Syntactic pattern recognition for X-ray diagnosis of pancreatic cancer.

Abstract

Presents new algorithms used for the recognition of morphologic lesions of selected abdominal organs shown in X-ray images. These methods are aimed to facilitate diagnosis of pancreatic cancer (PC) and chronic pancreatitis (CP), based on the analysis of X-ray images acquired by endoscopic retrograde cholangio- pancreatography (ERCP). This method uses catherization of the major duodenal (Vater's) papilla combined with oral administration of a contrast medium to obtain roentgenograms of the pancreatic and biliary ducts. This method is commonly found to be most suitable for morphologic examination of the pancreas, as it provides high diagnostic efficacy, enabling recognition of PC as well as CP. The purpose of this work is detection of the most specific pathological changes, characterizing ducts affected by PC, including the occurrence of local dilations, stenoses, and side branches in the pancreatic duct as well as cysts or cavernous bulges. For CP, in most cases, the ducts are characterized by abnormal lateral branchings of first, second, or third order and, as in the case of PC, by local enlargements or constriction narrowings. The authors use attribute context-free grammars, which enable quick detection of pathological changes in shape, to recognize these signs.