Identifying Staging Markers for Hepatocellular Carcinoma before Transarterial Chemoembolization: Comparison of Three-dimensional Quantitative versus Non–three-dimensional Imaging Markers

Volumetry-based tumor size thresholds are prognostic discriminators with regard to postprocedural prognosis in patients with hepatocellular carcinoma who will undergo treatment with transarterial chemoembolization.

Abstract

Purpose

To test and compare the association between radiologic measurements of lesion diameter, volume, and enhancement on baseline magnetic resonance (MR) images with overall survival and tumor response in patients with unresectable hepatocellular carcinoma (HCC) treated with transarterial chemoembolization (TACE).

Materials and Methods

This HIPAA-compliant retrospective, single-institution analysis was approved by the institutional review board, with waiver of informed consent. It included 79 patients with unresectable HCC who were treated with TACE. Baseline arterial phase contrast material–enhanced (CE) MR imaging was used to measure the overall and enhancing tumor diameters. A segmentation-based three-dimensional quantification of the overall and enhancing tumor volumes was performed in each patient. Numeric cutoff values (5 cm for diameters and 65 cm³ for volumes) were used to stratify the patient cohort in two groups. Tumor response rates according to Response Evaluation Criteria in Solid Tumors (RECIST), modified RECIST (mRECIST), and European Association for the Study of the Liver (EASL) guidelines were recorded for all groups. Survival was evaluated by using Kaplan-Meier analysis and was compared by using Cox proportional hazard ratios (HRs) after univariate and multivariate analysis.

Results

Stratification according to overall and enhancing tumor diameters did not result in a significant separation of survival curves (HR, 1.4; 95% confidence interval [CI]: 0.7, 2.5; P = .234; and HR, 1.6; 95% CI: 0.9, 2.8; P = .08, respectively). The stratification according to overall and enhancing tumor volume achieved significance (HR, 1.8; 95% CI: 0.9, 3.4; P = .022; and HR, 1.8; 95% CI: 1.1, 3.1; P = .017, respectively). As for tumor response, higher response rates were observed in smaller lesions compared with larger lesions, when the 5-cm threshold (27% vs 15% for mRECIST and 45% vs 24% for EASL) was used.

Conclusion

As opposed to anatomic tumor diameter as the most commonly used staging marker, volumetric assessment of lesion size and enhancement on baseline CE MR images is strongly associated with survival of patients with HCC who were treated with TACE.

^© RSNA, 2014

Online supplemental material is available for this article.

Introduction

Hepatocellular carcinoma (HCC) is a growing public health problem worldwide. With more than 700 000 newly diagnosed patients per year, HCC continues to be a major oncologic challenge, primarily in Asian countries, with rising incidences in Europe and the United States (1,2). In patients with intermediate- to advanced-stage disease, catheter-based intraarterial therapies such as transarterial chemoembolization (TACE) have been included in several treatment guidelines and can now be seen as the mainstay of therapy, with the capability to prolong patient survival while preserving a relatively high quality of life (3).

The importance of cross-sectional imaging for the diagnosis, staging, and treatment response assessment in HCC cannot be overstated. For instance, all commonly used staging systems, such as the Barcelona Clinic Liver Cancer staging system and the Cancer of the Liver Italian Program system, take into account tumor size of the dominant nodule, as well as lesion multiplicity as seen on preprocedural images to select suitable candidates for surgical treatment or local-regional therapies (1,4–6). The importance of diameter-based cutoffs as discriminators for treatment recommendations has been propagated by the Milan criteria, which consolidated the 5-cm threshold as a selection criterion for liver transplantation (7). The growing availability of cross-sectional imaging has facilitated early diagnosis of HCC, leading to a higher detection rate of smaller lesions. This development was taken into account by the authors of the Barcelona Clinic Liver Cancer staging system and was implemented by further stratifying this threshold to include different lesion sizes as prognostic discriminators. However, the recently developed Hong Kong Liver Cancer classification challenged this concept and maintained the 5-cm threshold as the only relevant size-based prognostic discriminator (8). In the area of postprocedural imaging, the broad availability of dynamic, contrast material–enhanced (CE) computed tomography (CT) and CE magnetic resonance (MR) imaging has contributed to the shift away from anatomic treatment response criteria such as Response Evaluation Criteria in Solid Tumors (RECIST), which are based on tumor diameter, toward the more functional modified RECIST (mRECIST), as well as three-dimensional (3D) quantitative tumor assessment techniques (9–12), which are based on enhancement. These newer models were shown to more accurately reflect tumor biology, necrosis, as well as progression patterns (13). However, this wealth of new knowledge has not yet been translated to baseline imaging.

Therefore, our purpose was to test and compare the association between radiologic measurements of lesion diameter, volume, and enhancement on baseline MR images with overall survival (OS) and tumor response in patients with unresectable HCC who were treated with TACE.

Materials and Methods

Philips Research North American (Briarcliff Manor, NY) provided the software graphical user interface for the in-house 3D tumor assessment software that was developed at the Johns Hopkins Hospital (Baltimore, Md). One author (M.L.) is an employee of Philips Research North America. The control of the data and the information in the manuscript were maintained by the remaining authors.

Study Cohort

This retrospective, single-institution analysis was conducted in compliance with the Health Insurance Portability and Accountability Act and was approved by the institutional review board of Johns Hopkins Hospital. Between December 2001 and December 2008, with waiver of informed consent, a total of 244 consecutive patients with unresectable HCC underwent their first TACE procedure as the first liver-directed therapy. A total of 74 patients were excluded because of missing or inadequate (no contrast agent applied, severe artifacts involving the tumor, missing phases) MR imaging. In addition, 66 patients were excluded because of extrahepatic, as well as intrahepatic (main or lobar) portal vein thrombosis. This exclusion criterion (portal vein thrombosis) was based on the utilized Hong Kong Liver Cancer staging system, which provided the imaging cutoffs for this study, as well as on the Barcelona Clinic Liver Cancer staging system. Both staging systems help isolate patients with portal vein invasion from other patients with HCC, primarily because of a substantially worse prognosis, when they are treated intraarterially. Also, patients with metastatic HCC (n = 7), as well as those with infiltrative disease (n = 15, tumors with ill-defined borders or invasion of adjacent structures [eg, diaphragm]) were excluded from the OS analysis. In addition, patients with missing laboratory parameters were excluded (n = 3). The remaining 79 patients, treated with conventional TACE or drug-eluting beads TACE, were included in the final analysis (Fig 1).

Figure 1:

Flowchart of the patient selection process. A total of 74 patients were excluded because of missing or inadequate MR imaging results. In addition, a total of 66 patients were excluded because of portal vein thrombosis. Patients with metastatic HCC (n = 7), as well as those with infiltrative disease (n = 15), were excluded from the OS analysis. In addition, patients with missing laboratory parameters were excluded (n = 3).

Evaluation and Staging

All included patients underwent a full clinical examination, as well as baseline laboratory tests (liver function tests, serum albumin level, prothrombin time, total bilirubin level). The Eastern Cooperative Oncology Group (ECOG) performance status was recorded in all patients. The stage of disease was assessed by using the Child-Pugh classification system, the Barcelona Clinic Liver Cancer staging system, as well as the Hong Kong Liver Cancer staging system (8). All patients were evaluated by a multidisciplinary liver tumor board and were considered to have unresectable HCC.

TACE Protocol

All TACE procedures were performed by the same interventional radiologist (J.F.G., with 18 years of experience in hepatic interventions). A consistent approach according to our standard institutional protocol was used. First, angiographic steps were performed with contrast agent injections from the celiac trunk and the superior mesenteric artery to define the hepatic arterial anatomy, to determine portal venous patency, and to evaluate tumor vascularity. Angiography was performed with the available angiographic system (Allura Xper FD20; Philips Healthcare, Best, the Netherlands) from the celiac axis, as well as selectively in the right or left hepatic artery. Injection rates varied depending on the blood vessel caliber and flow, and the power injector (Medrad, Warrendale, Pa) rate was 1–3 mL/sec. The contrast agent used was ioxilan (Oxilan; Guerbet, Roissy, France). For conventional TACE, patients were treated with selective (lobar or segmental) and superselective injections. A solution containing 50 mg of doxorubicin (Pfizer Laboratories, New York, NY) and 10 mg of mitomycin C (Pfizer Laboratories) in a 1:1 mixture with iodized oil (Lipiodol; Guerbet, Paris, France) was infused and was followed by administration of 100–300-μm-diameter microspheres (Embospheres; Merit Medical, South Jordan, Utah). Substantial arterial flow reduction to the tumor was defined as the technical end point of embolization. Specifically, the end point was defined by the amount of heartbeats (ideally, two to five heartbeats) that were necessary to clear the contrast agent column (intraarterially visible contrast agent at the tip of the microcatheter). For drug-eluting beads TACE, patients were treated with selective (lobar or segmental) to superselective (sublobar) injections. Two milliliters of a deformable microsphere (LC Bead; Biocompatibles, BTG, Surrey, England) with a diameter of 100–300 μm was loaded with 100 mg of doxorubicin hydrochloride (25 mg/mL) and mixed with an equal volume of the nonionic contrast material. Up to 4 mL of drug-eluting beads was administered. Complete occlusion of the tumor-feeding blood vessels was avoided to maintain the arterial pathway for retreatment. Additional details of the TACE protocol have been reported previously (14).

MR Imaging Technique

All included patients underwent imaging with a standardized MR imaging protocol before and after the initial TACE. The mean time between baseline imaging and TACE was 14.5 days (standard deviation, 13.6 days; range, 1–50 days), and the mean time between the TACE procedure and follow-up MR imaging was 31.4 days (standard deviation, 17.7 days; range, 14–94 days). MR imaging was performed with a 1.5-T MR imaging unit (Magnetom Avanto; Siemens, Erlangen, Germany) by using a phased-array torso coil and 0.1 mmol per kilogram of body weight of intravenous gadopentetate (Magnevist; Bayer, Wayne, NJ). The protocol included breath-hold unenhanced and CE T1-weighted 3D fat-suppressed spoiled gradient-echo imaging (repetition time msec/echo time msec, 5.77/2.77; field of view, 320–400 mm; matrix, 192 × 160; section thickness, 2.5 mm; receiver bandwidth, 64 kHz; flip angle, 10°) in the late hepatic arterial phase (20 seconds, triggered using the time-to-peak technique), the portal venous phase (70 seconds), and the delayed phase (3 minutes).

Imaging Data Evaluation

Image evaluation was performed by two independent readers (a radiologist [R.D.] with 7 years of experience in abdominal MR imaging and a radiology resident [J.C.]) who did not perform the TACE procedure. One targeted lesion per patient was selected and used for analysis. A target lesion was defined as the largest, dominant lesion that was treated during the first session of TACE. The inclusion of multiple target lesions per patient was omitted to evaluate the role of dominant lesions only, as described in other works (15,16).

The non-3D measurements of the target lesions were performed on the basis of RECIST (largest tumor diameter) and mRECIST (largest enhancing diameter), as described in the literature (9,12,17,18). All measurements made by the two readers were performed by using standardized electronic calipers and Digital Imaging in Communications and Medicine files. Prior to the measurements, images were examined in axial, coronal, and sagittal reconstructions to visually identify the largest tumor expansion (for diameter and enhancement, respectively). The respective section with the largest expansion of the tumor was then used for individual manual measurements. The measurements were then averaged for the survival analysis. Any ambiguity in regard to the selected plane (axial, coronal, sagittal) of measurement was resolved in consensus. Figure 2 provides an overview of all anatomic and enhancement-based assessment methods.

Figure 2:

MR assessment techniques. A, Unidimensional measurement of the largest anatomic lesion diameter. B, Unidimensional measurement of the largest, unidimensional enhancing diameter. True tumor enhancement was defined as tumor areas with a hyperintense MR signal or increased contrast material uptake in the arterial phase T1-weighted MR imaging sequence that did not exhibit an elevated signal at native T1-weighted MR imaging. C, Segmentation-based tumor volume, with the red area indicating area of maximum enhancement or viable tumor. D, Quantification of enhancing lesion volume. Information about color coding is in Appendix E1 (online). On A, the yellow line is a ruler instrument measuring the largest diameter of the tumor, and on B, the yellow line is a ruler instrument measuring the largest enhancing tumor tissue.

Three-dimensional quantitative image analysis was performed by using a software prototype (Medisys; Philips Research, Suresnes, France) by the same readers during a different session by using fully anonymized image data sets 1 month after the conventional measurements were performed (10). The technical specifications of the utilized software prototype are explained in Appendix E1 (online). The software used semiautomatic 3D tumor segmentation in arterial phase CE MR imaging before the initial TACE session. The overall tumor volume was directly calculated on the basis of this segmentation and is expressed in cubic centimeters (Fig 2, C). The resulting 3D segmented volume was used for quantitative analysis of lesion enhancement (Fig 2, D). The accuracy, reader-independent reproducibility of semiautomatic tumor segmentation, as well as the radiologic-pathologic validation of the system have been reported elsewhere (19–22). The 3D quantitative calculation of enhancement was based on image subtraction (10,23). Briefly, the 3D segmentation mask used to calculate the overall tumor volume was transferred onto the subtraction image, and a region of interest was placed in extratumoral liver parenchyma as a reference to calculate the relative enhancement values within the tumor (24). The patient-specific, average signal intensity within the region of interest was defined as a threshold to estimate the overall enhancement within the 3D segmented volume. Subsequently, enhancing regions were expressed in cubic centimeters (enhancing volume) and were visualized by using a color map overlay on the arterial phase MR image (10). Additional methodological specifications for these techniques, as well as the region of interest specifications can be found in Appendix E1 (online).

Cutoff values were defined for each radiologic method to stratify the patient cohort into two groups: those with smaller tumors or lesser enhancement and those with larger tumors or greater enhancement. The definition of cutoff values was based on the most commonly used threshold of 5 cm overall tumor diameter. As such, the same cutoff value was used for the measurements of the largest diameter of the enhancing portion of the lesions. The cutoff for the volumetric threshold was extrapolated from the diameter threshold on the basis of the following equation: V = (4/3) · πr³, where V is the volume and r is the radius for the calculation of spheroid volumes, resulting in a volumetric threshold of 65 cm³, which was then used for the calculation of the overall, as well as the enhancing, tumor volumes.

The results of the baseline MR imaging stratification (groups with diameters above or below the 5-cm cutoff threshold and volumes above or below the 65-cm³ cutoff threshold) were related to the tumor response rates measured on postprocedural MR imaging. For the assessment on the first follow-up MR image obtained after the first TACE session, we used the common techniques of RECIST, mRECIST, and the European Association for the Study of the Liver (EASL) guideline (all of which are non-3D) to identify responders and nonresponders. Table 1 illustrates the utilized definitions for tumor response.

Table 1.

Response Assessment Definitions

Note.—CR = complete response, PD = progressive disease, PR = partial response, SD = stable disease.

Statistical Analysis

OS was defined from the date of the first TACE session until death. Patients who were lost to follow-up, were alive at the end-of-observation date (October 30, 2013), or were treated surgically were censored. The agreement of manual measurements between the radiologic readers was assessed by using linear regression analysis to investigate the correlation of results. The Pearson correlation coefficient (R²) was calculated, residual analysis was performed, and residual standard errors (RSEs) were calculated for each reader. Cutoff values as described above were used for each method to stratify the patient cohort in two groups: those with smaller tumors or lesser enhancement and those with larger tumors or greater enhancement. Survival curves were estimated with the Kaplan-Meier method and were plotted for each method and group. The median OS and the 95% confidence interval (CI) were calculated. The differences in the survival curves between the groups within each method were assessed by using the log-rank test. A P value of .05 or less was considered to indicate a significant difference. The predictive value of each parameter (eg, lesion diameter, volume, or enhancement) and technique was assessed by using Cox proportional hazard ratios (HRs). In the first step, a univariable Cox regression model was used to evaluate the association of OS with clinical factors assessed at baseline (Table 2): sex, age, ECOG performance score, multiplicity of lesions, cause, and the Child-Pugh class. In the second step, adjusted Cox proportional HRs for all radiologic measurements were estimated from the Cox regression model, which simultaneously included the respective radiologic method, as well as the clinical factors that were found to be significantly associated with patient OS (P ≤ .05) (25). Staging systems that have already included tumor size (Hong Kong Liver Cancer and Barcelona Clinic Liver Cancer staging systems) were not considered for these analyses. In addition, a stepwise logistic regression model with a forward selection process that was based on likelihood ratios was used to identify favorable combinations of tumor characteristics with regard to OS. All statistical computations were performed by using commercial statistical software (Prism Version 6, GraphPad Software, San Diego, Calif; and SPSS Version 22.0, IBM, Armonk, NY).

Table 2.

Baseline Patient Characteristics and Treatment History

Note.—Numbers in parentheses are percentages, and percentages were rounded. BCLC = Barcelona Clinic Liver Cancer staging system, HBV = hepatitis B virus, HCV = hepatitis C virus, HKLC = Hong Kong Liver Cancer class.

Results

Patient Characteristics

Table 2 summarizes baseline characteristics of the included patients. Mean patient age was 62.3 years (standard deviation, 12.3 years). Mean lesion size was 7.76 cm (standard deviation, 4.14 cm). Almost one-half of the patients showed an unaffected clinical performance prior to the first TACE session (49% [39 patients] were assigned an ECOG performance score of 0, 39% [31 patients] were assigned an ECOG performance score of 1, and 12% [nine patients] were assigned an ECOG performance score of 2).

Treatment, Clinical Outcome, and Interreader Agreement

Table 2 provides an overview of the treatment history, as well as the frequency, of TACE for the patient collective. All TACE sessions were technically successful, and no major toxicity was reported. Median OS of the entire population was 16.4 months (95% CI: 11.4, 21.5). For the non-3D quantitative measurements, the agreement between the radiologic readers was good for measurement of the largest tumor diameters (R²= 0.91), as well as for the largest unidimensional extent of enhancing tissue (R²= 0.86). No substantial differences between the individual measurements and the averaged values were observed for both readers (for reader 1, RSE was 8.9 for the largest diameter and RSE was 10.6 for the largest unidimensional extent of enhancing tissue; for reader 2, RSE was 9.1 for the largest diameter and RSE was 11.0 for the largest unidimensional extent of enhancing tissue).

Univariate Analysis

With the univariate analysis of clinical parameters used to characterize the patient cohort, we identified ECOG performance status (ECOG score, < 0; HR, 3.6; 95% CI: 1.5, 8.5; P = .004), as well as the Child-Pugh class (Child-Pugh class B; HR, 1.8; 95% CI: 1.1, 3.1; P = .028), as significantly correlated with OS (P < .05). All remaining baseline parameters (eg, tumor multiplicity, age, sex) that were included in the univariate analysis did not correlate with patient survival, when used to stratify the patient cohort (P > .05). As a result of the univariate analysis of the imaging parameters, for the stratification of the patient cohort according to both overall and enhancing tumor diameter, a significant stratification of the survival curves was not achieved (P = .234 and P = .08, respectively) (Fig 3). As for the 3D quantitative techniques, for both parameters, a significant separation was achieved, with P = .022 for the overall tumor volume and P = .017 for the enhancing tumor volume. Accordingly, patients with an overall tumor volume smaller than 65 cm³ showed a significantly higher median OS of 26.7 months (95% CI: 0.4, 52.9), as opposed to patients with larger dominant lesions who survived for a median of 15.4 months (95% CI: 14.1, 16.8). This separation was even greater when stratification was determined according to the enhancing tumor volume. Patients with enhancing tumor volumes smaller than 65 cm³ showed a median OS of 29.7 months (95% CI: 14.5, 44.9), as opposed to patients with a greater extent of enhancing tumor volume, who survived for a median of 15.0 months (95% CI: 10.4, 19.6) (Table 3).

Figure 3:

OS curves for the respective thresholds for utilized image assessment techniques. Kaplan-Meier analysis results are shown for each subgroup and technique on the basis of cutoff values. According to univariate analysis, only the volumetric techniques provided a significant separation of the survival curves (P < .05).

Table 3.

Univariate and Multivariate Analysis of the Median OS according to Measured Parameters

^*Numbers in parentheses are percentages, and percentages were rounded.

^†Numbers in parentheses are 95% CIs.

^‡Significant difference.

Multivariate Analysis

Adjusted for the clinical factors with a significant effect on OS, the multivariate analysis for the radiologic technique showed significant values for the stratification methods that were based on enhancing diameter (P = .026; HR, 1.9; 95% CI: 1.0, 3.4), overall tumor volume (P = .006; HR, 2.2; 95% CI: 1.3, 3.9), and enhancing tumor volume (P = .013; HR, 1.9; 95% CI: 1.2, 3.4). For the identification of potentially favorable combinations of the imaging-based measurements, significance was not achieved, and the stepwise forward selection identified the stratification according to the enhancing tumor volume as the technique with the highest association with outcome, while excluding the other parameters from the equation. When we further stratified patients with an overall tumor volume above 65 cm³ (n = 46) by using the enhancing tumor volume, those with less than 65 cm³ of enhancing tumor volume (n = 11) demonstrated a trend toward a higher median OS of 29.7 months (95% CI: 8.1, 51.3), as opposed to those with more than 65 cm³ of enhancing tumor volume (n = 35; median OS, 15.0 months; 95% CI: 10.4, 19.6). However, significance was not achieved with this result (P = .067).

Relationship of Baseline Imaging Parameters to Postprocedural Tumor Response

The baseline stratification methods were related to the postprocedural tumor response. The primary finding was that smaller lesions demonstrated slightly better response rates to TACE, when enhancement-based response criteria were applied. Table 4 provides a detailed overview of the response assessment in each group. Of note, target lesions with smaller overall volumes on baseline images (< 65 cm³) showed lower RECIST response rates (0%, zero of 33) as compared with larger lesions (13%, six of 46). However, smaller lesions exhibited considerably higher response rates according to mRECIST (27%, nine of 33) and EASL (45%, 15 of 33) guidelines, as compared with larger lesions (15% [seven of 46] for mRECIST; and 24% [11 of 46] for EASL guidelines). A similar trend was observed for lesions with a lesser extent of enhancing volume (< 65 cm³) that exhibited low RECIST response rates (5%, two of 44), while showing slightly better response rates according to the mRECIST (20%, nine of 44) and EASL (39%, 17 of 44) guidelines.

Table 4.

Response Rates at Follow-up Imaging

Note.—Numbers in parentheses are percentages, and percentages were rounded.

Discussion

The main finding of this study was that volumetry-based tumor size thresholds are prognostic discriminators with regard to postprocedural prognosis in patients with HCC who will undergo treatment with TACE. While the quantification of volumetric lesion enhancement was identified as having the highest association with outcome on baseline images, conventional measurements of lesion diameter—as used in most staging systems—could not be confirmed as a reliable stratification instrument in this subset of patients.

Substantial efforts have been made to properly understand the specific biological characteristics of the mostly hypervascular HCC lesions with regard to baseline imaging prior to local-regional therapies (13,26). We hypothesized that tumors with a greater extent of enhancing tissue have a correspondingly higher proliferative potential, as compared with lesions with only minimal enhancement, which will then reflect on patient survival. This hypothesis was confirmed by our results, which demonstrated significantly better OS rates in patients with a lesser extent of viable tumor tissue. Interestingly, the stratification according to overall tumor volume was similarly associated with patient survival compared with the quantification of enhancing lesion portions. This is not surprising, primarily because both values are clearly interconnected. However, the results of the subgroup analysis in patients with volumetrically larger tumors confirmed the trend toward a further separation of survival rates according to lesion enhancement. As such, patients with anatomically larger lesions showed better survival rates if the overall extent of enhancing tumor volume was below the identified volumetric threshold. Most important, both volumetric techniques have demonstrated significant benefits over the conventional, unidimensional measurements. This can be mainly explained by the nature of nonvolumetric measurements. These methods are based on the assumption that tumor growth happens in a symmetrical, spherical manner, with a homogenous uptake of MR contrast material. However, most liver tumors are prone to volumetric asymmetry and frequently demonstrate inhomogeneous patterns of tumor enhancement. This is particularly the case in HCC lesions, where baseline CE MR imaging frequently demonstrates nonhomogeneously growing nonspherical tumors with segmental enhancement and, in some cases, even scattered foci of enhancing tumor tissue, which cannot always be measured by using unidimensional techniques. Principally, the volumetric quantification of overall and enhancing lesion volume is biologically more representative and provides a unique advantage of a whole-tumor analysis (10,11). It is capable of providing a representative quantification of tumor characteristics regardless of tumor shape or enhancement distribution. In addition, segmentation-based 3D quantitative measurements are known to provide technical benefits with regard to reproducibility and interobserver agreement, which might very well result in an improved stratification of patients (20). In addition to these results, the assessment of tumor response provided important insight into the role of baseline lesion characteristics with regard to expected response rates after TACE. Our results suggest that smaller lesions are likely to exhibit a better response to TACE; however, the differences in response rates between the groups were only minimal. It can thus be stated that tumor response to TACE is not influenced by lesion size, with slight benefits for smaller lesions and tumors with a smaller amount of enhancing tissue.

This study had some limitations: First, our analysis relied on a retrospective assessment of dominant index lesions only and did not include nontarget lesions. However, several studies established the use of the dominant lesion as an independent, prognostically sufficient predictor of survival after intraarterial therapy (15,16). This approach is based on the assumption that the largest lesion can reflect the overall tumor burden and, in fact, represents the most aggressive focus of the disease. In addition, this approach reduces the clinical effort of a radiologic reading while maintaining a reasonable accuracy of the method. Second, an important limitation is the definition of a single cutoff value per method. Ideally, an extended analysis would introduce a stepwise approach or at least one additional cutoff, as described within the Barcelona Clinic Liver Cancer staging system, with the goal of including very small lesions with a diameter of less than 3 cm into a separate group. However, our medium-size patient cohort with largely intermediate-advanced–stage disease did not offer a sufficient number of individuals that would allow for a meaningful subgroup analysis in this regard. Furthermore, our analysis sought to test the 5-cm threshold value, which is the most frequently described value in the literature. An additional but necessary limitation of this study was the exclusion of patients with extrahepatic disease, as well as portal venous invasion. The final verdict on the value of intraarterial therapies in this subgroup of patients has yet to be delivered, and it was thus decided to exclude this subgroup of patients from the analysis.

In summary, our findings support the use of volumetry-based biomarkers in baseline MR imaging in patients with HCC who will undergo treatment with TACE. The association of lesion enhancement on baseline imaging with patient outcome deserves further studies and may potentially be included in newer staging algorithms for HCC.

Advances in Knowledge

• ■ Three-dimensional (3D) volumetric thresholds (65 cm³ of the overall or enhancing tumor volume) for the staging of hepatocellular carcinoma (HCC) lesions on contrast material–enhanced MR images prior to transarterial chemoembolization (TACE) are prognostically more accurate as compared with the commonly used, diameter-based cutoffs (5 cm).

• ■ The stratification of prognosis according to overall and enhancing tumor diameters did not result in a significant separation of survival curves (hazard ratio [HR], 1.4; 95% confidence interval [CI]: 0.7, 2.5; P = .234; and HR, 1.6; 95% CI: 0.9, 2.8; P = .08, respectively).

• ■ With univariate and multivariate analysis, the stratification of prognosis according to overall and enhancing tumor volume achieved significance (HR, 1.8; 95% CI: 0.9, 3.4; P = .022; and HR, 1.8; 95% CI: 1.1, 3.1; P = .017, respectively).

Implication for Patient Care

• ■ When diameter-based techniques were compared with 3D volumetric thresholds, significantly better stratification of survival was achieved with 3D volumetric thresholds, and they can therefore be considered as reliable prognostic discriminators for future staging systems.

APPENDIX E1

Tumor Segmentation

To optimize the accuracy of the tumor segmentation, a systematic approach was used to determine the imaging sequence to be used for segmentation. The following describes the approach, and then information on the segmentation software itself is provided. A semiautomatic 3D segmentation of the tumor was performed by using a multiphasic CE MR sequence performed before TACE. The segmentation was performed during multiphasic CE MR imaging that included arterial (20 seconds) and portal venous (70 seconds) phases. For the final analysis, the arterial phase was chosen over the portal venous phase because all lesions in the analysis demonstrated much better enhancement in the early phase. The tumor segmentation was performed by using in-house software (Medisys; Philips Research). This software uses non-Euclidean geometry and theory of radial basis functions, which allows the segmentation of 3D objects with straight edges and corners (19). The algorithm creates image-based masks located in a 3D region whose center and size are defined by the user, yielding the nomenclature “semiautomatic.” After identifying an initial control point, the user can interactively expand or contract the 3D mask. Adjustments of the overall 3D volume of the mask can be interactively performed by placing additional control points. The shape and the spatial localization of the final 3D segmented mask (Fig 2, C) are registered to the coordinates within the MR imaging data set and—on image registration—may be applied to other MR images in the same patient. With the 3D nature of the segmentation, the tumor volume can be directly calculated.

Calculation of Enhancing Volume

To calculate the volume of the enhancing tumor tissue, the following steps were performed:

• 1The unenhanced MR image was subtracted from the arterial phase image to remove background enhancement (23). This step is of particular importance to achieve an accurate assessment of lesions with hemorrhagic necrosis and helps mitigate false-positive enhancement from contrast enhancement.
• 2The 3D segmentation mask from the arterial phase CE MR image was transposed onto the subtracted image set from above.
• 3A 3D region of interest (1 cm³) was placed in extratumoral liver parenchyma of the subtracted image set to calculate the relative enhancement values within the tumor volume as a reference for normalization (Fig 2, D) (24). Additional information on the selection of regions of interest is provided in the next section (Appendix E1).
• 4A threshold that was based on image enhancement was used to define viable tumor tissue as voxels within the 3D mask, where the enhancement exceeded the average +2 standard deviations value of the region of interest. Additional information on the calculation of the region of interest–based threshold is provided in the next section (Appendix E1).
• 5A normalized color map overlay on the arterial phase MR image was used to demonstrate regional tumor enhancement heterogeneity (Fig 2, D), with red representing maximum enhancement or viable tumor and blue representing below the threshold, no enhancement, or necrotic tumor tissue (10). Additional information on color coding is provided below.

Definition of the Region of Interest and Color Coding

As opposed to fully automated segmentation techniques, a semiautomated approach allows the combination of software-based image processing with manual adjustments by a radiologic reader. The goal of the region of interest selection in this study was to achieve an intuitive approach, resembling the reference standard of a radiologic reading. Practically, a radiologic reader compares enhancement properties of the tumor with the nontumoral liver tissue rather than extrahepatic tissue. Several region-of-interest localizations (including extrahepatic regions of interest within nonenhancing soft tissue, eg, Psoas muscle) were considered; however, they appeared to be counterintuitive and failed to provide consistent results. The abnormal enhancement patterns of cirrhotic livers were taken into account when selecting the localization of a region of interest. Accordingly, regions of interest were placed within visually nonenhancing tissue on the postsubtraction arterial phase image. Furthermore, to avoid corrupted regions of interest within focally cirrhotic liver tissue, signal intensity statistics were calculated for every 3D (1 cm × 1 cm × 1 cm = 1 cm³) region of interest with the goal of achieving a maximum of signal homogeneity. This was performed as follows:

• 1A 1-cm³ region of interest was placed in a localization, as described above (ipsilateral liver lobe, nonenhancing extratumoral areas).
• 2The software provided the minimum and maximum voxel brightness values within the cubic region of interest. The numeric output was in patient-specific arbitrary units for each region of interest. The software furthermore calculated the mean brightness value (MBV), standard deviation, as well as the coefficient of variation (CV). Empirically, a CV of less than 30% was seen as acceptable, while a CV greater than 30% was rejected, leading to region of interest repositioning.
• 3The MBV ± 2 standard deviations was selected as a cutoff (threshold), with all values above seen as real contrast enhancement.
• 4On the basis of the selected region of interest and the MBV, a patient-specific (normalized) 3D color map was overlaid onto the tumor tissue enclosed by the segmentation mask. The color blue was identified as areas with equal or lower signal intensity as the MBV ± 2 standard deviations, while all signal exceeding this value was coded as an equally distributed histogram of tissue enhancement. The range of enhancement in arbitrary units was coded in color shades (aqua, yellow, red), with red representing the maximum signal intensity for a particular patient.