Classification Of Benign And Malignant Pulmonary Nodules Based On Convolutional Neural Networks

The latest domestic statistics show that the most serious malignant diseases in terms of morbidity and mortality are lung cancer in China at present.Lung nodules have complex morphology,and the classification based only on manual experience is prone to miss and misdiagnosis;therefore,it is important to develop a deep learning model for CT images to realize the automatic auxiliary classification and diagnosis function for benign and malignant lung nodules.The main research of this paper is as follows:Firstly,an overview of the social background of the research and its significance is given,the current status of domestic and foreign research in the related task area is analyzed,and the main problems in the current task of classifying benign and malignant pulmonary nodules are analyzed,and the research objectives and directions are expressed.Secondly,to address the problems of small size of medical image dataset and unbalanced sample categories,the lung nodule image data samples are collected by center-nearest neighbor slicing on real lung CT images,and some image data enhancement work is performed to expand the dataset capacity.Thirdly,a two-dimensional convolutional neural network model based on multiscale feature extraction is designed for automatic classification of benign and malignant lung nodules to address the problem of low efficiency of traditional manual feature extraction.The focal loss function is partially improved to enhance the training effect of the model.Finally,to address the problem that 2D convolutional networks are difficult to fully learn 3D features of lung nodules,we design a Two-branch Multi-dimensional Information Fusion Net(TMF-Net)for the classification of benign and malignant lung nodules,and use3 D convolutional networks to supplement 2D convolutional networks to extract feature information of lung nodule targets in different dimensions and fuse them to maximize the retention of 3D spatial features of lung nodules.Meanwhile,an attention mechanism is introduced in the two-branch network to improve the model’s attention to important features and further strengthen the feature learning ability of the model.The experimental results confirm that the model has good performance in the task of classifying benign and malignant lung nodules.