phase CT images. (A) A 49-year-old woman (height, 164 cm; body weight, 58 kg) in the healthy validation group had liver volume of 1113.4 cm3, spleen volume of 118.6 cm3, and liver-to-spleen volume ratio of 9.39. The volume indexes were within the population-based (liver volume, 824.5–1700.0 cm3; spleen volume, 81.1–322.0 cm3; liver-to-spleen volume ratio, 3.96–13.78) and the personalized (liver volume, 858.0–1428.9 cm3; spleen volume, 60.3–240.6 cm3; and liver-to-spleen volume ratio, 4.53–17.40) reference intervals. (B) A 56-year-old woman (height, 152 cm; body weight, 54 kg) with liver cirrhosis caused by hepatitis C virus had liver volume of 807.3 cm3, spleen volume of 234.6 cm3, and liver-to-spleen volume ratio of 3.44. The personalized reference intervals in this patient were 807.7–1341.6 cm3 for liver volume, 39.3–208.8 cm3 for spleen volume, and 4.82–19.32 for liver-to-spleen volume ratio. Both population-based and personalized reference intervals led to classification of liver volume and liver-to-spleen volume ratio as being outside the reference intervals. The spleen volume was within the population-based reference interval but was outside the personalized reference interval.
Compared with the healthy validation group, patients in the viral hepatitis group less frequently had volume indexes within the population-based and personalized reference intervals (range, 64.6%–86.7%; P < .001 for all volume indexes) (Table 3). Compared with the population-based reference intervals, in the viral hepatitis group the personalized reference intervals led to classification of more patients as having volume indexes outside the reference interval for liver volumes (21.5% [34 of 158] vs 13.3% [21 of 158], P = .01), spleen volumes (29.1% [46 of 158] vs 22.2% [35 of 158,; P = .01), and liver-to-spleen volume ratios (35.4% [56 of 158] vs 26.6% [42 of 158], P < .001) (Fig 3).
Comparison of the Model for Median Liver Volume with the Standard Liver Volume Models
Figure 4 depicts the effects of age, height, and weight on the volume indexes. In terms of the variables that affect the volume indexes based on their associations with the median (50th percentile) of the volume indexes, liver volume had positive associations with height (coefficient, 2.69; 95% CI: 1.39, 3.98; P < .001) and weight (coefficient, 12.87; 95% CI: 12.01, 13.73; P < .001), had a negative association with female sex (coefficient, −35.22; 95% CI: −54.64, −15.79; P < .001), and was not associated with age (coefficient, 0.63; 95% CI: −0.12, 1.37; P = .1). Spleen volume had positive associations with height (coefficient, 1.03; 95% CI: 0.56, 1.5; P < .001) and weight (coefficient, 1.73; 95% CI: 1.41, 2.04; P < .001) but a negative association with age (coefficient, −1.54; 95% CI: −1.79, −1.28; P < .001), whereas an association between sex and spleen volume was not detected after adjustment for other variables (coefficient, 0.93; 95% CI: −5.98, 7.85; P = .79). Liver-to-spleen volume ratio was associated with age (coefficient, 0.07; 95% CI: 0.06, 0.08; P < .001), sex (coefficient, −0.32; 95% CI: −0.63, −0.02; P = .03), and height (coefficient, −0.03; 95% CI: −0.05, −0.01; P = .001).
![Effect of age, height, and weight on the 2.5th, 50th, and 97.5th percentiles of (A) liver volume, (B) spleen volume, and (C) liver-to-spleen volume ratio. Data points indicate the volume index of each individual. Green, blue, and red lines represent 2.5th percentile, 50th percentile, and 97.5th percentile regression lines adjusted for the effect of other factors. Gray boxes represent the population-based reference intervals of volume indexes. For demonstration purposes, the regression lines for a factor of interest were drawn by using the slopes and intercepts of the factors calculated from the multivariable quantile regression models with the other factors being fixed using the proportion of female sex, mean body mass index (weight [kg]/height [m2]), and mean age in the reference group. The effect of female sex on the 2.5th, 50th, and 97.5th percentiles regression lines was downward (female) shifting of the regression slopes by its coefficients compared with male sex (not shown in figure).](/cms/10.1148/radiol.2021204183/asset/images/medium/radiol.2021204183fig4.gif "Effect of age, height, and weight on the 2.5th, 50th, and 97.5th percentiles of (A) liver volume, (B) spleen volume, and (C) liver-to-spleen volume ratio. Data points indicate the volume index of each individual. Green, blue, and red lines represent 2.5th percentile, 50th percentile, and 97.5th percentile regression lines adjusted for the effect of other factors. Gray boxes represent the population-based reference intervals of volume indexes. For demonstration purposes, the regression lines for a factor of interest were drawn by using the slopes and intercepts of the factors calculated from the multivariable quantile regression models with the other factors being fixed using the proportion of female sex, mean body mass index (weight [kg]/height [m2]), and mean age in the reference group. The effect of female sex on the 2.5th, 50th, and 97.5th percentiles regression lines was downward (female) shifting of the regression slopes by its coefficients compared with male sex (not shown in figure).")
Figure 4: Effect of age, height, and weight on the 2.5th, 50th, and 97.5th percentiles of (A) liver volume, (B) spleen volume, and (C) liver-to-spleen volume ratio. Data points indicate the volume index of each individual. Green, blue, and red lines represent 2.5th percentile, 50th percentile, and 97.5th percentile regression lines adjusted for the effect of other factors. Gray boxes represent the population-based reference intervals of volume indexes. For demonstration purposes, the regression lines for a factor of interest were drawn by using the slopes and intercepts of the factors calculated from the multivariable quantile regression models with the other factors being fixed using the proportion of female sex, mean body mass index (weight [kg]/height [m2]), and mean age in the reference group. The effect of female sex on the 2.5th, 50th, and 97.5th percentiles regression lines was downward (female) shifting of the regression slopes by its coefficients compared with male sex (not shown in figure).
The quantile regression model for median liver volume was as follows: 0.63 × age – 35.22 (if female) + 2.69 × height + 12.87 × weight – 87.43.
The scatterplots of the estimated standard liver volumes (21,23,31) against the observed liver volume in the healthy validation group are presented in Figure E1 (online). The agreement between the estimated standard liver volume and the observed liver volume for our model (CCC, 0.72; 95% CI: 0.68, 0.76) was stronger than those for the Western model (CCC, 0.53; 95% CI: 0.49, 0.58) and Middle Eastern model (CCC, 0.55; 95% CI: 0.5, 0.59) (23,31) but was similar to that for the Japanese model (CCC, 0.70; 95% CI: 0.65, 0.74) (21), which suggests differences in liver volume distribution across different ethnic groups.
Discussion
Although recent advances in deep learning algorithms may enable the routine clinical application of CT volumetry, reference intervals guiding volumetric assessment of the liver and spleen have not been established. We determined the reference intervals for liver volume, spleen volume, and liver-to-spleen volume ratio by using population-based and personalized approaches in a large sample of healthy adults. Both population-based and personalized reference intervals were validated in the healthy validation group, with approximately 95% of healthy donors classified as having volume indexes within the reference interval. However, the personalized reference intervals led to classification of more patients with viral hepatitis as having volume indexes outside the reference interval than the population-based reference intervals (34 vs 21 for liver volume [P = .01], 46 vs 35 for spleen volume [P = .01], 56 vs 42 for liver-to-spleen volume ratio [P < .001]). This indicates that the personalized reference intervals, which account for variations in volume indexes depending on personal factors, may allow for more precise classifications of volume indexes than the population-based reference intervals.
Besides regression models, we generated a web calculator (https://i-pacs.com/calculators) that allows for instant determination of personalized reference intervals for the volume indexes for a given individual after inputting age, sex, height, and weight information. This calculator facilitates the clinical use of personalized reference intervals of the volume indexes. When clinical information to calculate the personalized reference intervals is missing, the following may be used as alternatives: the population-based reference intervals; the population-based reference intervals according to age, height, and weight strata (Table E4 [online]); and the personalized reference intervals calculated with missing height or weight variables (Table E6 [online]).
We measured volume indexes at CT by using a deep learning algorithm for automated liver and spleen segmentation. The algorithm produced minor segmentation errors in only 5% of individuals; the segmentation errors could be rapidly corrected by radiologists and were associated with errors of less than 10% in the measured volume indexes. The CT data analysis with the deep learning algorithm can be performed as a background process before the radiologists’ CT review. Thus, if the deep learning algorithm is successfully incorporated in clinical workflows, radiologists may be able to review CT images along with deep learning–processed organ segmentation and volume results with little additional time and effort, and they may determine the status of liver and spleen volumes based on their measured volumes instead of visual volume estimation. In this context, the volume index reference intervals presented in our study may be used for identifying abnormal liver and spleen volumes associated with various pathologic conditions. The clinical utility and diagnostic performance of the volume reference intervals in detecting specific diseases should be further evaluated in future research.
We evaluated factors associated with the volume indexes. As expected, liver and spleen volumes were significantly correlated with body weight and height (P < .001), in line with the results of previous studies (2,18,21–23). We also found that women generally had smaller livers than men (P < .001). However, there was insufficient evidence to conclude that age was associated with liver volume (P = .1). Previous studies have reported different patterns of age-associated changes in liver volume among children and adults (22,24). The liver volume-to-body weight ratio has a negative correlation with age among children but generally remains constant among adults of all ages (22,24), an established observation supported by our findings. Unlike liver volume, spleen volume correlated negatively with age, which contradicts the findings of a previous study reporting no association between spleen volume and age among adults (18). We assumed that methodologic limitations of the previous study might explain these conflicting results; the previous study included a smaller study sample than ours (n = 1230 versus 2989) and, therefore, might have been relatively underpowered. Spleen volume in the previous study was estimated by using measurements of spleen length, possibly decreasing the accuracy of volume measurements compared with our CT-based measurements.
In addition to reference intervals, we proposed a quantile regression model for the median (50th percentile) liver volume as a model for standard liver volume and validated it by showing a good agreement (CCC, 0.72) between the estimated standard liver volume and the observed liver volume in the healthy validation cohort. We also found that, among the reported models for standard liver volume, including the Middle Eastern model (CCC, 0.55) and the Western model (CCC, 0.53), the strongest agreement between the estimated standard liver volume and the observed liver volume was found in the Japanese model (CCC, 0.70) derived from the same Asian population as our study (21,23,31). This finding indicates a potential influence of ethnicity on liver volume and also suggests that our reference intervals may not be directly generalizable to different ethnic groups.
Our study had limitations. First, the reference group and the healthy validation group were derived from the same source population (ie, liver donors from the same institution), which may have potentially led to an optimistic validation result. Second, despite a wide age range (18–62 years), the donors in the reference group were mostly young. Because of the skewed distribution of age in the reference group, our results might not have fully accounted for the effect of age on the volume indexes. Third, in our study, we built separate quantile models for each volume index as a function of personal factors. This approach did not account for a potential association between liver and spleen volumes and an interaction among the personal factors, which may have reduced the models’ prediction performance. Fourth, we excluded donors with hepatic steatosis and other potentially nonhealthy conditions to establish the reference intervals in the individuals confirmed to be healthy. However, this may have introduced a selection bias to our study. In addition, we could not evaluate the effect of hepatic steatosis on the volume indexes in our study.
In conclusion, we proposed population-based and personalized reference intervals for liver volume, spleen volume, and liver-to-spleen volume ratio. The proposed reference intervals enable evidence-based assessments of liver and spleen volumes and would be clinically useful, especially when deep learning–based automated CT volumetry is implemented in clinical practice. The generalizability of the established reference intervals needs to be further validated in the different ethnic groups and people with varying somatotypes.
Author Contributions
Author contributions: Guarantor of integrity of entire study, S.S.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, D.W.K., J.H., S.S.L., J.H.K.; clinical studies, D.W.K., J.H., S.S.L., J.H.K., Y.L., B.K.K.; experimental studies, J.S.Y., H.I.S.; statistical analysis, D.W.K., J.H., S.S.L., N.Y.K., Y.S.S.; and manuscript editing, D.W.K., J.H., S.S.L.
* D.W.K. and J.H. contributed equally to this work.
Supported by a National Research Foundation of Korea grant funded by the Korean government (MSIT) (2020R1F1A1048826) and the Bio and Medical Technology Development Program of the NRF, which is funded by the Ministry of Science and ICT, South Korea (NRF-2016M3A9A7918706).
References
- 1. . Zakim and Boyer's hepatology: A textbook of liver disease. Philadelphia, Pa:Elsevier Health Sciences,2018.
- 2. . Assessment of liver fibrosis severity using computed tomography-based liver and spleen volumetric indices in patients with chronic liver disease. Eur Radiol 2020;30(6):3486–3496.
- 3. . Hepatosplenic volumetric assessment at MDCT for staging liver fibrosis. Eur Radiol 2017;27(7):3060–3068.
- 4. . Changes in liver and spleen volume in alcoholic liver fibrosis of man. Hepatology 1989;9(4):589–593.
- 5. . Accurate computed tomography-based portal pressure assessment in patients with hepatocellular carcinoma. J Hepatol 2014;60(5):969–974.
- 6. . Measurement of spleen size and its relation to hypersplenism and portal hemodynamics in portal hypertension due to hepatic cirrhosis. Am J Gastroenterol 1996;91(12):2580–2583.
- 7. . Splenomegaly: investigation, diagnosis and management. Blood Rev 2009;23(3):105–111.
- 8. . Liver/spleen volume ratio as a predictor of prognosis in primary biliary cirrhosis. J Gastroenterol 2008;43(8):632–636.
- 9. . An ordinal model to predict the risk of symptomatic liver failure in patients with cirrhosis undergoing hepatectomy. J Hepatol 2019;71(5):920–929.
- 10. ; Edinburgh Liver Surgery and Transplantation Experimental Research Group (eLISTER). The value of residual liver volume as a predictor of hepatic dysfunction and infection after major liver resection. Gut 2005;54(2):289–296.
- 11. . Standardized measurement of the future liver remnant prior to extended liver resection: methodology and clinical associations. Surgery 2000;127(5):512–519.
- 12. . Deep learning algorithm for automated segmentation and volume measurement of the liver and spleen using portal venous phase computed tomography images. Korean J Radiol 2020;21(8):987–997.
- 13. . Determination of normal splenic volume on computed tomography in relation to age, gender and body habitus. Eur Radiol 1997;7(2):246–248.
- 14. . Estimating splenic volume: sonographic measurements correlated with helical CT determination. AJR Am J Roentgenol 2003;181(6):1615–1620.
- 15. . 3D deeply supervised network for automated segmentation of volumetric medical images. Med Image Anal 2017;41(40):54.
- 16. . Fully automatic liver attenuation estimation combing CNN segmentation and morphological operations. Med Phys 2019;46(8):3508–3519.
- 17. . Automated CT and MRI liver segmentation and biometry using a generalized convolutional neural network. Radiol Artif Intell 2019;1(2):180022.
- 18. . Spleen size is significantly influenced by body height and sex: establishment of normal values for spleen size at US with a cohort of 1200 healthy individuals. Radiology 2016;279(1):306–313.
- 19. . The normal range: it is not normal and it is not a range. Postgrad Med J 2018;94(1117):613–616.
- 20. Clinical and Laboratory Standards Institute. Defining, establishing, and verifying reference intervals in the clinical laboratory; approved guideline. CLSI Document C28-A3.3rd ed. Wayne, Pa:Clinical and Laboratory Standards Institute,2008.
- 21. . Estimation of standard liver volume in Japanese living liver donors. J Gastroenterol Hepatol 2006;21(11):1710–1713.
- 22. . Calculation of child and adult standard liver volume for liver transplantation. Hepatology 1995;21(5):1317–1321.
- 23. . Body surface area and body weight predict total liver volume in Western adults. Liver Transpl 2002;8(3):233–240.
- 24. . Estimation of standard liver volume for liver transplantation in the Korean population. Liver Transpl 2004;10(6):779–783.
- 25. . Quantile Regression. Cambridge, England:Cambridge University Press,2005.
- 26. . Quantile regression and its applications: a primer for anesthesiologists. Anesth Analg 2019;128(4):820–830.
- 27. . Continuous reference intervals for 38 biochemical markers in healthy children and adolescents: Comparisons to traditionally partitioned reference intervals. Clin Biochem 2019;73(82):89.
- 28. . The World Health Organization fetal growth charts: concept, findings, interpretation, and application. Am J Obstet Gynecol 2018;218(2S):S619–S629.
- 29. . CT indices for the diagnosis of hepatic steatosis using non-enhanced CT images: development and validation of diagnostic cut-off values in a large cohort with pathological reference standard. Eur Radiol 2019;29(8):4427–4435.
- 30. . An algorithm for the grading of activity in chronic hepatitis C. The METAVIR Cooperative Study Group. Hepatology 1996;24(2):289–293.
- 31. . Standard formula for liver volume in Middle Eastern Arabic adults. Transplant Proc 2010;42(9):3600–3605.
Article History
Received: Nov 2 2020Revision requested: Jan 4 2021
Revision received: May 28 2021
Accepted: June 4 2021
Published online: Aug 17 2021
Published in print: Nov 2021
