Method And System Implementation For Texture Feature Analysis Of Lung CT Images

The lung is one of the important organs of the human body,and the treatment of lung diseases is a hot spot for clinical application research.Given the anatomical characteristics of the lung,the use of medical imaging data is now an important way to diagnose lung diseases,and computed tomography(CT)images have brought broad prospects for clinical applications due to their image quality and other factors.However,in the diagnostic process,when the amount of data is too much the diagnostic results are often affected by the diagnostic level of the doctor,the doctor’s visual fatigue and the individual differences of the patient,which makes it difficult to guarantee the objectivity of the diagnostic results.Therefore,with the support of efficient feature analysis methods,extracting key information from medical images to characterize regions of interest(ROI),and introducing feature analysis based Radiomics into the field of medical image assisted diagnosis,it is expected to overcome the above shortcomings.For the collected plateau lung CT images,accurate segmentation of ROI and feature analysis are carried out to explore the degree of difference in image pixels reflected by the pathological changes of pulmonary tuberculosis in CT images.Based on this,the specific work of this paper is as follows:(1)A method of lung parenchyma segmentation based on the U-Net model is proposed to address the problem of complex lung CT image data in highland people,and the problems of left and right lung adhesions,gross contours and blurred boundaries that are very likely to arise when segmenting lung parenchyma.In the pre-processing part of the lung CT image data,the Otsu method combined with mathematical morphology is used to complete.Firstly,the Residual block is introduced in the coding stage to replace the normal convolution with the residual convolution to improve the accuracy and speed of network convergence;secondly,the Augmented Attention Module is introduced at the jump connection to enhance the features of the lung substance contour information to improve the segmentation accuracy;finally,considering the processing efficiency of the network,the number of network layers is optimized,and the depth of the network model is reduced to 4 layers to ensure the unity of segmentation quality and efficiency as much as possible.The correctness and validity of this method are verified using the publicly available dataset LUNA16,and the experiments show that the selected evaluation indicators have been improved to varying degrees.Applying the trained network to the lung parenchyma segmentation of CT images of highland population,the proposed method has some advantages over other existing methods.(2)To address the problem of poor accuracy of automatic diagnosis relying on morphological or geometric features in the identification and diagnosis of pulmonary tuberculosis in CT images,a method of sensitive feature analysis of pulmonary tuberculosis based on texture features is proposed.Firstly,multi-level feature extraction based on Py Radiomics model to extract ROI feature data of lung CT images with texture features as the main feature and perform comprehensive analysis;secondly,the selection and optimization of the original feature data is carried out,combining Mann-Whitney U test and LASSO algorithm to obtain a set of feature vectors that can comprehensively describe the image information;then,multiple classifier models are trained to compare and select the model with the best evaluation index;finally,the SHAP model is introduced to interpret and evaluate the classifiers to investigate the highly sensitive features and the degree of influence that affect the determination of CT images of pulmonary tuberculosis in the data of this study.The experimental results show that this method gives the differences in features between CT images of pulmonary tuberculosis and normal lung CT images through more comprehensive and precise characterization,and precisely obtains the degree of differences in pulmonary tuberculosis pathological alterations reflected in image pixels,which can provide a quantitative aid for diagnosis in actual clinical pulmonary tuberculosis pathological examination.(3)A lung CT image segmentation and feature analysis system is designed and developed,which mainly contains two functional modules: lung parenchyma segmentation and image feature analysis.The former module realizes the automatic segmentation of lung substance of CT images,including image import,pre-processing,segmentation,image saving and other functions;the latter module realizes the extraction of image features after segmentation and analysis operations,including image import,feature extraction,feature optimization,feature analysis,data visualization,data saving and other functions.The above functions have passed system testing and are user-friendly to a certain extent.