Discrimination between chronic pancreatitis and pancreatic adenocarcinoma using artificial intelligence-related algorithms based on image cytometry-generated variables

Abstract

The incidence of pancreatic adenocarcinomas (PA) is increased in the setting of chronic pancreatitis. Distinguishing chronic pancreatitis from pancreatic adenocarcinomas is often difficult, and is based on routine brush cytological specimens provided during endoscopic retrograde cholangiopancreatography (ERCP). Reactive epithelial changes in chronic pancreatitis may appear similar to those of a well-differentiated cancer. Brush cytology specimens were obtained during ERCP from 49 patients with diseases for which the differential diagnosis included chronic pancreatitis and/or pancreatic adenocarcinoma. Image cytometry was performed involving the assessment of between 200-400 Feulgen-stained nuclei per case; for each case, 40 quantitative cytometric variables were generated. Data analysis was performed using artificial intelligence methods of data classification that produced decision trees and production rule systems. Different classification models were produced for a subset of 34 patients. The best models were identified by the use of a sampling technique (leave-one-out), and were tested on the remaining 15 patients. These models were based on 5 of the 40 variables associated with a significant discriminatory function. Pancreatic adenocarcinoma was diagnosed in the training data set of 34 patients during a leave-one-out process with an estimated sensitivity of 91% and specificity of 87%. Both sensitivity and specificity were 80% in the independent test set of 15 patients. We conclude that inflammatory and malignant pancreatic epithelia exhibit distinct morphological features that can be distinguished by decision tree-based classifiers employing image-cytometric numerical data.