Research On Diagnosis Model Of Solitary Pulmonary Nodules Based On Ct Images

In recent years, due to the worsening environmental pollution and other reasons, lung disease, has become one of the factors affecting people's health, of which the mortality rate of lung cancer is cancer ranks the first place. Early detection, early diagnosis and early treatment to improve lung cancer survival rate is an important tool. Early lung disease usually presents in the imaging of solitary pulmonary nodules, the CT image through the examination of the lung nodule detection and identification has become an important way to diagnose lung disease. Types of lung disease due to pathological changes in the diversity and complexity, how to improve the qualitative diagnostic accuracy of CT has become a problem we face. Computer-aided diagnosis of lung disease detection and diagnosis significantly reduces the workload of physicians, improve the work efficiency, to make the diagnosis more objective-oriented, and improve diagnostic accuracy.In this paper, spiral CT lung scan images for the object, to lung CT images of medical signs for the starting point, use of computer diagnostic technology to identify benign and malignant pulmonary nodules and to study its clinical diagnosis of secondary significance.Firstly, we selected by surgery or puncture the lung nodules examined samples of a total of 193 cases of the image, based on the characteristics provided by physicians in conjunction with previous research data of 21 optimized CT imaging features of pulmonary nodules as benign solitary pulmonary nodule Diagnosis of malignant classification model input vector.Secondly, in order to study the effect of computer diagnostics, we use the current commonly used neural network model is constructed of a three-layer BP networks, experiments showed that diagnostic accuracy rate was 71.5%. Also for the relatively small number of samples of benign nodules in this situation, we chose to study better for the small sample of support vector machine model, and through parallel web search method to find the optimal parameters of the experiment come to its diagnostic accuracy rate of 68.9%. Comparative diagnostic accuracy rate of 80.3% doctor's diagnose, can be seen from the diagnostic accuracy of computer diagnosis of physician diagnosis of reference value.Finally, through statistical comparison of two diagnostic methods for the classification results, the specificity, the support vector machine with neural network was significantly higher than 79.6% of 34.7%. Drawing through the ROC curve comparing neural networks and support vector machine model of the classification results, the area under the curve of 0.714 larger than the support vector machine neural network,0.655, showing that support vector machines better. Results show:Computer diagnostic significance for physician-assisted diagnose, can help doctors diagnose play the role of two other model is more reliable diagnosis of support vector machines.