Detection Of Upper Gastrointestinal Early Cancers Based On Gastroscopic Images

Electronic gastroscopy as the convenient means of checking early uppergastrointestinal cancer is relied mainly on experience of digestive endoscopists forpreliminary identification of early cancer. The computer-aided diagnosis method caneffectively improve the accuracy of early upper gastrointestinal cancer diagnosed fordoctor. The research is relied mainly on generally electronic gastroscopic image, on thebasis of the image processing to detect early cancer of the upper digestive tract forimproving the survival rate of patients.Firstly，the obtained gastroscopic images are pre-processed, which the reflectiveand dark region are removed, and the effective region is extracted; the effective regionis segmented with grayscale canny operator and expansion operator. Then for theclassification and identification of the divided region, we obtain texture and colorfeatures from each region in each color space. Secondly, for classification offset inSVM with unbalanced data sets, the ODR-BSMOTE algorithm is introduced to obtainthe optimal feature; Finally, the clinical gastroscopic images of the esophagus andstomach are used to verify the effectiveness and feasibility of the method, and the fourindicators which are sensitivity, specificity, accuracy, and geometric average correctrate (G-mean) are used to assess the performance of the method. The accuracy rate andG-mean of classification of early esophageal cancer are, respectively,87%and86%,and these results for early gastric cancer are85%and88%, respectively in ourexperiment. Compared to similar methods, the paper solves these problems whichinclude the selection and optimization of the features and unbalanced data sets, andfurther improves the accuracy of classification and recognition of early cancers, ourmethod can improve more accurate identification of the lesion region, and reduce thetime to process gastroscopic images, which is more suitable for clinical application.