The Study And Application On Lung CT Image Processing Based On Ensemble Classifier And Random Forest Algorithm

Object: In recent years,accompanied by foggy weather continues to increase,interstitial lung disease prevalence and mortality continued to rise,early clinical symptoms are not obvious,only 30% of patients with pulmonary biopsy can detect symptoms of interstitial lung disease.Although the development of interstitial lung disease is easier to diagnose late,but has long been the significance of early diagnosis.Pulmonary CT(Computed Tomography,CT)on lung tissue and interstitial more detailed display of its morphological changes,especially in the determination of interstitial lung disease such as lung lesions in the lung disease has a unique diagnostic value The In clinical diagnosis,due to the parenchymal part of the lung and other trachea,bronchial tissue adhesion,resulting in the determination of the lesion area when there is a certain amount of fuzzy information.Combined with image preprocessing,feature extraction,classifier and other related algorithms to achieve the classification of lung CT images and lung parenchymal division,for the diagnosis of the disease to provide more effective information on the disease,is conducive to follow-up treatment of interstitial lung disease.Methods:This study is aimed at different types of lung disease in CT images similar pathological features,the gradient information of CT images of lungs was extracted by PHOG,to take the "voting" of the integrated thinking training classifier into a strong classifier to build a robust classification model,and then to the complex lung CT images to achieve sick and healthy two different types of identification of the effective distinction.Secondly,the study of lung CT images of lung parenchyma and non-pulmonary parenchymal parts of the existence of unclear boundaries of the problem,the use of lung CT image texture changes large,gray contrast obvious features,using the Gray Level Co-occurrence Matrix Algorithm to Obtain the Texture Feature and Combine the Gray Feature to Form the Characteristic Matrix,select random forest as a classifier,a segmentation algorithm of super-pixel and random forest is proposed to realize the exact segmentation of pulmonary parenchyma.Results: To test the performance of the algorithm model,the University of Geneva was selected as an open interstitial lung disease database for ILDs.Experimental results show,the accuracy rate of lung CT image classification model based on integrated classifier algorithm is 94.55% and the sensitivity is 86.44%,and the ideal classification result is obtained;based on the random forest of lung CT image segmentation algorithm in healthy lung CT image accuracy rate as high as 99.09%,lung fibrosis,hair glass,emphysema,pulmonary nodules sick image segmentation accuracy of more than 90%.Conclusion: In the classification of lung CT images,this paper presents a classification model based on integrated classifier,can achieve high accuracy of lung CT image health and prevalence of two types of classification,and robustness is better.Although the sensitivity of lung CT image classification algorithm model still needs to be further improved,it is of practical significance to determine the clinical diagnosis and treatment of pulmonary disease.In the aspect of pulmonary parenchymal division,proposed a method based on random forest classifier,can accurately and efficiently achieve the different types of pathological features of lung CT images of the lung parenchyma segmentation.In the severe lung CT image segmentation and algorithmic operation efficiency need further study,it has a good application prospect for the further work of detecting and quantifying CT images of lungs.