Feature Extraction and Classification of Xinjiang High Morbidity Esophageal Cancer Based on Tamura and Wavelet Transform

Goal: The incidence of esophageal cancer in Xinjiang is remaining obstinately high. In order to improve the accuracy and efficiency of esophageal cancer diagnosis, this paper discusses the techniques and methods involved in the computer-aided diagnosis of esophageal carcinoma, which also to provide evidence for early detection, diagnosis and treatment of esophageal cancer. Methods: Selected normal esophagus, mushroom esophageal carcinoma, constricted esophageal carcinoma and ulcerative esophageal carcinoma, each 100 pieces and the regions of interest (ROI) were segmented manually. Then extracted texture features based on Wavelet Transform (WT) and Tamura algorithm and constructed Support Vector Machine (SVM) and BP Neural Network classifier to classify these features. Results: A total of 26 dimensional features were extracted from each image. The best classification accuracy of normal esophagus, mushroom esophageal carcinoma and ulcerative esophageal carcinoma reached 98.667%; It was 95.667% for normal esophagus, constricted esophageal carcinoma and ulcerative esophageal carcinoma; And 87.667% for normal esophagus, mushroom esophageal carcinoma and constricted esophageal carcinoma; 89.25% when classified normal esophagus and three kinds of advanced esophageal carcinoma. The classification efficiency of comprehensive features which consist of Tamura and WT features was better than that of single Tamura or WT feature. The Classification performance of SVM was superior to BP neural network. Conclusion: In this study, texture features based on Tamura and WT were extracted and classified by SVM and BP neural network classifier from ROI of normal esophagus and three kinds of advanced esophageal carcinoma. Results show that the classification accuracy of this study is high, which lays a foundation for the development of computer aided diagnosis system for esophageal carcinoma.