Classification Of Malignant Pulmonary Nodules Based On Convolutional Neural Network

Lung cancer is a catastrophic disease in the world,ranking first in cancer mortality.Research shows that early screening,accurate diagnosis,and active treatment can significantly reduce the mortality rate of patients with lung cancer.Therefore,early screening studies of lung cancer are of great significance.The early form of lung cancer is pulmonary nodules,which usually exist in three types: solitary type,vascular adhesion type and proximal pleural type.It usually behaves in circles or like circles in medical images.The lung tissue is complex,so it is difficult for clinicians and interpreting physicians to distinguish accurately between pulmonary nodules and thoracic blood vessels and bronchi.Vascular adhesion type and proximal pleural type are more difficult in the screening of lung cancer.In recent years,with the development of medical imaging technology,the introduction of computed tomography(CT)technology has made early screening for lung cancer possible.However,there are thousands of chest CT images,which seriously increase the workload of the radiologists.Convolutional neural network(CNN)is based on big data,and it does not require manual extraction of features and high degree of automation.It has become a new direction in the field of image processing in the near future.How to use the combination of convolutional neural network(CNN)and medical images to assist doctors in diagnosis has become a research hotspot in medical image processing.Therefore,this article will focus on the use of convolutional neural networks to achieve the complete process of radiologist readings.The specific idea is: After the patient has taken a chest CT image,the image containing the pulmonary nodule is first screened by a convolutional neural network;then sent to a full convolutional neural network for pulmonary nodule detection;And finally the detected pulmonary nodule is sent to a convolution.The classification of malignancy of lung nodules in neural networks can be achieved,which can reduce the burden of doctors in reading and assist the clinician in the early screening of lung cancer for the purpose of realizing reliable,stable and high-precision screening of lung cancer.The work of this paper is mainly focused on the classification of pulmonary nodules malignancy based on convolutional neural networks.Firstly,this paper reviews the research status and development of deep learning in the field of imaging.Then it studies the pulmonary nodule picture screening,pulmonary nodule detection and the classification of malignant degree of pulmonary nodules.It focuses on the study of the improved convolutional neural network to realize the classification of chest image dichotomy,pulmonary nodule detection and lung nodule classification,which obtains satisfactory results.The main contributions of this article are as follows:(1)For the two classifications of chest images,we compared a variety of convolutional neural network models through a large number of experiments,and finally proposed a convolutional neural network algorithm which is suitable for lung nodule image screening for chest CT images.The method not only achieves high-precision screening of pulmonary nodules without segmenting or compressing the size of the picture,but also has good results in the screening of pulmonary nodules with a size of less than 3 mm.(2)In the detection of pulmonary nodules,we firstly performed the extraction of the lung parenchyma and the edge repair operation to eliminate the noise interference outside the pulmonary parenchyma,which is helpful for our algorithm to achieve a higher accuracy of lung nodules location.In order to improve the utilization of the network,we shared with the above convolutional neural network the part of the convolution used to extract the target feature when we designed the localization network.Through experiments,the network can accurately locate lung nodules and achieve better positioning results.Combined with regional prediction network and target discriminant network,we designed a pulmonary nodule localization network,which can accurately locate pulmonary nodules through experiments,and achieve a good localization effect.(3)Aiming at the diagnosis of benign and malignant lung nodules,the paper put forward the idea of classification based on the degree of malignancy.By simulating the classification of pulmonary nodules by four radiologists,an integrated learning method is proposed to realize the classification of lung nodules by fusion of multiple neural networks,and good classification results were obtained.