Research Of Lung Images Feature Extraction And Classification Based On SVM

Lung cancer is the most common visceral malignant tumor disease, the morbidity andmortality of lung cancer are one of the highly risks to human health, while the number ofpatients with lung cancer in China is top one in the world. Compare to other cancers, thebiological characteristics of lung cancer is very complex, the early onset usuallyasymptomatic or just with mild symptoms, the pathogenesis time is short with highlymalignant and difficult to find, easy to transfer, but later lung cancer is difficult to cure.Therefore, detection and treatment are the major and key way to improve the survival rate oflung cancer patients in the early growth stages of tumor. Studies have shown that if lungcancer in the early was detected and treated timely, the patient survival rate may rose from14%to49%. In recent years, with the improvement of imaging technology, clinical medicinehas proven that low dose CT scanning is the most effective imaging method for earlydetection of asymptomatic lung cancer. These lung diseases manifests solitary pulmonarynodules in medical imaging, Therefore, the detection and identification of solitary pulmonarynodules is the most important means of lung diseases diagnosis.With the enhancement of computer software and hardware bases, the development ofdigital image processing technology, continue in-depth study of the pattern recognitionmethod, the computer-aided diagnosis system can provide strong support for detection anddiagnosis of lung cancer. The lung image processing algorithms can be carried out withcomputer, then it is used to detect the pathological changed characteristics, and help doctorsto find out the suspicious objects for further analysis and judge, on the one hand, it can greatlyreduce the workload of doctors and improve efficiency, on the other hand, it can makediagnostic imaging more objective and improve diagnostic accuracy. Therefore, usingcomputer to aided diagnosis pulmonary nodules, extract the characteristics of lung nodules,detect and distinguish the pulmonary nodules has great significance and research value.In order to realize accurately intelligent diagnosis of lung CT, noise reduction processing,feature extraction and classification is the key, therefore, this study focused on the threeaspects of noise reduction processing, feature extraction and classification for in-depthresearch. The specific research of this dissertation is deployed from the following aspects:(1) Pathological analysis of lung CT images: This dissertation primary introduces thecharacteristics of medical images, analyzes the medical signs of pulmonary nodules in lungCT images and introduces some of the medical knowledge of the pulmonary nodules.(2) Lung CT image preprocessing: due to the inevitable noise which collected in theprocess of Lung CT images, and the particularity of the Lung CT images’ imaging principle,edge blurring of the target images is appeared, in order to minimize the effect of subsequent feature extraction and classification of noise and edge blurring, the Lung CT images shouldbe preprocessed. This dissertation use maximum between-class variance method which canadaptive extract threshold to carry out binarization processing for the Lung CT images, thenuse binary image morphological method to change the image as the background pixel to0, thetarget pixel to1and the target area has no empty hole, finally carry out multiplicationoperation between binary image and original image. This method can determine theboundaries of the target image well and completely eliminate background noise.(3) Feature extraction: This dissertation extracts the characteristic values from multipleangles such as gray-scale feature, texture feature, morphological characteristic etc, and selectsrepresentative texture feature values based on Gray-leval co-occurrence matrix and Grayscale-gradient co-occurrence matrix, morphological characteristics based on invariant moments ascharacteristic data of subsequent classification for classification study of the Lung CT imagein the fifth chapter.(4) Classification and analysis of Pulmonary CT image based on support vector machine:This dissertation selects support vector machine which can well deal with the small sampleproblem as classifier,46normal lung CT images and83abnormal lung CT images werecollected from the203Hospital of PLA is classified in this experiment, The classificationaccuracy can reach89.3714%, and then the validity of the classification is verified.In summary, the proposed morphological methods is used to reduce noise and twoclassification of the lung CT image is carried out by the Support Vector Machine(SVM) inthis dissertation, while a better classification result is obtained. The future work should focuson the in-depth study of lung pathology, feature selected, classification algorithms optimizedand establish a perfect processing method of the lung images, these harder works is needed inthe future for further improving the classification accuracy.