Variation in textural parameters of hepatic lesions during contrast medium injection

Abstract

Background

Textural parameters extracted using quantitative imaging techniques have been shown to have prognostic value for hepatocellular carcinoma (HCC).

Purpose

To evaluate whether the contrast medium timing of the image acquisition affects the reproducibility of textural parameters in HCC and hepatic tissue.

Material and Methods

This retrospective study included 17 patients with 37 HCC lesions. Perfusion computed tomography (CT) was obtained after 50 mL contrast medium injection. HCC lesions were segmented for analysis. The gray-level co-occurrence (GLCM) textural analysis parameters, homogeneity, energy, entropy, inertia, and correlation were calculated. Variation was quantified by calculating the SD of each parameter during respective perfusion series and the inter lesion variation as the SD among the lesions.

Results

The average change in texture parameters in both HCC and hepatic tissue per second after injection was 0.01% to 0.3% of the respective texture parameter. In HCC, the average variation in homogeneity, energy, and entropy within each lesion after contrast medium injection was significantly less than the variation observed among the lesions (23% to 74%, P < 0.001). Significant differences in energy, entropy, inertia, and correlation between hepatic tissue and HCC were observed. However, when considering the intra-individual variation of hepatic tissue over time, only the HCC parameter energy was significantly outside that 95% confidence interval (P < 0.02).

Conclusion

The contrast medium timing does not affect the reproducibility of textural parameters in HCC and hepatic tissue. Thus, contrast medium timing should not be an issue at CT texture analysis of HCC.

Keywords

Computed tomography quantitative computed tomography abdomen gastrointestinal liver adults

Introduction

Hepatocellular cancer (HCC) is the second largest cause of global cancer-related death and has the highest health-related economic cost worldwide (1). Current available treatment includes chemotherapy, surgical resection, liver transplantation, and various locoregional treatments such as percutaneous ethanol injection (PEI), radiofrequency ablation (RFA), trans-arterial chemoembolization (TACE), and radioembolization (2).

Several imaging-based methods such as magnetic resonance imaging (MRI), contrast-enhanced computed tomography (CECT), and perfusion CT (PCT) have been used for both initial diagnosis and to assess treatment response, as these are non-invasive and repeatable. Current guidelines establish imaging as the primary way of diagnosing HCC, and biopsy for diagnosis is usually not required (3).

Features extracted using quantitative imaging techniques such as texture analysis have been shown to correlate to treatment response and molecular genetic markers for HCC (4 –6). They have also been used for the quantification of liver cirrhosis, fibrosis, and steatosis (7 –9). These features offer a method of quantifying image properties that have typically been described subjectively, such as image irregularity and coarseness. They could also reveal additional information not observable by a human examiner. Texture analysis, through the calculation of gray-level co-occurrence matrices (GLCM), quantifies the relationship between individual gray levels of voxels in a selected region of the image volume. The GLCMs are then used for the calculation of different textural parameters, which are used to quantify the heterogeneity in the selected region. The heterogeneity is believed to reflect conditions of tumor microbiology such as vascularization, tumor necrosis, local hypoxia, and differences in the rate of cell proliferation. Differences of textural parameters have been observed between cancerous tissue from normal tissue for several types of cancers and have also been shown to have prognostic value (10 –13).

However, the calculated textural parameters need to be reproducible in order to be useful in clinical practice. Previous studies have shown variation of textural parameters dependent on image acquisition parameters and due to the use of different CT scanners (14,15). A recent study on esophageal cancer has shown a variance in textural parameters depending on the use of contrast media (11). Although good reproducibility of textural parameters has been shown for CT scans of colorectal cancer and lung cancer acquired at two different timepoints, respectively (16 –19), the variation in textural parameters during contrast medium injection has to our knowledge not been studied. This needs to be elucidated as the timing of the image acquisition during contrast medium injection could affect the reproducibility of textural parameters.

The aim of the present study was to quantify and compare the variation in textural parameters in HCC lesions and hepatic tissue during contrast medium injection to the inter-individual variation.

Material and Methods

Patients

This is a retrospective observational study on patients with histologically or radiologically verified HCC recruited during 2013–2014. The patients had an additional PCT exam performed in addition to their clinical routine examination. Inclusion criteria were radiological signs of HCC and no previous treatment for HCC of any kind. Patients with more than four HCC lesions or lesions >7 cm in diameter were excluded.

Ethical approval for the study was granted by the regional ethical review board in Stockholm (approval nos. DNR 2013/1072-32 and DNR 2013/405-31) and written informed consent was obtained.

Imaging

Patients were scanned using a single Siemens Definition Flash perfusion CT scanner (Siemens AG, Erlangen, Germany). The total number of scan cycles was 26–28. The scan time was 43–54 s with a cycle time of 1.5 s between each scan. Tube voltage was 80 kV and tube current was 160 mAs. Scans were done with the use of intravenous contrast medium at a dose of 50 mL of 400 mg iodine/mL. The flow rate was 6 mL/s and the duration of the injection was 8.3 s. The scanning direction was alternating craniocaudal and images with a slice thickness of 1.5 mm were used for further analysis.

Segmentation of lesions and hepatic tissue

The lesions were first marked out on the images by a consultant radiologist. A senior consultant radiologist with 27 years of experience then manually segmented the lesion on the center timepoint of the PCT series using the slice with the largest area of the respective lesion. ImageJ 1.50e (Bethesda, MD, USA) was used for the segmentation. The resulting region of interest (ROI) was then copied to the other timepoints. The positioning was adjusted in order to compensate for movement due to breathing when necessary. A junior radiologist then segmented the hepatic tissue close to the segmented lesion on all timepoints. The median area of the ROI was 2.7 cm² for lesions and 5.8 cm² for non-malignant hepatic tissue.

A ROI was placed in the abdominal aorta in all timepoints and the timepoint where 160 HU exceeded was used as the reference timepoint for inter-individual comparison.

Texture analysis of lesions and non-malignant hepatic tissue

The ROIs of the lesions and the hepatic tissue were analyzed using the GLCM-based parameters energy, entropy, homogeneity, inertia, and correlation according to a previously described method (11).

The Hounsfield unit information of all included pixels in the segmented ROIs of the lesion and hepatic tissue was rescaled to 256 gray levels. The calculation of GLCM data was done using an in-house developed plug-in for ImageJ 1.50e. Offsets for all eight possible directions was calculated using 256 bins in order to obtain the textural parameters.

The measured textural parameters were then normalized to 1 at the timepoint where the attenuation of the ROI placed in the abdominal aorta reached 160 HU to reduce the effect of inter-individual variation.

Variability in the texture parameters over time was defined as the SD of each respective texture parameter over all timepoints for each lesion or hepatic tissue ROI. Variation in texture parameters between different HCC lesions was defined as the SD of each respective texture parameter at each timepoint for all lesions or hepatic tissue ROIs.

Statistical analysis

The percent change was calculated as the difference of the compared value and the reference value, divided by the reference value ${(value}_{comparison} - {value}_{reference}) /$ ${value}_{reference}$ . The intra-lesion values were used as the reference when intra- and inter-lesion values were compared. The lesion values were used as the reference when lesion and tissue values were compared.

The difference between the means was tested using the Mann–Whitney U test with tie correction. Linear regression was used to determine drift of textural parameters over all timepoints.

Statistics were calculated using R 3.4.3 (R Foundation, Vienna, Austria). P values < 0.05 were considered statistically significant.

Results

Patients

A total of 17 patients were included. One patient underwent two PCT scans with a time interval of three months. Each scan was analyzed separately. A total of 37 HCC lesions were identified. Histopathological data verifying the diagnosis of HCC were available for eight of the participants. In the other patients, the European Association for the Study of the Liver (EASL) criteria were used for the diagnosis (3). The maximum diameter of the lesions was in the range of 10–55 mm. All participants except one had liver cirrhosis. The number of patients in Child-Pugh class A, B, and C was nine, seven, and one, respectively.

Variation in textural parameters during contrast injection

Injection of contrast medium increased the texture parameter entropy in HCC lesions and decreased the parameters energy and correlation in HCC lesions and hepatic tissue (Table 1). However, the observed slope coefficients (i.e. the respective increase over time) were small, in the range of 0.01%–0.3% of the respective texture parameter. In addition, no detectable systematic changes over time in textural parameters were observed when normalizing all textural parameters to 1 at the timepoint where the attenuation of the ROI placed in the abdominal aorta reached 160 HU (Figs. 1 and 2).

Table 1.

Median slope coefficient for linear regression over all timepoints for each textural parameter for all lesions.

	Median	(IQR)	Parameter (median (%))
Slope, lesions
Homogeneity	–0.00009	(–0.00022 to 0.00005)	0.076 (0.12)
Energy	–0.0000014	(–0.0000034 to –0.0000001)	0.0011 (0.12)
Entropy	0.00065	(0.00014–0.00216)	7.05 (0.01)
Inertia	0.8	(–0.3 to 1.7)	304 (0.25)
Correlation	–0.0000046	(–0.0000161 to –0.0000004)	0.0015 (0.31)
Slope, tissue
Homogeneity	–0.00004	(–0.00023 to 0.00000)	0.092 (0.05)
Energy	–0.0000008	(–0.0000034 to –0.0000001)	0.00082 (0.10)
Entropy	0.00085	(–0.00005 to 0.00310)	7.30 (0.01)
Inertia	0.3	(–0.3 to 1.1)	198 (0.17)
Correlation	–0.0000040	(–0.0000152 to –0.0000001)	0.0024 (0.17)

IQR, interquartile range.

Fig. 1.

Normalized mean textural parameters of all lesions over time. The x-axis indicates the timepoint relative to the 160 HU enhancement in the abdominal aorta, the y-axis the normalized value in comparison to the measured value at the 160 HU enhancement in the abdominal aorta. HU, Hounsfield unit.

Fig. 2.

Normalized mean textural parameters of all hepatic tissue ROIs over time. The x-axis indicates the timepoint relative to the 160 HU enhancement in the abdominal aorta, the y-axis the normalized value in comparison to the measured value at 160 HU enhancement in the abdominal aorta. HU, Hounsfield unit; ROI, region of interest.

In HCC lesions, the variability in texture parameters over time after contrast medium injection within each lesion was significantly smaller than the variation between different HCC lesions for the textural parameters homogeneity, energy, and entropy (Table 2). This difference was not observed in hepatic tissue, where no significant difference was detected over time in comparison to the variation between different individuals.

Table 2.

Mean SD for each textural parameter over all timepoints for each lesion/tissue and between all lesions/tissue for each timepoint.

SD	All timepoints	Between lesion/tissue	Change (%)	P
Lesions
Homogeneity	0.010	0.014	43	0.01
Energy	0.000187	0.000460	146	<0.001
Entropy	0.09	0.34	279	<0.001
Inertia	91	106	16	0.50
Correlation	0.00046	0.00060	29	0.08
Tissue
Homogeneity	0.009	0.012	32	0.26
Energy	0.000108	0.000142	32	0.31
Entropy	0.11	0.15	34	0.23
Inertia	73	78	7	0.96
Correlation	0.00050	0.00064	28	0.42

Statistically significant values in bold.

When comparing the variability in texture parameters after contrast medium injection in lesions with that of hepatic tissue, only the parameter entropy varied significantly. The average of each texture parameter varied over time, more in energy and entropy for lesions than that of hepatic tissue (Table 3).

Table 3.

Mean SD for each textural parameter over all timepoints for each lesion/tissue (Intra-lesion) compared with hepatic tissue (Intra-tissue) and between all lesions for each timepoint (Inter-lesion) compared with hepatic tissue (Inter-tissue).

SD	Intra-lesion	Intra-tissue	Change (%)	P
Homogeneity	0.010	0.009	–8	0.99
Energy	0.000187	0.000108	–43	0.65
Entropy	0.09	0.11	22	0.02
Inertia	91	73	–20	0.18
Correlation	0.00046	0.00050	7	0.32
Homogeneity	0.014	0.012	–15	0.26
Energy	0.000460	0.000142	–69	<0.001
Entropy	0.34	0.15	–57	0.001
Inertia	106	78	–27	0.11
Correlation	0.00060	0.00064	6	0.93

Statistically significant values in bold. SD, standard deviation.

No significant differences in variability of textural parameters in lesions and in hepatic tissue were detected between patients of different Child-Pugh class or due to the presence of liver cirrhosis.

Comparison of textural parameters between non-malignant tissue and lesions

Significant differences between the mean over all timepoints of each of the textural parameters energy, entropy, inertia, and correlation for hepatic tissue and HCC lesions were observed (Table 4). However, only the textural parameter energy differed more than the intra-individual variation of hepatic tissue over time, i.e. the 95% confidence interval (CI) of its SD (95% CI = 0.0013–0.0009, P < 0.02).

Table 4.

Mean textural parameters over all timepoints for hepatic tissue in comparison to lesions.*

	Tissue	Lesion	P	>1.96 SD
Homogeneity	0.089	0.080	0.12	N/A
Energy	0.0009	0.0013	0.02	Yes
Entropy	7.32	7.03	0.02	No
Inertia	263	322	0.04	No
Correlation	0.0024	0.0015	0.002	No

Statistically significant values in bold. *Whether the observed difference is > 1.96 times the observed intra-hepatic tissue SD for significantly different parameters is also presented.

SD, standard deviation.

No significant differences in the mean of the textural parameters in lesions and in hepatic tissue were detected between patients of different Child-Pugh class or due to the presence of liver cirrhosis.

Discussion

Contrast media is used to enhance the differences between normal and pathological tissue. It is therefore highly possible that the timing of the imaging should affect the textural analysis. However, to our surprise no drift of textural parameters of importance was observed in this study. When evaluating the change over time, the slope coefficient for each of the included textural parameters was only 0.01%/s to 0.3%/s of its average value. The SD of the textural parameters homogeneity, energy, and entropy during contrast enhancement was significantly smaller for each lesion (i.e. intra-lesion variability) than the variation was between the lesions (i.e. inter-lesion variation). This difference was not observed in hepatic tissue. No detectable systematic changes in normalized textural parameters were observed (Figs. 1 and 2).

Our results are in contrary to the previously observed differences in esophageal tumors when comparing the textural parameters of native and CECT (11). This might indicate that textural parameters in different tumor types are affected differently by contrast media. As mentioned earlier, the lack of major variation in texture parameters with contrast medium injection in HCC was a bit surprising. Normal liver tissue has a dual blood supply from portal and arterial blood inflow, but HCCs are predominantly supplied by only arterial blood flow. This changed blood supply results in typical contrast dynamics at imaging with hyper-attenuation in the arterial phase (wash-in) and hypo-attenuation in the delayed phase (wash-out). Theoretically, this should result in major changes in texture parameters during contrast enhancement. However, the observed differences in mean textural parameters between hepatic tissue and HCC and their variability and variation during contrast injection support the hypothesis that the differences in blood flow are reflected in the behavior of textural parameters.

The observed textural parameters of lesions, except for energy, did not differ from surrounding tissue more than 1.96 times the SD of the intra-tissue variation of the surrounding tissue. This means that the differences of textural parameters between lesions and hepatic tissue were within the 95% CI when taking the observed variation in hepatic tissue over scanning time into account. This means that texture analysis does not yet seem to have the potential to be used for diagnosing, at least not when using single texture parameters. Combinations of different texture analysis parameters might be useful after more extensive research.

Variation in scanning parameters is also a factor that could affect textural parameters. The use of different scanners (15) and differences in imaging settings have been shown to affect the reproducibility of textural parameters (14). Therefore, the use of a single scanner using the same image acquisition settings for all scans was a strength of the present study. In comparison to standard CECT, the PCT scan used in the present study has a lower tube voltage and lower total radiation dosage for each scan cycle. This results in noisier images, which could potentially mask the minute changes of textural parameters due to the enhancement of the contrast medium. In addition, the lower amount of contrast medium used for each patient could result in less contrast enhancement, which affects the rescaling of Hounsfield unit information into GLCM gray levels. This could increase the observed heterogeneity in comparison to examinations using a regular dosage of contrast medium, as the lower enhancement might bin areas into a lower GLCM gray level than they would have been when using a regular dosage of contrast medium.

To our knowledge, this is the first study to evaluate the variation in textural parameters during contrast injection. Although the variation in textural parameters between examinations over a longer term has been studied and shown low variation of textural parameters at different timepoints (14,16,18,19), the present study shows that the variation in textural parameters is low enough during contrast injection so that precise timing during contrast-enhanced scanning is not required. However, the observed variation in textural parameters during the scanning time suggest that there exist inherent variations which have to be considered at statistical analysis of textural parameters for comparison of observed differences between different tissues. The use of an 80-kV tube current and low radiation dose protocol could have amplified the observed variation as this results in noisier images as previously discussed.

A limitation of this study was the use of a single two-dimensional (2D) slice for textural analysis as this results in fewer included voxels. It has previously been shown that analyzing a three-dimensional volume has a greater reproducibility than when using single axial slices (14). Another limitation is the use of a single observer for the placement of the ROI used for segmentation. However, the analysis of a single 2D slice in combination with the use of a highly experienced radiologist should limit the risk of segmentation errors. The same ROI was also copied and pasted for all timepoints reducing variation in lesion delineation at different timepoints. The effects of different segmentation methods and CT acquisition settings on textural parameters could potentially also be studied in order to further quantify the effects. However, this was beyond the scope of the present study.

In conclusion, the present study showed no additional variation textural parameters due to the use of contrast media for hepatic tissue or HCC lesions implying that reproducible contrast medium timing should not be an issue at PCT texture analysis of HCC.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received the following financial support for the research, authorship, and/or publication of this article: This work was supported by a research internship grant (Forskar-AT) from Karolinska Institutet.

ORCID iDs

Yi-Hua Zhang

Anders Svensson

Per Stål

References

Wallace

Preen

Jeffrey

, et al. The evolving epidemiology of hepatocellular carcinoma: a global perspective. Expert Rev Gastroenterol Hepatol 2015; 9:765–779.

Raza

Sood

GK.

Hepatocellular carcinoma review: Current treatment, and evidence-based medicine. World J Gastroenterol 2014; 20:4115–4127.

European Association for the Study of the Liver, European Organisation for Research and Treatment of Cancer. EASL-EORTC clinical practice guidelines: management of hepatocellular carcinoma. J Hepatol 2012; 56:908–943.

Brenet Defour

Mulé

Tenenhaus

, et al. Hepatocellular carcinoma: CT texture analysis as a predictor of survival after surgical resection. Eur Radiol 2019; 29:1231–1239.

Park

Kim

Choi

S-Y

, et al. Prediction of therapeutic response of hepatocellular carcinoma to transcatheter arterial chemoembolization based on pretherapeutic dynamic CT and textural findings. AJR Am J Roentgenol 2017; 209:W211–W220.

Ganeshan

Miles

Young

RCD

, et al. Hepatic entropy and uniformity: additional parameters that can potentially increase the effectiveness of contrast enhancement during abdominal CT. Clin Radiol 2007; 62:761–768.

Naganawa

Enooku

Tateishi

, et al. Imaging prediction of nonalcoholic steatohepatitis using computed tomography texture analysis. Eur Radiol 2018; 28:3050–3058.

Bharti

Mittal

Ananthasivan

Characterization of chronic liver disease based on ultrasound images using the variants of grey-level difference matrix. Proc Inst Mech Eng H 2018; 232:884–900.

Lubner

Jones

Kloke

, et al. CT texture analysis of the liver for assessing hepatic fibrosis in patients with hepatitis C virus. Br J Radiol 2019; 92:20180153.

10.

Ganeshan

Miles

Young

RCD

, et al. Hepatic enhancement in colorectal cancer: texture analysis correlates with hepatic hemodynamics and patient survival. Acad Radiol 2007; 14:1520–1530.

11.

Zhang

Y-H

Herlin

Rouvelas

, et al. Texture analysis of computed tomography data using morphologic and metabolic delineation of esophageal cancer-relation to tumor type and neoadjuvant therapy response. Dis Esophagus 2018; 32:doy096.

12.

Ganeshan

Goh

Mandeville

, et al. Non-small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology 2013; 266:326–336.

13.

Zhang

Graham

Elci

, et al. Locally advanced squamous cell carcinoma of the head and neck: CT texture and histogram analysis allow independent prediction of overall survival in patients treated with induction chemotherapy. Radiology 2013; 269:801–809.

14.

Zhao

Tan

Tsai

W-Y

, et al. Reproducibility of radiomics for deciphering tumor phenotype with imaging. Sci Rep 2016; 6:23428.

15.

Mackin

Fave

Zhang

, et al. Measuring computed tomography scanner variability of radiomics features. Invest Radiol 2015; 50:757–765.

16.

Balagurunathan

Kumar

, et al. Test–retest reproducibility analysis of lung CT image features. J Digit Imaging 2014; 27:805–823.

17.

Zhao

James

Moskowitz

, et al. Evaluating variability in tumor measurements from same-day repeat CT scans of patients with non-small cell lung cancer. Radiology 2009; 252:263–272.

18.

Tixier

Hatt

Le Rest

, et al. Reproducibility of tumor uptake heterogeneity characterization through textural feature analysis in 18F-FDG PET. J Nucl Med 2012; 53:693–700.

19.

Wang

Zhong

, et al. Reproducibility with repeat CT in radiomics study for rectal cancer. Oncotarget 2016; 7:71440–71446.