Study On Classification Models Of Benign And Malignant Pathological Types Of Pulmonary Nodules Based On Artificial Intelligence

In recent years,pulmonary CT imaging information and pulmonary tumor markers(TMs)information have been widely used in the intelligent diagnosis of benign and malignant pulmonary nodules.Using the advantages of Artificial Intelligence(AI)in data mining,the potential information related to the pathological types of pulmonary nodules can be extracted from pulmonary CT images and pulmonary tumor markers to achieve good pulmonary nodules diagnosis and classification of malignant multiple pathological types.Artificial intelligence technology is helpful to reduce the work pressure of clinical medical workers,thereby further improving the clinical diagnosis efficiency and accuracy of pulmonary nodules.In the design and research of a multi-resolution 3D multi-classification deep learning model for the diagnosis of benign and malignant pulmonary nodules and multiple pathologies,firstly,the 3D multi-resolution method is used to completely extract the 3D volume data information of pulmonary nodules in the pulmonary CT images;secondly,using the dual path network(DPN)as the main network,a multi-resolution 3D multi-classification deep learning model for the diagnosis of multiple pathological types of benign and malignant pulmonary nodules is constructed;thirdly,the model training and verification are carried out on the four classification data sets of inflammation,squamous cell carcinoma,adenocarcinoma and benign other;finally,after experimental testing,the multi-resolution 3D multi-classification deep learning model constructed in this thesis achieves an accuracy(ACC)of 0.805,receiver operating characteristic(ROC)area under the curve(AUC)is 0.8755.In the study of machine learning multi-classification model using tumor markers to construct multiple pathological types of benign and malignant pulmonary nodules,9 pulmonary tumor markers of carcinoembryonic antigen(CEA),carbohydrate antigen 50(CA50),carbohydrate antigen 125(CA125),carbohydrate antigen 242(CA242),carbohydrate antigen 724(CA724),non-small cell carcinoma-associated antigen(Cyfra21-1),neuron-specific enolase(NSE),serum gastrin releasing peptide precursor(ProGRP)and squamous cell carcinoma antigen(SCCA)are used,a multi-layer perceptron(MLP)softmax multi-classification machine learning model for benign and malignant pulmonary nodules is constructed.By performing model verification analysis on the four-class pulmonary tumor marker data set of inflammation,squamous cell carcinoma,adenocarcinoma and benign other,the accuracy rate of 0.887 is obtained,and the AUC value is 0.97.In the study of the fusion of deep learning and machine learning to build multiple pathological diagnosis models of benign and malignant pulmonary nodules,take the weight transfer method in transfer learning,transfer the weights of the trained models in the multi-classification deep learning model constructed by pulmonary CT images and the machine learning multi-classification model constructed by tumor markers,and finally weight fusion is performed at the end of the two single-modal models.Finally,based on artificial intelligence technology,the fusion of CT modalities and pulmonary tumor marker information is used to construct a classification and prediction model of multiple pathological types of benign and malignant pulmonary nodules.After verification,the multi-classification network model combining deep learning and machine learning can obtain an accuracy rate of 0.906,and the AUC value is 0.95.Experimental results show that the classification of multiple pathological types of benign and malignant pulmonary nodules is feasible.In this thesis,the multi-classification model that combines CT image information,pulmonary tumor marker information,and pathological type and other modal information is more suitable for practical clinical applications.Compared with the classification model of multiple pathological types of benign and malignant constructed using CT image information or pulmonary tumor marker information alone,it has better performance,can effectively improve the clinical diagnosis efficiency of pulmonary nodules,and help doctors formulate the diagnosis and treatment plan.The research work is a groundbreaking and challenging work with great research significance.