Computer-aided Diagnosis Technology For Esophageal Cancer X-ray Image Of Xinjiang Kazak Nationality

Objective: Esophageal cancer is one of the fastest rising type of cancers in China.The Kazak nationality is the high-risk group in Xinjiang Uygur Autonomous Region. This paper studied the methods and technologies of computer-aided diagnosis for esophageal cancer based on X-ray images, which aims at assisting radiologists in interpreting digital X-ray image features and improving the quality of diagnosis. Methods: The experiments were conducted in the MATLAB platform. Firstly, the regions of interest were selected under the guidance of the radiologist. And the preprocessing methods, including median filter and histogram equalization, were applied on the X-ray images. This step can improve the quality of the images. Secondly, six algorithms, including gray-level histogram, gray level co-occurrence matrix, gray level gradient co-occurrence matrix,Tamura texture, wavelet frequency and Hu invariant moments, were employed to the image feature extraction. Thirdly, two-step feature selection method consisting of ROC area selection and principal component analysis were used to select the features with strong classification ability. In addition, the Lib-SVM and 10-fold cross validation method based on the RBF kernel function were applied to classify the normal and abnormal esophageal images, fungating type, infiltrating type and ulcerative type esophageal images.And the classification performance were evaluated by the classification accuracy and AUC values. Results: Experimental results show that 66 features are extracted based on the algorithm above. According to the principle, AUC value greater than 0.7, 28 features are selected. Furthermore, the principal component analysis was used to select the features with the best classification performance. The top-ten principal components are selected for their cumulative percent reached to 90.33%. The classification performance of gray level co-occurrence matrix outperforms other algorithms when the single feature was used. The classification accuracy and AUC value for normal and abnormal images, and three kindsof abnormal images are 92.67% and 92.70%, 90.67% and 91.00%, respectively. And the processing time are 2.40 s and 2.80 s, respectively. When the comprehensive feature was applied, the classification accuracy and AUC value are 94.17% and 94.20%, 92.33% and92.30%, respectively. And the processing time are 7.70 s and 9.70 s. When the features with AUC value greater than 0.7 were used, the classification accuracy and AUC value are94.17% and 94.20%, 92.33% and 92.30%, respectively. And the processing time are 2.80 s and 4.50 s. When the the principal component analysis was employed to the features, the classification accuracy and AUC value are 95.33% and 95.00%, 94.59% and 94.00%,respectively. And the processing time are 1.50 s and 2.30 s, respectively. Conclusion: This paper combines the two-step selection method with SVM to analysis the four types esophageal images. The proposed methods achieve high classification performance in both of diagnostic quality and the processing time, which can provide the valuable reference to radiologists and lay a foundation for computer diagnosis system of esophageal cancer for Kazak nationality.