Computer-Aided Diagnosis of Digital Radiographs for Pneumoconiosis Using Wavelet-Based Texture Features
Purpose: To automatically detect pneumoconiosis using a computer-aided diagnosis (CAD) system on digital chest radiographs.Methods: Lung fields were first extracted by combining the traditional Otsu-threshold method with a morphological reconstruction on the digital chest radiographs.Then the wavelet transform was conducted on the images to calculate 28 energy texture features, upon which decision tree (DT) and support vector machine (SVM) were trained as a classifier respectively to discriminate a digital chest radiograph as normal or abnormal.Performances of both classifiers with full and selected texture feature sets and SVMs with different kernel functions for classifying chest radiographs on the selected feature set were assessed and compared by receiver operating characteristic (ROC) curve analysis.Results: SVM with the polynomial kernel function had the highest performance when using the selected texture features with an area under the ROC curve of 0.967 ± 0.015, compared to 0.877 ± 0.039 (P<0.001), 0.955 ± 0.019 (P=0.367), 0.947 ± 0.022 (P=0.242) and 0.904 ± 0.031 (P=0.007) by the DT and SVM with the kernel function of linear function, radial basis function and sigmoid function, respectively.Conclusions: When used to differentiate between normal and pneumoconiosis chest radiographs based on the wavelet transform-based energy texture features, SVM with the polynomial kernel function outperformed DT and SVMs with other kernel functions.
computer-aided diagnosis wavelet transform decision tree support vector machine texture feature
Biyun Zhu Hui Chen Budong Chen Yan Xu Wenfang Wu Kuan Zhang
Institute of Biomedical Engineering, Capital Medical University, Beijing 100069, China Department of Radiology, Beijing Friendship Hospital, Capital Medical University, Beijing 100053, Ch
国内会议
北京
英文
322-329
2012-12-01（万方平台首次上网日期，不代表论文的发表时间）