Comparison of gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid-enhanced MRI and contrast-enhanced CT evaluation of liver tumors: a prospective study.

Introduction
Hepatectomy is the main treatment for various types of liver tumor as long as the background liver is noncirrhotic. For patients with recurrent hepatocellular carcinoma (HCC) or colorectal liver metastasis (CRLM) within the remnant liver, which is the most common site of recurrence [1–3], long-term survival is expected after multidisciplinary re-treatment, which may include repeat hepatectomy or chemotherapy. Especially due to the development of systemic chemotherapy, the opportunity of curative resection of initially unresectable disease or repeat local therapy including hepatetomy and ablation. As a result, time-to-interventional failure, which is a better surrogate of overall survival than recurrence-free survival [4, 5], could be prolonged [6–8]. Because curative resection plays a key role in treatment of primary and recurrent tumors, precise preoperative estimation of tumor number is essential.
Gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid-enhanced magnetic resonance imaging (Gd-EOB-MRI) detects liver tumors ≤ 15 mm in diameter with higher sensitivity than contrast-enhanced computed tomography (CE-CT) [9], and is widely used for preoperative assessment of tumor number in patients with HCC [10–13] or CRLM [14–18]. In addition, it is helpful in assessing the suitability of resection in patients with pancreatic cancer, as hepatic metastases frequently manifest only as tiny hepatic lesions and their presence precludes resection [19–21]. However, Gd-EOB-MRI may overestimate a hepatic lesion as malignant owing to its high sensitivity. Based on several studies reporting positive predictive value (PPV) of CT (92–99%) and MRI (93–99%) [14, 15, 22–24], we can only estimate that the false-positive rate of each modality is < 10%. In most previous studies, hepatic lesions were categorized as benign, probably benign, equivocal, probably malignant, and malignant [14, 16, 18, 22]. However, the details of those categorized as equivocal have not been well described.
This prospective study aimed to evaluate the outcomes of hepatic lesions detected by CE-CT and Gd-EOB-MRI, including the incidence of false-positive lesions. We also examined the characteristics of equivocal lesions.

Methods
Patients aged ≥ 20 years who were to undergo hepatectomy for any type of hepatic tumor within 8 weeks after having undergone both CE-CT and Gd-EOB-MRI were eligible for inclusion. Among patients who underwent preoperative chemotherapy, only the imaging after chemotherapy was evaluated. We excluded patients who could not undergo either CE-CT or Gd-EOB-MRI for reasons such as impaired renal function or contrast allergy. Our inclusion criteria did not specify a maximum number of lesions. However, the patients with numerous hepatic lesions that were difficult to count accurately were excluded from the analyses. All participants provided written informed consent. Enrollment started in August 2020 and ended in July 2023. Institutional review board approval was obtained (#4160).

CT protocol
CT was performed using an Aquilion Precision (Canon Medical Systems, Otawara, Japan), Aquilion 64 (Canon Medical Systems), or Revolution system (GE Healthcare, Chicago, IL, USA). Contrast was administered (600 mgI/kg) over 30 s. A bolus tracking technique was used, with a region of interest (ROI) placed in the abdominal aorta. Image acquisition was triggered when the attenuation reached 150 Hounsfield units. The early arterial phase was acquired 6 s after the trigger under breath-hold, followed by the late arterial phase 12 s later, and the portal venous phase another 12 s later. The equilibrium phase was obtained 2 min and 50 s after the trigger. Images were reconstructed using a slice thickness of 5 mm for interpretation; thin-slice images (0.5–0.62 mm thickness) were available for reference when necessary.

MRI protocol
MRI was acquired using a 3 T Ingenia Elition system (Philips Healthcare, Best, Netherlands) with a 32-channel phased-array body coil. The following sequences were obtained: T2-weighted imaging, chemical shift imaging (including in-phase and opposed-phase T1-weighted imaging), and diffusion-weighted imaging with a b-value of 1,000 mm²/s. Contrast-enhanced imaging was performed using Gd-EOB-DTPA (Primovist; Bayer Healthcare, Berlin, Germany), which was administered intravenously (0.025 mmol/kg) at a flow rate of 1.5 mL/s, followed by a saline flush. T1 high-resolution isotropic volumetric excitation scans were acquired 20, 60, and 180 s after contrast injection, corresponding to the arterial, portal venous, and transitional phases, respectively. The hepatobiliary phase images were obtained 20 min after Gd-EOB-DTPA administration and were acquired in 3 planes (axial, sagittal, and coronal). Hepatobiliary phase images were reconstructed with a slice thickness of 4 mm and a reconstruction interval of 2 mm, resulting in overlapping slices.

Evaluation of liver lesions
Each imaging study was assessed by 2 experienced board-certified radiologists. A single radiologist served as the first reader (reader A). Two different radiologists served as the second reader (reader B), and each interpreted approximately half of the cases. Hepatic lesions were scored as 1, 2, 3, 4, or 5, which corresponded to categorization as benign, probably benign, equivocal, probably malignant, or malignant, respectively. The criteria for each score are summarized in supplementary Table 1. Lesions that were scored differently by the 2 readers were scored as the higher number. Those scored as 1 or 2 were defined as benign, and those scored as 4 or 5 were defined as malignant (Fig. 1).

Operative procedure
The decision to perform hepatectomy was based on tumor burden and background liver function assessed by the indocyanine green retention rate at 15 min. Preoperative chemotherapy was administered to CLRM patients with unresectable disease and those with maximal tumor size > 5 cm or tumor number ≥ 5 (classified as H2 or H3 in the Japanese Classification of Colorectal, Appendiceal, and Anal Carcinoma, third English edition) [25]. Type of hepatectomy was selected using the Makuuchi criteria for HCC [26]. For metastatic liver tumors, nonanatomic partial hepatectomy was the first choice [27] and anatomic segmentectomy or hemihepatectomy was indicated when the tumors were adjacent to major intrahepatic vessels. Intraoperative ultrasonography was used during surgery and the lesion number corresponding to the preoperative imaging assessment was allocated to each lesion to be resected as shown in Fig. 1. After surgery, each specimen was cut by surgeons and pathologists. Every lesion was marked with the number corresponding the one allocated by preoperative images.

Standard reference
Pathologic diagnosis was set as the standard reference. Hepatectomy specimens were cut at 5-mm intervals to identify the targeted lesions. Pathologic diagnoses were rendered by experienced pathologists who subspecialized in hepatobiliary tumor evaluation. The lesions left unresected were considered benign provided that they remained unchanged for ≥ 1 year after hepatectomy. False positive rate was defined as the number of lesions finally confirmed as benign among those assessed as malignant before surgery, or 1 − PPV. Sensitivity, accurate diagnostic rate, and PPV was calculated as the follows.
Sensitivity =[the number of lesions detected with any score]/the number of lesions finally confirmed as malignant.
Accurate diagnostic rate =[the number of lesions assessed as malignant with sore 4 or 5 by the imaging]/[the number of lesions finally confirmed as malignant].
PPV =[the number of lesions with final diagnosis of malignancy]/[the number of lesions with imaging diagnosis of malignancy scored as 4 or 5].
For the calculation of sensitivity and PPV, the equivocal lesions were also included as long as they were scored as 4 or 5 by any other judgement or their final diagnosis was malignant.

Statistics
All statistical analyses were performed using JMP Pro (Version 17.0.0). SAS Institute Inc. Continuous data are expressed as medians with range and were compared using the Mann–Whitney U test. Sensitivity was compared using the McNemer test. PPV was compared using the Fisher exact test. Inter-rater reliability was assessed using the Cohen kappa coefficient. P < 0.05 was considered significant.

Results

Patient characteristics
Among the 121 patients who met inclusion criteria during the study period, 16 were excluded for the following reasons: interval between the date of imaging and hepatectomy exceeded 8 weeks (11 patients), hepatectomy with curative intent was abandoned because of advanced disease (2 patients), difficulty in identifying each targeted lesion owing to numerous tumors (2 patients), and postoperative death with unresected lesions in the remnant liver (one patient). Therefore, 105 patients were enrolled and analyzed. The patient recruitment was performed only after the referral and hepatecomy could be intended. As a result, the informed consent was obtained only after the examination of either CE-CT or Gd-EOB-MRI in 91 patients (87%). Their characteristics are summarized in Table 1. Median age was 70 years (range, 31–90) and 70 patients (67%) were men. Tumor type was CRLM in 72 patients, HCC in 29, and other (adenoma, angiosarcoma, metastases of gastrointestinal tumor, and metastases of renal cell carcinoma) in 4. There was no patient with 2 or more different types of tumors. Twenty patients (19%) had previously undergone hepatectomy for tumor before enrollment. Thirty CRLM patients and one with metastases from a gastrointestinal stromal tumor underwent preoperative chemotherapy. In these 31 patients, the median interval from the final date of chemotherapy to CE-CT and Gd-EOB-MRI was 23 days (range, -46-89) and 15 days (range, -49-95), respectively (P = 0.4535). In 9 patients, either CT or MRI was performed before chemotherapy was terminated. CE-CT was performed before Gd-EOB-MRI in 55 patients; the median interval between the 2 examinations in these patients was 10 days (range, 1–54). In the other 50 patients, Gd-EOB-MRI was performed first; the median inter-examination interval in these patients was 7 days (range, 1–42). The median interval from CE-CT to hepatectomy was 22 days (range, 1–55); the corresponding interval from Gd-EOB-MRI was 25 days (range, 2–56).

Eight hundred eighty-six lesions were recognized before surgery either by CE-CT or Gd-EOB-MRI. Among these, 298 were identified as malignant (294 based on pathologic examination and 4 based on follow-up imaging). Twenty-one lesions not recognized by either CE-CT or Gd-EOB-MRI were identified as malignant by pathologic examination (CRLM, 19; angiosarcoma, 1; gastrointestinal stromal tumor metastasis, 1). Among the 886 lesions recognized on preoperative imaging, only one had an inter-reader discrepancy in diagnosis of benign versus malignant. This lesion, a 15-mm nodule on the surface of segment 8 in a patient with renal cell carcinoma liver metastases, was detected only on Gd-EOB-MRI. It could not be identified during surgery and was counted as a false-positive by reader A, who judged it as malignant. The kappa value for the level of agreement between the 2 readers in their interpretation of CE-CT and Gd-EOB-MRI was 0.767 and 0.579, respectively (Table 2). After excluding the lesions not recognized by either of the 2 readers (CE-CT, 458 lesions; Gd-EOB-MRI, 335 lesions), the kappa value for the agreement in CE-CT and Gd-EOB-MRI was calculated as 0.841 and 0.919, respectively.

Supplementary Table 2 summarized the number of lesions detected by each imaging modality among the lesions finally proved as malignant. Long interval > 4 weeks was not associated with larger difference in the number of lesions between the two modalities than short interval < 4 weeks.
There were 26 lesions that were not recognized by any of the imaging evaluation and only identified by intraoperative ultrasonography and/or pathologic evaluation. Of them 21 lesions were malignant with 19 CRLM and the others were angiosarcoma and metastasis of gastrointestinal stromal tumor. The median size of these 21 lesions was 4 mm (range, 1–19).

Sensitivity and PPVs
Figure 2 shows the final diagnosis of hepatic lesions according to imaging diagnosis, which was classified as not recognized, benign, equivocal, or malignant. False negatives occurred mainly because of non-recognition; however, some were caused by incorrect diagnosis. The details of judgement by each reader for finally benign and malignant lesions are presented in Supplementary Table 3.
For both readers A and B, Gd-EOB-MRI was superior to CE-CT for detecting malignancy (Table 3). None of the lesions judged as benign by either reader were actually malignant. There were 16 lesions that were false positive by any of the 4 judgement (CE-CT and Gd-EOB-MRI by reader A, and CE-CT and Gd-EOB-MRI by reader B). Table 4 shows the summary of them. Eleven of them were surgically removed and four were pathologically confirmed as substantial tumor with 2 confirmed as hemangioma, one, as adenoma, and another as granuloma. All other lesions were non-substantial because they were not recognized by intraoperative ultrasonography or by histopathologic examination.

Tumor size distribution of false-negative and false-positive lesions
Among the 315 tumors finally diagnosed as malignant, the proportion of tumors incorrectly diagnosed (including non-recognition) increased as the tumor size decreased. In addition, tumor size significantly differed between correctly and incorrectly diagnosed lesions (Fig. 3). However, the size of false-positive and correctly diagnosed malignant lesions was comparable, except in the Gd-EOB-MRI evaluation by reader B (false positive vs. true positive; CE-CT by reader A, 8 mm [3–22] vs. 18 mm [2-252], P = 0.0813; CE-CT by reader B, 8 mm [8–8] vs. 16.5 mm [2-252], P = 0.2010; Gd-EOB-MRI by reader A, 9.5 mm [3–25] vs. 13 mm [1-258], P = 0.1407; Gd-EOB-MRI by reader B, 7 mm [3–25] vs. 13 mm [1-258], P = 0.0147).

Final diagnosis of equivocal lesions
One hundred forty-two lesions in 53 patients were judged as equivocal in at least one of the imaging studies by either reader. Type of liver disease in these 53 patients was CRLM in 40 patients, HCC in 10 patients, and other (renal cell carcinoma metastases, gastrointestinal tumor metastasis, angiosarcoma) in 3 patients. Proportion of HCC in the patients with equivocal lesions (10/53 [19%]) was smaller than that in those without equivocal lesions (20/52 [38%]) (P = 0.0263). The proportion of patients who received preoperative chemotherapy was significantly higher in patients with equivocal lesions than in patients with lesions categorized as other than equivocal (45% vs. 13%; P = 0.0004). Size of equivocal lesions was smaller than that of truly judged malignant lesions either by CE-CT (Reader A, 6 mm [2–47] vs. 18 mm [2-252], P < 0.0001; Reader B, 6 mm [2–47] vs. 16.5 mm [2-252], P < 0.0001) or by Gd-EOB-MRI (Reader A, 6 mm [2–23] vs. 13 mm [2-258], P < 0.0001; Reader B, 4.5 mm [2–23] vs. 13 mm [1-258], P < 0.0001). Of the 142 lesions, 3 were evaluated as equivocal on all 4 assessments with CE-CT or Gd-EOB-MRI by reader A or B), and these were all pathologically confirmed as malignant. Of the other 139, 59 were finally confirmed as benign and 80 as malignant (Table 5). The benign lesions included 7 false-positive lesions, all of which were recognized as equivocal by only one of the 4 assessments. Sixty-eight lesions (49%) were recognized as malignant by any of the other assessments; among these, 7 were confirmed as benign and 61 as malignant. Among the 53 lesions not recognized by any of the other assessments, 35 were confirmed as benign and 18 as malignant. It was quite rare for a benign lesion to be judged as malignant by the other imaging modality, and vice versa.

Additionally, 7 lesions were recognized as false-positive by both CE-CT and Gd-EOB-MRI either by reader A or B. In most cases, at least one of the imaging modalities provided the correct diagnosis.
Four of the equivocal lesions were shown to be malignant because of an increase in size after hepatectomy. The details of these lesions are summarized in Supplementary Table 4.

Discussion
The present study showed that Gd-EOB-MRI was significantly better at detecting malignant hepatic tumors than CE-CT. This superiority has been reported previously. The sensitivity of CE-CT and Gd-EOB-MRI for detecting these tumors in our study was 70% and 90%, respectively, which is comparable with previously reported values (62–87% and 82–97%, respectively) [9, 14–18, 22, 24, 28–30].
The PPV of both imaging modalities in our study was over 95%, which is also comparable with those reported in previous studies [14, 15, 22–24]. The fact that the CE-CT PPV was higher than the Gd-EOB-MRI PPV in reader B’s evaluation suggests that Gd-EOB-MRI might have a higher false-positive rate. However, the incidence of false positives was quite rare compared with false negatives. Importantly, no lesion classified as benign by MRI was later found to be malignant, highlighting MRI’s reliability in confidently ruling out malignancy. Furthermore, approximately 70% of truly benign lesions deemed equivocal on CT were not recognized or classified as benign on MRI, underscoring the potential role of MRI in reducing unnecessary follow-up imaging or invasive procedures for equivocal lesions.
The ability of imaging to detect tiny hepatic tumors remains satisfactory. Tanaka et al. reported that Gd-EOB-MRI was superior to CE-CT for detecting tumors ≤ 15 mm; however, for tumors ≤ 5 mm, Gd-EOB-MRI and CE-CT detected only 54% and 10%, respectively. [9] In our series, the true positive rates were 50% and 10%, respectively. The false-positive lesions were not exclusively small size; one-third were > 1 cm in size. The actual diagnosis of false-positive lesions is usually hemangioma, cyst, biliary adenoma, or necrotic tissue [14, 15, 17, 22, 23, 28, 31]. Sofue et al. and Asato et al. reported that vascular structures and small areas of hepatobiliary phase enhancement can mimic malignancy on MRI, potentially leading to false positives [14, 17]. In our study, 12 of 16 false-positive lesions were not a substantial tumor. Nine of these 12 lesions were false-positive only on Gd-EOB-MRI, which were detected as tumors ranging from 3 to 17 mm in diameter on hepatobiliary phase imaging. Yamada et al. reported that false-positive lesions on EOB-MRI are more likely to be smaller and located near the liver surface or vasculobiliary structures than true CRLM [24]. The non-substantial false positive lesions in our series would mostly be peripheral tributary of vasculobiliary structures, but we were unable to confirm this in our study. In some previous studies, such lesions were explained as intrahepatic vasculature, thrombosed vessels, partial-volume averaging, unexplained sub-centimeter areas on the hepatocyte-phase images, or unexplained sub-centimeter areas on hepatocyte-phase images [14, 17].
Lesions are mainly detected based on hyperintensity on diffusion-weighted imaging and hypointensity in the hepatobiliary phase, while qualitative diagnosis is made using characteristics observed on single-shot T2-weighted imaging. Since qualitative diagnosis relies on subjective criteria, we initially expected discrepancies between radiologists to arise from differences in the interpretation of lesion characteristics. However, the primary cause of disagreement in our study was differences in lesion detection itself, not differences in qualitative assessment. One possible explanation for this is that readers A and B worked in the same institution and shared a common diagnostic framework, leading to a high level of agreement in malignancy assessment. As a result, discrepancies in the classification of benign versus malignant lesions were minimal. To enhance diagnostic accuracy, systematic double reading by multiple radiologists and multimodal imaging approaches should be encouraged. Moreover, regular collaboration between radiologists and surgeons, integrating intraoperative findings and histopathological results, is crucial for reducing variability and improving lesion detection.
The incidence of detection of equivocal lesions was unexpectedly high in our study. These were mostly non-HCC lesions and had been imaged after chemotherapy administration. In addition to tiny hemangioma, dysplastic nodules and tumors with marked response to chemotherapy are encountered as equivocal in cases with HCC and CRLM, respectively. Recently, HCC without hepatitis B or C virus infection has become the majority indicated for hepatectomy. As a result, it has become rare that patients with severe cirrhotic liver in which differentiation of HCC from dysplastic nodules often matters. The assessing radiologists had been informed of the clinical course of each patient, including whether preoperative chemotherapy had been administered. Since post-chemotherapy tumor viability assessment is challenging, dynamic Gd-EOB-MRI and diffusion-weighted imaging may be useful tools, although no definitive parameters have yet been established.
In our study, lesions categorized as equivocal were typically small (median diameter 4–6 mm) and lacked definitive imaging features to confidently classify them as benign or malignant. These lesions frequently showed mild hyperintensity on diffusion-weighted imaging and hypointensity on hepatobiliary phase imaging, while their T2-weighted signal was of intermediate intensity—not as bright as that of simple cysts or hemangiomas. Enhancement patterns were also atypical, and lesion margins were often indistinct. Such ambiguous features were particularly common in patients who had received preoperative chemotherapy, suggesting that treatment-related changes may contribute to diagnostic uncertainty.
In clinical practice, reference to pre-chemotherapy Gd-EOB-MRI can often facilitate more accurate evaluation of such lesions. For instance, lesions showing typical malignant features before treatment and subsequently decreasing in size are more likely to be interpreted as post-treatment changes of malignant tumors. In contrast, our study included only Gd-EOB-MRI and CE-CT performed within 8 weeks before surgery, and earlier imaging was not considered. This design limitation may have contributed to a higher number of lesions being categorized as equivocal compared to real-world clinical settings.
When using imaging to establish a diagnosis, an equivocal diagnosis is inevitable in a certain proportion of lesions. However, as with false-positive lesions, it was infrequent for a lesion to be judged as equivocal both by CE-CT and Gd-EOB-MRI. Our results suggested that equivocal lesions by CT with the judgement of malignant by Gd-EOB-MRI were likely to be correctly malignant, while equivocal lesions by Gd-EOB-MRI were most likely unrecognized by CE-CT and were rarely judged correctly by the same reader (Supplementary Table 3).
Only 20 lesions in our study (14%) were judged as equivocal on both CE-CT and Gd-EOB-MRI. Among these, only 3 were judged as equivocal by both readers, and all were CRLM with a moderate-to-marked pathologic response to chemotherapy (proportion of remnant cancer cells, < 50%). In most cases, other radiologic judgment of benign or malignant was correct. It is notable that at least one of the other judgments was malignant in 76% of lesions confirmed as malignant, in contrast to 11% of lesions confirmed as benign (Table 5). Our findings suggest that a combination of both imaging modalities, along with careful follow-up of equivocal lesions, is essential to avoid missing malignancies that may only become evident over time, although such cases would be uncommon. The relatively higher number of equivocal lesions observed on MRI may be partially explained by the study design, in which lesion identification and scoring were performed using only post-chemotherapy imaging. In clinical practice, the presence of a lesion on pre-chemotherapy MRI that later decreases in size would strongly suggest metastasis; however, when only post-treatment images are available, small lesions without characteristic features are more likely to be categorized as equivocal. In addition, certain lesions that had shrunk and calcified after chemotherapy were more easily interpreted as benign on CT, but appeared indeterminate or were even difficult to identify on MRI, further contributing to the number of equivocal assessments.
Furthermore, while CT failed to detect some lesions over 20 mm in size, MRI successfully detected all such lesions. This indicates that MRI should avert missing clinically significant lesions. Therefore, Gd-EOB-MRI should be considered essential for preoperative evaluation of hepatic tumors. One potential reason why CE-CT failed to detect some of these larger lesions is that their detectability may be compromised by background liver conditions, such as fatty liver, which can obscure the contrast between the tumor and surrounding liver parenchyma.
This study had several limitations. Not all lesions recognized by the imaging studies were fully evaluated pathologically. All lesions with a pathologic diagnosis of malignancy were corresponded to lesions identified on preoperative imaging; however, not all benign lesions were pathologically confirmed. Our standard surgical procedure for metastatic liver tumors is nonanatomic partial hepatectomy. In this study, major hepatectomy (right or left hemihepatectomy) was performed in only 21 patients (20%). Most lesions with a preoperative diagnosis of benign were not resected. Therefore, the number of truly benign lesions especially of small size might have been overestimated considering the high incidence of non-substantial lesions among the false positive lesions in the present series. However, all patients were followed for more than 1 year after hepatectomy, and only 4 lesions were determined to be malignant based on follow-up, suggesting the feasibility of radiologic assessment of both benign and malignant tumors. Slow growing tumors that would remain stable for around one year might have caused underestimation of false negativeness, but such incidence is generally rare. High proportion of the patients with HCC should also be noticed to generalize our results globally because HCC indicated for hepatectomy is more likely solitary compared with CRLM, as a result, the necessity to assess the diagnosis of multiple lesions is less frequent. Additionally, the second reader consisted of 2 radiologists, each of whom interpreted approximately half of the cases. While the possibility of inter-reader variability cannot be entirely excluded, their comparable expertise and consistent use of standardized institutional protocols likely minimized diagnostic discrepancies. Finally, although we aimed to minimize artifacts through standard breath-hold instructions and optimized imaging protocols, respiratory motion during MRI acquisition may still have affected image quality in some cases. This is an inherent limitation of MRI and may have contributed to variability in lesion detection, particularly for small or subcapsular lesions. Moreover, lesions located in the lateral segments of the liver are more susceptible to cardiac motion artifacts, which may further obscure small lesions in these regions.
In conclusion, MRI exhibited better detection of liver lesions than CT, while the false-positive rate was low for both modalities. It was not rare to encounter equivocal lesions by imaging study, but it was quite rare that a lesion was judged as equivocal with both modalities and by the different readers. Routine use of both CE-CT and Gd-EOB-MRI and evaluation by 2 or more radiologists is highly recommended to enhance the accuracy of diagnosis of liver tumors.

Supplementary Information
Below is the link to the electronic supplementary material.