Eur Radiol DOI 10.1007/s00330-015-4203-4
VASCULAR-INTERVENTIONAL
Parametric response mapping of contrast-enhanced biphasic CT for evaluating tumour viability of hepatocellular carcinoma after TACE Jan B. Hinrichs 1 & Hoen-Oh Shin 1 & Daniel Kaercher 1 & Davut Hasdemir 1 & Tim Murray 2 & Till Kaireit 1 & Carolin Lutat 1 & Arndt Vogel 3 & Bernhard C. Meyer 1 & Frank K. Wacker 1 & Thomas Rodt 1
Received: 4 November 2015 / Revised: 22 December 2015 / Accepted: 30 December 2015 # European Society of Radiology 2016
Abstract Objectives To determine the feasibility and role of parametric response mapping (PRM) for quantitative assessment of regional contrast-enhancement patterns in hepatocellular carcinoma (HCC). Methods Biphasic CT of 19 patients receiving repetitive conventional transarterial chemoembolisation (cTACE) for intermediate stage HCC were retrospectively analysed at baseline and follow-up at 3, 6, and 9 months. Voxelbased registration of arterial and porto-venous phases, with segmentation of the largest target lesion was performed. Frequency distribution plots of density-pairs of segmented voxels were generated. To differentiate necrotic, hypervascular and non-hypervascular tumour, and lipiodol/calcification, thresholds of 30, 100, and 300 HU were applied. Changes in density frequency plots over
time were analysed and compared to response and assessment criteria (WHO, RECIST, EASL, mRECIST) and survival. Results PRM was feasible in all cases. Tumour volumes and hypervascular/non-hypervascular volume ratio showed significant longitudinal decrease (p < 0.05). Hypervascular volume at baseline was inversely correlated to survival (R = -0.57, p = 0.005). The only predictive parameter following cTACE to show significant survival difference was the change of the viable/non-viable ratio (p = 0.044), whereas common response assessment criteria showed no significant difference in survival. Conclusions PRM allows a quantitative and more precise assessment of regional tumour vascularisation patterns and may be helpful for TACE treatment planning and response assessment.
Jan B. Hinrichs and Hoen-Oh Shin contributed equally to this work. * Jan B. Hinrichs
[email protected]
Arndt Vogel
[email protected] Bernhard C. Meyer
[email protected]
Hoen-Oh Shin
[email protected]
Frank K. Wacker
[email protected]
Daniel Kaercher
[email protected]
Thomas Rodt
[email protected]
Davut Hasdemir
[email protected] Tim Murray
[email protected]
1
Department of Diagnostic and Interventional Radiology, Hannover Medical School, Carl-Neuberg-Str. 1, 30625 Hannover, Germany
Till Kaireit
[email protected]
2
Department of Diagnostic and Interventional Radiology, Beaumont Hospital, Dublin, Ireland
Carolin Lutat
[email protected]
3
Gastroenterology, Hepatology and Endocrinology, Hannover Medical School, Hannover, Germany
Eur Radiol
Key Points • PRM allows more precise assessment of tumour vascularisation compared to conventional evaluation • PRM is beneficial for cTACE treatment planning and response assessment • PRM allows a quantitative assessment of regional contrast enhancement patterns Keywords Hepatocellular carcinoma . Parametric Response Mapping . Multi-detector computed tomography . Transarterial chemoembolization . HCC viability assessment
Abbreviations BCLC Barcelona Clinic Liver Cancer Score CT Computed tomography cTACE conventional transarterial chemoembolisation EASL European Association for the Study of Liver HCC Hepatocellular carcinoma HU Hounsfield unit IR Interventional radiology MDCT Multi-detector computed tomography mRECIST modified Response Evaluation Criteria in Solid Tumours MRI Magnetic resonance imaging PRM Parametric response mapping ROI Region of interest RECIST Response Evaluation Criteria in Solid Tumours SIRT Selective internal radiotherapy TACE Transarterial chemoembolization WHO World Health Organization
Introduction The key finding in diagnosis of hepatocellular carcinoma (HCC) in the cirrhotic liver is represented by a vascularisation pattern including arterial hyper-enhancement and portovenous wash-out indicating viable tumour tissue [1, 2]. Patients presenting with intermediate or advanced stages of HCC at diagnosis are not candidates for potentially curative surgery or ablation [3–5]. Transarterial chemoembolisation (TACE) represents a standard component of local therapy in non-surgical patients with intermediate stage HCC, as defined by Barcelona Clinic Liver Cancer (BCLC) stage B, and has been reported to show significant survival advantages [5, 6]. Today, there are different TACE regimes reported in the literature [7, 8]. Overall, TACE aims to reduce viable tumour tissue and to induce tumour necrosis [9]. The precise detection of residual viable intra-hepatic tumour tissue or tumour recurrence is one of the most challenging aspects of successful treatment planning and of high importance in HCC surveillance [10]. Dynamic contrast-enhanced multi-detector
computed tomography (MDCT) or magnetic resonance imaging (MRI) are frequently used to monitor the therapeutic response to TACE [11, 12]. World Health Organization (WHO) criteria or Response Evaluation Criteria in Solid Tumours (RECIST 1.1) were introduced to report treatment response [13, 14]. The European Association for the Study of Liver (EASL) recommend measuring changes in the area of tumour enhancement on contrast-enhanced imaging [15, 16] and more recently, modified RECIST criteria were introduced, assessing change in the degree of tumour viability reflected by contrast enhancement [15–17]. All tumour treatment response criteria rely on the subjective assessment of tumour size and/or vascularisation patterns, and thus may be imprecise. A novel method to analyse images and to report treatment response is called parametric response mapping (PRM) [18]. PRM represents a semi-automated and quantitative method to measure viable tumour tissue reflected by changes in contrast enhancement in multi-phasic imaging. The PRM imaging approach compares contrast enhancement before and after treatment on a per voxel basis and classifies treatment response of each voxel within a region of interest (ROI) [18, 19]. Choi et al. reported on the feasibility and usefulness of PRM in HCC after TACE. However, they limited assessment to arterial phase CT data [19]. In another study recurrence of HCC could be detected by use of PRM [20]. Because of the importance of arterial and porto-venous enhancement patterns of HCC, we believe that the combination of the arterial and the porto-venous phase in PRM analyses may provide useful additional information concerning initial evaluation, treatment planning, and response assessment. In this study, we assess the feasibility of PRM based on arterial and porto-venous phase CT data and evaluate the potential to assess regional changes in vascularisation patterns indicating viability of HCC. Furthermore, the prognostic relevance of this technique in patients undergoing repetitive conventional transarterial chemoembolisation (cTACE) is analysed.
Material and methods Study design The local ethics committee approved this retrospective study. Nineteen patients [three women, 16 men, 64.2 ± 7.7 years (mean ± standard deviation)] with intermediate stage HCC were included in this study. Of the included patients, 18 had a Child-Pugh A score, with a single patient having a ChildPugh B score. Between January 2006 and September 2012 these patients were treated with repeated cTACE in our institution. Inclusion criteria for this retrospective study were as follows: four or more biphasic MDCT in our institution
Eur Radiol
(including a baseline scan preceding cTACE treatment); execution of cTACE procedure between the MDCTs; minimal tumour-size of ≥2 cm of at least one lesion. Exclusion criteria were as follows: previously treated HCC; recurrent HCC; changes of the therapeutic regime or additional treatments such as ablation within the analysed follow-up period.
CT imaging All CT examinations were performed on a 64-row multi-detector CT (GE Lightspeed VCT, GE Healthcare, Chalfont St. Giles, UK) with a tube voltage of 100 kVp or adjusted according to the body mass index up to 120 kVp, dose modulation was applied adopting the mAs to a fixed noise ratio. A detector collimation of 64 × 0.625 mm was used, with a reconstructed slice thickness of 1 mm, and a reconstruction interval of 1.25 mm. The entire abdomen was scanned during a single breath-hold. A mechanical injector was used for contrastmedium administration through a peripheral venous access of at least 18 G. The total injected volume was 138 mL, comprising of 88 mL (350 mgI/mL) of contrast agent (Imeron 350, Bracco, Milan, Italy) followed by a 50 mL saline flush administered at a flow rate of 3.5 mL/s. For automatic bolus tracking we used a region of interest placed in the descending aorta with a threshold of 100 Hounsfield units (HU) and a 6-s start delay to ensure an optimal contrast within the abdominal arteries. The porto-venous phase was acquired 90 s after contrast bolus injection, according to Choi et al. [2, 21]. For evaluation of the therapeutic effect after cTACE, follow-up CT with the same protocol was performed every 3 months.
Transarterial chemoembolisation cTACE was performed according to our standard operating procedures every 3 months. The median delay between TACE and follow-up CT was 10 weeks. A 5 F introducer sheath (Avanti+, Cordis, Waterloo, Belgium) was placed in the right common femoral artery. A superior mesenteric artery angiogram to examine for anatomical variations of the hepatic arterial supply and to indirectly assess the portal vein was performed, followed by a celiac angiogram using an appropriate diagnostic catheter. A micro-catheter was placed in the tumour-feeding arteries for selective cTACE administration; if necessary, supra-selective administration was performed. The end-point of embolization was loss of hypervascularisation on control angiograms. The chemoembolisation included cisplatin and doxorubicin with 10 mL of lipiodol (Guerbet, Aulnay-sous-Bois, France) as the embolising agent, with up to 10 mL of additional lipiodol administered as required in patients with persisting hypervascular areas on control angiogram.
Image processing and parametric response mapping To achieve a voxel-to-voxel correspondence of the HU values in the arterial and porto-venous phase, the CT scan of the porto-venous phase was spatially deformed to match the arterial phase, which remained unchanged (Fig. 1). Registration was performed using a volume of interest only including the liver, and employing an open-source ANTS software package (2011, Release 1.5, Penn Image Computing and Science Laboratory, University of Pennsylvania, USA). The following steps were performed for registration: first, the arterial and porto-venous phases were corrected for movement and rotation differences in the two scans; second, a deformable registration was applied. This second operation was required to compensate for small elastic deformation to allow for point to point matching of tumour tissue in both phases. Technical details of the applied registration techniques have been described in literature before [22]. Following the registration process the largest tumour lesion was defined as the target lesion, as most patients had at least several tumours. The mean size of the target lesion was 61.9 mm (range: 20.6 to 162.1 mm). Manual segmentation of the target lesion based on the arterial phase CT data was performed using the software MEVIS lab® (MeVis Medical Solutions, Bremen, Germany; Fig. 2). The segmentation result was used as a mask to limit the voxel-to-voxel evaluation to the target lesion only. Based on this segmentation, a voxelbased density distribution was generated using the software Matlab2013a (MathWorks, Natick, MA, USA). Depending on their arterial (x-axis) and porto-venous (y-axis) HU density, the frequency distribution of the coincided voxels of the arterial (x-axis) and the porto-venous phase (y-axis) data were depicted on a scatter plot using a logarithmically scaled colour coding (Fig. 3). On the generated scatter plots thresholds of 0, 30, 100, and 300 HU applied to define tumour components according to HU values reported by Park et al. and Furlan et al. and measurements in our clinic [23, 24]. Voxels with an arterial density of 0–30 HU and a porto-venous density of 0–30 HU were categorised as necrosis. Voxels with an arterial and porto-venous density above 300 HU were categorised as calcification/lipiodol. The remaining voxels with arterial and porto-venous density values of 30–300 HU were categorised as viable tumour tissue. These voxels were subdivided into hypervascular tumour components (arterial density 100–300 HU and porto-venous density of 30–100 HU) and nonhypervascular tumour components (the remaining viable tumour tissue). The absolute number of voxels for each of the tumour component was calculated using Mathematica 9.1 (Wolfram Research, Champaign, IL, USA). Based on the absolute number of voxels in each of the described areas, the following parameters could be calculated from the density distribution plots for each target lesion: total tumour volume (cm3), i.e. all the segmented tumour volume: viable and non-
Eur Radiol
Fig. 1 Representative example of the registration process. Porto-venous images demonstrating hypodense tumour (A; arrow) due to wash-out in the right liver lobe are registered using ANTs to the arterial images (B) which shows the corresponding hyperdense area (B; arrow). The right image (C) shows the result of the registration process. The porto-venous
phase was spatially deformed to match the arterial phase, which remained unchanged; note the change of vessels in the left liver lobe. In general, there was no gross change in the image appearance due to the voxel based registration process
viable; viable tumour tissue volume (cm3), i.e. all of the viable tumour tissue with HU from 30–300; hypervascular tumour volume (cm3), i.e. segmented tumour components with HU values of 100–300 in the arterial phase and 30–100 in the porto-venous phase (indicating venous wash-out); viable/ non-viable tumour ratio representing a parameter for tumour devascularisation (non-viable included negative density values, necrosis, calcification/lipiodol), and the percentage of hypervascular tumour (as % of viable tumour tissue).
as RECIST, WHO, mRECIST, and EASL [16, 25]. We chose up to a maximum of two lesions in the baseline CT as target lesions to evaluate treatment response with the aforementioned criteria [25–27]. In all statistical tests, a p-value of 0.05 was defined as the level of statistical significance. Statistical analyses were conducted using SPSS Statistics®, Version 22 (IBM, USA).
Results Statistical analysis Descriptive statistical analyses of patients’ demographics were performed. Pairwise Wilcoxon signed-rank test was used to measure changes in the tumour lesions from baseline to follow-up CTs. Kaplan-Meier analysis was used to compute survival analysis. Analysis of the quantitative tumour component data was performed with regard to the longitudinal course under cTACE treatment. Furthermore, comparison of this data to overall survival was performed and the prognostic effect was compared to established response evaluation criteria such
Fig. 2 Representative images of the segmentation of a large tumour in the right liver lobe. (A) Manual segmentation is performed based on the arterial phase images as indicated by the enhancement in the aorta. The right image (B) shows the segmented area of the tumour. The result of the segmentation was used to mark the area of the tumour for further analysis with the Parametric Response Mapping
All patients underwent cTACE and CT without adverse events. The median survival was 972 days (range 368 to 2048 days). The 1-year survival rate was 100 %; the 3- and the 5-year survival rate were 47 % and 26 %, respectively. Results of the analysed conventional tumour response criteria (RECIST, WHO, mRECIST, EASL) are summarised in Table 1. Post-processing with PRM was technically feasible in all examined patients (n = 19). A complete analysis of an individual patient – including registration, segmentation, and generation of density plots based on the data of all four CT studies took approximately 14 h in this experimental setting. PRM allowed analysis of the different tumour tissue components as summarised in Table 2. The density plots allow visualisation of complex changes of the different tumour components. Baseline tumour volume decreased from 42.4 cm3 to 22.9 cm3 (46 % reduction) to the third follow-up CT. After 3 TACE procedures a significant reduction of viable tumour tissue (p = 0.014), of hypervascular tissue (p = 0.04), and of the viable to necrotic tissue ratio (p = 0.024) was noted. Examples of two different density plots following cTACE are shown in Figs. 4 and 5. Analysis of prognostic relevance was performed based on Kaplan-Meier survival curves. Hypervascular-volume at baseline was inversely correlated to survival (R -0.57, p = 0.005). No significant survival difference for the reduction of hypervascularised tumour tissue in total (cm3, p = 0.438) or
Eur Radiol
Discussion
Fig. 3 Example of a voxel-based density map for parametric response mapping. The frequency distribution of the coincided voxels of the arterial (x-axis) and the porto-venous phase (y-axis) data were depicted on a scatter plot using a logarithmically scaled colour coding. Voxels with arterial and porto-venous density of 0–30 HU were categorised as necrosis (green box). Voxels with an arterial and porto-venous density above 300 HU were categorised as calcification/lipiodol (defined by the blue lines). The remaining voxels with arterial and porto-venous density values of 30–300 HU were categorised as viable tumour tissue (orange box). These voxels were subdivided into hypervascular tumour components (arterial density 100–300 HU and porto-venous density of 30–100 HU; red box) and non-hypervascular tumour components (the remaining space of the orange box). For longitudinal analysis, the absolute number of voxels was calculated for each category
for the reduction of hypervascularised tumour relative to the total tumour volume (% of viable tissue, p = 0.918) was found. Interestingly, a significant survival difference was found for the ratio of viable to necrotic tumour tissue (p = 0.044) indicating that only the induction of significant tumour necrosis leads to improved survival (Fig. 6). All other investigated response criteria (WHO, RECIST, EASL, mRECIST) failed to show a significant survival difference in this study, although a prognostic relevance has been reported for EASL and mRECIST in patients treated with TACE previously [15, 16, 28].
Table 1 Median baseline and follow-up measurements according to the particular tumour response criteria
Baseline CT 1 CT 2 CT 3
As shown in this study, Parametric Response Mapping allows monitoring HCC for treatment effects following cTACE using biphasic CT data. PRM including density distribution plots enabled a valid longitudinal assessment of regional tumour contrast enhancement patterns and viability together with a quantitative analysis of the different tumour components. PRM based on arterial and porto-venous phases, as reported in our study, has the potential to increase accuracy in assessing and monitoring HCC. The quantitative information provided using this method includes information on the enhancement in arterial and in porto-venous CT phases, integrating two parameters of paramount importance for HCC assessment in the cirrhotic liver: arterial hypervascularisation and porto-venous wash-out [29]. Furthermore, quantitative volumetric data on different lesion components is provided. PRM can provide objective measurement parameters additional to the visual radiological assessment that, when performed by a radiologist knowing the procedure and pathology, will only address these issues in a descriptive manner. Because of the quantitative nature of the method presented in this study, PRM may be especially helpful to measure treatment response in studies investigating intra-arterial IR procedures or pharmacological treatment. Precise information for therapy planning and assessment of therapeutic efficacy is also crucial in everyday clinical work. A more detailed assessment can facilitate planning and allows individualised treatment strategies. Regarding the relevance for longitudinal assessment, it should be remembered that beside potentially curative treatment options such as surgical resection, radiofrequency ablation, and liver transplantation, a number of alternative non-curative treatment options are available [3, 30–33]. Conventional TACE, drug-eluting bead TACE, selective internal radiotherapy (SIRT), chemo-saturation, and percutaneous isolated chemoperfusion represent the diversity of transarterial treatment options for local tumour control of HCC, whilst therapies such as sorafenib offer systemic treatment [30, 34–38]. PRM may lead to a more precise
WHO (mm2)
RECIST (mm)
EASL (mm2)
mRECIST (mm)
2264.8 2259.4 (-0.23 %) 1704.4 (-24.7 %) 1504.3 (-33.6 %)
71.3 65.7 (-7.9 %) 55.8 (-21.7 %) 56.6 (-20.6 %)
1182.9 999.7 (-15.5 %) 733.0 (-38.0 %) 693.9 (-41.4 %)
57.2 52.8 (-7.7 %) 49.9 (-12.8 %) 44.6 (-22.0 %)
Median values and %-values of the measurements obtained from baseline imaging and follow-up CT (1-3) after TACE according to the various tumour response criteria. In the context of the definitions of these response criteria we chose up to two target lesions (including the lesion used for PRM) per patient for calculation of the response assessment. CT = computed tomography, EASL = European Association for the Study of Liver, RECIST = response evaluation criteria in solid tumours, mRECIST = modified response evaluation criteria in solid tumours, WHO = World Health Organization.
Eur Radiol Table 2 Parametric response mapping parameters
Parameter
Baseline
CT 1
CT 2
CT 3
p-value (BL –CT3)
Tumour volume (cm3) Viable tumour tissue (cm3) Hypervascular tumour tissue (cm3) Viable/necrotic tissue ratio, % Hypervascular/viable tumour ratio, %
42.42 41.57 6.53
40.43 40.03 4.1
30.91 30.63 2.29
22.9 19.9 1.96
0.013 0.014 0.004
92.6 13.78
90.12 13.57
34.45 12.46
32.49 11.94
0.024 0.358
Median values of the parametric response mapping parameters obtained from baseline imaging and follow-up CTs after TACE. BL = baseline
assessment of the effectiveness of a treatment, and may indicate the need to change the therapeutic regime for nonresponders earlier compared to conventional response assessment. This may reduce the morbidity and cost associated with prolonged and ineffective treatments. The overall survival is the generally accepted major end point for clinical trials in HCC [39]. Longer follow-up times are needed to assess the overall survival, however, and for the majority of clinical scenarios there is insufficient survival data to inform treatment decisions. Whether the current treatment should be continued or changed is, therefore, predominantly determined by radiological assessment of tumour response rates to treatment, which often serve as surrogate parameters. Fig. 4 Hypervascular tumour in the right liver lobe (A; arrow), following three cTACE procedures significant accumulation of lipiodol is seen as well as shrinkage of the tumour (B; arrow). After treatment, the hypervascular component of the tumour is reduced (B). Corresponding PRM analysis (C, D) reflects the decrease of tumour volume (i.e. shrinkage), especially of the hypervascular component. The number of voxels with unaltered high density in between arterial and porto-venous phase correlating to lipiodol increases
These assessments are weaker end points for a number of reasons, notably as they depend on the skills of the investigator and the appropriateness of the reference criteria [39]. Nevertheless, radiological parameters serve as accepted surrogate parameters in monitoring tumour response without yet knowing the optimal assessment method both in terms of survival and reliability [15]. Thus, evaluation of valid and accurate response parameters indicating patient’s survival after local tumour therapy - such as PRM - is important. Shim et al. showed that a tumour response criteria which takes tumour enhancement into account is superior to size criteria alone for categorising tumour response to TACE [16]. The enhancement of the tumour in CT images is interpreted as
Eur Radiol Fig. 5 Hypervascular diffuse central tumour, at Baseline (A, arrow) and Follow-up 3 (B). Following three cTACE procedures, shrinkage and a less hypervascular tumour are seen (B; arrow). Corresponding PRM analysis (C, D) shows decrease of voxel numbers (i.e. shrinkage) and shift of the centroid to less hypervascular density values
vascularisation and correlates with overall survival as shown for mRECIST and EASL [15, 16]. Interestingly, EASL and mRECIST failed to show significant survival differences in our cohort. In contrast, the reduction of viable to necrotic tumour tissue was the only parameter significantly correlating with better survival indicating that the PRM approach might
Fig. 6 Kaplan-Meier survival curves showing prognostic relevance of the viable/necrotic tissue ratio. A significant (p = 0.044) increase in overall survival following TACE treatment for patients with an abovemedian viable/necrotic tissue ratio is seen
be more precise than EASL or mRECIST within the investigated patients. Liu et al. reported EASL and mRECIST to be more accurate than RECIST to determine treatment response in combination to sorafenib in combination with TACE [28]. Thus, we expect comparable results when evaluating other therapies such as DEB-TACE and sorafenib using the presented PRM approach, which needs to be elucidated in further studies. The beneficial effect of PRM for these specific treatments might be even greater as these techniques interfere with the vascularity of the tumour to a higher degree compared to the cTACE technique described in our study [37]. Our study has a couple of limitations. First, the number of study patients included in our study was limited, and the study is retrospective. Furthermore, the studied patients were homogenous, and all investigations were performed with the same system. On the other hand, only in this retrospective manner were survival data and homogenous image quality available. Further studies investigating the efficacy of the presented technique will be necessary in order to prove its applicability in clinical practice, especially concerning different HCC stages and varying CT scanners. However, we believe there is no reason why the general mode-of-action of this technique should not work under these conditions. Another potential limitation of our experimental setting is the timeconsuming post-processing of about 14 h. We mainly attribute
Eur Radiol
this to the experimental software used for post-processing. Commercial software bares the potential to perform postprocessing at least semi-automated and to render significantly faster. The chosen threshold values to categorise the different tumour components were partially derived from literature [23, 24] and partially from own initial measurements. However, in the end the threshold values were determined arbitrarily to a certain level, and there is room for major improvements. A possibility to generate more precise threshold values is represented by analysis of optimal thresholds based on a bigger patient collective or the implementation of an input function, which could be derived from the bolus tracking images. The fact that the technique produced relevant measurement improvements even with the reported thresholds is an indicator of the robustness of the method. In conclusion, parametric response mapping in this study was shown to be beneficial for technical outcome analysis and response assessment in patients undergoing cTACE procedures. Once the technical limitations are overcome, the approach has the potential to provide additional information to current response assessment criteria and provide objective information on tumour composition.
6.
7.
8.
9.
10.
11.
12. 13. 14.
15. Acknowledgements This study was presented in parts at the ECR 2015. The scientific guarantor of this publication is Frank Wacker. The authors of this manuscript declare relationships with the following companies: Siemens Healthcare and ProMedicus (Bernhard Meyer, Frank Wacker; outside the submitted work). The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. The authors state that this work has not received any funding. One of the authors has significant statistical expertise. Institutional Review Board approval was obtained. Written informed consent was waived by the Institutional Review Board. None study subjects or cohorts have been previously reported. Methodology: retrospective, experimental, intra-individual comparison, performed at one institution.
References 1.
2.
3.
4.
5.
Hennedige T, Venkatesh SK (2013) Imaging of hepatocellular carcinoma: diagnosis, staging and treatment monitoring. Cancer Imaging 12:530–547 Choi J-Y, Lee JM, Sirlin CB (2014) CT and MR imaging diagnosis and staging of hepatocellular carcinoma: part II. Extracellular agents, hepatobiliary agents, and ancillary imaging features. Radiology 273:30–50 El Serag HB, Marrero JA, Rudolph L, Reddy KR (2008) Diagnosis and treatment of hepatocellular carcinoma. Gastroenterology 134: 1752–1763 Bruix J, Sherman M (2005) Management of hepatocellular carcinoma - Bruix - 2005 - Hepatology - Wiley Online Library. Hepatology. doi:10.1002/hep.20933/pdf Llovet JM, Bruix J (2003) Systematic review of randomized trials for unresectable hepatocellular carcinoma: chemoembolization improves survival. Hepatology 37:429–442
16.
17.
18.
19.
20.
21.
22.
23.
24.
Lo C-M, Ngan H, Tso W-K et al (2002) Randomized controlled trial of transarterial lipiodol chemoembolization for unresectable hepatocellular carcinoma. Hepatology 35:1164–1171 Marelli L, Stigliano R, Triantos C et al (2007) Transarterial therapy for hepatocellular carcinoma: which technique is more effective? A systematic review of cohort and randomized studies. Cardiovasc Intervent Radiol 30:6–25 Niessen C, Wiggermann P, Velandia C et al (2013) Transarterial chemoembolization - status quo in Germany. Fortschr Röntgenstr 185:1089–1094 Golfieri R, Cappelli A, Cucchetti A et al (2011) Efficacy of selective transarterial chemoembolization in inducing tumor necrosis in small (
