Pancreatic cancer detection with a non-contrast MR protocol: is it reliable?

Purpose The pancreatic cancer (PC) is the 4th leading cancer-related death, becoming the second one by 2030, with a 5 year survival rate of 8%. Considering its increased incidence in high-risk categories compared to the general population, we aimed to validate a non-contrast MR protocol, to detect PC in its earliest phase, which could be suitable as a screening tool in high-risk patients. Materials and methods In this retrospective study, we selected 200 patients (> 40 years) from our radiological database, which performed upper abdominal MRI between 2012 and 2017. 100 were negative for pancreatic lesions and 100 positive for pancreatic lesion (< 30 mm). The latter group included: 40 PDAC (pancreatic adenocarcinoma), 42 BD-IPMN (Branch Duct- Intraductal Papillary Mucinous Neoplasm), 10 PNET(pancreatic neuroendocrine tumor), 4 SCN(serous cystic neoplasm), 3 IPS(intrapancreatic spleen), 1 MCN(mucinous cystic neoplasm). Three readers (R1, R2 and R3) with a high, medium and low experience, respectively, analysed, first, the non-contrast MR sequences (single-shot T2w breath-hold, GE T1w FS, DWI and 2D/3D MRCP), and then the standard MR protocol, independently, randomly and anonymously. Readers identified or excluded the presence of pancreatic lesion, in both reading sessions. These results were compared with the histopathological diagnosis, and then divided into 3 different classes of lesions: all lesions, pancreatic adenocarcinoma and solid lesion. Mcnemar’s test was used to compare the results. The inter-observer agreement was determined according to the kappa statistic in both protocols, and then the inter-protocol agreement was calculated. Results The non-contrast MR protocol has reached statistical parameters values ranging between 83% in SE (sensitivity) by R3 and 99% in NPV (negative predictive value) by R1. The standard MR protocol has reported slight increasing statistical parameters compared to those of the proposed one. However, there are not significant statistical differences between the both protocols. The proposed non-contrast MR protocol has reported the highest NPVs in the PDAC group detection (R1: 99%, R2: 99%, R3: 98%). In all groups of lesions, the agreement between the two protocols was excellent for each Reader ranging from 96 to 98%. Conclusion The proposed non-contrast MR protocol showed high PC detection values and a time execution ≤ 20 min. Therefore, it can be proposed as a screening tool in high-risk patients.


Introduction
Pancreatic cancer (PC) is the fourth leading cause of cancer-related death in the USA and many Western countries [1] with 47.050 deaths in 2020 in USA [2]. During 2008 and 2016, the death rate increased by 0.4% per year [2]. Although, it is predicted to become the second death leading cause by 2030 [3]. Unfortunately, the late PC diagnosis makes less than 20% of patients fit for surgery, with a 5 year survival rate of 8% [4]. On the contrary, an early diagnosis, with a PC < 10 mm without nodal involvement, carries out a significantly better prognosis, and a 5 year survival rate up to 100%, in some series [5].
Given the low incidence of PC in the general population, population-based screening is not considered cost-effective [6]. However, selected patients with increased risk, of at least 5-10 times higher than the general population, could benefit from screening. Patients with chronic pancreatitis [1], Familial pancreatic cancer (FPC) and some Inherited cancer syndromes [1,[7][8][9]] present a raised lifetime risk of PC with an early-onset of PC in family members [10]. The high incidence of PC in these groups justifies the screening, which should be started between 40 and 50 years, and performed until the patients become "unfit for surgery" [9].
Screening program for PC appears to constitute a costeffective intervention [11], especially for Patients with Hereditary Pancreatitis [12]. According to some Authors, EUS (endoscopic ultrasound) and MRI are the most suitable tools for pancreatic imaging in small lesions detection [13]. However, MRI is more easily accepted as a screening tool, with a better cost-benefit balance [9,14].
Some Centers suggest a yearly interval follow-up [14][15][16], although this could be insufficient to detect PC in the earliest phase, and a shorter time window could be desirable, as suggested by some Authors [17]. However, the standard MR protocol is time consuming and costly. Therefore, a shorter and cheaper assessment of the pancreas with a high negative predictive value (NPV) should be desirable.
The aim of this study is comparing a non-contrast MR protocol to the standard one in the small PC detection.

Materials and methods
As a retrospective study, patient informed consent was waived. This study has obtained Institutional Review Board approval on 20.06.2019 n. 699/CE Marca.

Patients' selection
In our radiological database, from January 2012 to March 2017, were found 4250 patients with upper abdominal MR study. From those, using the clinical folders, radiology resident "in training" selected 1365 patients following these criteria: -Age > 40 years, -Without history of previous abdominal surgery either gastro-intestinal intervention or chemo-radio-therapy for upper abdominal tumours -Availability of a complete MR study, including preand post-contrastographic sequences DWI (Diffusion Waited Images) and MRCP (Magnetic Resonance Cholangiopancreatography).
Then, using the radiological reports, were selected only patients with: -PC smaller than 30mm, -Solid lesions, not PC, < 30 mm -Cystic pancreatic lesions < 30 mm -Patients without pancreatic lesions, who had performed MR imaging for other conditions, without evidence of a solid mass elsewhere in the field of study.
The resulting pool of 658 patients was included. Then, it was divided in two parts: one containing patients with pancreatic lesions (387), one those without pancreatic lesions (271). 100 patients were selected in each group, randomly. The patients' recruitment was interrupted when the predefined number was reached, in order to recreate the different conditions found in patients at high-risk to develop a PC [9,18].
The main focus of a screening test is to find lesions in their early stages; therefore, we chose the threshold of < 30 mm for pancreatic lesions, in order to exclude cystic lesions > 30 mm, considered as a worrisome feature [19]. As a retrospective study, the patients' clinical folders did not contain adequate information allowing Authors to select only patients belonging to a poll of high-risk people for PC. However, patients older than 40 yrs have been selected, in order to analyse a pancreatic parenchyma as similar as possible to the high-risk ones [9].
The other group contained 100 patients without pancreatic or abdominal disease.

Diagnosis confirmation
Pathology evaluation of surgically resected served as a standard of reference for the diagnosis of PDACs (n = 40). About PNETs, 6 were resected, while 4 had somatostatin analogue nuclear medicine study confirmation, and followed-up due to the small size. IPSs were characterized by Nuclear Medicine. For cystic lesions, benignity was confirmed by at least 2 years follow-up.
A Radiologist, expert in abdominal MRI and pancreatic pathologies, revised the MRI exams of the negative patients group, to confirm that no abdominal or pancreatic disease were present.
For our non-contrast MR protocol, we selected the following sequences from the standard protocol: axial and coronal SSFSE T2w BH, axial FS-GRE T1w, DWI (2 b values 50-800), 2D/3D MRCP (Table 1). 3 Readers, with different level of experience in MRI of the pancreas (R1: > 15 yrs; R2: > 5 yrs; R3: "in-training") independently and randomly, reviewed the cases anonymously, unaware of symptoms and clinical history of the patients. Readers were aware that solid, cystic or no pancreatic lesions could be observed. Image analysis has made by using the workstation of our PACS system (Fuji Synapse, version 4.1.6.0, Fujifilm, Valhalla, NY, USA).

Imaging analysis
The radiology resident "in-training" was responsible for the management of cases review, showing only the chosen sequences for the analysis. In patients with multiple studies, the baseline MR study was chosen for the reading session. The assessment of the images quality was not conducted before the reading session, as well as to recreate as much as possible a realistic situation.
Reading session was performed in two steps: first, the Readers analysed the proposed non-contrast MR protocol; secondly, one month later, with a new random order, they analysed the standard MR protocol.
The presence of pancreatic solid lesion was based either on a discernible mass-like lesion in the pancreatic parenchyma or based on secondary features. Those are: focal main pancreatic duct (MPD) dilatation, focal stenosis of MPD, even if without upper dilation, loss of regular pancreatic lobulations, dilatation of choledochus, focal hypointensity on FS-GRE T1w [22], focal hyperintensity on high b values DWI or hypovascular lesion during dynamic enhancement ( Fig. 1).
Pancreatic cystic lesions were evaluated following the International consensus guidelines [19]: BD-IPMN as a cystic lesion > 5 mm communicating with MPD on MRCP, while MCNs and SCN were evaluated by morphological appearance (oligocystic for MCN, microcystic for SCN) [19].
Based on these criteria, in each reading session, Readers indicated whether or not a pancreatic lesion is present, suggesting the solid or cystic nature.

Statistical analysis
Continuous variables are expressed as mean ± SD, and categorical data as frequencies or percentage.
In each reading session and for each Reader, sensitivity (SE), specificity (SP), positive predictive value (PPV) and NPV were calculated for pancreatic lesions detection.
Results of both reading sessions were compared with the final diagnosis according to three different classes of lesion: solid lesions, PDAC and all lesions; Mcnemar's test was used to compare the results of the protocol in the different classes of lesions between Readers. Inter-protocol agreement for each Reader was determined according to the kappa statistic, for the identification of pancreatic lesions according to the three different classes of lesions. Moreover, the inter-reader agreement was calculated on a three-category (R1 vs. R2 vs. R3) basis, in both protocols. Poor, moderate, good and excellent agreement are indicated by κ values ≤ 0.40, from 0.41 to 0.60, from 0.61 to 0.80 and > 0.80, respectively. A p value < 0.05 was considered as statistically significant. Statistical analysis was performed with a commercially available software package (SPSS, version 20.0 for Windows, SPSS Inc., Chicago, IL, USA) and STATA 14 software (StataCorp, College Station,TX).

Results
No statistical differences were observed in the age of the two groups of patients (Patients with pancreatic lesions: mean 67.6 ± 10 years; Patients without pancreatic lesions: mean 64.8 ± 10 years). The mean size of the pancreatic lesions group was 15.6 mm ± 6.045 mm.
Results of reading sessions are reported on Table 2. On Table 3   Readers in the different classes of lesions, in both MR protocols. On Table 4 is shown the inter-protocol agreement for each Reader. Finally, on Table 5 is illustrated the agreement between the Readers. On Table 3, all Readers reached very high values in all statistical parameters in all classes of lesions, in both protocols. In particular, in the proposed MR protocol, the statistical parameters values ranged between 83% in SE by R3 and 99% in NPV by R1. All Readers have reported a slight increase of all statistical parameters in the standard MR protocol compared to those of the proposed one. However, there are not significant statistical differences between both protocols, in the three categories of lesions, for each Reader. Furthermore, the proposed MR protocol has reported the highest NPVs in the PDAC group (R1: 99%, R2: 99%, R3: 98%). Moreover, in the PDAC identification, R2 has shown higher SE and NPVs in the proposed MR protocol compared to those of the standard one (SE: 97% vs. 95%, NPV: 99% vs. 98%). Similarly, R2 has reported higher SP and PPVs in solid lesion group in the proposed MR protocol compared to those of the standard one (SP: 98% vs. 97%, PPV: 95% vs. 92%). None of those differences were statistically significant. Moreover, in solid lesions group, R2 and R3 show lower SE and NPV results compared to those of R1, in both protocols, according to their different experience.  R1  R2  R3  R1  R2  R3   PDAC  40  39  39  37  40  38  39  IPMN  42  0  1  1  0  1  0  PNET  10  5  2  1  1  1  1  Other lesions  8  2  2  1  0  0  1  Negatives  100  0  1 1 0 1 2 In both protocols, all Readers reached lower SE values in the solid lesions group compared to those of PDAC and all pancreatic lesions group. These results are mostly related to the poor PNET detection in the reading session ( Table 2).
On Table 4, in all groups of lesions, the agreement between the two protocols was excellent for each Reader ranging from 96 to 98%.
On Table 5, an excellent agreement was found between the three Readers in both protocols. In particular, the k statistic mean results were in the proposed and standard MR protocol: all lesions detection as of 0.86 and 0.89, PDAC detection as of 0.84 and 0.89, and solid lesions detection as of 0.86 and 0.89, respectively.
Cystic lesions were used as a confounding factor in the pancreatic solid lesion detection, especially for the PDAC, therefore, no statistical results were reported for those lesions.

Discussion
Our results show high values in all analysed parameters, especially SE and NPV, in both protocols.
No statistical differences were found between the two protocols either in the Readers capability to detect pancreatic lesions or in their agreement.
According to these results, especially the high SE and NPV values, our short protocol can be suitable in a high-risk patients' screening program [23], thanks to the excellent results in detecting PDAC, the most frequent malignant PC [24].
Our imaging protocol was performed with a standard 1.5 T MR machine, easily available in the Hospitals. However, even if not significant statistical differences were found between Readers, is to underline that our Hospital is the second reference centre for pancreatic diseases in a large regional area of about 5 millions inhabitants, taking care of hundreds of patients with pancreatic pathologies and more than 70 pancreatic resections yearly. Thus, our results suggest that an experienced Radiologist in pancreatic imaging should perform the reading session.
The proposed non-contrast MR protocol does not require a long acquisition time (about 20 min in/out room) and the use of contrast media, avoiding gadolinium brain deposition [25].
MRI has an high sensitivity to identify solid pancreatic lesions, thanks to the high contrast between normal parenchyma and pancreatic lesions in FS-GRE T1w [22]  and DWI images [21]. Although DWI images can have some limitations in small pancreatic solid lesions detection [26], the combination of the different sequences allowed us to have the reported excellent results. Consequently, it is highly recommended as screening modality for PC. According to some screening program experience [11][12][13][14], growth rate is correlated to the survival outcome. Mean volume doubling time (VDT) of pancreatic cancer ranges from 132,3 days [27] to 144 days [28], although in Ahn et Al. it varied from less than one month to more than four years [27]. Moreover, Mortenson et Al. observed that the surface area of a PC may grow yearly from 1.17 to 7.15 times [17], suggesting a 3 months follow-up interval. A recent study, based on genomic sequencing on cancer cells in seven PC patients, reports an average of 11.7 years elapsed from tumor initiation to overt cancer development and an average of 6.8 years elapsed between the development of overt cancer and the development of metastatic disease. Thus, there is a large time window for detection of PC while the disease is in its earliest and treatable stages [29].
However, the predicted progression rate suggests that once a PC is detectable by diagnostic tests (stage I), its growth and progression to stage IV disease can occur within 1 year [30]. Based on these observations, 1 year screening program interval time can be insufficient to diagnose the PC in the early and resectable stage (Fig. 2). For this reason, a short and, relatively, cheap imaging test is requested for high-risk patients, allowing them to be screened twice a year. Moreover, a non-contrast MR protocol avoids the spending time to read and explain to the patients the contrast media informed consent, the vein cannulation and the post-procedure activities, reducing also any contrast-media adverse reaction and the care-time of the patient.
About the PNET results, according to WHO classification 2010 [31] and Kuo et Al. [32] in PNETs ≤ 2 cm the most significant predictors of disease-specific survival are grading and growth. Although SEER database excludes PNETs to be considered benign, and small size does not preclude malignant behaviour [32], Bettini et al. found that out of 51 R0 resected incidental non functional PNETs (NF-PNETs) ≤ 2 cm, only 6% were malignant and in the long term follow-up (median 47 months) there were no deaths directly related/caused by the specific disease [33]. It is also reported that there is no direct benefit in early surgery Even after EUS no worrisome features were appreciated. At followup after 10 months a large hypovascular lesion occupies the uncinate process (arrow: C), with multiple hepatic metastases (arrows: D) approach for PNETs ≤ 2 cm [34]. So in a systematic screening programme, it is possible to find in a further followup study small missed PNETs without affecting long-term patient's prognosis (Fig. 3) [35][36][37].
Furthermore, in patients with cystic lesions or cystic PNETs, DWI is the sequence for suspecting malignant degeneration: DWI is helpful for distinguishing mucin plug from solid nodule in cystic lesions [38], but mostly there is a significant correlation between diffusion restriction and invasiveness/malignancy, as demonstrated by Kim [39]. The most "time consuming" sequence is 3D MRCP, which, due to the triggered acquisition, can last several minutes in uncooperative patients. Some studies have demonstrated that in the visualization of pancreatic ductal system there are not significant differences between 2 and 3D MRCP [40,41], thus, it can be concluded that a 2D SSFSE is a suitable sequence in a short pancreatic screening protocol. Recent development of compressed sensing MRCP allow to obtain 3D MRCP in a single breath hold, thus reducing the time requested [42].
Our study has several limitations, as a retrospective design, might have a possible selection bias, which we have tried to overcome making a randomized selection of cases including a large variability of pancreatic alterations. The selection of all cases comes from a radiology database, which does not include cases that could have been positive on EUS or other non-radiologic workups. The cohort of patients is not a high-risk PC category, however, only patients over 40 yrs were included, trying to replicate similar pancreatic parenchyma morphology. The only group which is not comparable is the Hereditary Pancreatitis group, where early advanced alteration of pancreatic parenchyma and ductal system are associated in a very young population [43]. In this group, a specific prospective study should be conducted. In this last group of high-risk patients some parenchymal alterations, due to chronic inflammation, such as focal stenosis of MPD or loss of regular pancreatic lobulations, could simulate a pancreatic lesion. As well as, patients with chronic pancreatitis could present the same pancreatic alterations described above, with a very challenging capability to differentiate suspect lesions from chronic alterations. Consequently, the number of patients recalled for a complete work-up could be higher than needed, thus we think it is necessary to perform the MRI in the same center, where previous exams are easily available and comparable, allowing the detection of minimal changes in the different studies.
Even if the PDACs included in our study were resectable, some small PDACs in a screening scenario could face locally advanced or metastatic, which means that they would not be considered as resectable before neo-adjuvant chemotherapy. However, thanks to the short acquisition time of the proposed non-contrast MR protocol, it can be offered twice a year, reducing the risk to detect a PDAC in an advanced stage. In comparison to EUS, MRI has the advantage of a whole exploration of the upper abdomen, allowing in the same setting the detection or exclusion of small liver metastases, especially thanks to DWI [44][45][46][47][48][49][50].
Although the most experienced radiologist (R1) had excellent results (SE 97%, SP 95%), even the less experienced radiologist (R3) obtained satisfactory results (SE 92%, SP 97%). Considering as standard of reference the breast cancer screening double reading which reports sensitivity of 72% and specificity of 97% [51,52], even if in a different situation, which is not a screening ones, those achieved results seam to be quite high and encouraging for PC detection.

Conclusion
The reported high PC detection values, the short time consuming (less than 20 min "in and out door") and the "free gadolinium injection", makes our proposed non-contrast MR protocol suitable to be used as a screening tool in high-risk patients, reducing the time lapse between the exams, with more chances to the patient to be diagnosed in the early phase.

Acknowledgements Not applicabile
Author contributions FM contributed to Conceptualization, Investigation, Methodology, Writing-original draft, Visualization. VP contributed to Investigation, Visualization, Methodology. LB contributed to Investigation, Methodology. SV contributed to Visualization. VC contributed to Statistical analysis. MF contributed to Methodology. GM contributed to Writing-review, Validation, Supervision, Project administration.
Funding Open access funding provided by Università degli Studi di Napoli Federico II within the CRUI-CARE Agreement. This research and its authors did not receive any specific grant neither from funding agencies in the public, commercial, nor from not-for-profit sectors or any other provider.

Data availability
The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials.

Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Ethics approval As a retrospective study, patient informed consent was waived. This study has obtained Institutional Review Board approval on 20.06.2019 n. 699/CE Marca.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.