Differentiation of pyogenic hepatic abscesses from malignant mimickers using multislicebased texture acquired from contrastenhanced computed tomography

Shi-Teng Suo， Zhi-Guo Zhuang， Meng-Qiu Cao， Li-Jun Qian， Xin Wang，Run-Lin Gao， Yu Fan and Jian-Rong XuShanghai， China

?

Differentiation of pyogenic hepatic abscesses from malignant mimickers using multislicebased texture acquired from contrastenhanced computed tomography

Shi-Teng Suo， Zhi-Guo Zhuang， Meng-Qiu Cao， Li-Jun Qian， Xin Wang，Run-Lin Gao， Yu Fan and Jian-Rong Xu
Shanghai， China

BACKGROUND： Pyogenic hepatic abscess may mimic primary or secondary carcinoma of the liver on contrast-enhanced computed tomography （CECT）. The present study was to explore the usefulness of the analysis of multislice-based texture acquired from CECT in the differentiation between pyogenic hepatic abscesses and malignant mimickers.

METHODS： This retrospective study included 25 abscesses in 20 patients and 33 tumors in 26 subjects who underwent CECT. To make comparison， we also enrolled 19 patients with hepatic single simple cyst. The images from CECT were analyzed using a Laplacian of Gaussian band-pass filter （5 filter levels with sigma weighting ranging from 1.0 to 2.5）. We also quantified the uniformity， entropy， kurtosis and skewness of the multislice-based texture at different sigma weightings. Statistical significance for these parameters was tested with oneway ANOVA followed by Tukey honestly significant difference （HSD） test. Diagnostic performance was evaluated using the receiver operating characteristic （ROC） curve analysis.

RESULTS： There were significant differences in entropy and uniformity at all sigma weightings （P＜0.001） among hepatic abscesses， malignant mimickers and simple cysts. The significant difference in kurtosis and skewness was shown at sigma 1.8 and 2.0 weightings （P=0.002-0.006）. Tukey HSD test showed that the abscesses had a significantly higher entropy and lower uniformity compared with malignant mimickers （P=0.000-0.004）. Entropy （at a sigma 2.0 weighting） had the largest area under the ROC curve （0.888） in differentiating abscesses from malignant mimickers， with a sensitivity of 81.8% and a specificity of 88.0% when the cutoff value was set to 3.64.

CONCLUSION： Multislice-based texture analysis may be useful for differentiating pyogenic hepatic abscesses from malignant mimickers.

（Hepatobiliary Pancreat Dis Int 2016；15：391-398）

texture analysis；

contrast-enhanced computed tomography；

liver；

pyogenic hepatic abscess；

malignant mimicker

Introduction

Hepatic abscess is a pus-filled mass inside the liver which is classified as pyogenic， amebic or fungal.［1］The incidence of hepatic abscess is relatively low but it is a potentially life-threatening disease.［2］Pyogenic hepatic abscesses are usually treated with antibiotics and percutaneous drainage. Modern medical imaging techniques such as computed tomography （CT） and magnetic resonance imaging （MRI） play a pivotal role in the diagnosis of hepatic abscess.［3］Hepatic abscesses usually appear as thick-walled lesions with low attenuation at CT and show increased peripheral rim enhancement at contrast-enhanced CT （CECT）.［4］However， imaging findings are often nonspecific because certain primary or secondary carcinoma of the liver may develop central necrosis which may mimic the appearance of hepatic abscesses.［5］Therefore， though various studies using CT or MRI havebeen conducted to differentiate hepatic abscesses from malignant mimickers， many overlapping image features still remain.［1， 3， 6， 7］

With recent advances in computer-aided diagnosis，texture analysis of CT has great potential to depict tissue morphological complexity and gives new insights into the complicated manifestations of diseases such as non-small cell lung cancer，［8-10］ground-glass nodules，［11］colorectal cancer，［12］gastric tumor［13］and hepatic tumor.［14， 15］Generally， texture analysis refers to various mathematic methods used to measure the distribution of pixels with different gray-level intensities.［16］Computer-based radiologic image texture analysis may maximize the information obtained from the medical images in clinical diagnoses in an objective， quantitative and reproducible way.［17］

CECT is widely used in the diagnosis of liver disease because it provides useful information about the enhancement patterns and the heterogeneity of lesions. Thus， although the images of hepatic abscesses and malignant mimickers are similar by eyes， the texture features extracted from CECT may indirectly capture the tissue heterogeneity and cell morphology. Microscopic biological changes within tissues could be reflected on texture heterogeneity.

The purpose of the study was to explore whether texture analysis of CECT helps differentiate pyogenic hepatic abscesses from malignant mimickers.

Methods

Study population

This study was approved by our Institutional Review Board. The requirement for written informed consent from patients was waived due to the retrospective study design. A systematic search of liver CECT image database was queried for clinically diagnosed hepatic abscess and malignant hepatic tumor with search key “cystic”，“l(fā)iquefactive” or “necrotic” between November 2011 and November 2013. Only cases with initial diagnosis of focal liver abscess or malignant hepatic tumor that underwent CECT before treatment were enrolled. Inclusion and exclusion criteria are listed in Table 1. The hepatic abscess group consisted of 25 abscesses in 20 patients and the malignant mimickers had 33 tumors in 26 subjects. The 26 patients with malignant mimickers included those with necrotic hepatocellular carcinoma （n=9）， massforming cholangiocarcinoma （n=6） and hepatic metastasis ［n=11， with primary malignant lesions in the stomach （n=6）， colorectum （n=3）， duodenum （n=1） and esophagus （n=1）］. Since the hepatic abscesses and malignant mimickers included in the study were all cystic focal liver lesions， to make comparison， we enrolled 19 patients with hepatic single simple cyst diagnosed in the same period. Anechoic lesions at ultrasonography were accepted as hepatic cysts if the patients were clinically asymptomatic and with negative serological tests.［18］The detailed patient characteristics for all groups are shown in Table 2.

Table 1. Inclusion and exclusion criteria for hepatic abscesses and malignant mimickers

Table 2. Patients characteristics

CT examination

CECT scans were performed with a 128-row helical quadruple-phase （i.e.， unenhanced， hepatic arterial， portal venous and delayed phases） CT scanner （Somatom Definition AS+； Siemens Healthcare， Forchheim， Germany）. A nonionic contrast agent （Ultravist 370 mgI/mL； Schering， Berlin， Germany） was administered （1.5 mL/kg body weight） at the rate of 4.0 mL/s， followed by 50 mL salineflush at the same rate， through an 18-gauge intravenous cannula inserted into the antecubital vein with an automatic dual-headed injector （Optivantage DH； Mallinckrodt， St Louis， MO， USA）. The arterial phase， portal venous phase and delayed phase scans were obtained at 25-30 seconds， 60-70 seconds， and 3 minutes respectively after the initiation of contrast administration.

Acquisition parameters were set as follows： tube voltage， 120 kV； effective tube current， 280 mAs； rotation time， 0.5 seconds； pitch， 1； section collimation， 1.5 mm；matrix， 512×512； field of view， 380 mm； and pixel spacing， 0.74 ×0.74 mm. Image reconstruction was performed with a soft tissue kernel and slice thickness of 1 mm and 10 mm.

Texture analysis

All DICOM data were transferred to a personal computer， and the CECT images of portal venous phase were chosen for texture analysis in consideration of the contrast between lesions and surrounding liver parenchyma. For each lesion， five consecutive 1-mm CECT axial slices in the central portion of the lesion were selected to avoid areas of artifact. Two radiologists independently delineated the lesion outline with a polygonal volume of interest （VOI） on the five slices for one lesion. VOIs were placed in the necrotic cavities of lesions excluding vascularized areas on CECT. Both of the two radiologists were blind to the clinical outcome.

Texture analysis mainly included two steps： （a） image filtration and （b） texture feature quantification， as previously used for non-small cell lung cancer［8-10］and colorectal cancer.［12］Image threshold was set to 0 to exclude air and fat tissue which were ＜0 Hounsfield units （HU）. Laplacian of Gaussian （LoG） band-pass filter was employed for image filtration in the spatial domain. LoG filter is a combination of Laplacian operator and Gaussian filter. Since Laplacian operator may detect edges as well as noise， convolution with a Gaussian kernel was first used to smooth the image. Different filter sigma parameters derive a series of images with different textural coarseness. The degree of roughness was decided by sigma from 1.0 to 2.5 in the study， where 1.0 indicates fine texture （features of 4 pixels in width）； 1.5， 1.8 and 2.0 mean various degrees of medium texture （features of 6， 8 and 10 pixels in width， respectively）； and 2.5 implies coarse texture （features of 12 pixels in width）.［9］After filtration，each pixel showing a negative number was set to zero.

Histogram-based texture features including entropy，uniformity， kurtosis and skewness were quantified on each VOI with and without image filtration. Entropy is a measure of gray level variation in a histogram； this is defined as zero when all data are the same and the number is bigger when the distribution is irregular. Uniformity refers to the distribution of gray level and the larger this number， the more uniform the distribution. Kurtosis is a measure of peakedness of the distribution， equals three if the histogram is Gaussian， larger than three if the histogram has a sharper peak. Skewness reflects asymmetry of the distribution， positive if more points lie to the left of the mean and negative if the opposite. Mean attenuation of unfiltered images within the VOI was also recorded and expressed in HU.

The entire image analysis algorithm was achieved by using an in-house program written in MATLAB （version R2011b； MathWorks， Natick， MA， USA）.

Statistical analysis

Interobserver variability of the two radiologists for texture feature measurements was analyzed by Bland-Altman test and the intraclass correlation coefficient （ICC）. The ICC range of 0.00 to 0.20 indicated poor degree of agreement； 0.21 to 0.40， fair； 0.41 to 0.60， moderate； 0.61 to 0.80， good； 0.81 to 1.00， excellent. Oneway ANOVA test was conducted to study the significance or otherwise of the difference among hepatic abscesses，malignant mimickers and simple cysts in terms of each texture feature. The Tukey honestly significant difference （HSD） test was used for multiple comparisons when comparing hepatic abscesses and malignant mimickers. The receiver operating characteristic （ROC） analysis was used to evaluate the sensitivity and specificity of texture features for the two groups. Diagnostic performance was determined by calculating the area under the ROC curve （AUC）. Higher values of AUC indicate better classification. Cutoff values were established by calculating the maximal Youden index （Youden index=sensitivity+specif icity-1）. The positive predictive value， negative predictive value， accuracy， sensitivity and specificity were obtained at cutoff value of each texture feature.

Statistical analyses were performed by using softwares SPSS （version 17； SPSS， Chicago， IL， USA） and MedCalc （version 11.4.2； MedCalc Software， Mariakerke， Belgium）. A P value less than 0.05 was considered statistically significant.

Results

Interobserver variability

ICC between the two radiologists were above 0.60 for all texture features， indicating a good degree of agreement. When a sigma 1.8 weighting was used， the interobserver reproducibility was excellent for skewness， entropy and uniformity （ICC=0.85， 0.93 and 0.94， respectively）， good for and kurtosis （ICC=0.71）. Excellent degree of agree-ment was also observed for mean attenuation of unfiltered images （ICC=0.98）. Fig. 1 displays the Bland-Altman plots with 95% limits of agreement for the texture parameters （sigma=1.8）.

Fig. 2. Texture analysis procedure for one slice. A polygonal region of interest delineating the largest cross-sectional lesion area on the CECT image was selected. Textures of different degrees of coarseness （5 filter levels with sigma weighting ranging from 1.0 to 2.5）were listed below the CECT image.

Fig. 3. Examples of CECT texture analysis （sigma=2.0） in different entities. A： a 73-year-old male diagnosed with a hepatic abscess； B：a 62-year-old female with a necrotic hepatocellular carcinoma； C：a 62-year-old female with a mass-forming cholangiocarcinoma；and D： a 54-year-old male with hepatic metastasis from the duodenum. The texture entropy （sigma=2.0） of the four cases were 3.79， 3.46， 3.31 and 3.49， respectively.

Texture analysis

Texture analysis procedure for one slice is shown in Fig. 2. The mean VOI of the three groups was as follows： hepatic abscess， 2421±2168 mm3（range 262-9131）；malignant mimicker， 1970±1867 mm3（range 382-7801）；simple cyst， 865±696 mm3（range 163-2794）. Fig. 3 shows CECT texture analysis of four cases clinically diagnosed with hepatic abscesses or malignant mimickers. Table 3 presents texture features for the 3 groups and results of one-way ANOVA test. There were significant differences in entropy and uniformity at all sigma weightings （P＜0.001）， and in kurtosis and skewness only at sigma 1.8 and 2.0 weightings （P=0.002-0.006） when hepatic abscesses， malignant mimickers and simple cysts were compared. Tukey HSD test showed that both entropy and uniformity at all sigma weightings were significantly different between hepatic abscesses and malignant mimickers （P=0.000-0.004）. Hepatic abscesses tended to have a higher entropy and a lower uniformity than malignant mimickers. However， no significant difference was noted for mean attenuation of unfiltered images （P=0.31） and kurtosis or skewness at all sigma weightings （P=0.55-0.99）. At all sigma weightings except 2.0 and 2.5， simple cysts had the best uniformity among the three entities. Fig. 4 displays the box plots of entropy in different subgroups at all sigma weightings.

Table 3. Mean attenuation， entropy， uniformity， kurtosis and skewness at all sigma weightings for hepatic abscesses， malignant mimickers and simple cysts

Tukey HSD test showed that the difference of entropy was comparable to that of uniformity between hepatic abscesses and malignant mimickers in all sigma weightings. ROC analysis revealed that entropy had larger AUCsthan uniformity except using a sigma 1.0 weighting （0.741 vs 0.743）. We focused on the role of entropy in diagnostic accuracy （Table 4 and Fig. 5） and found that entropy had the highest efficacy by using a sigma 2.0 weighting， the AUC was 0.888 in differentiating hepatic abscesses from malignant mimickers. When we set up the cutoff value to 3.64， the sensitivity and specificity for the diagnosis of malignant mimickers was 81.8% and 88.0%， respectively. The AUC of mean attenuation of unfiltered images was only 0.704， and the sensitivity and specificity for hepatic abscess were 78.8% and 76.0% respectively at a cutoff value of 27.30.

Fig. 4. Box plots of entropy in different subgroups for unfiltered images （0） and filtered images at all sigma weightings （1.0-2.5）. Top and bottom of boxes： 25%-75% percentiles of data； line in box： median value； ●： outliers.

Fig. 5. ROC analysis of entropy （at all sigma weightings） and mean attenuation of unfiltered images considering the sensitivity and specificity for the differentiation between hepatic abscesses and malignant mimickers. The ROC curves demonstrated that using a sigma 2.0 weighting， entropy had the highest AUC of 0.888，with a sensitivity of 81.8% and a specificity of 88.0% when the cutoff value was set to 3.64.

Discussion

Because some necrotic hepatic malignancies mimic hepatic abscesses clinically and radiologically， it is of utmost importance to differentiate these two lesions. The typical CT imaging features of hepatic abscess are presence of air，double-target sign and rim enhancement. Clinical manifestations like fever， rigor and severe abdominal pain also aid the diagnosis.［19］However， hepatic malignancy such as necrotic hepatocellular carcinoma， mass-forming cholangiocarcinoma and hepatic metastasis may share similar imaging features especially rim enhancement. Thus， it still remains a challenge to differentiate the two entities particularly in the absence of typical clinical symptoms.

Texture analysis is an image processing method that is used to quantitate the tissue heterogeneity by computing the distribution of texture coarseness and irregularity within an area or a volume. The texture analysis used in this study was based on the one described by Ganeshan and co-workers［8， 9， 20］who assessed the potential value of texture analysis on regular CT or CECT to improve the prognostic value of standard imaging. Texture analysis showed that tumor heterogeneity is correlated with the intratumoral histopathological changes， such as variation in cellularity， angiogenesis， extravascular extracellular matrix， and areas of necrosis.［16］We hypothesized that histopathological features of the two etiologically different entities at the microscopic level manifest themselves as different texture heterogeneity patterns on CECT.

Table 4. Receiver operating characteristic （ROC） analysis of mean attenuation of unfiltered images and entropy at all sigma weightings in the differentiation of hepatic abscesses from malignant mimickers

Our data support our hypothesis. We found thatmultislice-based texture analysis on CECT was useful in differentiating hepatic abscesses from malignant mimickers. In general， hepatic abscess was more radiologically heterogeneous than malignant mimicker. At a medium texture degree （sigma=2.0）， the classification accuracy of entropy could reach to 84.5%. This can be explained from the aspect of the histopathological mechanism and biochemical components of the two entities. Hepatic abscess is an infection caused by microorganisms and the abscess cavity is filled with pus， which is a yellowish brown， viscous fluid that consists of inflammatory cells，microorganisms， necrotic tissue and proteinaceous exuded plasma with high viscosity and cellularity.［5］In contrast， spontaneous necrosis of liver cancer is caused by the tumor blood supply shortage due to the rapid proliferation and enlargement of cancer cells. On account of ischemic damage， the tumor is gradually replaced by central coagulative necrosis， circumferential fibrosis and dense infiltrates of inflammatory cells. Normal cellular morphology changes even fades， while the general outline of tissue structure still remains. The necrotic portion of tumor is mainly solid with a relatively stable structure compared to the abscess cavity， which is in a semi-fluid state with more cellularity. Therefore， the abscess cavity seems to have more tissue heterogeneity than tumor necrosis， leading to a higher entropy and lower uniformity on CECT images. Fig. 6 presented photomicrographs of histological sections from hepatic abscess and necrotic hepatocellular carcinoma specimens. At fine and medium texture degrees （sigma=1.0， 1.5 and 1.8）， simple cysts had the most uniform texture features among the three entities， which is mainly because simple cysts have watery contents without structures.

Some issues about the texture analysis should be discussed. First， the heterogeneity visible on CECT represents not only biological heterogeneity but also photon noise. LoG filter was first used and at different sigma weightings， CECT texture analysis reduced the photon noise and enhanced biologic heterogeneity at different levels.［16］We found that even without any filtration， there were significant differences in entropy and uniformity between hepatic abscesses and malignant mimickers. This is probably because scanning parameters and reconstruction algorithm are equal， photon noise has similar influence on image equality of individuals， and the influence is minor as compared with biological heterogeneity. Second， multislice images were assessed instead of a single axial image which was the case in previous studies.［8， 10， 13］The commonly-used largest cross-sectional image area allows only a limited evaluation of the distribution of mean attenuations and is prone to sampling bias， which is less representative of tissue heterogeneity.［21］

Fig. 6. Histology （HE staining； original magnification ×40） of hepatic abscess （A） and necrotic hepatocellular carcinoma （B）. The abscess cavity （*） is filled with pus containing inflammatory cells，microorganisms， necrotic tissue and proteinaceous exuded plasma with high cellularity. The necrotic portion of tumor is mainly made up of coagulative necrosis of carcinoma （▼）， and surrounded by viable carcinoma cells （★） and circumferential fibrosis （arrow）.

This study does have several limitations. First， due to the retrospective design， many patients had to be excluded. Although we recruited the patients based on our inclusion and exclusion criteria， selection bias may also exist. Second， histopathological correlation of texture features was not evaluated since not all patients received biopsy or surgical resection. Thus， our proposed associations remain speculative. Of all the patients without histopathological results， 11 （55.0%） patients had hepatic abscess and 9 （34.6%） had metastasis. Nevertheless， we thought that for hepatic abscess， a clinical follow-up period of longer than 3 months and decreased lesion size on follow-up CT or MRI after antibiotic treatment and percutaneous drainage was sufficient to exclude other potential diagnoses；［7］while for metastasis， primary tumor in other organs and increased lesion size or/and increased lesion number on a 3-month follow-up CT or MRI was sufficient for the diagnosis of malignancy.［19］Third， the portal venous phase was obtained using a time delay of 60-70 seconds for each patient. An individualbased scanning time window may better reveal the texture characteristics， which should be investigated in the future study. Fourth， we included only pyogenic hepatic abscess and ruled out amebic or fungal abscess. We also excluded pyogenic hepatic abscess in early stage when inflammation， hyperemia and edema dominated but no pus formed. Fifth， we did not compare other qualitative images such as enhancement patterns， or other image modalities such as MRI. It has been reported that diffusion-weighted MRI and gadolinium-ethoxybenzyldiethylenetriamine penta acetic acid-enhanced MRI have added value in the differential diagnosis of hepatic abscesses and malignant mimickers.［5， 7， 19］

To our knowledge， this was the first study assessing the potential usefulness of texture features of tissue heterogeneity extracted from CECT images to differentiate pyogenic hepatic abscesses from malignant mimickers. Entropy and uniformity are helpful to differentiate thesetwo entities. Larger prospective studies with histopathological results are needed to further confirm the relationship between CECT texture features and disease microenvironment characteristics.

Contributors： SST and ZZG proposed the study and contributed equally to this work. SST， ZZG and CMQ performed research and wrote the first draft. SST and ZZG collected and analyzed the data. All authors contributed to the design and interpretation of the study and to further drafts. XJR is the guarantor.

Funding： This study was supported by grants from Medical Engineering Cross Research Foundation of Shanghai Jiaotong University （YG2013MS37 and YG2012MS16） and the National Natural Science Foundation of China （81201172， 81371660 and 81371622）.

Ethical approval： This study was approved by the Institutional Review Board of Renji Hospital， School of Medicine， Shanghai Jiaotong University， Shanghai， China.

Competing interest： No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.

References

1 Méndez RJ， Schiebler ML， Outwater EK， Kressel HY. Hepatic abscesses： MR imaging findings. Radiology 1994；190：431-436.

2 Chen SC， Huang CC， Tsai SJ， Yen CH， Lin DB， Wang PH， et al. Severity of disease as main predictor for mortality in patients with pyogenic liver abscess. Am J Surg 2009；198：164-172.

3 Mortelé KJ， Segatto E， Ros PR. The infected liver： radiologicpathologic correlation. Radiographics 2004；24：937-955.

4 Mortelé KJ， Ros PR. Cystic focal liver lesions in the adult： differential CT and MR imaging features. Radiographics 2001；21：895-910.

5 Chan JH， Tsui EY， Luk SH， Fung AS， Yuen MK， Szeto ML， et al. Diffusion-weighted MR imaging of the liver： distinguishing hepatic abscess from cystic or necrotic tumor. Abdom Imaging 2001；26：161-165.

6 Brown KT， Gandhi RT， Covey AM， Brody LA， Getrajdman GI. Pylephlebitis and liver abscess mimicking hepatocellular carcinoma. Hepatobiliary Pancreat Dis Int 2003；2：221-225.

7 Park HJ， Kim SH， Jang KM， Lee SJ， Park MJ， Choi D. Differentiating hepatic abscess from malignant mimickers： value of diffusion-weighted imaging with an emphasis on the periphery of the lesion. J Magn Reson Imaging 2013；38：1333-1341.

8 Ganeshan B， Panayiotou E， Burnand K， Dizdarevic S， Miles K. Tumour heterogeneity in non-small cell lung carcinoma assessed by CT texture analysis： a potential marker of survival. Eur Radiol 2012；22：796-802.

9 Ganeshan B， Goh V， Mandeville HC， Ng QS， Hoskin PJ， Miles KA. Non-small cell lung cancer： histopathologic correlates for texture parameters at CT. Radiology 2013；266：326-336.

10 Ravanelli M， Farina D， Morassi M， Roca E， Cavalleri G， Tassi G， et al. Texture analysis of advanced non-small cell lung cancer （NSCLC） on contrast-enhanced computed tomography：prediction of the response to the first-line chemotherapy. Eur Radiol 2013；23：3450-3455.

11 Lee SH， Lee SM， Goo JM， Kim KG， Kim YJ， Park CM. Usefulness of texture analysis in differentiating transient from persistent part-solid nodules （PSNs）： a retrospective study. PLoS One 2014；9：e85167.

12 Ng F， Ganeshan B， Kozarski R， Miles KA， Goh V. Assessment of primary colorectal cancer heterogeneity by using whole-tumor texture analysis： contrast-enhanced CT texture as a biomarker of 5-year survival. Radiology 2013；266：177-184.

13 Ba-Ssalamah A， Muin D， Schernthaner R， Kulinna-Cosentini C，Bastati N， Stift J， et al. Texture-based classification of different gastric tumors at contrast-enhanced CT. Eur J Radiol 2013；82：e537-543.

14 Gensure RH， Foran DJ， Lee VM， Gendel VM， Jabbour SK，Carpizo DR， et al. Evaluation of hepatic tumor response to yttrium-90 radioembolization therapy using texture signatures generated from contrast-enhanced CT images. Acad Radiol 2012；19：1201-1207.

15 Yap FY， Bui JT， Knuttinen MG， Walzer NM， Cotler SJ， Owens CA， et al. Quantitative morphometric analysis of hepatocellular carcinoma： development of a programmed algorithm and preliminary application. Diagn Interv Radiol 2013；19：97-105.

16 Davnall F， Yip CS， Ljungqvist G， Selmi M， Ng F， Sanghera B， et al. Assessment of tumor heterogeneity： an emerging imaging tool for clinical practice？ Insights Imaging 2012；3：573-589.

17 Lambin P， Rios-Velazquez E， Leijenaar R， Carvalho S， van Stiphout RG， Granton P， et al. Radiomics： extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012；48：441-446.

18 Oru? E， Y?ld?r?m N， Topal NB， K?l??turgay S， Akg?z S， Savc? G. The role of diffusion-weighted MRI in the classification of liver hydatid cysts and differentiation of simple cysts and abscesses from hydatid cysts. Diagn Interv Radiol 2010；16：279-287.

19 Choi SH， Lee CH， Kim BH， Lee J， Choi JW， Park YS， et al. “Nondefect” of arterial enhancing rim on hepatobiliary phase in 3.0-T gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid-enhanced liver magnetic resonance imaging： distinguishing hepatic abscess from metastasis. J Comput Assist Tomogr 2013；37：849-855.

20 Ganeshan B， Miles KA， Young RC， Chatwin CR. Texture analysis in non-contrast enhanced CT： impact of malignancy on texture in apparently disease-free areas of the liver. Eur J Radiol 2009；70：101-110.

21 Ng F， Kozarski R， Ganeshan B， Goh V. Assessment of tumor heterogeneity by CT texture analysis： can the largest cross-sectional area be used as an alternative to whole tumor analysis？Eur J Radiol 2013；82：342-348.

Accepted after revision June 23， 2015

Author Affiliations： Department of Radiology （Suo ST， Zhuang ZG， Cao MQ， Qian LJ， Fan Y and Xu JR）， Department of Hepatic Surgery （Wang X） and Department of Pathology （Gao RL）， Renji Hospital， School of Medicine， Shanghai Jiaotong University， Shanghai 200127， China

Jian-Rong Xu， MD， PhD， Department of Radiology， Renji Hospital， School of Medicine， Shanghai Jiaotong University，No. 160 Pujian Rd.， Shanghai 200127， China （Tel： +86-21-68383570； Fax：+86-21-50896639； Email： xujianrong_renji@163.com）
? 2016， Hepatobiliary Pancreat Dis Int. All rights reserved.

10.1016/S1499-3872（15）60031-5
Published online November 9， 2015.

October 8， 2014