Integration of different criteria for borderline resectable pancreatic cancer using classification tree analysis: the use of radiological tumour-vascular interface in correlation with surgical and pathological outcomes

Abstract

Aim: To integrate various criteria for borderline resectable pancreatic cancer (BRPC) based on radiological parameters using classification tree analysis.

Materials and methods: The institutional review board approved this retrospective study and waived the requirement for informed consent. Two hundred and thirty-five tumour-vein interfaces and 67 tumour-artery interfaces in 245 patients with surgically confirmed pancreatic ductal adenocarcinoma who underwent both preoperative computed tomography (CT) and magnetic resonance imaging (MRI) were assessed by two independent readers. Radiological parameters for evaluation of the tumour-vascular interface were boundary, length of interface, degree of circumferential interface, and contour deformity of affected vessels. Classification tree analysis was performed to determine parameters associated with vascular invasion using pathological and surgical results as the reference standard.

Results: In the classification tree analysis for the tumour-vein interface, contour deformity and degree of circumferential interface were the first and second determining factors, respectively, for both surgical and pathological vascular invasion. For the tumour-artery interface, boundary and degree of circumferential interface were the first and second determining factors for surgical invasion, while contour deformity and length of interface were the first and second determining factors for pathological invasion. The BRPC group of modified criteria arbitrarily formed based on the results had similar surgical (74.1-81.6%) and pathological (54.3-63.3%) venous invasion compared to that of the National Comprehensive Cancer Network (NCCN) criteria, and the lowest surgical (33.3%) and pathological (6.7%) arterial invasion compared with those in previously established criteria for BRPC (43.3-55.6% and 22.2-26.1%, respectively).

Conclusion: Various criteria for BRPCs were integrated using classification tree analysis, and a modified criterion for BRPC, which provides satisfactory results, was established.