Research Of Lung Image Classification Based On Deep Learning

Lung cancer is one of the most common malignant tumors in China and the world today,and it is the leading cause of cancer death.With the continuous advancement of urbanization and industrialization,the problem of air pollution has become more and more serious,and the incidence of lung cancer has continued to rise due to the high smoking rate among residents.The incidence of lung cancer continues to rise.Therefore,reducing the morbidity and mortality of lung cancer is a major problem that needs urgent solution in China and the world at present.The best treatment period for lung cancer is early,but early lung cancer often has no clinical symptoms and is difficult to detect.At this stage,low-dose spiral CT can be used to collect lung images from people over 45 years old,which can better achieve early screening of lung cancer.In early screening,the main manifestation of lung cancer on the CT image of the lung is the pulmonary nodule.The doctor can diagnose the patient's disease based on the lung nodules in the CT image.However,the amount of data in CT images is huge.It is a painstaking task for doctors to screen lung nodules from a large number of CT image data,and it is easy to miss the lung nodules due to fatigue or lack of experience.And computers can help people with a single,repetitive and large amount of work.Therefore,using computers to assist doctors in detecting and classifying lung images is beneficial to help reduce the workload of doctors,reduce the leakage of lung nodules,and improve the efficiency of lung cancer diagnosis and treatment.Traditional computer-aided diagnosis of lung images often fails to achieve high accuracy,so it is not good for doctors to diagnose the classification of lung images.In recent years,with the excellent achievements of deep learning in image processing,it has become a general trend to apply deep learning algorithms to medical image processing,and has achieved good results.In this paper,the deep learning algorithm is used to perform the nodule detection and classification of lung images.The experimental comparison shows the effectiveness of computer-aided diagnosis using deep learning method.The deep learning method adopted in this paper is based on the improvement of Deep Lung system.By analyzing the problems existing in Deep Lung,this paper proposes a lung image classification algorithm based on deep learning.This paper implements a lung parenchymal extraction algorithm.The detection network uses Deep Lung's 3D DPN Faster R-CNN network with U-Net structure.Aiming at the problem that Deep Lung does not effectively utilize the semantic label of U-Net output and the fixed input size of classification input,this paper proposes an algorithm that effectively utilizes semantic label information to guide the original CT image by detecting the size and position of the nodule output of the network.The cropping makes the nodule information useful,and then uses the semantic information to process the cropped image.Then,through the spatial pyramid pooling,the input size is unified.Finally,the 3D DPN is used to further extract and classify the input features to obtain the final classification result.This paper uses LIDC-IDRI and LUNA16,two lung image public data sets,for training and testing of deep learning networks.Since the lung image often contains many tissues that are not related to the lung leaves,it will affect the detection effect of the network.Therefore,the image is preprocessed before network training.In the experimental results,the algorithm in this paper achieved good detection accuracy in nodule detection,achieved 91.04% classification accuracy in nodule classification,and obtained 82.06% accurate classification in patient-level classification diagnosis,close to the accuracy rate marked by the doctor.The deep learning algorithm used in this paper can be used for computer-aided diagnosis of lung images.