Application Of Computer-aided Eus Images Processing On Diagnosis Of Pancreatic Cancer

Pancreatic cancer is a highly malignant disease, which has a tendency to invadeand metastasize early. Moreover, the5-year survival rate with surgical resection isbelow5%[1].To review related studies of current imaging modalities of pancreas todiagnose pancreatic cancer, computed tomography (CT) is the primary method fordiagnosing and staging pancreatic malignancy. CT scan of the pancreas is wellcomplemented by endoscopic ultrasonography (EUS), which is more sensitive fordetecting early lesions; Moreover, diagnostic accuracy of EUS (85%) is significantlyhigher than CT scan (50%)[2]. However, the operator's experience and subjective factorshave a great impact on the results of early diagnosis of pancreatic cancer based on EUS;especially in the presence of chronic pancreatitis cases, the inflammatory statusobserved in patients with CP can interfere PC diagnosis, even experienced endoscopistswill produce false negative[3]. In addition, the application of the EUS guided fine needleaspiration (EUS-FNA) diagnostic procedure is limited in community hospitals.Furthermore, the EUF-FNA puncture accuracy rates were significantly affected bywhether guidance from pathologists was received[4]. Therefore, the development of anobjective and quantitative diagnostic method based on EUS imaging for the earlydifferential diagnosis of PC is crucial to overcome these diagnostic difficulties.Computer-aided diagnostic (CAD) techniques can assist radiologists in theidentification of lesions and improve diagnostic accuracy, particularly when used incombination with other physiological and biochemical methods. CAD techniques wereused as early as the1960s for the diagnosis of bone cancer[5]. In1998, the U.S. Food andDrug Administration (FDA) approved the first CAD system, the Image Checker Systemfrom R2Technology Inc., for use in the early detection of breast cancer. To date, someCAD research findings have been verified by the U.S. FDA; the application of CADtechniques was shown to improve the diagnostic accuracy and reduce the number ofmisdiagnosis[6]. Based on these successful experiences, we previously implemented theuse of digital image processing techniques for the successful differentiation of EUSimages depicting pancreatic cancer (PC) from EUS images of non-cancerous samples,including normal samples and samples exhibiting signs of chronic pancreatitis (CP).The diagnostic accuracy reached98%[7]. This study selected EUS images of3groups of patients with AIP, common chronicpancreatitis and pancreatic cancer, to validate the value of CAD technology in diagnosisof pancreatic. To identify this value, EUS images of AIP and common CP wereextracted separately to differentiate with PC based on a support vector machine.Meanwhile, the function of EUS and EUS-FNA in diagnosis of CP and PC was analyzedto compare with the CAD. Two parts of this study were as follows:1. The diagnosis of pancreatic cancer using computer aided EUS imagesprocessing1.1. Differential diagnosis of pancreatic cancer and autoimmune pancreatitisusing computer-aided diagnosis of EUS imagesObjective: To investigate the value of computer-aided diagnosis in diagnosing andclassifying the pancreatic cancer and autoimmune pancreatitis with medical imageprocessing technology to extract the texture features of endoscopic ultrasonography.Methods: Consecutive patients undergoing EUS-FNA of solid pancreatic masswere enrolled. Gold standard for final diagnosis included cytology from EUS-FNA forPC. In patients with AIP, clinical evaluation methods and the Asian criteria for AIP[8]were used for the final diagnosis. Representative regions of interest (ROIs) wereselected manually in EUS images and extracted rectangular sub-images from the ROIsfor texture analyses using image analysis software. Then, the distance betweenclass(DBC) algorithm and a sequential forward selection (SFS) algorithm were used fordata screening, to obtain a better combination of features; and, later, a support vectormachine (SVM) predictive model was built, trained, and validated to calculate thediagnostic sensitivity, specificity, accuracy, negative predictive value and positivepredictive value.Results: Patients with PC(n=262) and patients with AIP(n=41) underwentEUS-FNA were enrolled. Using computer-based techniques,105features wereextracted from the all EUS sub-images. From these texture features,16better featureswere selected, with a classification accuracy of95.90%. A predictive model was thenbuilt and trained. After performing200random tests, the average classified accuracy,sensitivity, specificity, positive predictive value and negative predictive value weredetermined as (91.16±0.15)%,(76.38±0.81)%,(95.29±0.14)%,(75.03±0.56)%and(95.36±0.14)%, respectively.Conclusions: Analysis of EUS images with digital image processing could accurately discriminate pancreatic cancer from autoimmune pancreatitis, which canhelp to improve the future diagnostic functioning of EUS.1.2. Differential diagnosis of pancreatic cancer and chronic pancreatitis usingcomputer-aided diagnosis of EUS imagesObjective: To investigate the value of digital imaging processing technology indistinguishing pancreatic cancer from chronic pancreatitis by analyzing the texturefeatures of EUS images.Methods: Just as1.1, consecutive patients undergoing EUS-FNA of solidpancreatic mass were enrolled. Gold standard for final diagnosis included cytology fromEUS-FNA for PC. In patients with CP, clinical evaluation methods and the RosemontCriteria[9]were used for the final diagnosis. Patients with CP were followed-up at leasttwo years. The methods related CAD technology is the same as the presentation in1.1.Results: Patients with PC(n=262) and patients with CP(n=126) underwentEUS-FNA were enrolled. Using computer-based techniques,105features wereextracted from the all EUS sub-images. Sixteen of these features were selected as abetter combination of features. A support vector machine (SVM) predictive model wasfirst built and trained by using these features as input variables for prediction of PC inhalf of the sub-images randomly, and then validated in the remainder of the data set.After200trials of randomised experiments, the average accuracy, sensitivity, specificity,the positive and negative predictive values of pancreatic cancer were (94.25±0.17)%,(96.25±0.45)%,(93.38±0.20)%,(92.21±0.42)%and (96.68±0.14)%, respectively.Conclusions: Computer-aided EUS image differentiation technologies provide anew and valuable diagnostic tool for the clinical determination of PC.2. Comparison of endoscopic ultrasonography, EUS-guided-fine needleaspiration and computer-aided diagnosis of EUS images for diagnosis ofpancreatic cancerObjective: To compare the value of EUS, EUS-FNA and CAD based on EUSimages for the differential diagnosis of PC and AIP, PC and CP respectivelyMethods: Inclusion criteria were as part1.1and1.2.Results: As part1.1and1.2, patients with PC (n=262), patients with AIP (n=41)and patients with CP (n=126) were selected. Reviewing EUS imaging, for PC, diagnosiswas malignant in209(79.77%) and benign in53(20.23%); whereas for patients withAIP, diagnosis was malignant in7(17.07%) and benign in34(82.93%); but for patients with CP, diagnosis was malignant in6(4.76%) and benign in120(95.24%); and for thedifferential diagnosis of PC and AIP, the accuracy, sensitivity, specificity were79.87%,79.77%,82.93%, respectively; but for PC and CP, they were84.79%,79.77%,95.24%,respectively. Meanwhile, Based on EUS-FNA, for patients with PC, diagnosis ofEUS-FNA which proceed for the first time was superior to EUS whose cytologicdiagnosis was malignant in231(88.17%) and benign in31(11.83%); but for AIP and CP,the diagnosis of EUS-FNA were all benign. Thus, for the differential diagnosis of PCand AIP based on EUS-FNA, the accuracy, sensitivity and specificity were89.77%,88.17%,100%, respectively; and for PC and CP, they were92.01%,88.17%,100%,respectively. Compared with the results of part1.1and1.2, the accuracy, sensitivity andspecificity of CAD were91.16%,76.38%,95.29%for PC and AIP respectively; andthey were94.25%,96.25%,93.38%for PC and CP, respectively.Conclusion: Compared with EUS and EUS-FNA, CAD is superior to EUS fordifferential diagnosis of pancreatic cancer and AIP, but is complementary to standardEUS-FNA; but CAD is superior for differential diagnosis of pancreatic cancer andchronic pancreatitis.This study came to the following conclusions:1. Differential diagnosis of pancreatic cancer and autoimmune pancreatitis usingdigital imaging processing technology is exciting. This technique is complementary tostandard EUS imaging and tissue acquisition techniques;2. Computer-assisted EUS image discriminates pancreatic cancer from chronicpancreatitis accurately. It could increase the accuracy of EUS diagnosis of tumors;3. Computer-aided EUS image differentiation technologies are easily, cheaply andnon-invasively. Further refinements of the merits of detecting and staging pancreaticcancer of EUS, such a model would be a novel and valuable research to differentiate PCin the future.