To determine the diagnostic utility of delayed hypointensity and delayed enhancing rim on magnetic resonance imaging (MRI) as indicators of hepatocellular carcinoma (HCC) in arterially enhancing nodules ≤5 cm in the cirrhotic liver and determine the features that best predict HCC.
Materials and Methods:
Gadolinium-enhanced MRI studies performed from January 2001 to December 2004 in patients with cirrhosis were evaluated for arterially enhancing nodules measuring ≤5 cm. Verification was via explant correlation, biopsy, or imaging follow-up. Sensitivity and specificity of diagnostic features of HCC were calculated. Features predictive of HCC were determined using the Generalized Estimating Equation approach.
In all, 116 arterially enhancing nodules were identified in 80 patients (<2 cm: n = 79, 2–5 cm n = 37). Sensitivity and specificity of delayed hypointensity for HCC measuring ≤5 cm, 2–5 cm, and <2 cm were 0.54 (40 of 74) and 0.86 (36 of 42); 0.72 (23 of 32) and 0.80 (4 of 5); and 0.41 (17 of 42) and 0.87 (32 of 37). For the delayed enhancing rim sensitivity and specificity were 0.64 (47 of 74) and 0.86 (36 of 42); 0.75 (24 of 32) and 1.0 (5 of 5); and 0.55 (23 of 42) and 0.83 (31 of 37), respectively. Lesion size (≥2 cm) and delayed enhancing rim, as main features and their interaction, were the most significant predictors of HCC.
HEPATOCELLULAR CARCINOMA (HCC) is the most common primary hepatic malignancy, with an incidence in the United States of 2.4 per 100,000 per year (1). Liver transplantation can markedly increase survival in patients with HCC if criteria for early staging are met (2, 3). Current United Network of Organ Sharing (UNOS) criteria for transplantation do not require tissue confirmation of HCC by biopsy but rely on imaging characterization, particularly arterial enhancement (4). However, pretransplant image staging has been shown to be less then 50% accurate when compared to posttransplant pathological staging with imaging over- and understaging HCC (5).
Due to the scarcity of livers available it is imperative to accurately characterize focal liver lesions prior to transplantation. However, using the current UNOS criteria of arterial blush, at a single transplant center in the United States, 33% of patients transplanted for HCC did not have tumor in the explant and 63% of the misdiagnosed tumors were arterially enhancing lesions 2 cm or smaller in diameter (6). Moreover, studies have shown that up to 93% of lesions less than 2 cm seen only on the arterial phase of imaging are nonneoplastic (7–11). Thus, arterial enhancement alone is not always specific for HCC (12).
Patients with limited HCC (TNM stage II: single HCC 2–5 cm or up to 3 HCCs, each less than 3 cm) can undergo liver transplantation. Single small HCC (<2 cm) can be successfully treated with resection or ablation and are more likely to have complete response to local therapies (13).
Hypointensity on portal venous and/or delayed phases of gadolinium-enhanced imaging has been reported as a specific indicator of HCC (14–16). However, small histologically well-differentiated “early HCC” may retain some conventional liver vasculature, and consequently not demonstrate delayed hypointensity (17). Delayed enhancing rim is also a known feature of HCC (18–20). Few studies have investigated the frequency of occurrence of delayed hypointensity or delayed enhancing rim in HCC in the cirrhotic liver.
The purpose of this study was to determine the diagnostic utility of delayed hypointensity and delayed enhancing rim on MRI as indicators of HCC in arterially enhancing nodules ≤5 cm in the cirrhotic liver and to determine the features that best predict HCC.
MATERIALS AND METHODS
This retrospective study was in compliance with our Institutional Review Board (IRB) rules and the Health Insurance Portability and Accountability Act (HIPAA). IRB approval was obtained for the study and informed consent waived due to its retrospective nature.
Our hospital MRI records were searched for all adults (≥18 years) with chronic liver disease who underwent liver MRI between January 2001 and December 2004. Changes of cirrhosis were confirmed by histology in 64 patients (43 percutaneous biopsies, 21 explants). In 12 patients pathologic confirmation was not available; however, morphological changes of cirrhosis and secondary portal hypertension were seen on MRI. In the four remaining patients, biopsy demonstrated fibrosis and inflammation without frank changes of cirrhosis. As screening is recommended in patients with bridging fibrosis (18) because transition from bridging fibrosis to cirrhosis is difficult to determine clinically, these patients were included in our study.
For inclusion, the patient must have had diagnostic quality contrast-enhanced 3D T1-weighted gradient-recalled echo (GRE) imaging with the arterial-dominant phase ideally timed (enhancement of hepatic arteries, slight enhancement of portal vein, and no enhancement of hepatic veins). However, if timing was earlier with minimal or no portal vein enhancement (n = 4), the patient was still included, as an enhancing nodule would accumulate contrast sooner than surrounding parenchyma. Studies that showed enhancement of the hepatic veins on the first phase of imaging were excluded.
Patients with known HCC who underwent partial hepatectomy or radiofrequency ablation were included and evaluated for additional lesions. Those treated with transarterial chemoembolization (TACE) or radiation therapy were excluded as these treatments may change the perfusion dynamics of the liver.
A total of 394 patients with cirrhosis and available verification of MRI findings were scanned from January 2001 to December 2004. In all, 274 patients were excluded because of no visualized arterially enhancing nodule ≤5 cm, nine for suboptimal dynamic phase imaging (missed arterial phase imaging n = 4; nondiagnostic study n = 5), and 31 for prior therapy (TACE n = 22, radiotherapy n = 9). Thus, the study population comprised 80 patients (mean age 57 years, range 36–78) with 116 arterially enhancing nodules measuring ≤5 cm. Twenty patients were women (mean age 59.4 years, range 36–78 years) and 60 were men (mean age 55.3 years, range 41–74 years). Fifty-six patients had one arterially enhancing nodule, 12 had two, and 12 had three.
MR imaging was performed on a 1.5-T system (Signa; General Electric Medical Systems, Milwaukee, WI). A 4-channel phased array receiver coil was used. T1-W imaging included dual-echo opposed-phase (OP)/in-phase (IP) GRE imaging (TR/TE <180/2.2–4.4 msec, flip angle 70°, slice thickness/gap 6/0 mm, matrix 256 (frequency) × 160 (phase)). T2-W imaging was performed using either a fat-suppressed respiratory-triggered fast spin-echo (FSE) sequence (TR range/TEeff <5000/80–100 msec, echo train length 18) (n = 54 patients) or a breath hold fast recovery FSE (FR-FSE) sequence (TR range/TEeff <2500/80–100 msec, echo train length 17) (n = 26 patients), slice thickness/gap 6/0 mm, matrix 256 × 224. A rectangular field of view was often used to reduce the number of phase-encoding acquisitions.
Dynamic imaging was performed before and after intravenous injection of gadopentetate dimeglumine (Magnevist; Bayer, Wayne, NJ) using a 3D GRE sequence with spectral fat saturation (TR/TE 4–6/1.2 msec, flip angle 12°, matrix 256–320 × 128–192, section thickness 4.0 mm with 2 mm overlap). Gadopentetate dimeglumine was used at a dose of 0.1 mmol/kg (maximum dose 20 mL) administered via a power injector in an antecubital vein at a rate of 2 mL/sec, followed by 20 mL of normal saline flush at the same rate. After unenhanced imaging, contrast arrival for arterial-dominant phase imaging were timed using automated contrast bolus detection, Smart-Prep (GE Medical Systems). Venous and extracellular phase images were obtained at ≈50–70 sec and 120–200 sec, respectively, after the initiation of contrast injection.
Studies were reviewed independently by two radiologists with 4 and 10 years experience interpreting liver MRI on a commercially available workstation (Advantage Workstation; GE Medical Systems). Lesions were preselected for review by one reader who noted the image number of the lesion on the arterial phase of postgadolinium imaging. If the second reviewer noted an additional lesion then the discrepancy was resolved by consensus (n = 2). All sequences for each study were presented to the reviewers in one setting. Readers were blinded to the clinical MRI interpretations and final diagnosis.
Arterially enhancing nodules measuring ≤5 cm were identified (Fig. 1a). Comparison to precontrast fat-suppressed T1-weighted sequence was made to ensure that hyperintensity was due to contrast enhancement and subtraction imaging (arterial-phase minus precontrast) was occasionally performed for lesions (n = 17) with hyperintense signal on precontrast imaging. For patients with multiple lesions the three largest lesions within this size limit were included. Both radiologists assigned a qualitative assessment score of lesion signal intensity with respect to the background liver as hyperintense, isointense, or hypointense. Lesion signal intensity was recorded on T1-W OP/IP, T2-W, and gadolinium-enhanced sequences. Any interpretation discrepancy (n = 6) was resolved by reassessment of lesion signal intensity relative to adjacent parenchyma by consensus of the two radiologists. Other findings of rim enhancement in the venous and/or delayed phases (Fig. 1b), and internal lipid, identified by comparing in-phase and out-of-phase T1-W imaging were also documented.
Verification of the identified lesions with pathology (explant or biopsy) or follow-up MRI was essential for inclusion. Lesions that demonstrated interval growth of 100% diameter increase in 1 year or 50% in 6 months (21) or had pathological confirmation were considered HCC. For classification as nonmalignant, lesions had to have pathological proof of diagnosis or at least 18-month MRI follow-up demonstrating stability or disappearance.
Explanted livers were cut sequentially in 5–10-mm sections in the axial plane to match the MRI studies. The MRI reports were reviewed at time of gross sectioning to guide nodule localization. The overall mean time between transplant and MRI was 7.2 months (range 1–33) including 13.1 months (range 1–33) for 14 nonmalignant lesions and 4.4 months (range 1–18) for 29 malignant lesions.
Of the 116 lesions included in the study, verification was by biopsy for 28, explant correlation for 43, and MRI follow-up for 45 lesions (mean 20.4, range 4–47).
Sensitivity and specificity of delayed hypointensity and delayed enhancing rim, as features diagnostic of HCC, were calculated on a lesion-by-lesion basis. Generalized Estimating Equations (GEE) (22) models were fit with the main imaging features of HCC and their interactions to determine the predictors of HCC. GEE is a statistical method that fits parameters to a generalized linear model when unknown correlation is present. The focus of a GEE analysis is on population level average effects rather than random effects. The main features were: size 2 cm or larger and less than 2 cm, T2-weighted hyperintense signal, T1-weighted hypointense signal, lipid content, delayed postcontrast hypointensity, and delayed enhancing rim. The sampling took place at the patient level, with clusters of lesions within individuals. Leaving one variable out at a time cross-validation was used to estimate sensitivity and specificity of the resulting models.
Of the 116 nodules included in the study, 74 (64%) were HCC (transplant n = 29, biopsy n = 22, MRI follow-up n = 23) and 42 were nonmalignant.
Of the 42 nonmalignant arterially enhancing nodules, 20 were verified at pathology (explant n = 14 [12 regenerative nodules and two high-grade dysplastic nodules]; biopsy n = 6 [two regenerative nodules, two high-grade dysplastic nodules, one fibrosis, and one focal nodular hyperplasia]) and 22 by follow-up MRI for a minimum of 18 months (mean 27.5, range 19–47).
Mean size of 74 HCC nodules was 2.2 cm (range 1.0–4.8 cm) and that of nonmalignant nodules 1.5 cm (1.0–4.6 cm). 32 of 74 (43%) HCC nodules measured 2–5 cm and 42 (57%) were smaller than 2 cm. For one patient with three nodules, evaluation of unenhanced T1-W and T2-W signal intensity was not possible secondary to motion artifact. Of 113 nodules with diagnostic T2-W imaging 56 were isointense in signal relative to liver parenchyma (HCC n = 25; nonmalignant n = 31) and 57 were hyperintense (HCC n = 46; nonmalignant n = 11). Internal fat was identified by demonstrating signal loss on out-of-phase imaging compared to in-phase and was seen in 14 nodules (HCC n = 12, nonmalignant n = 2).
Lesion-based sensitivities and specificities of delayed hypointensity and delayed enhancing rim for HCC are shown in Table 1. For analysis, hypointensity on either venous or delayed phase imaging was considered as delayed hypointensity. Delayed hypointensity was highly specific (0.87) for arterially enhancing HCC <2 cm but only seen in 17 of 42 (40%) of these small tumors. The 34 of 74 HCC in our study did not demonstrate delayed hypointensity including 25/42 <2 cm and 9/32 2–5 cm HCCs (Fig. 2). The sensitivity of delayed enhancing rim for HCC was slightly higher than that of delayed hypointensity, particularly for lesions <2 cm.
Table 1. Sensitivity and Specificity of Delayed Hypointensity and Delayed Enhancing Rim as Features of Arterially Enhancing HCC ≤5 cm
Size [cm] Parameter
Delayed enhancing rim
Delayed enhancing rim
Delayed enhancing rim
Four nonmalignant nodules demonstrated hypointensity on delayed phase imaging. Two of these nodules demonstrated no change on MRI follow-up over 26 and 35 months. Two had biopsy follow-up yielding one regenerative nodule and one regenerative nodule with small cell dysplasia.
Four nonmalignant lesions had a delayed enhancing rim. Two of these nodules were not seen on follow-up MRI done at 19 and 21 months. One was determined to be a high-grade dysplastic nodule at explant correlation. One had biopsy follow-up yielding high-grade dysplastic nodule.
Two benign lesions demonstrated both hypointensity and delayed enhancing rim. One lesion had biopsy yielding dysplastic nodule. One nodule had biopsy follow-up yielding fibrosis and 28-month MRI follow-up that demonstrated minimal change (Fig. 3).
A GEE model fitted with the individual features of HCC showed that delayed enhancing rim (P = 0.026) was a significant predictor of HCC (Table 2). To identify the best individual and combinations of features predictive of HCC, the least significant features were removed from the model (Table 3). The delayed enhancing rim and its interaction with size (≥2 cm) (P = <2 × 10−16) were the best predictors of HCC (Table 3).
Table 2. P-values of All Main Imaging Features of HCC Fitted in the GEE Model to Predict HCC
P-value equal to or less than 0.05 is statistically significant. Sensitivity: 0.732 (52/71) 95% CI (0.614, 0.831); specificity: 0.628 (27/42) 95% CI (0.480, 0.785).
Table 3. P-values of Main Imaging Features and Their Interactions Fitted in the GEE Model to Predict HCC
P-value equal to or less than 0.05 is statistically significant. Cross-validated sensitivity and specificity for this model: Sensitivity: 0.803 (57/71) 95% CI (0.691, 0.888); specificity: 0.619 (26/42) 95% CI (0.456, 0.764).
HCC is thought to arise in the cirrhotic liver as a result of stepwise progression from regenerative nodule to low then high-grade dysplastic nodule, to dysplastic nodule with microscopic HCC, to small HCC. Imaging features reflect a shift in the dominant blood supply of these nodules as they progress to HCC (20). With progression to malignancy, the blood supply shifts from portal venous to abnormal hepatic arterial supply (neoplastic angiogenesis) resulting in arterial enhancement (23, 24). Thus, arterial hypervascularity of a nodule is suspicious for HCC (25–27). However, arterial enhancement, especially in nodules measuring <2 cm, is nonspecific for HCC and is seen in nonmalignant lesions (23, 28).
Since most arterially enhancing lesions ≥2 cm are highly likely to be HCC, current recommendations suggest that these lesions do not require pathological confirmation if they display imaging features typical for HCC (arterial hypervascularity and delayed hypointensity) on CT or MRI (9). Results from this study confirm that a high proportion (86%, 32/37) of arterially enhancing nodules measuring 2–5 cm are HCC.
Arterially enhancing lesions <2cm have a lower but still significant probability of being HCC. In this study, 53% (42/79) of such lesions were HCC. In the study by Marrero et al (16), 19% of arterially enhancing nodules <2 cm were HCC, and 23% (6 of 26) in the study by Jeong et al (21). At this level of incidence, more specific criteria for characterization of these nodules as HCC is needed.
Previous studies suggest that T2-W imaging does not provide additional diagnostic value compared to 3D gadolinium-enhanced imaging (29–31) for the diagnosis of HCC in the cirrhotic liver. Similar to other studies (32, 33), our results suggest that T2-W imaging has better sensitivity in larger 2–5 cm lesions (0.80, 25/31) but decreases with smaller HCC <2 cm (0.53, 21/40).
As nodules progress to HCC and develop increasing arterial blood supply, there is a simultaneous decrease in their portal venous supply with resultant decreased enhancement during venous and extracellular phases (34). Delayed hypointensity has been shown to be highly predictive of HCC. Marrero et al (16) showed sensitivity of 80% and specificity of 95% of delayed hypointensity for HCC in arterially enhancing nodules <2 cm, while Burrel et al (7) found arterial enhancement with delayed hypointensity to be 89% sensitive for HCC in 1–2 cm nodules. In our study, delayed hypointensity of arterially enhancing lesions <2 cm as an indicator for HCC was less sensitive at 41% (17/42) but demonstrated good specificity of 87% (32/37). For HCC measuring 2–5 cm, delayed hypointensity had a sensitivity of 72% (23/32) and specificity of 80% (4/5). The lower sensitivity in this study compared to previous work may be secondary to patient inclusion criteria. We included patients with biopsy and long-term MRI follow-up compared to Burrel et al (7), who included only explant patients, many of whom had known concomitant HCC possibly increasing the likelihood that visualized lesion was HCC. A study by Yu et al (35) which also included biopsy and MRI follow-up reported similar results for delayed hypointensity in HCC ≤2 cm with sensitivity of 43% and specificity of 89%.
In all, 34 of 74 HCC in our study did not demonstrate delayed hypointensity including 25/42 <2 cm and 9/32 2–5 cm HCCs. This likely relates to internal histological variations and possibly residual portal venous supply (20, 34). Also, as portal venules have been suggested as the main drainage of HCC (36), immature portal venous drainage may prevent observation of delayed hypointensity.
The presence of a tumor capsule has been associated with the peripheral enhancing rim of HCC (14, 17, 34, 35). This capsule is histologically composed of an inner layer of fibrous tissue and an outer layer of compressed vessels (17, 34). The fibrous tissue will retain contrast longer than surrounding parenchyma, resulting in a delayed enhancing rim (35). This delayed rim enhancement has been shown to be highly specific for HCC in arterially enhancing lesions (14). A transient rim of enhancement in the venous phase, referred to as the “corona of HCC,” is another possible cause for the peripheral enhancing rim. This is due to enhancement of compressed hepatic sinusoids surrounding the tumor. These sinusoids receive the portal venous drainage of HCC (36).
Comparing delayed hypointensity and delayed enhancing rim, our data suggest that the presence of a peripheral rim has slightly higher sensitivity but essentially similar specificity for the diagnosis of HCC of all sizes. Currently the AASLD emphasizes delayed hypointensity in the diagnosis of HCC, recommending that for 1–2 cm arterial hypervascular lesions, if hypointensity in the portal venous or delayed phase is confirmed on two dynamic studies, either CT scan, contrast ultrasound, or MRI with contrast, the lesion can be treated as HCC without pathological confirmation (15). However, our study shows that the delayed enhancing rim is a better predictor of HCC both as a main feature or in combination with size ≥2 cm as compared with delayed hypointensity.
Overall specificity of both delayed hypointensity and delayed enhancing rim was (0.86, 37/42) in this study. As biopsy verification of three of the lesions demonstrating hypointensity and two of the lesions with delayed enhancing rim yielded fibrosis, regenerative nodule or dysplastic nodule, sampling error could have increased the false-positive results. Also, as suggested by Yu et al (35), underlying fibrosis could also contribute to a false-positive rate since delayed hypointensity and delayed enhancing rim may be overestimated due to visual misperception caused by delayed enhancement of surrounding fibrotic parenchyma (Fig. 3a,b).
The study has several limitations. The study design is retrospective. The explanted livers were not always cut with the radiologist physically present but were always compared to the MRI report. Selection bias may have been introduced, as only lesions/patients suspicious for HCC were more likely be followed and hence be included in the study. As mentioned above, sampling error could have occurred in benign lesions verified by biopsy, leading to a false-negative diagnosis of HCC. Also, evaluation of imaging features was done qualitatively, and readers may have different thresholds for these features.
In conclusion, in arterially enhancing nodules of all sizes, delayed hypointensity improves the specificity of diagnosis of HCC but is frequently absent in very small HCC (<2 cm). Moreover, the presence of a delayed enhancing rim, irrespective of the presence of delayed hypointensity, is a strong predictor of HCC.