Potential conflict of interest: Nothing to report.
This study evaluates the performance of diffusion-weighted magnetic resonance imaging (DWI) for the detection of hepatocellular carcinoma (HCC) in pre–liver transplantation patients, compared and combined with contrast-enhanced T1-weighted imaging (CET1WI), using liver explant as the standard of reference. We included 52 patients with cirrhosis (40 men, 12 women; mean age, 56 years) who underwent DWI and CET1WI within 90 days of liver transplantation. Magnetic resonance images were analyzed for HCC detection in three separate sessions by two independent observers: DWI images (DW-set), CET1WI (CE-set), and all images together (All-set). Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), per-patient accuracy, and per-lesion PPV were calculated for each image set. A total of 72 HCCs were present in 33 patients at explant (mean size, 1.5 cm [range, 0.3-6.2 cm]). Per-patient sensitivity and NPV of CE-set were significantly higher than those of DW-set when using pooled data between observers (P = 0.02 and 0.03, respectively), whereas specificity, PPV, and accuracy were equivalent. Per-lesion sensitivity was significantly higher for CE-set versus DW-set (59.0% versus 43.8%; P = 0.008, pooled data from two observers). When stratified by lesion size, the difference was significant only for lesions with a size between 1 and 2 cm (42.0% for DW-set versus 74.0% for CE-set; P = 0.001). The addition of DWI to CET1WI improved sensitivity for the more experienced observer. Conclusion: DWI is outperformed by CET1WI for detection of HCC, but represents a reasonable alternative to CET1WI for detection of HCC with a size above 2 cm. The addition of DWI to CET1WI slightly increases the detection rate. (HEPATOLOGY 2012;56:140–148)
Contrast-enhanced magnetic resonance imaging (MRI) after bolus injection of gadolinium chelates is routinely used in many centers for the detection and characterization of hepatocellular carcinoma (HCC) lesions, mainly based on the increased arterial supply in most HCCs. The reported sensitivity of MRI for HCC detection varies between 55% and 77.8%.1-7 Contrast-enhanced MRI is limited, however, by the possibility of false negatives (mainly for small lesions) and false positives (mainly related to nontumorous arterioportal shunts), which may decrease its diagnostic accuracy.
Diffusion-weighted MRI (DWI) has recently gained interest in liver imaging, showing improved detection of liver lesions compared with T2-weighted imaging (T2WI),8-12 and enabling lesion characterization using apparent diffusion coefficient (ADC).10, 12-19 However, there are limited data on the use of DWI for the detection of HCC.20-26 To our knowledge, only one study to date has correlated DWI with liver explant findings.26 This study, which included a limited number of patients, showed lower sensitivity of DWI for HCC detection compared with contrast-enhanced T1-weighted imaging (CET1WI). Conversely, another recent study showed higher sensitivity of DWI for HCC detection compared with CET1WI.25 Based on our clinical experience, we hypothesize that DWI may add useful information to CET1WI for HCC detection.
The objective of our study was to assess the performance of DWI for the detection of HCC in pre–liver transplantation patients, compared and combined with CET1WI (using extracellular gadolinium chelates), using liver explant as the standard of reference.
This single center study was Health Insurance Portability and Accountability Act compliant. Approval for this retrospective study was obtained from local institutional review board. A waiver of informed consent was obtained. Our institutional liver transplantation database was retrospectively queried to identify patients who underwent liver transplantation from January 2005 to March 2008. The search yielded 175 patients. The following patients were excluded: no liver MRI or MRI with a delay longer than 90 days before liver transplantation (n = 80), interval transarterial chemoembolization (TACE) between MRI and explant (n = 20), no DWI (n = 10), poor DWI quality (n = 9), and poor quality of CET1WI (n = 4). The final cohort included 52 patients: 40 men (mean age, 56.8 years [range, 35-77 years]) and 12 women (mean age, 50.2 years [range, 44-67 years]). All patients had cirrhosis, with the following etiologies: chronic hepatitis C (n = 25), chronic hepatitis B (n = 8), autoimmune hepatitis (n = 5), primary biliary cirrhosis (n = 3), alcohol abuse (n = 1), nonalcoholic steatohepatitis (n = 1), and cryptogenic cirrhosis (n = 9). The mean interval between MRI and explant was 38 days (range, 1-89 days). A total of 24 patients received TACE prior to MRI.
MRI of the liver was performed using different state-of-the-art 1.5-T systems (Avanto, Sonata, Symphony; Siemens Healthcare, Erlangen, Germany) and torso phased-array coils. For all sequences, we used parallel imaging (factor 2) and a field of view of 300-400 mm (with an 80% rectangular field of view).
Breath-hold (n = 30) or respiratory-triggered navigator echo technique (n = 22) fat-suppressed single-shot echoplanar imaging DWI was performed in the axial plane with tridirectional diffusion gradients using three b values (50, 500, and 1,000 seconds/mm2). The other parameters were repetition time (TR) 1,800-2,300 milliseconds (breath-hold)-1 respiratory cycle (respiratory-triggered), echo time (TE) 67-82 milliseconds, matrix 144 × 192, slice thickness/gap 7/1.4 mm, number of signals averaged two (breath hold) or four (respiratory triggered), acquisition time <25 seconds for breath-hold acquisition and at least 2 minutes for respiratory-triggered acquisition. Voxel-based ADC (apparent diffusion coefficient) maps using a monoexponential fit of signal intensity were automatically generated by the scanner.
Standard MR Sequences.
Routine breath-hold sequences included coronal single-shot T2-weighted HASTE (TR/TE, 1,200/90; matrix, 192 × 256; slice thickness/gap, 7/1 mm; one average); axial fat-suppressed turbo spin echo T2WI (TR/TE, 3,570/101; matrix 192 × 256; slice thickness/gap, 8/1.6 mm; one average); two-dimensional T1 in- and out-of-phase T1WI (TR/TE, 126/4.4 [in-phase]-2.2 [out-of-phase]; flip angle, 80°; matrix, 179 × 256; slice thickness/gap, 8/2.5 mm; one average); and axial contrast-enhanced T1WI using three-dimensional (3D) fat-suppressed spoiled gradient-recalled echo sequence (VIBE) before and after dynamic injection of 0.1 mmol/kg of gadopentetate dimeglumine (Magnevist; Bayer Healthcare Pharmaceuticals, Wayne, NJ) followed by a 20-mL saline flush with a power injector, with images acquired at the arterial, portal venous, and equilibrium phases. Acquisition parameters for VIBE sequence were TR/TE, 3.3-4.5/1.4-1.9; flip angle, 12°; one average; matrix, 128 × 192 (interpolated to 256 × 256); and interpolated slice thickness, 2-3 mm.
To determine the timing for the hepatic arterial phase, a 1-mL test bolus of contrast material was administered to determine time to peak arterial enhancement.
Two observers with different experiences (J. P. and S. K., 1 year and 8 years of experience in body MRI, respectively) retrospectively and independently reviewed the MR images on a workstation (Syngo, Siemens). The observers were blinded to the initial MRI reports and pathologic results. The observers randomly analyzed MR images in three different sessions: (1) DWI (with ADC maps) plus unenhanced T1WI and T2WI sequences (DW-set); (2) CET1WI plus unenhanced T1WI and T2WI sequences (CE-set); and (3) all images together (All-set). Each of the sessions was separated by at least 3 weeks to minimize recall bias. The observers were asked to record only lesions suspected to be HCC. Detected HCCs were circled on hard copies of diagrams of liver anatomy (with Couinaud segments delineated) and were recorded with the corresponding image number, liver segment, and lesion size (measured on portal venous or equilibrium postcontrast phases or on b 50 diffusion images for those lesions seen only on DWI). A lesion was diagnosed as HCC on standard imaging sequences if the lesion fulfilled any two of the four following criteria: (1) arterial enhancement, (2) portal venous or equilibrium phase washout, (3) capsule or pseudocapsule on portal venous and/or equilibrium phase, and (4) mild to moderate hyperintensity on T2WI (when compared with surrounding liver parenchyma).6, 27 However, the readers were given the freedom to diagnose lesions with atypical features (such as continuous enhancement, arterial enhancement without wash-out) as HCC as long as these lesions did not present imaging characteristics of benign lesions such as arterioportal shunts28 or hemangiomas.29, 30 We did not attempt to diagnose regenerative or dysplastic nodules in this study. A lesion was diagnosed as HCC on DWI if it showed the following: mild to moderate hyperintensity compared with liver parenchyma on DW images at b 50, restricted diffusion (remained hyperintense) at b 500 and/or b 1,000, with ADC visually lower or equal to that of surrounding liver parenchyma.12 ADC values were not measured in this study. A maximum number of five HCCs per patient was recorded on the basis of the largest size. All data (including lesion location and size) were transcribed from hard copies to electronic format by a third observer 3 (M.-S. P., 7 years of experience in abdominal MRI), who was responsible for MR-pathologic correlation (see below).
The third observer (M.-S. P.) correlated MRI findings as diagnosed by the first two observers with the pathologic findings based on the size and segment location of the lesions on explant. All 52 explanted livers were initially sectioned into 5-8 mm contiguous slices in the coronal plane. HCCs were identified grossly as those that were distinct from surrounding regenerative nodules in terms of size, texture, color, or degree of bulging beyond the cut surface of the liver. Livers were photographed, and all lesions other than ordinary regenerative nodules were sampled for histologic examination. Using the diagnostic criteria of the International Working Party's Terminology of Nodular Hepatocellular Lesions,31 the routine hematoxylin and eosin–stained slices from the nodules were classified as follows: regenerative nodule; dysplastic nodule, low grade; dysplastic nodule, high grade; small HCC (<2 cm); or HCC (≥2 cm). The HCCs were categorized as well-differentiated, moderately differentiated, and poorly differentiated. Microvascular invasion was noted.
SAS version 9.0 was used for all statistical computations. Generalized estimating equations based on a binary logistic regression model were used to compare the three sets of images. The model included imaging modality and observer as fixed classification factors, and the correlation structure was modeled by assuming observations to be correlated only when derived for the same patient. For the analysis of diagnostic accuracy on a per-subject basis, a patient was classified as positive for HCC if at least one HCC lesion was seen at pathology and was negative for HCC otherwise. Patients were defined as test-positive for HCC whenever an observer diagnosed at least one lesion as HCC using a given imaging modality and test-negative for that given combination of reader and modality otherwise. Per-patient sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were then computed in terms of concordance between the classification of patients as positive versus negative at pathology and the classification of patients as test-positive versus test-negative for a given combination of reader and modality. In addition, sensitivity and PPV were calculated on a per-lesion basis for each observer and for both observers averaged. Interobserver agreement was assessed in terms of kappa coefficients. All reported P values are two-sided significance levels without correction for multiple comparisons and were declared statistically significant when less than 0.05.
On the explanted livers, 72 HCCs with an average size of 1.5 cm (range, 0.3-6.2 cm) were present in 33 out of 52 patients (63.4%). Thirty-three HCCs were <1 cm, 25 HCCs were 1-2 cm, and 14 HCCs were >2 cm in size. Three patients had five HCCs, three patients had four HCCs, five patients had three HCCs, eight patients had two HCCs, and 14 patients had one HCC. Tumor differentiation was as follows: 24 were well-differentiated, 36 were moderately differentiated, and 12 were poorly differentiated.
Detection of HCC With MRI
There was no significant interaction between observer and modality in terms of their impact on any aspect of per-patient diagnostic accuracy (Table 1) (P >0.2). Although the sensitivity and NPV of DW-set were lower than those of CE-set, the difference did not reach significance for either observer. However, the pooled data between both observers showed the sensitivity and NPV of CE-set to be significantly higher than those of DW-sets (P = 0.02 and 0.03, respectively) likely due to sample size. Specificity, PPV and accuracy were equivalent between datasets. The addition of DWI did not improve the diagnostic performance of CET1WI for either observer and for pooled data.
Table 1. Diagnostic Sensitivity, Specificity, PPV, NPV, and Accuracy for HCC Detection Per Patient in 52 Patients With 72 HCCs
DW-set includes unenhanced T1WI, T2WI, and DWI. CE-set includes unenhanced T1WI, T2WI, and CET1WI. All-set includes all sequences. Significant P values appear in boldface type. Observer 1 was less experienced (1 year body MRI); observer 2 was more experienced (8 years body MRI).
P (DW-set versus CE-set)
P (CE-set versus All-set)
P (DW-set versus CE-set)
P (CE-set versus All-set)
P (DW-set versus CE-set)
P (CE-set versus All-set)
There was no significant interaction between observer and modality in terms of their impact of per-lesion sensitivity and PPV (Table 2) (P = 0.28). Lesion detection was significantly higher for both observers using CE-set versus DW-set (Fig. 1). The pooled data between the two observers showed that per-lesion sensitivity of CE-set (59.0% [85/144]) was significantly higher than that of DW-set (43.8% [63/144]; P = 0.008). The addition of DWI improved sensitivity only for the more experienced observer, who was able to detect seven additional HCCs (Fig. 2). There were no differences between data sets in per-lesion PPV.
Table 2. Diagnostic Sensitivity and PPV for HCC Detection Per Lesion in 52 Patients With 72 HCCs
There was a significant difference in diagnostic sensitivity between DW-set and CE-set only for HCC lesions measuring 1-2 cm (Table 3). Both data sets were equally good at detecting large lesions (>2 cm), with sensitivity approaching 90% for DW-set and 97% for CE-set (Fig. 3). In addition, both data sets were equally poor at detecting small HCCs (size <1 cm), with sensitivity below 32%. However, the calculated confidence intervals of the difference between pooled DW-set versus CE-set showed that the sensitivity of CE-set was up to 17.7% better than DW-set for lesions <1 cm and up to 14.1% better for lesions >2 cm, which could be clinically significant, despite the lack of statistical significance, likely due to sample size.
Table 3. Diagnostic Sensitivity for HCC Detection Stratified by Lesion Size in 52 Patients with 72 HCCs
For the first observer, there were 14 false positive lesions in 14 patients on CE-set (Fig. 4) and four false positive lesions in one patient on DW-set. For the second observer, there were four false positive lesions in four patients on CE-set and one false positive lesion in one patient only on DW-set. False positives on CE-set were presumed arterioportal shunts, whereas the false positives on DW-set were all related to EPI artifacts.
There was substantial agreement for DW-set (kappa 0.64) and CE-set (kappa 0.67) and almost perfect agreement for All-set (kappa 0.88) between the two observers on a per-patient basis. There was moderate agreement for DW-set (kappa 0.477) and CE-set (kappa 0.524) and substantial agreement for All-set (kappa 0.603) between the two observers on a per-lesion basis.
In this MR-explant correlation study, we have demonstrated that DWI is outperformed by CET1WI for the detection of HCC on a per-patient and per-lesion basis, but could represent a reasonable alternative to CET1WI for the detection of large HCCs (>2 cm). The sensitivity stratified by size in our study showed that the sensitivity of DW-set for HCCs >2 cm was high (89.3%) and that of combined (All-set) images was 100%. For HCCs <1 cm, however, the sensitivity was low for both modalities. For HCCs 1-2 cm, CET1WI showed significantly higher sensitivity than DWI (74% versus 42%). We have also observed that the addition of DWI to CET1WI slightly increases the detection rate (seven additional HCCs detected by the more experienced observer).
It is well established that multiphasic dynamic gadolinium-enhanced imaging has a good to excellent diagnostic accuracy for the detection of HCC depending on lesion size, with limited sensitivity for the detection of small lesions.1-7
Several studies have assessed the role of DWI for lesion detection and characterization, including HCC.10, 12-19, 32 For example, in a prior study from our group, we demonstrated higher sensitivity of DWI compared with standard breath-hold T2WI sequence for HCC detection (80.5% versus 53.9%, respectively; P < 0.001).12 Only few studies have specifically focused on HCC detection in the cirrhotic liver, especially in comparison with contrast-enhanced imaging.20-26 Only one of these studies has correlated DWI with liver explant findings,26 and showed lower sensitivity of DWI for HCC detection, compared with CET1WI (45%-55% sensitivity for DWI, compared with 92%-100% for CET1WI, depending on the reader). The study included only a small number of cases (37 patients with 29 HCCs) and did not assess the additive value of DWI over CET1WI. Recently, the study by Piana et al25 which assessed the role of DWI versus CET1WI (using extracellular gadolinium chelates) for the detection of HCCs >1 cm in a large number of patients (91 patients and 109 HCCs) reported higher sensitivity for DWI compared with CET1WI for HCC detection, in contradiction with our results. In their study, the sensitivity of conventional MRI criteria (wash-in/wash-out) for the diagnosis of HCC was 59.6% for both radiologists, compared with 81.7%-72.5% for DWI alone (depending on the reader). Their study did not include transplant patients, and did not assess false positives. Few other studies have also generally shown additional benefit of DWI compared with CET1WI. Xu et al.20 demonstrated significantly higher sensitivity of breath-hold DWI combined with CET1WI compared with CET1WI alone for detection of HCCs ≤2 cm (in 37 patients). Nishie et al.24 evaluated the added value of DWI to super-paramagnetic iron oxide (SPIO)-enhanced MRI. The average area under the curve of the three readers for the SPIO + DWI set was significantly higher than that of the SPIO set alone (0.870 versus 0.820; P = 0.025).
The limited sensitivity of DWI for detection of HCC may be explained by the difficulty to differentiate tumors from surrounding cirrhotic liver due to similar diffusion properties and ADC values,16 or may possibly be related to tumor grade.22 Lower ADC values of cirrhotic liver have been reported compared with the normal liver, possibly due to restricted water diffusion and decreased blood flow in fibrotic liver.33 One more reason for this high false negative rate may be the limited spatial resolution, and EPI related artifacts which may obscure lesion visualization.
There are several limitations to our study. First, it was a retrospective study with a relatively small number of patients. Results should be verified in a larger prospective study, for example using liver specific agents. Second, we used different MR systems and different sequences for DWI, which may result in varying image quality and signal-to-noise ratio.34 Third, we included post-TACE patients, in whom lesion appearance and detectability could be affected by treatment. However, in our practice, the majority of patients receive TACE as a bridge for liver transplantation.
In conclusion, we have demonstrated that DWI is outperformed by CET1WI for the detection of HCC on a per-patient and per-lesion basis, but represents a reasonable alternative to CET1WI for detection of HCCs >2 cm. We have also demonstrated that the addition of DWI to CET1WI slightly increases the detection rate. Therefore, we believe that DWI should be added to routine MR protocols tailored toward HCC detection, but cannot replace contrast-enhanced MRI at this point.