Histogram Analysis of Hepatobiliary Phase MR Imaging as a Quantitative Value for Liver Cirrhosis: Preliminary Observations

The diagnosis of liver fibrosis and cirrhosis in patients with chronic liver disease is critical, as cirrhotic patients are at higher risk of developing end stage liver disease, portal hypertension, and hepatocellular carcinoma (HCC).1, 2 These sequelae are important causes of morbidity, mortality, and increasing health care costs.3, 4 The cumulative incidence of HCC is significantly higher in patients with severe fibrosis than in those with no or mild fibrosis.5 Thus, the early detection and accurate staging of hepatic fibrosis or cirrhosis has become a critical issue in practice. Currently, liver biopsy is still the gold standard for the assessment of liver fibrosis. However, liver biopsy is invasive, difficult to repeat, and associated with significant patient morbidities. In addition, liver biopsy is associated with interobserver variability and sampling errors.6, 7, 8 The search for noninvasive imaging methods to assess liver fibrosis and cirrhosis has become an important endeavor. Although the presence of cirrhosis can be determined on the basis of liver contour abnormalities and portal hypertension stigmata using ultrasonography (US), computed tomography (CT), and conventional magnetic resonance (MR) imaging, these findings are insensitive in the detection of mild fibrosis and early cirrhosis.9, 10, 11

MR imaging has emerged as a promising modality for the assessment of diffuse liver disease. MR imaging provides excellent tissue contrast and high sensitivity to the effect of contrast agent,12, 13, 14 allowing for visualization of hepatic texture that offers the potential of noninvasively diagnosing cirrhosis. Previous studies have reported that fibrosis grades can be accurately assessed using superparamagnetic iron oxides (SPIO) or double contrast material (SPIO plus gadolinium chelates) enhanced MR imaging.15, 16 Gadoxetate disodium is an MR imaging contrast agent that was developed for evaluating the hepatobiliary system.17, 18 After intravenous injection, gadoxetate disodium is gradually taken up by hepatocytes and is eventually excreted via the biliary system.17 It is known that after gadoxetate is intravenously injected, it accumulates in hepatocytes and causes T1 shortening, which increases liver signal intensity (SI).19 In patients with fibrosis, gadoxetate disodium accumulates and causes T1 shortening preferentially in the spared liver parenchyma, resulting in fibrotic bands appearing relatively hypointense.20, 21 For this reason, it may be possible to diagnose cirrhosis on gadoxetate-enhanced MR images on the basis of hepatic texture alterations.

A histogram is a useful tool in hepatic texture analysis (TA),22, 23 as it depicts the distribution of SI levels. We postulated that cirrhotic liver would show a wider range of SI values than normal liver on hepatobiliary phase MR imaging due to fibrotic bands. To our knowledge, there have not been studies exploring cirrhosis with histograms of hepatobiliary phase images using gadoxetate disodium-enhanced MR imaging. The purpose of this study was to investigate whether histogram analysis of hepatobiliary phase MR imaging could be used as a quantitative index for determination of liver cirrhosis.

Of 37 cirrhotic patients, 35 patients showed definite morphological changes of cirrhosis on visual evaluation. The mean pixel size of the NLF group (0.64×0.42 mm) was comparable to the cirrhotic group (0.63×0.41 mm) (p>0.05). The mean SI±SD of the NLF group was 4146.7±167.0 (maximum value 10715.3; minimum value 374.8), while that of the cirrhotic group was 1960.6±143.5 (maximum value 11065.9; minimum value 216.1). There were no statistically significant differences in liver SNR between the normal group and the cirrhotic group (mean SNR±SD; 37.10±12.7 for NLF group and 32.34±15.3 for cirrhosis group, p=0.184).

The mean SI, SD, CV, and corrected CV of the NLF and cirrhotic groups are shown in Fig. 2. The SD was 167.0±142.1 (mean SD±SD) in the NLF group and 143.5±202.2 in the cirrhotic group. There was no statistically significant difference in SD between the NLF and cirrhotic groups (p=0.307). In the NLF group, the CV value was 0.041±0.009 (mean CV±SD; maximum value 0.08; minimum value 0.03). In the cirrhotic group, the CV value was 0.071±0.020 (mean CV±SD; maximum value 0.17; minimum value 0.03). The CV values of the cirrhotic group were significantly higher than those of the NLF group (p<0.001). The CV for the NLF and cirrhotic groups showed a nonnormal distribution (p<0.05, Shapiro-Wilk tests), showing skewness of 1.11 and 0.94, respectively (Fig. 3). The more skewed data for the NLF group than cirrhotic group might be explained by an inhomogeneous group of patients. The kurtosis of the NLF group was 2.03, while that of the cirrhotic group was 1.95. The Spearman's correlation test indicated that CV values were strongly correlated with the presence of cirrhosis with a value of 0.729 (p<0.001). The corrected CV values of the cirrhotic group were also higher than the NLF group (0.468±0.20 vs. 0.349±0.22) (p<0.001).

Fig. 2
Box plots show the results of analysis of (A) the SD, (B) CV values, and (C) corrected CV for the NLF and cirrhotic groups. Each box stretches from the 25th percentile at lower edge to the 75th percentile at upper edge; the median is shown as a line across the box. There are two adjacent values below and above the box: the largest value is below the upper inner limit and the smallest value is above the lower inner limit. Outside values below or above the inner limit but within the outer limit are outliers. Star marks in Fig. 2A indicates outlying values meaning wide range of pixel values in cirrhosis. SD, standard deviation; CV, coefficient of variation; NLF, normal liver function.

• Click for larger image

Fig. 3
Histograms of NLF and cirrhosis groups. (A) The CV for NLF group showed a normal distribution, showing a skewness value of 1.11. (B) The CV for cirrhosis group showed a normal distribution, showing a skewness value of 0.94. CV, coefficient of variation; NLF, normal liver function.

• Click for larger image

Sensitivities and specificities at various CV values were calculated for distinguishing normal from cirrhotic patients (Table 3). The cut-off value among CVs for distinguishing normal from cirrhotic patients with the best accuracy was 0.052 (sensitivity 83.8% and specificity 88.5%).

Table 3
Various Cut-Off Values of CV to Distinguish Normal Liver Function from Cirrhosis

• Click for larger image

Our results suggest that histogram analysis of hepatobiliary phase MRI can be helpful for differentiating patients with NLF and those with cirrhosis. Among the parameters of histograms, CV and corrected CV values of histograms were significantly different between NLF and cirrhotic groups. Our study suggests that histogram analysis holds the potential to provide a quantitative index of liver cirrhosis. Cirrhosis induces several morphologic changes including enlargement of the caudate lobe and the left lateral segment of the liver, atrophy of the right hepatic lobe and the left medial segment, nodularity of the liver surface, coarse liver architecture, ascites, splenomegaly, and development of collaterals.9, 24, 25, 26 However, most are based on morphologic changes in the liver that occur late in progression of liver disease or secondary signs of portal hypertension. These findings have limited utility for the detection of early cirrhosis and the grading of advanced fibrosis. The diagnosis of hepatic fibrosis of an intermediate degree is difficult because hepatic parenchymal textural alterations are often subtle. Although there was overlap of CV values between the NLF and cirrhotic groups, we were able to reliably differentiate between the two groups with the CVs of hepatobiliary phase imaging. Our study also introduced a simple and practical method for determining the distribution of pixel values. With our method, the SD and average value of ROIs can be measured with the picture archiving and communication system (PACS) and the CV can be easily calculated. Accordingly, we suggest that CVs of the hepatobiliary phase can be used as a screening measurement tool for evaluating hepatic parenchymal texture and aid in the diagnosis of cirrhosis in daily practice.

Biopsy remains the clinical reference standard for the diagnosis of liver fibrosis. Serum biomarkers for liver fibrosis have been described, but their clinical effectiveness has not been established. The development of noninvasive imaging methods to detect fibrosis has become a major interest for clinicians who hope for more efficient and effective serial follow-up of patients, documentation of temporal changes, and assessment of therapy response. Some investigators have proposed that it may be possible to diagnosis and grade fibrosis with MR imaging by measuring viscoelasticity27, 28, 29 or water diffusion.21, 30 A different approach is to directly visualize fibrotic tissue. In previous reports, fibrosis has been depicted as hyperintense reticulations against a hypointense liver background after the administration of SPIOs or combination of SPIO and gadolinium-based contrast material.15, 16, 31 The hyperintense reticulations, which are postulated to represent septal fibrosis, can be observed in cirrhotic liver tissue. Gadoxetate disodium-enhanced MR imaging has been suggested as a reliable method for staging hepatic fibrosis. Watanabe, et al.21 reported that contrast enhancement index on gadoxetate disodium-enhanced MR imaging is more reliable for staging hepatic fibrosis than diffusion-weighted MR imaging and clinical parameters. Our results suggest that liver cirrhosis can be accurately evaluated using TA of the hepatobiliary phase. With a cut-off value of 0.052, sensitivity and specificity values greater than 84% were achieved. Combined texture and secondary imaging features would likely further improve diagnostic performance, although this was not evaluated in this study. Moreover, as contrast agent enhanced MRI is part of routine techniques for liver imaging, there is no need to acquire additional sequences such as diffusion-weighted sequence or MR elastography.

A histogram is a graphical representation of the distribution of data.22, 23 Histograms can be computed in the PACS program, which makes it possible to assess the distribution of each pixel value. In PACS, the SI value within a ROI is shown on the monitor by an automatic calculation of the selected ROI. The histogram illustrates how the pixels in an image are distributed by graphing the number of pixels, while the SD represents the variation in the SI values. The vertical axis shows the total number of pixels with a given value. The CV is a normalized measure of dispersion of a probability distribution that can be used when comparing between data sets with different units or widely different means. The CV values of the hepatobiliary phase on gadoxetate disodium-enhanced MRI that are calculated on the ROI histogram reflect the homogeneity or heterogeneity of the hepatic parenchyma. Smaller CV values are thought to represent normal liver, while larger CV values reflect the possibility of cirrhotic liver in which with fibrotic bands are thought to cause reflections of various signal intensities. Although differences in the CV values between NLF and cirrhotic groups were observed, there was definite overlap of CV values between NLF and cirrhotic groups. This was presumably because variable stages of fibrosis were included in this study and small vessels or bile ducts could have been included in the ROIs. The evaluation of fibrosis, therefore, depends on the choice of ROI, which should be improved in the future.

Texture refers to the distribution of brightness and darkness within the image. TA is based on the quantification of complex mathematical patterns, which exist in the gray-level distribution of the pixels of images. TA evaluates the spatial location and SI of each pixel in the examined area.32 There are many features that can be applied for TA. Among them, the statistical measurement methods represent the statistical aspect of texture evaluation, concerning the interrelationships among pixels. Other texture feature analysis approaches are based on models. Jirák, et al.32 reported that the best discriminative feature for differentiating cirrhosis from normal was kurtosis; other differentiating features included difference entropy, gradient skewness, and skewness. Combination of texture parameters such as mean gray-scale value, contrast, angular second moment, or entropy have also been used for differentiating hepatic fibrosis from normal liver.33 CV and corrected CV have been suggested as accurate quantitative scores for liver texture features.15 Our TA is based only on the variability of SI on hepatobiliary phase images. More sophisticated texture features can be assessed and may facilitate better performance. Regarding the sequence of MR imaging, the equilibrium phase after administration of gadolinium-based contrast agent has been suggested to reflect the degree of fibrosis most accurately.33 Compared to equilibrium phase of gadolinium-based contrast agent, hepatobiliary phase images using gadoxetate disodium have better contrast-to-noise ratio (CNR), resulting in superior conspicuity for fibrotic bands in cirrhotic liver. Therefore, hepatobiliary phase imaging may be the optimal sequence for evaluating TA.

Our study was limited by its retrospective design and inclusion of cirrhotic patients who had advanced fibrosis with obvious imaging stigmata of liver disease or based on clinical findings. Second, we could not analyze the data according to the grade of fibrosis because no histopathologic proof was obtained for cirrhotic patients. Further studies are needed to determine if this method can be used in grading hepatic fibrosis. Third, the enhancement of hepatobiliary phase depends on several factors although organic anionic transport polypeptides 8 (OATP8, also known as OATP1B1/3) are the dominant determinant.34 OATP8 is thought to be responsible for uptake by hepatocytes of gadoxetic acid.35 OATP8 represent members of the OATP superfamily that mediate the sodium-independent hepatic uptake of a diverse range of organic compounds including bile acids, thyroid hormones, bilirubin gluconides, steroid conjugates, and numerous drugs such as statins or sartans.36 Therefore, hepatic uptake of gadoxetic acid may be influenced in patients taking these organic compounds because of competitive uptake. Also the SNR of the hepatobiliary phase may be influenced by modulating the flip angle. Increasing the flip angle in the hepatobiliary phase T1-weighted imaging increases the SNR of the liver, while decreasing the SNR of non-hepatocyte-containing tissues such as fibrosis results in an increased lesion to liver CNR with subsequent improved lesion conspicuity.37 This effect is due to an increase in T1-weighting, resulting in an increase in contrast between hepatic fibrotic tissues and functioning hepatic parenchyma.38, 39 Thus the quantitative value in our study may not be applied equally when flip angle is modulated from its conventional setting (10-15°). Fourth, the pixel size of hepatobiliary phase was not exactly the same in each patient. However, the pixel size was comparable in both groups and there was no statistical difference therein between the two groups. Fifth, we did not evaluate the interobserver variability of the measurement of ROIs. This issue should be further studied in a larger series. Finally, we excluded patients with impaired hepatic enhancement because this group of patients cannot be effectively imaged with gadoxetate disodium-enhanced MRI. Considering that hepatic function is closely related to hepatic uptake of gadoxetate, this may have introduced a component of selection bias to our study.

In conclusion, the histogram analysis of hepatobiliary phase MRI reflects the homogeneity or heterogeneity of the hepatic parenchyma. The CV values of histogram on hepatobiliary phase can be used as a quantitative value that can determine the presence of liver cirrhosis.