Efficacy comparison of multi-phase CT and hepatotropic contrast-enhanced MRI in the differential diagnosis of focal nodular hyperplasia: a prospective cohort study

Focal nodular hyperplasia (FNH) is a benign lesion, composed of hyperplastic hepatocytes separated by fibrous septa with common central scar (see Additional file 1). FNH is most probably a reactive proliferation of hepatocytes due to preexistent vascular malformation [1].

It is the next most frequent benign liver lesion following hemangioma and usually develops in unchanged liver parenchyma [2]. It is encountered in 0.3%–6% of the general population [3] but the incidence is increasing, partly due to progress in radiological imaging.

Different clinical behaviour and pathological features influence the importance of differentiating FNH from other hypervascular liver lesions such as hepatocellular adenoma (HCA), hepatocellular carcinoma (HCC), and hypervascular metastases as it is critical to ensure proper treatment.

Clinical symptoms and biochemical parameters in FNH are nonspecific. Diagnosis based on ultrasonography and core needle biopsy may not be conclusive. Furthermore, correct diagnosis of FNH in computed tomography (CT) and magnetic resonance imaging (MRI) may not be possible even in about 30% and 20% of cases respectively due to atypical radiological features [4, 5]. However, MRI with the use of organ-specific contrast agents optimizes diagnosis [6‐9]. There are two hepatotropic contrast agents available: gadoxetic acid and gadobenate dimeglumine (Gd-BOPTA). Hepatotropic contrast agents applied in MRI have a vascular-interstitial distribution during the first few minutes and are partially excreted by kidneys to urine. 3–5% of Gd-BOPTA and 50% of the gadoxetic acid is taken up by hepatocytes with normal metabolism and secreted into bile [10]. Due to contrast agent uptake by hepatic cells, it is possible to observe the enhancement of the liver parenchyma in so-called hepatobiliary phase (HBP). Enhancement persists for 1–4 h after administration of Gd-BOPTA and 20–40 min after gadoxetic acid [11]. Lesions lacking active hepatocytes are less enhancing than surrounding parenchyma or not enhancing at all. Parenchymal cells constituting FNH accumulate hepatotropic contrast agents in opposition to hepatic hemangiomas (HH) and the majority of metastases and hepatocellular carcinomas (HCC).

To compare the efficacy of two imaging modalities (CT vs. MRI) in the assessment of multiple liver lesions in different locations a conventional receiver operating characteristic (ROC) is not sufficient. However, the alternative free-response receiver operating characteristic (AFROC) methodology incorporates lesions location and confidence level of a reader [12]. AFROC curve is a plot of the fraction of correctly diagnosed lesions in their true location (lesion location fraction, LLF) and false positive fraction (FPF) [12].

The aim of this study is to compare the efficacy of multi-phase multi-detector CT and MRI with the use of hepatotropic contrast agent Gd-BOPTA in the differentiation of FNH from other focal liver lesions (FLLs) in patients with equivocal foci detected in ultrasonography.

In this prospective study, we included 207 patients with equivocal FLLs detected in ultrasonography and no contraindications to CT and MRI including administration of iodine contrast agent and Gd-BOPTA respectively. Multi-phase multi-detector CT and dynamic contrast-enhanced liver MRI with the use of hepatotropic contrast agent Gd-BOPTA were performed within 4 weeks. 43 patients did not show for either initial MRI or follow-up. Seven patients were excluded due to artifacts in MRI preventing further evaluation of lesions (Fig. 1).

157 patients who underwent both CT and MRI were included in the analysis.

In 89 patients final diagnosis was based on the histopathological examination: 21 patients were diagnosed with FNH, 35 – HCC, 4 – HCA and 22 with metastases. In the remaining 68 patients, the final diagnosis was based upon clinical and imaging follow-up revealing HH in 36 patients, FNH in 18 patients and liver metastases in 21 patients. In 15 patients diagnosed with FNH, the central scar was seen in histopathological examination. In the rest of FNH specimens areas of congestion were observed.

Table 1 lists all clinical symptoms for non-FNH and FNH group that lead to initial ultrasound examination and subsequently examined in CT and MRI.

In 38 patients liver cirrhosis was recognized on the basis of core needle biopsy, 35 of those patients were diagnosed with HCC, one with renal cancer metastasis and two with colorectal cancer metastases.

In FNH-group, slight elevation of serum GGTP level was seen in 10 patients, in all remaining cases biochemical examinations showed no significant changes, none of the patients was diagnosed with cirrhosis. In 6 patients with oncologic history, FLLs were detected at the follow-up abdominal ultrasonography. In 21 cases FNHs were detected incidentally (Table 1).

Patients characteristics in FNH group and non-FNH group are listed in Table 2. The first group included 118 patients with 335 non-FNH foci (Table 3). The second group included 39 subjects with 45 FNH foci.

Women were significantly more frequent in FNH- group (31/39) than in non-FNH group (64/118), χ² = 7.82, p = 0.0052. Statistically significant difference was also seen in patients’ age and lesion diameter: in FNH-group the mean age was 36 years (18÷56) vs 56 years (21÷79) in non-FNH group (Z = 6.97, p < 0.0001), in FNH-group the mean diameter of a focus was 37 mm (10÷85 mm) vs 29 mm (5÷80 mm) in non-FNH group (Z = −4.03, p < 0.0001).

Interobserver reproducibility and agreement for all parameters were high with Kappa values of 0.85÷1.0.

In analyzed patients, CT revealed 335 lesions and MRI – 380. MRI discovered 21 more HH, 14 more HCC and 10 more metastases. Both in CT and MRI the number of diagnosed FNH foci was equal. 6 patients had bifocal FNH.

Radiological signs (Fig. 2 and Additional file 2, Additional file 3 and Additional file 4) appearing statistically more often in FNH than in other FLLs in CT and/or MRI included:

Diagnostic efficacy of above mentioned radiological signs and other analyzed features is presented in Table 4.

Homogeneous enhancement in HAP in both CT and MRI is a feature of a good diagnostic efficacy (78% and 77% accordingly). However, low positive predictive value (PPV) (36% and 32% in CT and MRI accordingly) strongly decreases its clinical usefulness.

Feature characterized by the highest values of sensitivity (100%) and specificity (94%) in the diagnosis of FNH in MRI was the enhancement of the lesion in HBP. In all cases of FNH, the focus was isointense (42 foci) or hyperintense (3 foci) in comparison to the adjacent liver parenchyma in HBP. The same enhancement pattern in HBP as in FNH observed in case of 21 HCC foci (isointense foci) resulted in PPV of 68% for this feature. Moreover, 4 of the enhancing in HBP HCC foci presented with homogeneous enhancement in HAP and were isointense in PVP and EP. However, in all those cases patients suffered from cirrhosis.

To increase the efficacy parameter further analysis was conducted on logic sums of different radiological signs and clinical data. This analysis was performed separately for CT and MRI (Table 4).

Based on MRI study we stated that simultaneous occurrence of two radiological signs (non-hypointense enhancement in HBP and homogeneous enhancement in HAP) was a criterion enabling the diagnosis of FNH with PPV of 82%. Furthermore, after exclusion of cirrhotic patients, the efficacy of HBP reached very high values: sensitivity of 100%, specificity of 99%, PPV of 94%, the negative prognostic value of 100% and accuracy of 100% (Table 4).

In post hoc analysis the McNemar’s test with Bonferroni-Hochberg’s correction was performed to compare individual radiological signs and logic sums. Presence of central scar and the logic sum of homogeneous enhancement in HAP and presence of central scar proved to diagnose FNH better than the logic sum of enhancement pattern in HAP and PVP (p = 0.0072 and p = 0.0006 accordingly). Logic sum of enhancement pattern of HAP and PVP and exclusion of cirrhosis recognized FNH better than the logic sum of homogeneous enhancement in HAP and presence of central scar (p = 0.006). However, the logic sum of enhancement pattern of HAP and PVP and exclusion of cirrhotic patients did not perform significantly better than the presence of central scar alone (p = 0.054) or the logic sum of enhancement pattern in HAP and PVP (p = 0.0824). All the results are presented in Additional file 5. In MRI the logic sum of enhancement pattern in HAP and HBP after exclusion of cirrhotic patients determined FNH better than enhancement pattern in HBP alone (p < 0.0001), the logic sum of enhancement pattern in HAP and HBP (p < 0.0001) and the logic sum of enhancement pattern in HAP and HBP and presence of central scar (p = 0.0037). Though, the logic sum of enhancement pattern in HAP and HBP after exclusion of cirrhotic patients qualified FNH worse than enhancement pattern in HBP after exclusion of cirrhotic patients (p = 0.0148). All the results are presented in Additional file 6.

In AFROC analysis AUC for CT was significantly smaller than for MRI for each reader (Table 5). The mean AUC for CT examination was 0.934 (standard error 0.009, 95% confidence interval 0.9164÷0.951) and for MRI examination 0.941 (SE 0.008, 95% CI 0.924÷0.957), (Fig. 3). The difference between AUCs for CT and MRI reached −0.007 (SE 0.003, 95%CI -0.012÷ −0.001), was statistically significant (p = 0.0145) and in favour of MRI examination.

The confidence level of 5 and 4 (very likely and likely) was considered as a positive diagnosis of FNH. In CT the 1st reader rated 33 lesions as FNH (73%, 32 lesions ranged as very likely and single lesion as likely to be FNH), the 2nd reader – 32 lesions (71%, 30 lesions – very likely and 2 lesions – likely) and the 3rd reader – 32 lesions (71%, 31 lesions – very likely and single lesion – likely). In MRI the 1st reader rated 44 lesions as FNH (98%, 37 lesions – very likely and 7 lesions – likely), the 2nd reader – 44 lesions (98%, 32 lesions – very likely and 12 lesions – likely) and the 3rd reader – 41 lesions (91%, 34 lesions – very likely and 7 lesions – likely).

In CT there were six false-negative lesions for the 1st reader (13%), one for the 2nd reader (2%) and two for the 3rd reader (4%, mean of 6.7%). In MRI there were only two false-negative lesions for the 1st reader (4%). The 2nd and 3rd reader diagnosed all FNH lesions correctly (mean 1.5%). All the false-negative lesions were misdiagnosed as HCC.In CT there were four HCC lesions and one metastasis misinterpreted as FNH by the 2nd reader. The 3rd reader also misinterpreted four HCC lesions as FNH. There were no false-positive lesions for the 1st reader. However, only the 1st reader misdiagnosed two HCC lesions and one HCA lesion as FNH in MRI. The misdiagnosed HCA had lobulated shape and showed enhancement in HAP and HBP, therefore mimicking FNH.

In the retrospective analysis, the readers discovered 31 of 41 missed lesions (69%) in CT examinations. The missed lesions were HH (12 lesions), HCC (11) and metastases (8). The decision was based on a consensus of the three readers.

In our study, as well as in available literature, FNH occurs most often in women in the 3rd-5th decades of life [1, 3]. The majority of our cases (70%) were females and mean age was 36 years. In 13% of the patients, FNH was bifocal, and this frequency is similar to the other studies where multi-focal lesions were reported in up to 15% to 20% [3, 15, 16]. FNH is also 10 times more common than HCA [17] and in our study, we diagnosed 4 HCA and 45 FNH. In the group of patients with FNH, over half of the patients were asymptomatic, one-third of the patients complained of abdominal pain, 10% had a neoplastic disease with a primary focus localized outside the liver, none of the patients suffered from liver cirrhosis (Table 1). Biochemical parameters, except for a slight elevation of GGTP activity in 10 patients, did not show any changes. Similar serum GGTP level elevation was also observed by Cherqui et al. [4]. To summarize, the clinical picture of our study group did not differ from the literature data.

In our material, only 53% of lesions showed a central scar in CT. Some authors emphasize that presence of a central scar is typical for FNH foci larger than 3 cm [16, 18]. The relation between the diameter of the lesion and presence of a central scar is confirmed by correlating radiological and microscopic images [19, 20]. Comparable to our study, Brancatelli et al. [16] observed a central scar in CT in half of the subjects in a group of 124 FNH (mean diameter of 41 mm), less often in lesions under 40 mm. In our study mean diameter of FNH with the scar was significantly larger than of foci without visible scar both in CT and MRI. Other authors reported the presence of a central scar in 75% of the cases in HAP in MRI, similar to our study (73%). On the other hand, lack of central scar is an atypical radiological manifestation of FNH, complicating the diagnosis [19, 20]. Our results show that presence of a central scar has high diagnostic specificity, however, the low sensitivity limits its usefulness considerably.

The use of contrast agents allows showing enhancement differences between focal lesions and normal liver parenchyma [5, 6, 8, 9]. Focal lesions such as FNH, hepatocellular carcinoma (HCC) and metastases from kidney or endocrine tumours present abundant vascularization. Most metastases, mainly from the gastrointestinal tract, are typically hypovascular. Generally, evaluation of vascularization of lesions in contrast-enhanced CT and MRI is comparable. In our material, agreement rate of enhancement pattern assessment of particular foci in CT and MRI was 94%. Homogeneous enhancement in HAP was visible in 84% of FHN lesions in CT and 89% in MRI. Those findings concord with results of other authors [21, 22]. Comparable percent of FNH lesions showed enhancement similar to the adjacent liver parenchyma in PVP. However, enhancement pattern in HAP and PVP have very low PPV (32–38%) and the logic sum of the enhancement pattern in HAP and PVP does not improve PPV sufficiently (67%). Moreover, the presence of central scar proved to be a better diagnostic feature than the enhancement pattern in both HAP and PVP. Only after exclusion of cirrhotic patients, diagnostic efficacy of the enhancement pattern in HAP and PVP proved to be comparable with that of a presence of central scar. This data demonstrate a need for more effective diagnostic methods in the differentiation of FNH.

Gd-BOPTA enables detection of lesions containing active hepatocytes. Usefulness of hepato-specific contrast-enhanced MRI in recognition of liver tumours has been well described [6‐8].

According to Kim et al., FNH is typically a focus presenting slight enhancement in HBP with a hypointense central scar [21]. In our material, the majority of FNH foci (93%) was isointense with the surrounding liver parenchyma in HBP and only 3 large lesions (≥35 mm) showed a discrete increase of the signal. All FNH foci presented with the enhancement pattern typical for the unchanged hepatocyte and were never hypointense.

In comparison, 4 adenomas, often requiring differentiation with FNH, showed weaker enhancement than the liver parenchyma in HBP what made them unequivocally excluded from further differential diagnosis. A similar observation was made by Grazioli et al. [8] and this is most probably due to the presence of regressive changes in adenomas and lack of bile ducts. However, Roux et al. in the analysis of 27 HCA lesions found 44% of them enhancing in HBP [23]. Our material is too small to compare with that collected by Roux et al.

On the other hand, both in the study of Kim et al. [24] and in ours, malignant lesions (HCC) were found among isointense foci in HBP and constituted 22% of all HCC lesions. Huppertz et al. reported that some highly differentiated HCC can enhance similarly to normal hepatocytes in HBP [25].

The liver-specific phase of MRI in the diagnosis of FNH had satisfactory sensitivity and specificity, only PPV parameter is of 68% (Table 4). However, the sum of two radiological signs: non-hypointense lesion in HBP and homogeneous enhancement in HAP, established as criterion enabling the diagnosis of FNH, was characterized by a higher value of PPV (82%). HCC foci, that show isointensity in HBP and strongly enhance in HAP, are a limitation of Gd-BOPTA-enhanced MRI.

In case of good clinical and radiological cooperation in interdisciplinary teams, improved diagnostics of FLLs is possible with evaluation of additional clinical data (Table 4). Although neoplastic disease in anamnesis is twice more frequent in patients in the non-FNH group, this does not influence differential diagnostics of FNH. However, patients with liver cirrhosis were not suspected of FNH [13]. Finally, after exclusion of cirrhotic patients, presence of non-hypointense (iso- or hyperintense) foci in HBP proved to be the most accurate feature in FNH diagnosis. Parameters of diagnostic efficacy, including PPV, range from 94% to 100% (Table 4). Introduction of these diagnostic parameters gave only three false positive results (non-cirrhotic patient with a small isointense focus in HBP and final diagnosis of HCC). A non-hypointense lesion in HBP after exclusion of cirrhotic patients turned out to diagnose statistically better than other features or their logic sums beside the logic sum of homogeneous enhancement in HAP and a non-hypointense lesion in HBP. According to McNemar’s test, both features diagnosed FNH similarly. However, enhancement pattern in HBP after exclusion of cirrhotic patients showed higher efficacy parameters (Table 4).

In AFROC analysis AUC for MRI was significantly larger than for CT. The authors did not find any other study applying AFROC analysis to compare the efficacy of multi-phase CT and hepatotropic contrast-enhanced MRI in the diagnosis of FNH. Chung et al. compared the efficacy of multi-phase CT and hepatotropic contrast-enhanced MRI in the diagnosis of FFL but assessed only efficacy parameters [26]. Chung et al. concluded that MRI may provide more certain diagnosis, especially in case of FNH. Soussan et al. collected data concerning confidence level of contrast-enhanced ultrasonography in comparison to MRI in the diagnosis of FFLs but did not perform FROC or AFROC analysis [27]. Sofue et al. compared the diagnostic performance of contrast-enhanced CT and a combination of contrast-enhanced CT and hepatotropic contrasted-enhanced MRI with use of AFROC analysis [28] with the conclusion that combination of CT and MRI was more effective than CT alone.

This study has few limitations. Firstly, diffusion-weighted imaging in MRI was not included in the analysis. However, the study was designed to assess the efficacy of morphological and contrast-enhanced examinations only. Secondly, not all lesions were histologically proven. In 43% of patients, the final diagnosis was based upon clinical and radiological follow-up. This group contained all HH cases and almost half of FNH cases, where the biopsy was not possible or necessary. Thirdly, besides HCA the studied group does not include rare entities like fibrolamellar carcinoma, nodular regenerative hyperplasia or cholangiocarcinoma. Although we included incidental findings and the examined group is relatively large, we did not encounter those rare neoplasms. Finally, although we have shown that MRI has higher efficacy in the diagnosis of FNH than CT, one has to take into consideration the cost efficiency of those examinations. This issue goes beyond the scope of this paper.

In the end, it has to be emphasized, that in course of cirrhosis if a new FLL is found, it is unlikely to be FNH. In such a case, even if the lesion presents radiological features of FNH, a liver biopsy should be performed.

Assessment of dynamic contrast-enhanced MRI with hepatobiliary phase after Gd-BOPTA administration is a useful diagnostic tool. The only limitation of this method is small and/or highly differentiated foci of HCC, strongly enhancing after contrast agent administration in HAP and isointense in HBP. It can be avoided thanks to the cooperation of referring physician and radiologist and excluding patients with known liver cirrhosis. MRI examination with the administration of hepatotropic contrast agent is a more effective method in the differential diagnosis of FNH than multi-phase multi-detector CT. Hepatotropic compounds enable assessment of both vascularization of focal changes and hepatocyte function in the course of one examination (extracellular phase and liver-specific study in the one-stop-shop examination).