Differenzial Diagnosis of Pancreatic Cancer from Normal Tissue Using Digital Imaging Processing and Pattern Recognition based on Support Vector Machine of EUS Images

Background:

Endoscopic ultrasonography (EUS) can detect morphologic abnormalities of pancreatic cancer with high sensitivity, but with limited specificity. In the present study, we applied digital imaging processing (DIP) technique to EUS images of pancreas to develop a classification model for differenzial diagnosis of pancreatic cancer.

Methods:

EUS images were obtained and correlated with cytological findings after fine-needle aspiration. Texture features were extracted from the region of interest (ROI) and multi-fractal dimension vectors were introduced in the feature selection to the frame of M-band wavelet transform. The sequential forward selection (SFS) process was used for the better combination of features. Using the area under the receiver operating characteristic (ROC) curve and other frequently used texture features based on the separability criteria, a predictive model was built, trained, and validated, according to the support vector machine (SVM) theory.

Results:

153 pancreatic cancer and 63 non-cancer cases were selected for the analysis. After introducing 2 order multi-fractal dimension vectors of feature, the better combination of features was selected. From 67 frequently used texture features, 20 better features were selected, resulting in a classification accuracy of 99.07% after being added to 9 other features. A predictive model was then built and trained. After 50 random tests, the average accuracy, sensitivity, specificity, positive predictive value, and negative predictive value for the diagnosis of pancreatic cancer were 97.98±1.23%, 94.32±0.03%, 99.45±0.01%, 98.65±0.02%, and 97.77±0.01%, respectively.

Conclusion:

The classification of EUS images for differentiating pancreatic cancer from normal tissue by DIP. Further refinements of such a model could increase the accuracy of EUS diagnosis of tumors.