Study On The Computer-assisted Diagnostic Of Early Esophageal Cancer Based On Endoscopy Images

To improve the treatment of esophageal cancer, the most effective way is to increase the penetration of conventional endoscopy, which means digestive endoscopist will have to bear more arduous task of diagnosis. Computer-aided diagnosis method can reduce the workload of doctors, reduce errors that may result from human factors. In this paper, image processing and pattern recognition methods are used to detecte the lesion regions in the electronic endoscopy images of early esophageal cancer, to improve the efficiency and accuracy of routine gastroscopy.Firstly, considering the common image processing algorithms do not work well with endoscopy images, we design a simple, fast method to extract the region of interest(ROI) based on the characteristic of big change between pixels in the lesion area. In addition, some new feature operators are proposed to describe the global and local characteristics of each region of interest. Finally, in order to reduce the rate of misdiagnosis as much as possible, the original sample weights initialization method of Adaboost is modified to make strong classifier process more emphasis on sensitivity during its training. Experimental results show that the sensitivity and specificity of our method are 93.6%, 77.2%, resepectively, which has obvious advantages to other related methods. Moreover, this method is simple and rapid, which make it more in line with real-time requirements of clinical applications.