Can preoperative liver MRI with gadoxetic acid help reduce open-close laparotomies for curative intent pancreatic cancer surgery?

Objectives To evaluate gadoxetic acid-enhanced liver MRI (EOB-MRI) versus contrast-enhanced computed tomography (CECT) for preoperative detection of liver metastasis (LM) and reduction of open-close laparotomies for pancreatic ductal adenocarcinoma (PDAC). Methods Sixty-six patients with PDAC had undergone preoperative EOB-MRI and CECT. LM detection by EOB-MRI and CECT and their impact on surgical planning, open-close laparotomies were compared by clinical and radiology reports and retrospective analysis of imaging by two blinded independent readers. Histopathology or imaging follow-up was the reference standard. Statistical analysis was performed at patient and lesion levels with two-sided McNemar tests. Results EOB-MRI showed higher sensitivity versus CECT (71.7% [62.1-80.0] vs. 34% [25.0-43.8]; p = 0.009), comparable specificity (98.6%, [96.9-99.5] vs. 100%, [99.1-100], and higher AUROC (85.1%, [80.4-89.9] vs. 66.9%, [60.9-73.1]) for LM detection. An incremental 7.6% of patients were excluded from surgery with a potential reduction of up to 13.6% in futile open-close laparotomies due to LM detected on EOB-MRI only. Conclusions Preoperative EOB-MRI has superior diagnostic performance in detecting LM from PDAC. This better informs surgical eligibility with potential reduction of futile open-close laparotomies from attempted curative intent pancreatic cancer surgery. Supplementary Information The online version contains supplementary material available at 10.1186/s40644-021-00416-4.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) has a poor prognosis with an overall 5-year survival rate of approximately 8% (range 3-32% depending on the stage) [1]. Generally, surgical resection is the only potentially curative treatment for pancreatic ductal adenocarcinoma (PDAC) [2] and unlike colorectal cancer, preoperative detection of liver metastasis (LM) precludes curative surgery. Since PDAC often presents late, only approximately 20% of patients are resectable at presentation, and in many, the disease has already metastasized [3,4]. Contrast-enhanced computed tomography (CECT) is the most commonly used imaging modality for staging pancreatic cancer with its ability to detect local extension and distant metastasis [5,6]. Unexpected adverse findings, in particular LM, have been reported intraoperatively in 8-26% with potentially resectable pancreatic cancer on CECT [7][8][9], resulting in aborted surgeries [10]. Thus, it is of great relevance to detect LM on preoperative imaging to avoid futile (openclose) laparotomies and resultant associated postoperative morbidity/mortality and added hospitalization costs [11].
Gadoxetic acid (Gd-EOB-DTPA) is a hepatobiliary MRI contrast agent [Eovist (US) or Primovist (Canada, Europe), Bayer HealthCare, NJ, USA] with both extracellular and liver-specific properties. Its uptake by hepatocytes and biliary excretion enhances the contrast between the liver parenchyma and hepatic lesions, thereby improving liver lesion detection [12]. Some studies have reported the incremental value of Gadoxetic acid-enhanced liver MR (EOB-MRI) over CECT for detection of LM, particularly in colorectal cancer [13,14]. It has recently been shown that even though EOB-MRI had similar performance to extracellular-enhanced MR imaging in detecting PDAC, it had greater sensitivity in detecting LM [15]. Although a few studies have evaluated the added value of EOB-MRI over CECT in identifying LM from pancreatic cancer [10,11], specifically PDAC [16], assessment of direct impact on surgical management of PDAC is not well reported.
This study aimed to compare the diagnostic performance of EOB-MRI versus CECT in the preoperative detection of LM in patients being considered for curativeintent PDAC surgery and its impact on surgical planning and reduction of futile (open-close) laparotomies.

Patients
This was a single-institution HIPAA compliant, retrospective study with research ethics board approval and patient consent waiver. All consecutive patients with PDAC who had undergone preoperative CECT and EOB-MRI as per institutional standard of care from January 2018 to October 2019 were included. Those with incomplete or unavailable clinical, imaging or surgical data were excluded. Patient demographic characteristics (age and sex), pancreatic tumor location, tumor histopathology, carbohydrate antigen 19-9 [CA19-9], intraoperative observations, surveillance imaging and clinical outcomes were collected from the institutional electronic patient record (EPR) system. One hundred twenty-one patients with PDAC were referred to our institution for pancreatic cancer management during the study period. Seven patients were excluded due to lack of preoperative imaging and forty-eight patients because of unavailability of EOB-MRI. The final study cohort comprised of 66 patients with PDAC (37 males and 29 females with a mean age of 64.8 ± 10.4 years, range 35-85 years) who had undergone both preoperative CECT and EOB-MRI. (Fig. 1) The patient demographics and clinicopathologic characteristics are detailed in Table 1. The absolute median time intervals between CECT and EOB-MRI (12 days, IQR , range 0-119 days) and between the latest preoperative imaging and surgery (10 days, IQR [7-23.5], range 2-101 days) were less than 2 weeks. Study data were managed using REDCap (Vanderbilt University Medical Center, TN, USA) electronic data capture tools hosted at our institution [17]. The original imaging report observations as well as secondary reader interpretations of CECT and EOB-MRI were compared to the reference standard for LM diagnosis.

Imaging techniques
CECT was performed on 64-MDCT scanners (Aquilion 64, Toshiba CA, USA) using arterial and venous phases. Scanner parameters were as follows: axial slice thickness, 2 mm; interval, 2 mm; detector configuration, 64 × 0.5 (32 mm); tube voltage, 120 kV; automated tube current of 100-500 mA, helical pitch, 53; pitch factor, 0.828; tube rotation time of 0.5 s; and slice thickness for reconstructed coronal and sagittal planes, 3 mm. The craniocaudal coverage for arterial phase was from top of the liver to below the duodenum, including the entire liver, and for venous phase was from top of the diaphragm to the iliac crest. A dose of 2 ml/kg to a maximum of 150 mL non-ionic contrast agent (Ultravist 370, GE Healthcare) was administered intravenously, through a 22-gauge intravenous catheter inserted into a forearm vein, with a power injector [Medrad® Stellant® Dual Head] at a rate of 3 mL/s. Arterial and venous phases were performed on abdominal aorta at 100 HU plus 25 s and 60 s, respectively. Multiplanar reformatted images in coronal and sagittal planes were reconstructed for all phases with a slice thickness of 3 mm.

Patient and image analysis
We analyzed the formal reports of the preoperative EOB-MRI and CECT from our Radiology Information System (RIS) for the diagnosis of LM verified as per the reference standard. The number, location, size, and final radiologist diagnosis of all reported liver lesions were recorded from the reports according to a 3-point confidence scale scoring system (benign, indeterminate, and malignant). The final imaging diagnosis (malignant, benign, or indeterminate) per patient was also categorized according to the most concerning liver lesion reported on CECT and EOB-MRI. The actual impact of EOB-MRI during preoperative surgical planning was determined by the number of patients excluded from curative surgery because of LM detected incrementally on EOB-MRI over CECT as documented in clinical charts.
Additionally, two subspecialized abdominal radiologists each with 20 and 10 years of experience, independently and in blinded fashion retrospectively evaluated the anonymized EOB-MRI and CECT examinations with a gap of more than 4 weeks between the two modalities to minimize recall bias probability. All images were reviewed on the imaging workstations (Coral PACS, Version: 2.5.0.0, JDMI Informatics, University Health Network, Toronto, Canada). All visible liver lesion locations based on Couinaud segmental anatomy were recorded and measured on the axial T1-weighted postcontrast hepatobiliary phase of EOB-MRI and portal venous phase of CECT. The lesion imaging characteristics and final imaging diagnosis were again recorded, utilizing the same 3-point confidence scale scoring system for both modalities using standard diagnostic imaging criteria used in clinical practice (Additional file 1) [18,19]. The reader diagnosis for LM was verified against the reference standard.
The potential impact of EOB-MRI on preoperative surgical planning based on the detection of LM on CECT versus EOB-MRI by the two readers was compared. Readers determined simulated eligibility for curative pancreatic cancer surgery based on LM detection. The differences in patient eligibility between CECT and EOB-MRI based on LM, which would have resulted in potential futile (open-close) surgery attempts, was calculated for each reader. Inter-reader agreements for LM detection and simulated decision-making process were also evaluated.

Reference standard
Histopathology, intraoperative observations, and/or interval follow-up were considered as the reference standard for the diagnosis of LM. Suspected LM were evaluated by either pre−/intra-operative biopsy or interval follow-up imaging. In the presence of multiple suspected LM with similar imaging characteristics, a preoperative biopsy was performed only from one of the lesions, and the remainder were evaluated on follow-up imaging for change in lesion size/morphology. On follow-up imaging, a suspicious liver lesion was considered as metastasis if it showed ≥20% increase in axial diameter within a 3-month interval.

Statistical analysis
A patient-level analysis of preoperative EOB-MRI and CECT original reports to measure the diagnostic performance of the modalities in detection of LM and the actual impact of EOB-MRI on surgical planning during the prospective clinical decision-making process was performed. Additionally, patient-level and lesion-level analyses to evaluate the diagnostic performance of CECT and EOB-MRI imaging in detection of LM and the impact of EOB-MRI versus CECT on simulated surgical eligibility and differences in potential futile (open-close) laparotomies, for which McNemar tests were carried out. Imaging scores were summarized for each reader by modality (EOB-MRI, CECT) on a 3-point scale (metastasis, indeterminate, benign). Inter-observer agreement in scoring hepatic lesions based on a binary classification treating indeterminate as benign was evaluated by reporting the concordance rate (%) of diagnosis between the readers for both EOB-MRI and CECT. To calculate the sensitivity and corresponding 95% confidence intervals (CIs) of EOB-MRI and CECT in detecting LM in correlation with the reference standard, we used generalized estimating equations (GEE) to account for repeated measurements within a patient [20]. The CIs were computed under the Clopper-Pearson exact method. At the patient level, the potential impact of EOB-MRI on surgical planning was summarized for each reader.
All tests were two-sided, and a P-value < 0.05 was considered an indicator of a statistically significant difference. Statistical analyses were performed using R version 3.5.0 [21].

Results
Curative-intent pancreatic cancer surgery was excluded in 15 patients [LM in 12, and vascular issues (occlusion     The reasons for open-close laparotomy were surface LM (n = 3), intraparenchymal LM (n = 2), and unexpected common hepatic arterial tumor encasement (n = 1). (Fig. 1 (Fig. 2). Excluding these two patients from surgical exploration based on EOB-MRI would have further lowered the actual open-close laparotomy rate to 6.1% (4/66) from 9.1% (6/66). Among patients excluded from surgery due to LM (12/66), 10/12 patients with LM were identified by EOB-MRI while CECT identified only 5/12 patients. Thus, an incremental 7.6% (5/66) of patients were excluded from surgery due to LM detected only on EOB-MRI and CECT detected only 50% of the patients with LM compared to EOB-MRI. Two patients with LM were reported as indeterminate on both EOB-MRI and CECT. One of these patients had pancreatic tumor invading to portal vein confluence, and surgery was excluded on those grounds with follow-up imaging also indirectly confirming LM. In the other patient, percutaneous biopsy of an indeterminate liver lesion confirmed LM, thereby excluding surgery.
In the lesion-level analysis of retrospectively performed blinded independent readers' observations 261 liver lesions were recorded 208 benign lesions (mean size, 7 mm; range 2-59 mm) and 53 metastases (mean size, 8 Table 4

Discussion
As per the current National Comprehensive Cancer Network (NCCN) guidelines (2021), the preferred imaging tool for pancreatic imaging is dual-phase pancreatic protocol CECT, whereas MRI is mostly used as a problem-solving tool (CT-indeterminate liver lesions, suspected tumor invisible on CECT, and contraindications for CECT, such as severe allergy to iodinated intravenous contrast material) [22]. However, there is a significant incidence of unnecessary laparotomy in up to 41% of patients with CT-defined resectable PDAC [5,8,9]. A meta-analysis assessing predicting resectability in patients with PDAC by CECT revealed a summary positive predictive value (sPPV) of 81%, indicating that 19% undergo surgical exploration only [5]. Considering the burden of cost and morbidity from futile surgery and the delay in the initiation of alternative treatments like chemotherapy, it is crucial to accurately select candidates for curative-intent surgery thereby decreasing the incidence of unnecessary laparotomies. Like MRI with extracellular contrast, MRI with liver-specific contrast could miss LM on dynamic arterial and portal phases. Yet, it can provide added value due to its hepatobiliary phase (HBP), in which both the hyper-and hypovascular metastases demonstrate hypointensity and increased conspicuity due to increased lesion-to-liver contrast. Fig. 2 A 78-year-old female with otherwise resectable pancreatic cancer. A small liver metastasis (arrows) was diagnosed in segment 6 on EOB-MRI (a-d), but was invisible on CECT (e). While the lesion (arrows) shows faint rim enhancement in arterial (a) and portal (b) phases, and diffusion restriction in DWI (c), it is clearly detectable on hepatobiliary phase (d). This impacted the surgical plan (compared to CECT alone) with cancellation of curative-intent surgery and potential avoidance of a futile laparotomy Generally, LM from PDAC are hypovascular and hypointense in arterial and portal-venous phases, and HBP could improve detection, thereby impacting surgical management. It has been shown that HBP can improve the detection of colorectal cancer LM, especially in the fatty liver or lesions smaller than 1 cm [13]. Our study demonstrates the beneficial impact of preoperative EOB-MRI on surgical planning due to its high diagnostic accuracy in LM detection with a potential reduction in the incidence of aborted curative-intent pancreatic cancer surgeries. At both patient and lesion levels, this study confirms superior sensitivity of EOB-MRI in depicting LM compared to that of CECT. Further, CECT was able to detect only half of the patients with LM compared to EOB-MRI. A few previous studies have reported the superior sensitivity of EOB-MRI in the detection of LM in pancreatic cancer [10,11], specifically PDAC [16]. Motosugi et al. found that EOB-MRI was equivalent to CECT in depicting PDAC, and had a higher sensitivity for detecting LM [16]. Chew et al. reported sensitivity and specificity of 76.2 and 95.8% respectively for EOB-MRI in detecting LM in 69 patients with resectable pancreatic cancer on CECT, comparable with our mean reader results [11]. Ito et al. demonstrated that a significant predictor of LM was the identification of possible lesions on EOB-MRI, while other preoperative clinical factors, such as tumor markers and CECT findings were not predictive of LM [10]. Recently, a systematic review [23] reported that MRI was more sensitive than CT in detecting pancreatic cancer LM. This study supports our results; however, not all reviewed studies compared EOB-MRI with CECT in patients with PDAC while only two included studies performed MRI with a liverspecific agent in PDAC. Besides, this systematic review did not evaluate the impact of EOB-MRI versus CECT towards reduction of futile laparotomies. In our study, over one-tenth of patients with CT-defined resectable PDAC were incrementally detected to have LM on EOB-MRI rendering them unresectable. Mean reader results also revealed an incremental number of LM on EOB-MRI with apparent disease-free liver on CECT in a significant number of patients. Unlike CECT, EOB-MRI was able to correctly diagnose all patients with intraparenchymal LM, but similar to CECT, it failed to identify tiny liver surface metastases in a few patients leading to aborted surgery. These observations suggest that apart from small surface LM, EOB-MRI is highly reliable for detecting intraparenchymal LM in PDAC.
At a lesion level, our study showed that EOB-MRI had very high PPV and NPV for diagnosis of LM from PDAC. Likewise, at patient level, EOB-MRI had higher sensitivity and NPV than CECT for accurately identifying patients with LM. The high PPV of EOB-MRI for metastatic liver disease guides appropriate exclusion of curative-intent surgery in those patients, while the higher NPV compared to CECT can provide greater confidence towards proceeding with curative-intent surgery should there be no other contraindications. At lesion level, EOB-MRI was also significantly better than CECT in detection and characterization of small LM (≤1.0 cm). This result is also compatible with previous studies that evaluated the detection of LM on EOB-MRI [13,24].
The incremental yield of EOB-MRI in detecting LM compared to CECT can reduce futile open-close laparotomies. Practically, in this study, surgery had been planned after the inclusion of both EOB-MRI and CECT findings. Therein EOB-MRI led to a two-fold exclusion of patients from attempted curative-intent PDAC surgery versus CECT due to detection of LM on EOB-MRI only. Additionally, the simulated rate of open-close surgery was significantly lower when the surgical plan was based on the detection of LM on EOB-MRI compared to CECT alone. This difference represents the potential incremental impact of EOB-MRI on clinical decision making and can reduce unnecessary open-close laparotomies. Additionally, reader interpretations of EOB-MRI demonstrated greater diagnostic confidence for lesion characterization with a lower incidence of indeterminate liver lesion diagnoses compared to CECT. Since conclusive preoperative imaging reports are, incontrovertibly, more helpful in making surgical decisions than those with indeterminate diagnosis, EOB-MRI further also helps in reducing diagnostic uncertainty regarding the presence or absence of LM during the preoperative clinical decision-making process.
Staging laparoscopy (SL) is not routinely utilized at our institution during the preoperative staging of PDAC. Although some NCCN member institutions routinely use laparoscopy before surgery or chemoradiation to rule out undetected metastases on imaging, there is a lack of consensus on the routine use of staging laparoscopy (SL) [25,26]. Further, in most studies, the utility of Fig. 3 A 55-year-old female with pancreatic cancer and indeterminate lesion on CECT but diagnosis of liver metastasis on EOB-MRI (arrows). There is a vascular blush in segment 6 on arterial phase of CT scan (a), measuring 9 mm. On the portal phase of CT scan (b), a subtle, indeterminate low-attenuation lesion is evident. On EOB-MRI, a mildly T2 hyperintense nodule (c) demonstrates peripheral enhancement with central hypoenhancement in arterial (d) and portal (e) phases and marked hypointensity on the hepatobiliary phase (f). The patient was excluded from surgery, and the percutaneous focal liver lesion biopsy confirmed metastatic adenocarcinoma Although some of the advantages of SL include low rates of morbidity and mortality, and detection of peritoneal and liver surface metastases, it cannot adequately evaluate the posterior surface of liver or detect deeply located intraparenchymal LM with a false-negative rate of up to 9% having been reported [27]. In terms of costeffectiveness, again, most studies have analyzed the costeffectiveness of SL against the diagnostic performance of CECT and not EOB-MRI [28,29]. In contrast, a retrospective indirect cost-effectiveness analysis of SL after MRI for pancreatic head carcinoma concluded that considering the paucity of missed metastases on MRI, the cost-effectiveness of SL is poor, and its yield is marginal.
The NCCN panel suggests that SL can be considered for patients staged with resectable pancreatic cancer at increased risk for disseminated disease (markedly elevated CA 19-9, large primary tumors, large lymph nodes) and those with a borderline resectable disease before administration of neoadjuvant therapy [22]. In the future, it could be promising to assess if EOB-MRI can reduce the need for SL, consequently saving time, cost and potential morbidities. Less than one-fifth of our patients had neoadjuvant chemotherapy, which may potentially increase the incidence of liver steatosis, and therefore decrease the sensitivity of CECT to detect LM. A previous study showed the superiority of EOB-MRI to CECT in the detection of small colorectal metastasis following chemotherapy, particularly in patients with hepatic steatosis [13]. They found a positive association between metastasis size of ≤1.0 cm detected only on EOB-MRI and the presence of hepatic steatosis. Our study did not find any significant increase in the presence of steatosis on chemical sift MRI versus CECT in those who had received chemotherapy. Therefore, we do not believe that neoadjuvant chemotherapy had significant influence on the results of our study.
Dual-energy CT (DECT), a relatively new imaging technology, allows characterization of tissue composition, such as calcium, iodine, and fat, and improves detection of pathologies by using two X-ray beams of different kVp energies [30,31]. One of the stated benefits of DECT over conventional CT scan can be better characterization of small hypoattenuating liver lesions (cysts vs. LM) using iodine images. However, given limited availability and higher costs of DECT, it is not widely available for use in daily clinical practice. In the future, perhaps further studies directly comparing DECT with conventional CT scan or DECT with EOB-MRI might be valuable in preoperative surgical staging for patients with PDAC.
We must acknowledge some limitations of our study. First, this is a retrospective analysis with a relatively small patient cohort at a single site. However, all the study cohort patients underwent both EOB-MRI and CECT for direct head-to-head comparison of LM detection and impact on surgical decision-making process. In order to justify routine inclusion of EOB-MRI in all patients under consideration for curative pancreatic surgery, perhaps a prospective multi-center study with a larger cohort is required. Second, we did not specifically evaluate DWI diagnostic performance separately against the hepatobiliary phase of EOB-MRI. Our study aimed to perform a comprehensive evaluation of EOB-MRI rather than assessing portions of the MRI examination. A systematic meta-analysis assessing LM detection in patients with colorectal or non-colorectal cancer has shown lower sensitivity of DWI compared to EOB-MR imaging or the combination of both sequences in [32] with the latter having the highest sensitivity on a perlesion basis. Finally, not all suspicious liver lesions were pathologically proven. If there were more than one suspicious liver lesion, tissue sampling would be performed first from the largest or the most accessible one, and further plans would be made accordingly. Disease progression on the subsequent follow-up imaging was used as a reference standard to confirm the metastatic nature of the unbiopsied lesions.

Conclusions
EOB-MRI provides superior diagnostic performance compared to CECT in the preoperative detection of LM from PDAC. Moreover, this incremental yield of EOB-MRI better informs the surgical decision-making process and can lower the incidence of failed curative-intent pancreatic surgery. Consideration should be given to inclusion of EOB-MRI in the preoperative assessment of PDAC deemed potentially resectable on CECT.