Pulmonary Nodule Detection Method Based On Deep Learning

Lung cancer is one of the most malignant tumors with the fastest growth rate of morbidity and mortality in all diseases and the greatest threat to human life and health.The incidence and mortality of lung cancer in men accounted for the first place in all malignant tumors,and the incidence rate among women was the second,with mortality accounting for the second place.According to the survey,only 19.7%of patients survived more than 5 years after lung cancer was diagnosed in China.Lung cancer is mainly caused by malignant pulmonary nodules.Early lung nodule detection and diagnosis technology can prevent lung cancer in time and improve the survival rate of lung cancer,which is of great significance.In recent years,artificial intelligence technology has developed rapidly.Deep learning technology has achieved breakthrough application in text,image,speech and other fields.This paper is mainly based on deep learning technology for lung nodule detection and recognition methods.The main contributions are as follows:1.An adaptive threshold adjustment strategy is proposed for image binarization.In the data pre-processing process,the lung parenchyma needs to be segmented,which reduces the interference of environment like the machine tool,the chest wall and bones in the original picture.Segmentation of the lung parenchyma is mainly based on the CT value of different tissues in the medical image to binarize the image.The traditional binarization method uses the CT value of the lung parenchyma as the overall threshold,but the fixed threshold in different environments does not work well.Through the dynamic adjustment strategy,each image is adjusted to find the most suitable threshold,so that the final segmented lung parenchyma is more complete.2.A 3D-DenseUnet convolutional neural network model is proposed for regional detection of pulmonary nodules.The 3D-DenseUnet model consists of two parts:a contracting path to obtain context information and a symmetric expanding path to pinpoint the location information of the target.The 3D-DenseUnet network model uses a special dense connection mechanism.Through dense connections and batch normalization operations,the network can detect more lung nodules,and the detection sensitivity of the model is significantly improved.3.A residual convolutional neural network model(RCN)is proposed for the classification of benign and malignant pulmonary nodules.The model is based on the residual connection to construct the convolutional layer to prevent the distortion caused by the data transmission in the deep network.It can accurately classify the candidate lung nodules after detection,and effectively reduce the false positives of candidate lung nodules.4.Using the above model and method,an automatic detection system for lung nodules was designed and implemented.The system is an application for automated data processing and diagnosis of pulmonary nodules for chest medical image data.It has the characteristics of strong expansion.In this paper,by constructing a neural network model,a method for detecting lung nodules from medical image data is proposed.The experimental results show that this method has great advantages in detecting and filtering false positives in lung nodules.The method can effectively assist the doctor in early diagnosis and treatment of lung cancer,and improve the survival rate of the patient.