Hepatocellular adenomas: Accuracy of magnetic resonance imaging and liver biopsy in subtype classification

Authors


  • Potential conflict of interest: Nothing to report.

Abstract

Hepatocellular adenomas (HCAs) are divided into genotype/phenotype subgroups associated with different evolutive profiles. Therefore, recognition of subtype is of clinical importance in patient management. Magnetic resonance imaging (MRI) is considered the most informative imaging modality and liver biopsy a key diagnostic tool whose role in HCA subtyping has never been extensively studied. The purpose of our study was to evaluate the diagnostic performance of MRI and liver biopsy with and without immunohistochemistry and to assess the interobserver agreement for MR classification in a consecutive series of resected HCAs. Forty-seven HCAs with preoperative MRI and biopsy were retrospectively included. MRI data were reviewed independently by two abdominal radiologists blind to the pathological results and classification. Subtyping of HCAs on liver biopsy was made blindly to clinical, biological, and imaging data and to final classification. Routine histological analysis was based on morphological criteria and immunohistochemistry was systematically performed when enough tissue was available. Final subtyping of HCA was based on the examination of the surgical specimen. Radiologists correctly classified HCAs in 85%. The interobserver kappa correlation coefficient was 0.86. Routine histological analysis led to 76.6% of correct classification and 81.6% when immunophenotypical characteristics were available. The additional value of immunophenotypical markers is best in HCAs containing steatosis. Agreement between MRI findings and routine histological analysis was observed in 74.5%, leading to a likelihood ratio of subtype diagnosis higher than 20.Conclusion: MRI and biopsy analysis are two efficient methods in subtyping HCAs and their association increases the diagnosis confidence. Interobserver variability in MRI criteria is very low. (HEPATOLOGY 2011;)

Benign hepatocellular lesions in noncirrhotic liver can be divided into two main groups according to their pathogenesis: regenerative formations, which are mainly focal nodular hyperplasias (FNHs), and neoplastic lesions, corresponding to hepatocellular adenomas (HCAs).1 For many years the most important issue has been differentiating these two groups because patient management is different for each. Indeed, most HCAs are still resected to prevent complications such as hemorrhage and malignant transformation. On the other hand, conservative treatment is the rule for most FNHs.2, 3

More recently, HCAs have been classified as heterogeneous lesions on the basis of molecular characteristics.4, 5 It is interesting to note that distinct phenotypical features have been identified.6 Three HCA genotype/phenotype subtypes have now been described: (1) hepatocyte nuclear factor 1 (HNF1α)-mutated HCAs, mainly characterized by steatosis and negative liver fatty acid protein (LFABP) expression; (2) gp130-mutated HCAs corresponding mainly to telangiectatic/inflammatory tumors with expression of acute inflammatory markers (serum amyloid protein [SAA] and C-reactive protein [CRP]); and (3) β-catenin-mutated tumors showing cytological abnormalities and an acinar pattern.4, 7-11 There is also a small group of HCAs with no specific morphological or immunophenotypical features which is called unclassified HCA.5, 6

Recent studies have identified several risk factors for hemorrhage and malignant transformation in HCAs.5, 6, 12, 13 Besides male gender, tumor size is an important risk factor for both complications, and a cutoff of 5 cm has been proposed.12 The risk also varies significantly among HCA subtypes. Most HCAs undergoing malignant transformation present mutations of the β-catenin gene.14, 15 Yet, some telangiectatic/inflammatory HCAs, whatever the β-catenin status, may undergo malignant transformation, whereas the HNF-1α-inactivated HCA subtype is known to be associated with a lower risk of malignant transformation.12 For instance, in a large series of cases the telangiectatic/inflammatory subtype was characterized by a higher risk of hemorrhage (30%) and malignant transformation (10%) compared to steatotic HCA.12 Moreover, in a recent study focusing on HCA with malignant transformation into hepatocellular carcinoma (HCC), 56% of them were telangiectatic/inflammatory, whereas only one was steatotic LFABP-negative.16 Therefore, identifying the HCA subtype is clinically important for patient management.

Magnetic resonance imaging (MRI) is considered the most informative imaging technique for classifying these entities because findings such as fat, sinusoidal dilatation and necrotic or hemorrhagic components can be identified.17-21 Two groups have already described specific MRI patterns involving diffuse fat distribution and sinusoidal dilatation in two HCA subtypes, steatotic LFABP-negative HCAs and telangiectatic/inflammatory HCAs, respectively.20, 21 In these two series, MRI data were reviewed and a consensus was reached by radiologists with no attempt to assess interobserver agreement of HCA subtyping. Finally, liver biopsy is a key diagnostic tool in most liver tumors. Nevertheless, the role of liver biopsy in subtyping HCA has not been extensively studied, especially since surrogate diagnostic immunomarkers have been developed.

Thus, the purpose of this study was to evaluate the diagnostic accuracy of both MRI and liver biopsy (with and without immunohistochemistry) and to assess interobserver agreement for the subtype classification of a consecutive series of resected HCAs.

Abbreviations: HCA, hepatocellular adenoma; HNF1α, hepatocyte nuclear factor 1α; LFABP, liver fatty acid binding protein; MRI, magnetic resonance imaging; SAA, serum amyloid A.

Patients and Methods

Patients.

Between June 1998 and May 2008, 167 patients with HCAs were surgically treated in our institution. Among them, patients with preoperative MRI and biopsy performed in our institution were retrospectively included in the study. This study was validated by the Ethics Committee and confidentiality of results was strictly respected.

A study coordinator (who did not participate in the readings) indicated the nodule that had been biopsied on MR images in patients with multiple HCAs. Thus, the same 47 nodules were reviewed on histology and MRI.

Imaging Investigations.

All MR imaging was performed in our institution with a 1.5-T magnet (Gyroscan Intera; Philips Medical Systems, Best, the Netherlands) with a maximum gradient strength of 40 mT/m and a slew rate of 200 mT/m/msec using multiarray torso coils for signal reception. All MR acquisitions included T1-weighted chemical shift sequences performed in-phase (repetition time ms, echo time ms, 145/4.6; flip angle, 80°; section thickness, 6 mm; reconstruction matrix, 256 × 256; number of signals acquired, one) and opposed-phase (145/2.3) and respiratory-triggered T2-weighted fat-suppressed turbo spin-echo imaging (1,600/70; flip angle, 90°; field of view, 34 cm; reconstruction matrix, 512 × 512; number of sections, 24; section thickness, 8 mm; number of signals acquired, two). Parameters for 3D fat-suppressed gradient-echo T1-weighted acquisitions were as follows: 3.3-4.5, 1.4-1.9; flip angle, 12°; matrix, 128-192 interpolated to 256 × 256; rectangular field of view, 34 cm; interpolated section thickness, 2-3 mm; slab thickness, 160-200 mm to ensure full coverage of the liver; and bandwidth, 488-490 Hz/pixel. Phase encoding was performed in a sequential manner. These sequences were performed during late arterial, portal venous, and equilibrium phases (at 20, 50, and 180 seconds, respectively) after intravenous administration of a gadolinium chelate (gadoterate meglumine, Dotarem; Laboratoire Guerbet, Aulnay-sous-Bois, France) at a dose of 0.1 mmol per kg of body weight, followed by a 20-mL saline solution flush (2 mL/sec). Mean exam time was 20-25 minutes.

All MR images were read on a PACS station. Hard-copy films were scanned and converted to electronic medical images In 15 patients (between 1998 and 2003).

Image Evaluation.

All MRI data were reviewed retrospectively and independently by two abdominal radiologists (M.R. and V.V. with 6 and 25 years of experience, respectively) blind to pathological results and classification. The following criteria were used for image analysis: (1) signal intensity of images on T1- and T2-weighted sequences compared to the intensity of the surrounding liver parenchyma (on T2-weighted images, signals were considered high intensity if they were at least as intense as the spleen signal); (2) homogeneous or heterogeneous appearance of the lesion on each sequence; (3) presence of fat deposits in the lesion defined as a signal dropout on opposed-phase or fat-suppressed T1-weighted MR images compared to in-phase T1-weighted MR images (absence, focal or diffuse distribution in the lesion); (4) lesion enhancement on multiphase examination (presence and type of enhancement in the arterial-dominant phase, lack of or persistent enhancement in the portal venous and equilibrium phases); (5) presence of hemorrhagic and/or necrotic components (if a hemorrhagic or necrotic component was found in the lesion, only the nonhemorrhagic and nonnecrotic tissue of the lesion was taken into account for evaluation of signal intensity). HCA was considered steatotic (suggesting HNF-1α-mutated HCA) when diffuse and homogeneous signal dropout was observed on chemical shift sequences.18 HCA was considered telangiectatic/inflammatory when the lesion exhibited a marked high intensity signal on T2-weighted sequences, associated with delayed persistent enhancement.18 HCA was considered unclassified when the lesion did not display the MRI pattern typical of steatotic or telangiectatic/inflammatory HCAs.

Pathological Analysis.

Final diagnosis and subtyping of HCAs was based on examination of the surgical specimen. All liver resections underwent macroscopic analysis and tissue sampling of both the tumoral and nontumoral liver was performed. Histological diagnosis of HCA was defined as a tumor composed of benign hepatocytes arranged in regular plates of one or two cells thick, outlined by a preserved reticulin's framework, with numerous unpaired arteries. No portal tracts were present. The following markers were used for immunohistochemistry: SAA (Dako, 1:25 dilution), LFABP (Abcam, 1:20 dilution), β-catenin (BD Biosciences, dilution 1:200), and glutamine synthetase (Chemicon, 1:500 dilution) to improve the diagnostic accuracy of β-catenin activation. HCA subtyping into telangiectatic/inflammatory (SAA-positive), steatotic (LFABP-negative), and unclassified HCA (HCA without any specific morphological or immunophenotypical features) was performed according to previously described criteria including morphological and immunophenotypical features.4, 12 β-Catenin activation was assessed by immunohistochemistry in all HCAs whatever the presence of cell atypias and was considered activated when nuclear staining of tumoral hepatocytes was observed. When discordances were observed between morphological and immunophenotypical features, morphological features, if characteristic, were considered for subtyping.4 The nontumoral liver was systematically reviewed.

Liver Biopsy.

Four senior radiologists with more than 10 years of experience in abdominal imaging performed liver biopsy at our institution. Patient sedation (10 mg of diazepam) was administered 1-2 hours before the procedure. After administration of local anesthesia, core needle biopsy samples were obtained with an 18G needle loaded into a semiautomatic biopsy system. Up to three core samples were obtained, and the adequacy of core needle biopsy was determined on the basis of the position of the needle in the target lesion and the size and color of the specimens. Biopsy of the tumor-free portion of the liver was performed in all patients.

Subtyping of HCA on liver biopsy was performed by an experienced pathologist (V.P.) blind to the clinical, biological, and imaging data and to the histopathological classification of the surgical specimen. Biopsies were fixed in formalin, embedded in paraffin, and stained with hematoxylin-eosin, picrosirius red, and reticulin staining. Analysis of morphological criteria was referred to as “routine histological analysis.” Immunohistochemistry was systematically performed for review when enough tissue was available. Analysis of morphological criteria and immunohistochemistry was referred to as “combined histological analysis.”

Statistical Analysis.

We first summarized the clinical and morphological features of our observations: quantitative variables were determined by mean and range; binary variables were determined by percentages.

Second, we analyzed and compared the percentage of correct diagnoses made with each diagnostic procedure. The percentage of correct diagnoses was computed for each radiologist as well as their corresponding binomial confidence interval and compared using Liddel's test. Interobserver agreement was assessed using the kappa value and the discrepancies among diagnostic techniques were determined. A similar analysis was performed for histological procedures, assuming that immunohistochemistry data were missing at random, and MRI findings by the senior radiologist and histological procedures were also compared.

The diagnostic value of each procedure was assessed including the sensitivity, specificity, and likelihood ratios (LRs) of each of the HCA subtypes. The LR summarizes the sensitivity and specificity of a diagnostic test in a single value and reflects the discriminant power of the test. Specifically, the LR of a positive test is the ratio of the probability of a positive test result in a patient with and without the disease being tested, i.e., LR = sensitivity/(1-specificity). In practice, if a pretest assessment of the probability (P1) that the investigated diagnosis is correct is made, P1 can be graphically combined with the LR to give the posttest probability (P2) that the diagnosis is correct using a nomogram. Alternatively, P2 can be computed manually because multiplying the pretest odds of the disease by the LR gives the odds of the disease following a positive test: (P1/(1−P1)) × LR = P2/(1−P2).

Finally, we assessed the diagnostic value of a procedure that would require concordant MRI and histological findings to make a diagnosis.

Statistical tests were two-tailed and considered significant with a P-value of 0.05. The 95% confidence intervals (CI) were calculated. All calculations were performed using Stata 10.0 (Statacorp, College Station, TX).

Results

Clinical and Morphologic Data.

A total of 47 patients were studied, including 41 women and 6 men. Mean age at inclusion was 46 years (16 to 67 years). Twenty-one women were taking oral contraceptives (mean duration 11.6 years, 1.5 to 20 years) and data were unavailable in two women. The histological examination of the resected specimen revealed 169 HCAs (mean 3.6 tumors per patient). Twenty-one patients had one nodule, 19 patients had between two and 10 nodules, and seven patients had more than 10 nodules.

The mean size of the 47 HCA surgical specimens was 6.8 cm (1.7-16 cm). HCAs were subtyped into telangiectatic/inflammatory in 34 (72%) cases, steatotic LFABP negative in 11 (23%), and unclassified in two (4%) cases (LFABP-positive, SAA-negative, β-catenin inactivated). It should be noted that eight telangiectatic/inflammatory HCAs had additional morphological features, including steatosis (>33% in seven cases) or the presence of cell atypias associated with β-catenin activation (one case). Hemorrhagic areas were observed in six HCAs (12.7%), including three telangiectatic/inflammatory HCAs, two steatotic LFABP-negative HCAs, and one unclassified HCA. Finally, morphological and immunophenotypical features of surgical specimens were in agreement in all cases, except one HCA which had the morphological features of a telangiectatic/inflammatory subtype but was SAA-negative (case 37). Detailed data are reported in Table 1. Steatotic LFABP-negative HCAs were predominantly composed of steatotic hepatocytes (mean 74.5%, ranging from 60%-90%). Nontumoral liver examination showed steatosis in six patients (all with telangiectatic/inflammatory HCAs), iron overload in two patients, granuloma in one, and multiple microscopic HCAs in one.

Table 1. Subtyping of Hepatocellular Adenomas on Biopsy Specimen and MRI with Surgical Specimen as Reference Method
Case No.Surgical SpecimenBiopsyImaging
Routine AnalysisCombined AnalysisJuniorSenior
  1. Superimposed morphological features: [1] steatosis < 33%, [2] steatosis > 66%, [3] hemorrhage, [4] necrosis, [5] cell atypias.

  2. Tel-Infl: telangiectatic/inflammatory HCA; Un: unclassified HCA; NA: not available.

  3. Immunohistochemical phenotype: LFABP-/SAA-: steatotic, LFABP+/SAA+: telangiectatic/inflammatory, LFABP+/SAA-: unclassified, β-cat+: nuclear positivity of tumoral hepatocyte.

  4. All steatotic HCAs on surgical specimen were LFABP-negative.

1Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
2Tel-Infl [1]SteatoticLFABP+/SAA+SteatoticSteatotic
3Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
4SteatoticSteatoticLFABP-/ SAA−SteatoticSteatotic
5Tel-Infl [1]Tel-InflLFABP+/SAA+Tel-InflTel-Infl
6Tel-InflUnLFABP−/SAA−Tel-InflTel-Infl
7Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
8SteatoticSteatoticLFABP−/ SAA−SteatoticSteatotic
9Tel-InflTel-InflLFABP+/SAA+UnUn
10Tel-InflTel-InflNATel-InflTel-Infl
11Tel-InflTel-InflNAUnTel-Infl
12Tel-InflTel-InflNATel-InflTel-Infl
13Tel-InflTel-InflLFABP+/SAA−Tel-InflTel-Infl
14Tel-Infl [2]Tel-InflNATel-InflTel-Infl
15SteatoticSteatoticLFABP−/SAA−SteatoticSteatotic
16SteatoticSteatoticLFABP−/SAA−SteatoticSteatotic
17SteatoticUnLFABP−/SAA+SteatoticSteatotic
18Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
19Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
20SteatoticSteatoticLFABP −/ SAA−SteatoticSteatotic
21Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
22Tel-Infl [5]UnLFABP+/SAA+/βcat+SteatoticSteatotic
23Tel-Infl [1]Tel-InflLFABP+/SAA+Tel-InflTel-Infl
24Tel-InflTel-InflLFABP+/SAA+UnTel-Infl
25Tel-InflTel-InflNATel-InflTel-Infl
26Tel-Infl [1]Tel/InflLFABP+/SAA+Tel-InflTel-Infl
27Steatotic [3]UnLFABP−/SAA+UnUn
28Tel-InflTel-InflNATel-InflTel-Infl
29Un (SAA-)UnLFABP+/SAA+UnUn
30Tel-InflUnLFABP+/SAA+Tel-InflTel-Infl
31Un [3], [4]Tel-InflNAUnUn
32Tel-Infl [1]Tel/InflLFABP+/SAA+/ βcat+SteatoticSteatotic
33Tel-Infl [1]UnLFABP+/SAA+SteatoticSteatotic
34Tel-InflTel-InflNATel-InflTel-Infl
35Tel-InflTel-InflLFABP+/SAA+UnTel-Infl
36SteatoticUnLFABP+/SAA−SteatoticSteatotic
37Tel-Infl (SAA-)UnLFABP+/SAA−UnclassifiedTel-Infl
38SteatoticSteatoticLFABP−/SAA−SteatoticSteatotic
39Steatotic [3]SteatoticLFABP−/SAA+SteatoticSteatotic
40Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
41SteatoticUnLFABP+/SAA−SteatoticSteatotic
42Tel-InflTel-InflNATel-InflTel-Infl
43Tel-InflTel-InflLFABP+/SAA+Tel-InflTel-Infl
44Tel-Infl [3]Tel-InflLFABP+/SAA+Tel-InflTel-Infl
45Tel-InflTel-InflLFABP+/SAA−Tel-InflTel-Infl
46Tel-Infl [3], [4]Tel-InflLFABP+/SAA+Tel-InflTel-Infl
47Tel-Infl [3], [4]Tel-InflLFABP+/SAA+/ βcat+UnUn

MRI Findings.

Junior and senior radiologists correctly classified HCAs on MRI in the subgroups in 76.6% (CI: 61%-88%) and 85.1% of the cases (CI: 71%-94%), respectively. Detailed results are summarized in Table 1. The two readers agreed on the classification in 43 out of 47 lesions (91.5%, CI: 79%-98%). In the four remaining cases (numbers 11, 24, 35, 37), the junior radiologist responded “unclassified” in four cases, whereas the senior radiologist responded telangiectatic/inflammatory. All lesions corresponded to telangiectatic/inflammatory HCA on histological analysis of the surgical specimen. The interobserver kappa correlation coefficient was found to be 0.85 (CI: 0.69-0.97). Tumor size was not statistically different between correctly and incorrectly HCAs classified by MRI (6.3 cm versus 6.8 cm, P = 0.71).

Histological Findings.

The mean length of the biopsy was 20.9 mm (6-50 mm). HCA subtyping based only on elementary histological features led to a correct classification in 76.6% (CI: 61%-88%). In 38 cases (81%) in which immunophenotypical features were available, subtyping was correct in 81.6% (CI: 65%-93%) (Table 1). Mean length of biopsy was not statistically different between correctly and incorrectly HCAs classified by pathology (20.6 mm versus 21 mm). Representative cases are illustrated in Fig. 1. There were nine cases (24%; cases 2, 13, 17, 22, 27, 29, 30, 33, 45) showing disagreement between routine histological analysis and combined histological analysis. Immunohistochemistry was in agreement with the final diagnosis of surgical specimens in six cases (67%) by correctly reclassifying five HCAs that were initially considered unclassified on routine histological analysis into three telangiectatic/inflammatory and two steatotic LFABP-negative. Data are provided in Table 1. Subtyping of hemorrhagic HCAs was possible in all cases.

Figure 1.

Biopsy of HCAs. Steatotic LFABP-negative HCA. (A) Hematoxylin and eosin (H&E) staining showing hepatocellular proliferation composed mainly of steatotic hepatocytes with unpaired arteries included (magnification ×25). (B) LFABP immunostaining showing negative HCA in contrast to the positive reaction in the adjacent nontumoral liver (magnification ×100). Telangiectatic/inflammatory HCA. (C) H&E staining showing an hepatocellular adenoma characterized by significant sinusoidal dilatation and a cluster of small arteries surrounded by inflammation (magnification ×25). (D) SAA immunostaining is observed in all tumoral hepatocytes with strong positivity (magnification ×200). Telangiectatic/inflammatory HCA with steatosis. (E) On H&E, steatotic hepatocytes are predominant in hepatocellular proliferation. Note the presence of cluster of arteries and focal sinusoidal dilatation (black arrow, magnification ×200). (F) SAA immunostaining shows heterogeneous cytoplasmic positivity of tumoral hepatocytes (magnification ×200).

Results by HCA Subtype.

The diagnostic results of the two radiologists, routine and combined histological analysis by HCA subtype are set out in Table 2. Representative cases are illustrated in Figs. 2-4. The distribution of patients having MR features suggestive of either steatotic LFABP-negative HCAs or telangiectatic/inflammatory HCAs are shown in Figs. 5 and 6. In addition, there were no statistical differences in correct subtyping between HCA <5 cm and those >5 cm (imaging: 80% versus 87.5% P = 0.45, routine biopsy: 66.7% versus 81.3%, P = 0.27, and combined biopsy analysis 75% versus 84.6%, (P = 0.48).

Figure 2.

MRI of a steatotic LFABP-negative HCA (white arrow) in segment IVb (same patient as Fig. 1A,B). In-phase (A) and opposed-phase (B) T1-weighted gradient-echo sequence showing a strong and homogenous dropout of signal intensity due to marked fat content. (C) On respiratory-triggered T2-weighted fat-suppressed turbo spin-echo imaging, the signal of the lesion is low intensity compared to the surrounding liver. Three-dimensional fat-suppressed gradient-echo T1-weighted acquisitions after intravenous injection of gadolinium (D-F) chelate: the signal of the lesion is spontaneously low intensity and shows slight enhancement on arterial phase imaging (D). On portal venous (E) and delayed phases (F) there is a low intensity signal.

Figure 3.

MRI of a telangiectatic/inflammatory HCA (white arrow) (same patient as Fig. 1C,D). In-phase (A) and opposed-phase (B) T1-weighted gradient-echo sequence showing relative high intensity signal on opposed-phase due to drop in signal intensity of the surrounding liver due to steatosis. On respiratory-triggered T2-weighted fat-suppressed turbo spin-echo imaging (C) a highly intense signal is present. Three-dimensional fat-suppressed gradient-echo T1-weighted sequence after intravenous injection of gadolinium chelates (D-F) the lesion was spontaneously high intensity with strong enhancement on arterial phase (D) and persistent enhancement on delayed phase (E,F).

Figure 4.

MRI of a telangiectatic/inflammatory HCA (white arrow) containing high fat content and misclassified as a steatotic lesion in segment VII (same patient as Fig. 1E,F). In-phase (A) and opposed-phase (B) T1-weighted gradient-echo sequence showing a strong and heterogeneous intensity drop out in the lesion related to fat content. On respiratory-triggered T2-weighted fat-suppressed turbo spin-echo imaging (C) the lesion is iso-intense compared to the surrounding liver. Three-dimensional fat-suppressed gradient-echo T1-weighted acquisitions after intravenous injection of gadolinium chelates (D-F). The lesion is spontaneously iso-intense and shows mild enhancement on arterial phase imaging (D). There was no persistent enhancement on portal and delayed phase (E,F), the lesion remains slightly hypo-intense.

Figure 5.

Distribution of patients having MR features suggestive of steatotic LFABP-negative HCAs. Of the 11 steatotic LFABP-negative HCAs, one did not show a diffuse and intense signal drop out on opposed-phase MR images and is not represented in the figure. There were four false-positive cases. All of them were telangiectatic/inflammatory HCAs.

Figure 6.

Distribution of patients having MR features suggestive of telangiectatic/inflammatory HCAs. Of the 34 telangiectatic/inflammatory HCAs, six did not show marked high intensity signal on T2-weighted sequences, associated with delayed persistent enhancement on MR images and are not represented in the figure. There were no false-positive cases, indicating that MR features are were specific for telangiectatic/inflammatory HCAs.

Table 2. Diagnostic Performances of Junior and Senior Radiologists on MRI, Routine Histological Analysis, and Combined Histological Analysis of Biopsy in HCA Subtyping Stratified by HCA Subtype
 Unclassified HCAs (n = 2)Steatotic LFABP-Negative HCAs (n = 11)Telangiectatic/Inflammatory HCAs (n = 34)
Se (%) CISp (%) CILR CISe (%) CISp (%) CILR CISe (%) CISp (%) CILR CI
  1. Se, sensitivity; Sp, specificity; LR, likelihood ratio; CI, confidence interval.

MR imaging Junior10015.8-10084.470.5-93.56.433.25-12.790.958.7-99.888.973.9-96.98.183.19-2170.652.5-84.910075.3-10019.61.28-301
MR imaging Senior10015.8-10093.381.7-98.6155.03-44.890.958.7-99.888.973.9-96.98.183.19-2182.465.5-93.210075.3-10022.81.49-348
Routine histological analysis501.26-98.78065.4-90.42.5.56-11.263.630.8-89.197.285.5-99.922.93.15-16782.465.5-93.292.364-99.810.71.62-70.8
Combined histological analysis00-97.583.868-93.81.46.12-17.881.848.2-97.710087.2-10044.32.8-70284.665.1-95.691.761.5-99.810.21.54-66.8

Concordance/Discordance Analysis.

Agreement between the evaluation of MRI findings by the senior radiologist and routine histological analysis was 74.5% (CI: 59%-87%) corresponding to a kappa value of 0.54 (CI: 0.34-0.74). There was disagreement in 12 cases (cases 6, 9, 17, 22, 30, 31, 32, 33, 36, 37, 41, 47). Correct classification was obtained at MRI and biopsy in seven and three cases, respectively. In 2/12 cases (17%) the lesions were misclassified by both.

Agreement between assessment of MRI findings by the senior radiologist and the combined histological analysis was 63.2% (CI: 45%-79%) (kappa = 0.36, CI: 0.13-0.60). There was disagreement in 14 cases (cases 2, 6, 9, 13, 22, 27, 29, 32, 33, 36, 37, 41, 45, 47). Correct classification was obtained at MRI and immunohistochemistry in seven and seven cases, respectively.

Diagnostic values and likelihood ratios when MRI and routine histological analysis agreed were further assessed according to HCA subtypes. MRI and routine histological analysis were in agreement and correct in classifying 25 of the 34 cases of telangiectatic/inflammatory HCA (sensitivity 73.5%; CI: 55%-88%). None of the HCAs were incorrectly classified as telangiectatic/inflammatory (specificity 100%; CI: 75%-100%). When there was agreement for a diagnosis of telangiectatic/inflammatory subtype, the LR was 20.4 (CI: 1.3-313).

MRI and routine histological analysis were in agreement and correct in classifying 7 of the 11 cases of steatotic LFABP-negative HCAs (sensitivity 63.6%; CI: 30%-90%). Only one HCA was incorrectly labeled steatotic (specificity 97,2%; CI: 85%-100%). When they agreed on the diagnosis of steatotic, the LR was 22.9 (CI: 3.1-167).

MRI and routine histological analysis were in agreement and correct in classifying one of the two cases of unclassified HCAs (sensitivity 50%; CI: 1.2%-99%). Only one HCA was incorrectly labeled as unclassified (specificity 97.8%; CI: 88%-100%). When they agreed on the unclassified diagnosis, the likelihood ratio was 22.5 (CI: 2.0%-244%).

Discussion

For the first time, our study evaluates the diagnostic value of MRI as well as routine and combined histological analysis including immunohistochemistry of biopsy samples in a large single-center series of surgically resected HCAs. The results show the high diagnostic value of these methods. Our population included 47 patients with HCAs, most of them being telangiectatic/inflammatory, and very few were unclassified. Percentages of the different subtypes were close to those previously reported in the literature.5, 6, 12

This is the first independent series to validate previously published MRI criteria.20, 21 Our study confirms the high sensitivity and specificity of MRI findings for the diagnosis of telangiectatic/inflammatory and steatotic LFABP-negative HCAs. Although Laumonier et al.20 had a specificity of 100% in steatotic LFABP-negative HCAs, ours was 88.9%. The difference may be due to the presence in our series of telangiectatic/inflammatory lesions containing significant steatosis leading to marked dropout on opposed-phase T1-weighted sequences. Unlike Laumonier et al.,20 who only reported focal signal dropout in 11% (n = 3) of telangiectatic/inflammatory HCAs on the chemical shift sequence, we report diffuse dropout in 21% (7/34 cases) of telangiectatic/inflammatory HCAs. However, in retrospect, the dropout was diffuse but slightly heterogeneous in the false-positive cases. Importantly, the specificity of telangiectatic/inflammatory subtyping in the present study was 100% compared to 87.5% reported by Laumonier et al.20

Lesions without MRI features suggesting steatotic or telangiectatic/inflammatory HCAs were labeled unclassified with a specificity of 93.3%. Although the presence of washout has been associated with β-catenin activation, our study does not confirm the specificity of this sign.20 We observed three HCAs with washout on MRI: two were unclassified without β-catenin activation and one was telangiectatic/inflammatory with β-catenin activation. Moreover, two other HCAs with β-catenin activation were telangiectatic/inflammatory and did not show any washout. Interobserver variability in HCA subtyping is another important key issue, and has not been previously evaluated. In our study, the kappa value was excellent (>0.80), indicating that MRI criteria are accurate and robust. Results obtained from readings performed by two radiologists with different levels of expertise were similar in all except four cases, emphasizing that these criteria are easy to learn and could be made generally available.

We evaluated the diagnostic performance of percutaneous biopsy of HCA with routine histological analysis as well as the additional diagnostic value of immunophenotypical markers. Indeed, whereas all centers usually perform morphological analysis, immunohistochemistry is not part of daily practice and is usually found in specialized centers. Correct diagnosis of HCA subtyping was obtained with routine and combined histological analysis in 76.6% and 81.6% of cases, respectively. The slight improvement in subtyping performance between routine and combined pathological analysis should be tuned down because the analysis was performed by a pathologist with experience in liver tumors. However, one can expect significant input of immunohistochemistry in the HCA subtyping on biopsy to be much higher for general pathologists. It is interesting to note that immunohistochemistry provided more information in steatotic LFABP-negative HCAs (sensitivity 81.8% versus 63.6%) than in telangiectatic/inflammatory HCAs (sensitivity 84.6% versus 82.4%). This increase in sensitivity may be explained, as previously observed, by the degree of steatosis, which may vary in LFABP-negative HCAs.5 An increase in specificity was also found, as one telangiectatic/inflammatory HCA was misclassified as steatotic on routine histological analysis (case 2) due to the presence of a marked steatosis in telangiectatic/inflammatory subtype, as previously reported.10 Thus, the specificity of combined analysis on biopsy was 100% in steatotic LFABP-negative HCA, with an LR of 44.3. These results strongly support the importance of immunophenotypical markers in the diagnosis of HCA with steatosis. This has clinical value because steatotic LFABP-negative HCAs have the most benign course, allowing more conservative management in these cases.12 In addition, β-catenin activation, using both β-catenin and glutamine synthetase markers, has to be screened on biopsy given that β-catenin-activated HCA display the highest risk for malignant transformation.14, 15

Immunohistochemistry was not available in 19% of cases due to insufficient histological material. This drawback is mainly because the study was retrospective and would probably have occurred less in prospective studies. To note, we only performed a single reading of biopsies because immunophenotypical subtyping obtained from immunohistochemistry is less related to observer's subjectivity and included internal controls.

MRI and routine histological analysis were in agreement in 74.5% of cases. In these cases, the LR was very high (>20) whatever the different HCA subtypes, allowing a very confident diagnosis. We also analyzed discordant cases between MRI and routine histological analysis. In nearly 60% of these cases the correct diagnosis was obtained with MRI.

In conclusion, MRI and biopsy are two accurate methods for subtyping HCA. The diagnostic value is increased when these methods are associated. Interobserver variability is very low for MRI criteria. Finally, immunohistochemistry increases the accuracy of the biopsy, especially in the subtyping of HCAs containing steatosis and showing β-catenin activation.

Ancillary