Lung Nodules Detection And Classification Of Benign And Malignant Lings In CT Images Based On CNN Model

Purpose: Lung cancer is the most common primary malignant tumor of the lung,therefore,early diagnosis and treatment of lung cancer is a key tool to reduce the mortality rate of patients.Lung cancer is mainly manifested by lung nodules in its early stage,and an important part of lung cancer diagnosis is the identification of pulmonary nodules,based on which the differentiation of benign and malignant pulmonary nodules is also an important basis for staging assessment of nodules.Traditional detection methods are cumbersome,complicated for feature extraction and have average detection effects.In this study,we propose some improvement methods and improve the detection accuracy based on the existing detection algorithms for pulmonary nodule detection and benign-malignant diagnosis using convolutional neural network with CT images as the research object.Research Method: Using the 1018 patient example CT images included in the LIDC-IDRI public dataset,a lung parenchyma segmentation method with improved U-Net is proposed to deepen the overall depth of the network by Res-Net.In addition,a new boundary penalty term is designed in the loss function of lung parenchyma segmentation in order to improve the segmentation effect of the network on the edges of lung parenchyma.In addition,the backbone network of CNN is improved by using two different sizes of residual feature extraction networks,which makes the target detection algorithm more adaptable to detect lung nodules and make benign and malignant diagnosis.Results: Using multi-scale convolutional neural network to classify ROI,the final accuracy of pulmonary nodule identification reached 92.5% and the sensitivity reached95.2%.The final accuracy of the model is greatly improved by training it several times,which can better assist clinical diagnosis of benign and malignant pulmonary nodules and provide an effective basis for nodule staging assessment.Conclusion: For the problem of pulmonary nodule detection,a deep convolutional neural network is proposed for the identification of pulmonary nodules,and the structure of deep residuals and jump connections makes the extracted features more comprehensive and exhaustive,and successfully reduces false-positive nodules,which has basically met the requirements of clinical applications.For the diagnosis problem of benign and malignant pulmonary nodules,the identified pulmonary nodules are classified by convolutional neural network,the sample size is enriched by using data augmentation,the problems caused by positive and negative sample imbalance are solved by using improved loss function and Re LU activation function,and the variable storage performance of dense convolutional neural network is optimized to reduce the number of parameters to be trained.