Simple Vascular Architecture Classification in Predicting Pancreatic Neuroendocrine Tumor Grade and Prognosis.

Abstract

Vascularity is a critical feature in the evaluation of pancreatic neuroendocrine tumor (PNET). When done by EUS, contrast agents are recommended. However, vascular architecture (VA) can also be evaluated by routine Doppler flow in EUS without contrast agents. Our aim was to provide a simple VA classification in EUS for PNET grade and prognosis. All pathologically proven PNET cases with EUS between 2012 and 2018 were retrospectively analyzed. The Doppler imaging was retrieved for VA classification. Predictive model construction was performed by machine learning algorithms. A total of 112 PNET cases were evaluated, among which 93 cases were subjected to VA classification. The VA was classified into type A (peritumoral with or without intratumoral vessels [A1 or A2]); type B (only intratumoral vessels); and type C (flow was absent). The VA classification was significantly correlated with tumor grades: 74% type A1 was G1, 73% type B was G2, and 58% type C was G3. Multivariate analysis indicated that elevated serum CA19-9 and type C classification were the independent predictors of G3 tumor. Five machine learning models were constructed, among which random forest was the best one with an AUC of 0.9972. Low-risk patients classified by this model exhibited better prognosis than high-risk patients (p = 0.0087). In the novel simple VA classification, peritumoral, intratumoral, and absent vessels are prone to be G1, G2, and G3, respectively. Combined with serum CA19-9 and lesion size, the VA classification could predict tumor grade and prognosis in PNET.