Research On Features Extraction Of Lung Images And Computer-aided Diagnosis System

Tuberculosis is one of the diseases that threaten the health of human beings.In the incidence of infectious diseases network report,the incidence and death of tuberculosis ranks in front of infectious diseases.In the present clinical medical field,the diagnosis still merely depends on doctors' radiograph reading.However,a large number of patients always lead to the radiologists' heavy task,high pressure,and the low diagnosis efficiency.This becomes an important reason why computer-aided diagnosis is capable of becoming a hot issue of research.This thesis designs and implements an automatic tuberculosis diagnosis system for lung X-ray radiography.The proposed system consists of three modules:lung field segmentation,features extraction,and diagnosis classification.The segmentation of lung field is based on Graph Cut algorithm.In order to achieve effective automatic segmentation,instead of specifying the type of seed points manually,the mask matching method is used here.In the feature extraction module,the combination of color and texture features is applied to describe the segmented lung field.First,divide the segmented image into several image blocks,and calculate the 6-dimensional edge information and the 24-dimensional fuzzy color histogram information in each image block.Second,the 24-dimensional information is added to the corresponding edge type information,thereby,each image block gets a 144-dimensional vector.After the process of normalization,quantization,and elimination of redundancy,the vectors are the feature vectors to be used for tuberculosis diagnosis.Finally,the support vector machine with an appropriate kernel function is employed to realize diagnosis system of tuberculosis.The thesis introduces features extraction of lung images and Computer-aided diagnosis method and elaborates lung segmentation,features extraction,and the design of SVM classifier.On the basis,this thesis presents the design of diagnosis system and its concrete implementation.The simulation of the proposed diagnosis system is based on the MATLAB platform.The test results show that the accuracy of the combination of the color and texture features is 87.6%,which ranks the highest among 10 different common features.Meanwhile,combining the proposed feature with CLD and GLCM can achieve an accuracy,approximately 89.32%,which is much higher than most of the existing tuberculosis systems.Therefore,the combination feature proposed in this thesis can provide doctors with diagnostic information helpfully,with the result of achieving the Computer-aided diagnosis of tuberculosis disease.