Improving preoperative detection of synchronous liver metastases in pancreatic cancer with combined contrast-enhanced and diffusion-weighted MRI

All patients older than 18 years with potentially resectable pancreatic cancer without liver metastases on CECT and additional Gd-enhanced MRI with DWI performed in our hospital from January 2012 to December 2016 were eligible for inclusion. Patients were recruited from the Radiology Information System. MRI was routinely performed in our center in all patients with potentially resectable disease or indeterminate liver lesions on CECT. Patients with locally resectable or borderline resectable pancreatic cancer were included. Resectability was established using criteria of the Dutch Pancreatic Cancer Group (PREOPANC trial, DPCG 2012). Exclusion criteria were local or systemic treatment for pancreatic cancer prior to imaging, locally advanced pancreatic cancer on CECT, incomplete liver imaging, and a time interval between CT and MRI or imaging and surgery of more than 2 months. The primary outcome parameters were the presence of liver metastases on Gd-enhanced MRI with DWI and the sensitivity of Gd-enhanced MRI with DWI for synchronous liver metastases. The secondary endpoint was the number of lesions suspicious for metastases detected by the different MRI sequences. Confirmation of liver metastases was obtained by histopathology, ¹⁸FDG-PET, and surgical findings. Explorative surgery was performed in all patients with (borderline) resectable tumors without histopathological proof or ¹⁸FDG-PET confirmation of metastases. Demographic characteristics were collected from the electronic medical records. Survival rates were obtained from the general practitioners in October 2015 and were updated in January 2018 from data in the electronic medical records.

CECT was performed in different hospitals and produced at different models of 16- and 64-row multidetector CT scanners. Only high-quality datasets with image acquisition in the portal-venous phase and slice thickness of 3–5 mm were included for analysis.

All MR imaging of the abdomen was performed in our academic tertiary referral center on a 3.0 Tesla system (Magnetom Skyra, Siemens Healthcare, Erlangen, Germany). The imaging protocol is displayed in Table 1. The protocol consisted of a T1-weighted axial in- and opposed phase gradient-echo VIBE, a half Fourier acquisition single-shot turbo spin-echo (HASTE), pre- and post-contrast T1-weighted 3D gradient-echo VIBE, and a respiratory triggered single-shot spin-echo echoplanar DWI in the transverse plane with monopolar diffusion gradients along three orthogonal directions with b-values of 0/50, 500, and 800 s/mm², using δ = 10.1 ms and Δ = 33.5 ms. Fifteen ml of gadoterate meglumine 0.5 mmol/mL (Dotarem, Guerbet, Villepinte, France) was injected in an antecubital vein at 2.5 ml/s with a saline flush (NaCl 0.9%) of 20 ml at 2.5 ml/s using a pump injector (Optistar Elite, Mallinckrodt, Dublin, Ireland). MR cholangiopancreatographic images were also obtained; these images were not used in this study.

MR images were consecutively reviewed by a radiologist (JH) with 14 years of experience in abdominal and pancreas imaging, on a commercial PACS workstation (Impax, Agfa Healthcare, Belgium). The observer was blinded to all clinical information, pathology reports, and the original radiology report, aside from the diagnosis of pancreatic cancer. In both modalities, liver lesions were scored using a 3-point scale: 1-benign, 2-indeterminate, 3-malignant (i.e., metastasis). Number, size, location, and imaging characteristics and the presumed diagnosis of the lesion were noted. Benign lesions were diagnosed using established imaging criteria [25‐27]. On CECT, hypodense lesions that show typical features of a simple cyst (fluid attenuation measurements, round-oval, well-defined borders, no contrast enhancement), a hemangioma (localization next to vessels, peripheral nodular enhancement, centripetal fill-in), or focal fatty infiltration (geographic hypodense area, angular margins, typical location) are classified as benign lesions. Indeterminate liver lesions on CECT included hypodense liver lesions that were too small to be characterized. Metastases are hypodense lesions with rim enhancement. On MRI, metastases of pancreatic cancer are typically of moderately high to isointense signal intensity on T2W-images and mildly hypointense to isointense on T1W-images. Metastases can either be hypo- or hypervascular, and show homogeneous or peripheral enhancement (ring or wedge-shaped) in the arterial phase, homogeneous enhancement or peripheral enhancement with complete or incomplete centripetal progression in the portal-venous and interstitial phase [28]. On DWI, a lesion was classified as malignant (i.e., metastasis) when it was (moderately) hyperintense at b = 0/50 s/mm² and remained hyperintense at the highest b = 800 s/mm² and a lesion was considered benign when it was hyperintense at b = 0/50 s/mm² and showed a substantial decrease in signal intensity at higher b-values (b = 500 and 800 s/mm²). If none of the criteria were met, a lesion was classified as indeterminate. For the analysis, indeterminate lesions were classified as benign, as in clinical practice indeterminate lesions that cannot be further classified will be regarded as benign unless proven otherwise by biopsy. Whenever more than ten malignant lesions (i.e., metastasis) per slice were present, the number of malignant lesions per slice was estimated in dozens.

All data were processed using SPSS software (version 20, SPSS, Chicago, IL). The sensitivity of Gd-enhanced MRI with DWI was calculated with a 95% confidence interval (CI). ANOVA test was performed to determine the differences between the group with liver metastases and the group without liver metastases. Paired samples t test was used to determine the difference between contrast-enhanced MRI and DWI regarding detection of malignant lesions. The differences between various MRI sequences regarding lesion detection were compared using the Friedman test. Post hoc analysis with Wilcoxon signed-rank tests was conducted with a Bonferroni correction applied. Survival analysis was performed using Kaplan–Meier curves with the day of diagnosis on imaging as entry date and log-rank test to test for statistical significance. A p value of less than 0.05 was considered statistically significant.

Sixty-six consecutive patients (median age 65 years, range 36–82 years) out of 93 patients with potentially resectable pancreatic cancer were eligible for inclusion. Twenty-seven patients were excluded for the following reasons: no confirmation of the presence or absence of malignant lesions (n = 4), local or systemic treatment prior to imaging (n = 3), artifacts or incomplete liver imaging (n = 8), and a time interval between imaging or imaging and surgery of more than two months (n = 12). Nineteen (29%) patients were diagnosed with liver metastases. Altogether 32 out of 47 patients without liver metastases underwent resection of the tumor. In the remaining 15 patients, the tumor was unexpectedly locally advanced (n = 12), metastasized intraperitoneally (n = 2), or the patient was too weak for surgery (n = 1). There was a significant difference in the survival between patients with liver metastases and without liver metastases (χ²(2) = 28.354, p = 0.000). Descriptives of included patients are described in Table 2.

Confirmation of liver metastases was obtained by histopathology in twelve patients; only in two cases transabdominal ultrasound with biopsy was successful. In the remaining patients, histopathology was obtained intraoperatively (n = 9) or by autopsy (n = 1). In six patients without histological proof, preoperative ¹⁸FDG-PET showed avid lesions in the liver, suggestive of liver metastases. In one patient multiple liver metastases were confirmed by intraoperative inspection and palpation of the liver and peritoneal metastases were histologically proven. The absence of liver metastases in the remaining 46 patients was confirmed intraoperatively by inspection and palpation of the liver (n = 43) and ¹⁸FDG-PET (n = 4). The mean time interval between CECT and Gd-enhanced MRI with DWI was 15 days (SD 12 days) and 26 days (SD 14 days) between Gd-enhanced MRI with DWI and surgery.

Gd-enhanced MRI with DWI detected malignant lesions in 16 out of 19 patients with liver metastases. The sensitivity of Gd-enhanced MRI with DWI was 84% (95% CI 60–97%). The positive predictive value was 94% (95% CI 69–99%), and the negative predictive value was 94% (95% CI 85–98%). There was one false positive on a per-patient basis, in this patient one liver lesion with perilesional ring enhancement and persistent high signal intensity on DWI was characterized as malignant on Gd-enhanced MRI with DWI. There was no evidence of liver metastases during surgery and follow-up CECT after 1 year. There were three false negatives on a per-patient basis. In the first case, one indeterminate lesion in liver segment six on CECT was characterized as benign on Gd-enhanced MRI with DWI. However, intraoperative biopsy-proven metastasis in segment two was not detected on MRI. In the second case, there were neither liver lesions on CECT nor Gd-enhanced MRI with DWI. In the last case, one lesion was indeterminate on Gd-enhanced MRI with DWI, yet showed high uptake on preoperative ¹⁸FDG-PET and thus was classified as metastasis.

In the negative-on-CT group, the per-patient prevalence of liver metastases was 20% (9/44). MRI was of additional value in 16% (7/44). In the indeterminate-on-CT group, the per-patient prevalence of liver metastases was 45% (10/22). MRI was of additional value in 90% of the patients (20/22).

On a per lesion basis, Gd-enhanced MRI with DWI detected 397 malignant lesions in 16 out of 19 patients with liver metastases. Contrast-enhanced MRI detected 156 malignant lesions, whereas DWI detected 397 malignant lesions (p = 0.051). In three patients, 20 to 50 malignant lesions were detected only by DWI. In one patient, even more than 100 malignant lesions were visible only on DWI (Fig. 1). Table 3 summarizes the detection rate of malignant lesions in the different sequences of Gd-enhanced MRI with DWI. There was a statistically significant difference in the number of malignant lesions detected by T2W-HASTE, T1W-VIBE precontrast, arterial phase, portal-venous phase, and DWI (χ²(2) = 32.861, p = 0.000). Post hoc analysis with Wilcoxon signed-rank tests was conducted with a Bonferroni correction applied, resulting in a statistically significant difference with a p value of 0.005. DWI detected significantly more metastases compared to T2W-HASTE (Z = − 3.181, p = 0.001), T1W-VIBE precontrast (Z = − 3.183, p = 0.001), arterial phase (Z = − 2.943, p = 0.003), and portal-venous phase (Z = − 3.063, p = 0.002). Figures 2, 3, and 4 show examples of three different patterns of liver metastases of pancreatic cancer on Gd-enhanced MRI with DWI.

Ninety-five percent of all liver metastases detected on Gd-enhanced MRI with DWI were subcentimeter lesions: 85% ≤ 5 mm, 10% 6–10 mm, and 5% > 10 mm. Nine patients (47%) had oligometastatic liver disease (i.e., ≤ 5 liver metastases [29]) and eleven patients had polymetastatic liver disease.