Pancreatic cancer : prevention through improved diagnostics in individuals at risk

With the intent of timely detecting malignant precursors or pancreatic cancer (PC) in an early stage, current guidelines recommend surveillance for patients with pancreatic cystic neoplasms (PCNs), such as intraductal papillary mucinous neoplasms (IPMN), and individuals at risk (IAR) for familiar/hereditary pancreatic cancer (PC). This surveillance is mainly performed with magnetic resonance imaging (MRI). However, despite the associated examination time and costs, MRI is still suboptimal regarding its accuracy in predicting malignancy. This thesis aims to contribute towards improved secondary prevention of PC through more efficient and accurate diagnostic methods in patients with IPMN and familiar/hereditary PC through (i) a shorter MRI protocol, (ii) new imaging features and radiomics models for the prediction of malignancy, (iii) a better understanding of the surveillance program for IAR. In Study I, we compared a short (SP) and a comprehensive pancreatic MRI protocol (CP), with an acquisition time of approximately 8 and 35 minutes, respectively, in a cohort of 154 patients with PCNs. Our results showed that the SP provided equivalent clinical information in evaluating mural nodules, as well as cystic and main pancreatic duct diameters, compared to the more time-consuming and expensive CP in the surveillance of PCNs. In Study II, we assessed whether two novel features, such as volumetry and elongation value (EV) and other routinely used resection criteria, could predict malignancy in a cohort of 106 patients operated for BD- and mixed-type IPMN. Cases with mass-forming PC were excluded as a possible cause of main pancreatic duct (MPD) dilatation. Our results showed that volumetry and EV were not predictive for malignancy. Only elevated serum levels of CA19-9, mural nodules and dilated MPD (in the absence of stricturing masses) were associated with malignancy. In Study III, we evaluated the performance of MRI-based radiomics models in the preoperative prediction of malignancy in 130 patients operated for BD and mixed-type IPMN after exclusion of mass-forming PC. The radiomics models were internally cross-validated. Our results showed that a “pure” radiomics model outperformed a model including standard clinical and imaging features, suggesting that it might effectively predict malignancy in BD-IPMN even without standard clinical/imaging information. In Study IV, we described the imaging findings and the performance of a mainly MRI-based surveillance program in a cohort of 278 individuals at risk (IAR) for familial/hereditary PC. Our results showed that focal pancreatic lesions were identified in over half of IAR, the vast majority being small cysts. No lesions with high-grade dysplasia were detected. Five patients had PC, with the same prevalence of stage I and stage IV. The sensitivity and positive predictive values for detecting PC and its precursors were low (60% and 37%, respectively). In conclusion, a short pancreatic MRI protocol can be safely used in the surveillance of PCNs without losing relevant clinical information. Although novel imaging features cannot predict malignancy in IPMN, radiomics have shown its potentiality. In the context of individuals at risk for familial or hereditary PC undergoing surveillance, the early diagnosis of PC is still challenging with “traditional” cross-imaging methods such as MRI