ABSTRACT
Objective To extract the texture features of endoscopic ultrasonography (EUS) by digital imaging processing(DIP) and pattern recognition,and then to investigate its value for differential diagnosis between pancreatic cancer and chronic pancreatitis.Methods Two hundred and two patients with pathologicaly confirmed pancreatic malignancy,who underwent EUS from Feb 2005 to Mar 2011,and 104 patients with chronic pancreatitis (including 34 cases of autoimmune pancreatitis),who underwent EUS from May 2002 to Aug 2011,were randomly recruited in this study.The optimal texture features of EUS images in this study were selected by the sequence forward search (SFS) algorithm.With the optimal feature combination,cases were automatically divided into pancreatic cancer and chronic pancreatitis based on the findings of support vector machine (SVM),which were compared with the real results.the sensitivity,specificity,accuracy,positive predictive value and negative predictive value were calculated.Results Nine categories and 105 texture features were extracted based on all EUS images,and 13 features were chosen as optimal combination.Images of 306 cases were randomly divided into training set ( 153 cases,101 cases of cancer,52 cases of chronic pancreatitis) and testing set ( 153 cases,101 cases of cancer,52 cases of chronic pancreatitis).The classifier was trained with the training set and tested with testing set.We proceeded 200 times randomly.the average accuracy,sensitivity,specificity,positive predictive value and negative predictive value were ( 86.08 ± 0.14) ％,(79.47 ± 0.32) ％,( 89.71 ± 0.18 ) ％,( 81.21 ± 0.26 ) ％,( 88.93 ± 0.14 ) ％,respectively.Conclusion Differential diagnosis of pancreatic cancer from chronic pancreatitis by Computer-assisted EUS image analysis,highly accurate,convenient,non-invasive and less costly,is a novel and valuable method of early diagnosis.