Mo1527 Double-Duct Sign in the Era of EUS: Is the Absence of a Mass on EUS Adequately Reassuring?

Abstract

Gastroenterolgy, The First People’s Hospital of Shanghai, Shanghai Jiao Tong University, Shanghai, China; Eletronic Engineering, Fudan University, Shanghai, China Background: The clinical manifestations and imaging features of mass forming autoimmune pancreatitis and pancreatic cancer are very similar; therefore, differential diagnosis of such focal lesions remains a clinical challenge. Endoscopic ultrasound (EUS) guided FNA, as an invasive procedure, has the unique ability to obtain specimens for cytological diagnosis, thus, plays a key role in the evaluation patients with inconclusive findings. Digital imaging processing technology has become an encouraging and effective agent to tissue characterization for differential diagnosis. Aim: In this study, we aimed to investigate the value of digital imaging processing technology to analyze EUS image in differentiating mass forming autoimmune pancreatitis from pancreatic cancer by extracting the texture features of the typical EUS images and then developing a predictive model to pursue a quantitative, inexpensive and noninvasive computer-aided diagnostic approach. Material and Methods: A total of 324 patients consisting of 288 pancreatic cancers and 36 mass forming autoimmune pancreatitis were enrolled in this study. All patients underwent EUSFNA and cytological analysis. The texture features were extracted from the selected typical EUS images by using computer-based image analysis software. Then, the processes of distance between class and the sequential forward selection were used to obtain a better combination of the texture features. Meanwhile, a support vector machine (SVM) predictive model was developed, trained and validated as well by using a half-half algorithm. Results: From the selected typical EUS images, 105 texture features were initially extracted, among which, 16 better features were chosen, with a classification accuracy of 95.94%. A predictive model was then built and trained. The total 324 patients were randomly divided into a training set and a testing set. The training set was used to train the SVM, and the testing set was used to evaluate the performance of the SVM. After performing 200 random tests, the average classified accuracy, sensitivity, specificity, positive predictive value and negative predictive value were determined as (92.05 0.13) %, (78.08 0.82) %, (94.21 0.16) %, (76.14 0.57) % and (95.12 0.12) %, respectively. Conclusions: This study reveals that digital imaging processing of EUS images is an inexpensive and noninvasive approach which is effective in differentiating pancreatic cancer from mass forming autoimmune pancreatitis, therefore, it may improve the diagnostic function of EUS without fine needle aspiration.