Functional imaging of hepatocellular carcinoma using diffusion-weighted MRI and 18F-FDG PET/CT in patients on waiting-list for liver transplantation

Background To compare the apparent diffusion coefficient (ADC) on diffusion-weighted imaging (DWI) with the standardized uptake values (SUV) measured by18F-FDG-PET/CT in naïve hepatocellular carcinoma (HCC) nodules, and to determine whether these markers are associated with tumours at high-risk of aggressiveness. Methods From 2007 to 2010, all patients with HCC on the waiting list for liver transplantation and who underwent both FDG-PET/CT and 1.5-T DWI-MRI (b values: 0, 200, 400, and 800 s/mm2) were included in this institutional review board-approved retrospective study. Tumour size, tumour ADC, tumour-to-liver ADC ratio (ADCT/L), maximal tumour SUV and tumour-to-liver SUV ratio (SUVT/L) were measured and compared to serum alpha-fetoprotein (AFP) levels, tumour size and differentiation grade on explanted specimens. Results A total of 37 HCC nodules in 28 patients were available for correlation between MRI and PET/CT, 7 of which (in 7 patients) showed a SUVT/L > 1.15. We did not find any correlation between tumour ADC or ADCT/L and tumour SUV or SUVT/L. To note, SUVT/L was positively correlated with AFP levels (R = 0.95, P ≤ 0.0001), while ADCT/L was not (P = 0.73). Twenty-four patients (with 32 nodules) underwent liver transplantation. In this subgroup, an increased SUVT/L ratio was associated with larger tumours (average size, 32 ± 14 mm; range, 18–60 mm; P < 0.0001) and with poor differentiation on pathology (grades 3 and 4; P = 0.04), while ADCT/L was neither associated with tumour size or differentiation grade. Conclusions ADC and SUV measures in HCC nodules are not correlated. SUVT/L ratio correlates with AFP levels, tumour size and poor differentiation, and should probably be integrated as a co-variable in a predictive outcome model of patients on the waiting-list for liver transplantation.


Background
While the prognosis of hepatocellular carcinoma (HCC) without treatment is sombre, with 5-year life expectancy below 7 %, optimal outcome is obtained in patients with low tumour burden, eligible for liver transplantation (LT) [1], especially within the Milan criteria [2]. Several studies have reported that patients transplanted beyond the Milan criteria can achieve long-term recurrence-free survival: pathological features of HCC tumours appear to play an important role as prognostic markers of disease-free survival; increased alpha-fetoprotein (AFP) levels, micro vascular invasion (MVI), differentiation grade, and satellite nodules all appear to be associated with impaired outcome following LT [3][4][5][6][7][8][9]. As a result, the sole assessment of the number and size of HCC lesions in LT candidates is unable to precisely address the risk of cancer recurrence after LT. As some pathological prognostic markers such as MVI cannot be assessed on biopsy specimens, growing interest has raised for the evaluation of HCC using functional imaging as an adjunct to biology and conventional imaging for optimal tumour grading.
Several studies have reported that pre-LT PET/CT with [ 18 F] 2-deoxy-2-fluoro-D-glucose (FDG) may play a role in predicting tumour recurrence [10][11][12][13][14][15] in highgrade HCC tumours, which have a reduced glucose-6phosphatase activity, resulting in progressive accumulation of the tracer within tumour cells as compared to adjacent liver. Hence, FDG standardized uptake value (SUV) in HCC, as compared to adjacent normal liver (tumour-to-liver ratio), reflects tumour aggressiveness, including histological tumour grade and AFP levels, which in turn correlate with outcome. For Lee et al., a longer disease-free survival was observed after LT in patients with a tumour-to-liver ratio < 1.15, i.e., with less than 15 % higher FDG uptake above normal liver background [16]. These data were later confirmed by Kornberg et al., demonstrating the pejorative predictive value of FDG-positive HCC [17].
Diffusion-weighted imaging (DWI) provides information on the random Brownian motion of water molecules within tissues, which reflects their cellularity. Some authors have reported that high-grade HCC tended to show decreased apparent diffusion coefficient (ADC) values, as compared to well-differentiated tumours [18], although this remains to-date debated [19]. Hence, both SUV and ADC could be considered as biomarkers of HCC tumours at risk of recurrence following LT, and could thereafter be tested as diagnostic and prognostic tools for HCC management. To the best of our knowledge, however, there have been no reports correlating both markers in HCC. The aim of our study was to measure and compare the ADC values on DWI with SUVs measured by FDG-PET/CT in naïve HCC patients on waiting-list for LT, and to determine whether these markers can be associated with tumours at high risk of aggressiveness using the pathology of explanted specimens as the reference standard.

Patients
The local patient database was searched retrospectively to identify all consecutive patients with HCC referred to our institution for LT between 2007 and 2010, who underwent both liver MRI and FDG-PET/CT prior to LT as part of their work-up. During this period, 78 patients met the inclusion criteria. We excluded patients in whom DWI sequences were not performed (n = 15), patients bearing liver lesions all < 10 mm on MRI (n = 13) and patients who underwent percutaneous tumour ablation or chemoembolization prior to either PET/CT or liver MRI (n = 22). As a result, 28 patients were eligible, including 23 men and 5 women, with a median age of 59 years (range, 46 to 69 years); 9 patients were Child-Pugh A, 15 Child-Pugh B, and 4 Child-Pugh C; the underlying liver disease in the study population is shown in Table 1. Based on pre-LT imaging data, 67.8 % (19/ 28) patients fulfilled the Milan criteria (2), and 82.1 % (23/28) fulfilled the UCSF criteria [20]. All were included in clinical trials, which were active at this time, and signed informed written consent, in accordance with our Institutional Review Board (Comité de Protection de Personnes, Ile-de-France IX).
Fifteen patients underwent neo-adjuvant treatments before LT, all of which were performed after both the liver MRI and PET/CT: 13 patients underwent lipiodoldoxorubicin based chemoembolization, either alone (n = 8) or combined with radiofrequency ablation (n = 3) and surgery (n = 2); 2 patients underwent radiofrequency  [21] diagnostic criteria on MRI was determined, thereafter defining whether each patient fulfilled the Milan and UCSF criteria [2,20]. Analyses were performed on a commercially available Siemens Workstation (Syngo MMWP, Siemens Healthcare, Forchheim, Germany). In addition, for each HCC nodule, the following parameters were measured: size, location according to Couinaud liver segmentation, T1 signal intensity (SI) relative to muscle, T2SI relative to spleen. ADC maps were obtained using a monoexponential fit of the signal decrease on the 4 b values on a voxel-by-voxel basis. A region of interest (ROI) encompassing the entire lesion was placed on the b0 image, and then, propagated to the ADC map to provide the mean tumour ADC (ADC T ). The ADC of the normal liver parenchyma (ADC L ) was also measured on the right posterior segment of the liver, using a 20 mm ROI positioned away from tumours, vessels, and bile ducts.
The tumour-to-liver ratio (ADC T/L ) was computed for each HCC lesion.

PET/CT imaging
The median delay between PET/CT and MRI was 3 days (range, −181 to 245 days). All images were acquired using an integrated PET/CT scanner (Gemini GXL16, Philips, Da Best, The Netherlands). After fasting for at least 6 h, patients were injected with an average 5.2 MBq/kg FDG, and images were acquired 1 h later. A low-dose CT scan was obtained for attenuation correction and localization purposes (100 kV, 80 mAs), followed by an emission scan from the top of the skull to mid-thigh with 9-11 bed shifts of 2 min each. Images were reconstructed using a line-of-response row action maximum likelihood algorithm (LOR-RAMLA) with matrix size of 144x144, and corrections for attenuation, scatter and random coincidences. CT was obtained without contrast enhancement.
PET, CT, and fused PET/CT images were analyzed on a proprietary workstation (Syntegra, Philips, Milpitas, CA). SUV was defined as tissue concentration of FDG (kBq/mL) divided by the injected activity per weight (MBq/kg). All images were analyzed by 2 observers (EI, EE) blinded to MRI analysis. For each FDG-avid HCC nodule, the following parameters were collected: location of the tumour according to Couinaud liver segmentation, maximum SUV of the tumour (SUV T ) and maximum SUV of the normal liver parenchyma (SUV L ). ROIs were drawn on 5 contiguous slices around each focus of tumour uptake in order to capture the maximum SUV; in the normal liver parenchyma, ROIs were drawn at mid height, making sure that VOI outlines were restricted to areas of physiologic uptake and avoiding neighbouring sites of disease. The tumour-to-liver ratio (SUV T/L ) was calculated for each lesion; when no clear focus of uptake was identified, SUV T was considered equal to SUV L , and SUV T/L was set at 1.

Liver transplantation
Of the 28 patients, 24 were transplanted during the study period, while 2 patients showed tumour progression and were dropped from the waiting list, 1 patient was excluded of LT owing to aortic stenosis and 1 patient declined LT after registration. The median delay between MRI and LT was 117 days (range 1 to 446 days). The median delay between PET/CT and LT was 104 days (range, 23 to 394 days). In all cases, a bilateral subcostal incision with an upper midline extension was performed. LT was performed using preservation of the inferior vein cava (IVC) and temporary porto-caval anastomosis except when IVC resection was necessary. Reconstruction was performed in the piggyback fashion with end-to-side cava-caval anastomosis on the joined stump of the three main hepatic veins. Portal, arterial and biliary reconstructions were performed using standard techniques.

Pathology examination -Correlation with imaging findings
Explanted liver specimens were cross-sectioned and analyzed by a pathologist (JCa) blinded from imaging results. After formalin fixation, each liver slice was further pared to less than 5 mm thickness. All liver nodules, whether benign or consistent with HCC lesions, were macroscopically located according to Couinaud classification. The largest macroscopic diameter of each HCC lesion was recorded. For all HCC, at least a complete slice of the tumour including both the centre and the periphery was sampled for microscopic analysis. After hematoxylineosin-saffran staining, each nodule was analyzed for the presence of tumour cells. Microscopic evaluation of all tumours included the assessment of the following: tumour differentiation according to Edmondson-Steiner classification, presence of a capsule, quantification of necrosis, presence of MVI, and presence of satellite nodules. Patients with aggressive HCC were defined as presenting either an undifferentiated HCC (Edmondson-Steiner grades 3 and 4), and/or a micro vascular invasion, or beyond Milan criteria on the explanted specimen. Then, radiologists, nuclear medicine physicians and pathologist ensured adequate correlation between imaging and pathological data in a consensus meeting. Co-localization of tumours on pathology and imaging was not only based on segmental classification but also on the identification of specific landmarks such as hepatic veins, portal branches and central or peripheral location. Only HCC nodules retrieved simultaneously on pathology and on MRI were considered for statistical analyses.

Statistics
The correlation between DWI-MRI and PET/CT variables (ADC T , ADC L , ADC T/L , T1SI relative to muscle, T2SI relative to spleen, SUV T, SUV L, SUV T/L ) was analyzed using Pearson's correlation coefficients based on pair wise complete data (37 nodules in 28 patients). In addition, because of possible erroneous ADC measures within the left lobe, induced by cardiac motion, this correlation analysis was repeated only for right lobe lesions. The association between imaging features and pathological outcomes (32 nodules in 24 LT patients) were analyzed using the following methods: 1) dichotomous outcomes (within/beyond Milan criteria, absence/presence of MVI, differentiated/ undifferentiated HCC) vs. dichotomous FDG uptake (PET/CT was considered positive if the ratio SUV T/L was > 1.15 [16]) were analyzed based on categorical analyses using Fisher exact test; 2) dichotomous outcomes vs. continuous imaging features (DWI-MRI and PET/CT variables) were analyzed based on univariate logistic regression; 3) continuous outcomes (differentiation status) vs. all imaging features were analyzed based on univariate linear regression. A P-value < 0.05 was considered statistically significant. All analyses were conducted using R statistics software package.

PET/CT analysis
The mean SUV T of the 37 HCC lesions retrieved on MRI was 2.9 ± 0.7 (range, 2.3 to 6.3). The mean SUV L was 2.7 ± 0.3 (range, 2.3 to 3.6). The mean SUV T/L was 1.08 ± 0.28 (range, 1.00 to 2.63). Overall, using the SUV T/L cut-off value of 1.15, PET/CT was considered positive in 7 HCC in 7 patients: the mean SUV T was 3.8 ± 1.2 in these nodules, accounting for a mean SUV T/L ratio of 1.44 ± 0.53 (range, 1.19 to 2.63, Fig. 1).

Correlation between DWI-MRI and PET/CT
ADC T was not correlated with either SUV T , SUV L , or SUV T/L ratio ( Table 2, Figs. 2 and 3). ADC L did not correlate either with SUV L , SUV T , or SUV T/L ratio. ADC T/L ratio did not correlate with either SUV T/L ratio, SUV T , or SUV L . Because ADC measurements may be less reliable in the left lobe of the liver due to cardiac motion, we considered in a second analysis only lesions located within the right lobe (n = 28): ADC T was not correlated with either SUV T (R = 0.25, P = 0.3) or SUV T/L ratio (R = 0.29, P = 0.21). ADC L did not correlate either with SUV T (R = 0.24, P = 0.4) or SUV T/L ratio (R = 0.15, P = 0.57). ADC T/L ratio did not correlate with either SUV T (R = −0.18, P = 0.46) or SUV T/L ratio (R = −0.12, P = 0.6).

Pathology
Gross examination revealed 32 HCC nodules in the 24 LT patients, accounting for a mean number of 1.3 lesions per patient (17 patients had 1 nodule, 6 had 2 nodules, and 1 had 3 nodules). The mean size of these HCC nodules was 21 ± 11 mm (range, 4 to 47 mm). Overall, 11 patients (46 %) met the Milan criteria based on the pathological analysis of the explanted specimens. Eight patients (33 %) showed at least one HCC lesion with MVI. The Edmondson-Steiner histological grade for LT was as follows: 2 lesions were grade 1, 7 lesions were grade 2, 17 lesions were grade 3, and 2 lesions were grade 4; four lesions could not be graded because of significant tumour necrosis.

Correlation between imaging variables and criteria of aggressiveness
In the 28 patients on the waiting list, the median AFP level for all Child Pugh stages was 4.7 ng/mL (range, 0.7 to 13,469.0 ng/mL). The median AFP levels did not differ between Child-Pugh stages A, B and C: 5.0 ng/mL (range, 0.7 to 283.0 ng/mL), 4.6 ng/ml (range, 2.0 to 13,469.0 ng/mL), and 13.0 ng/mL (range, 2.2 to 33.0 ng/ mL), respectively (P = NS). SUV T/L ratio significantly correlated with the AFP level (R = 0.95, P < 0.0001) whereas no apparent correlation was detected between AFP and ADC T or ADC T/L ratio (P = 0.08 and P = 0.73, respectively). The mean AFP level was significantly higher in patients with SUV T/L ratio > 1.15 than in PET/ CT-negative patients (1951.9 ± 5078.9 ng/mL vs. 26.3 ± 64.0 ng/mL, respectively; P < 0.0001). In the 24 transplanted patients, the mean HCC tumour size was also significantly larger in patients with SUV T/L ratio > 1.15 than in PET/CT-negative patients (32 ± 14 mm vs. 23 ± 7 mm; P < 0.0001). Tumour size did not correlate with ADC T values (P = 0.25). In addition, positive PET/CT with SUV T/L ratio > 1.15 was associated with poor differentiation on pathology (P = 0.04), but not with MVI or Milan criteria (P = NS). None of the MRI variables -including ADC T , ADC T/L , T1SI relative to muscle, and T2SI relative to spleen -correlated with pathological outcomes, including differentiation grade, MVI and Milan criteria.

Discussion
DWI-MRI and FDG-PET/CT provide distinct functional and morphological information. To date, MRI or enhanced CT are essential for HCC lesion detection and characterization according to practice recommendations, while FDG-PET/CT is not. This study sought to evaluate the correlation between metabolic and DWI parameters, and to compare these parameters with established biomarkers of tumour aggressiveness and prognostic assessment. Increased FDG uptake within HCC nodules was associated with larger and less differentiated tumours, whereas ADC values observed in this study population appear not to correlate with AFP, differentiation on pathology and Milan criteria. c d e f Fig. 3 Poorly differentiated HCC in a patient with left lobe HCC, showing no restricted ADC on DWI-MRI and increased uptake on FDG-PET/CT. Dynamic 3D gradient echo T1 WI shows a non enhancing nodule on arterial phase (a, arrow), remaining hypo intense on portal phase (b, arrow) but associated to a thin peripheral enhanced rim. The nodule shows high signal intensity on DWI image with b = 0 s/mm 2 (c, arrow) decreasing on DWI b = 800 s/mm 2 images (d, arrow), although its signal intensity remained higher than the liver. The tumour ADC was 1.38 × 10 −3 mm 2 /s (e, arrow), similar to that of the adjacent liver. The lesion proved positive on FDG-PET/CT (f) AFP levels > 1000 ng/mL showed at high risk of recurrence [9]. Our study confirms that PET/CT findings correlate with AFP levels. PET/CT was considered positive if the SUV T/L ratio was > 1.15, which occurred in 7 patients on the waiting list (25 %), of whom 2 demonstrated tumour progression and did not undergo LT.
Higher-grade and less-differentiated tumours (according to Edmondson-Steiner classification) were associated with higher levels of FDG accumulation. Decreased levels of glucose-6-phosphatase observed in poorly differentiated HCC lead to increased FDG accumulation [10,22]. Our results confirm the performance of FDG-PET/CT in predicting tumour aggressiveness using the explanted specimen as the reference standard. In addition, PET/CT performed once as a screening tool even as far as a median 104 days prior to LT still correlates with pathological features of HCC. The ADC value, measured by DWI, usually reflects tumour cellularity as well as interstitial tissue assessment. The mean ADC T values observed in this study are comparable to those reported in the literature [18,23]. Similarly, ADC T/L ratios observed in this study are not high, reflecting the low contrast between the tumour lesions and adjacent liver on ADC maps [24][25][26]. This could be partly explained by the restricted diffusion observed in liver fibrosis [27]. There have been conflicting reports regarding the link between ADC values and HCC differentiation: for some, although differences are not significant in terms of parametric ADC values, a higher signal intensity of HCC lesions in DWI is usually associated with higher pathological grades [18,19]. However, the reasons accounting for an increase in mean ADC values vary: tumour necrosis is responsible for reduced cellularity and increased ADC [28]; on the opposite, increased necrosis is also observed in highly aggressive HCC lesions with active portions of the tumour mostly located at the periphery of the lesions [29]. Last, tumour vascularization impacts ADC measurements, especially when low b-values are included, such as in this study where 4 b-values were used, including b = 0 s/mm 2 . As multiple b-value acquisitions were not studied, the specific influence of the perfusion component of diffusion in predicting tumour aggressiveness could not be tested here. In addition, the ADC measurements encompassed the entire HCC lesions, reflecting all components of the tumour observed on a single slice rather than on the entire tumour volume, which could also have limited the performances of ADC measurements. FDG uptake on the other hand correlates with the metabolic activity of viable tumour cells. Although necrosis decreases the overall tumour uptake, the SUV is measured from the area of highest uptake within the lesion set; the SUV only considers the most aggressive tumour components. Comparing mean ADC to maximum SUV does not reflect a true comparison of the most active component of each lesion. Moreover, a majority of HCC lesions observed in our study showed similar SUV values as compared to the adjacent liver, thus reducing the dispersion of SUV values as compared to ADC measurements. The use of novel tracers such as 18 F-fluoromethylcholine (FCH) could allow a better distribution of uptake values in HCC lesions even in well-differentiated HCC lesions.
In contrast with previous publications, this study focused on patients on waiting list for LT, using pathology of explanted specimens as the reference standard in a large majority of them. Moreover, there have been conflicting reports on the correlation between DWI and PET/CT in HCC, as tissue cellularity and glucose metabolism may represent 2 different facets of tumour biology [30]: some studies have suggested a positive correlation between tumour SUVs and ADC values [31][32][33][34][35][36][37]. Other authors have reported an inverse relationship between these parameters, especially when discriminating malignant from benign lesion, or when predicting the aggressiveness of some tumours [31,[33][34][35][36][37][38][39]. Hypercellular areas could increase impedance to water diffusion, resulting in low ADC values and high FDG uptake, while lower cellularity areas could show increased ADC with decreased FDG uptake. These relationships were reported mainly in extra-digestive tumours possibly because the influence of motion on ADC measurements is more limited in this setting. This further suggests the need to test DWI-MRI together with multitracer PET (FDG and/or FCH) in dedicated PET/MR platforms allowing simultaneous acquisitions and limited influence of motion on parametric imaging analysis [40].
Several limitations of the presented study have to be mentioned. First, the patient population was small in this monocentric retrospective study. There is currently no national recommendation for systematic FDG-PET/ CT in routine before LT. Second, only patients on LT waiting list were considered eligible, which prevented inclusion of more aggressive and larger tumours. The study design further excluded patients who were treated with chemoembolization, radiofrequency, and chemotherapy before DWI-MRI because these treatments alter ADC measures [41,42]. Interestingly, 15 patients were treated between MRI and LT, which could possibly have impacted the correlation between imaging and pathology [41,42]. However, this did not affect the SUV and ADC correlation as all correlations were made on imaging examinations performed prior to any treatment. There was no image fusion between PET/CT and DWI-MRI; however, to our knowledge and to-date, no software enables the correct simultaneous fusion between b-maps, ADC maps and PET/CT images.
PET/CT was obtained without contrast. However this did not impact the co-registration of MRI images: all MRI lesions were detected on both FDG-PET/CT and on pathology. Additionally, it is important to note that only patients meeting the EASL-AASLD criteria were included, limiting the analysis to lesions larger than 10 mm with early enhancement on arterial phase. However, the aim of this study was not to assess the performance of each modality for the detection of HCC in patients on waiting list for LT, for which MRI plays a key role, but to study the possible predictive nature of FDG-PET/CT or DWI sequences on tumour aggressiveness. The median delay between FDG-PET/CT and MRI-DWI was 3 days with a maximum of 245 days. FDG-PET/CT and DWI-MRI datasets were evaluated separately without knowledge of the findings of the other imaging modality and without knowledge of histopathology findings.

Conclusions
Although mean ADC and FDG-PET/CT maximum SUV of HCC in patients on waiting list for LT are not correlated, both tumour size, poor differentiation and AFP levels correlate with SUV T/L , suggesting that FDG-PET/ CT can predict tumour aggressiveness and should probably be integrated as a covariable in a predictive outcome model of patients with HCC candidate to LT, while MRI must be used for lesion detection and characterization.