Research On Multi-Strategy Lung Nodule Detection And Classification Algorithm Based On Deep Learning

Lung cancer is one of the fastest growing cancers in the world.Up to now,the number of new cases and deaths of lung cancer rank first.Its incidence rate and mortality are still rising.In general,it is difficult to detect lung cancer in the early stage,and if appropriate treatment measures are not taken in time,it will seriously affect the survival rate of patients.The early sign of lung cancer is pulmonary nodules.Accurate medical treatment can effectively diagnose and save the lives of patients.Most of the pulmonary nodules have different shapes and volume changes,and have complex pulmonary environment,so the detection of pulmonary nodules is difficult.At the same time,in order to reduce the workload of radiologists,CT images should be used as much as possible.Therefore,early detection and benign and malignant diagnosis of pulmonary nodules are particularly important.The main research work of this paper is as follows:（1）A U-Net lung nodule detection model and algorithm（Dilated Residual U-Net,DRUNet）based on residual and dilated convolution module is adopted to obtain high detection sensitivity.In the detection phase,the residual learning module is added after the convolution layer of U-Net network,which can effectively prevent the problems of over fitting and gradient disappearance in the process of network training.In order to improve the convergence speed and performance of network training,a batch normalization layer is added to the convolution unit.In addition,dilated convolution（expansion convolution）is introduced,which can reduce the feature loss,increase the visual field perception domain of lung nodule image,enhance the ability of target detection to analyze the image background and target distribution,improve the accuracy of target detection,and make improvements in the network architecture and design ideas.Finally,appropriate loss function and other parameters are selected to significantly improve the detection performance of the network.（2）A lung nodule classification model and algorithm（Multi-Stream Res Ne St,MSRes Ne St）for pulmonary nodule classification based on the fusion of multi-stream and attention mechanism is proposed.In the classification stage,the multi-stream convolution network multistream CNN used for fine-grained classification and the residual network Res Ne St introduced spatial attention mechanism are combined to further improve the classification accuracy of lung nodules and realize the classification of malignant grade of lung nodules.The MS-CNN network adds a multi-stream mechanism,better integrates the ability of context information to learn image features and extract features,and can better obtain image features;The attention mechanism can extract image features to the greatest extent,the network performance is significantly improved,and the learning rate and loss function are improved.The experimental results show that the spatial attention mechanism can also significantly improve the classification effect of the proposed algorithm.In this paper,the models and algorithms of lung nodule detection and classification in the field of medical images are studied.The experimental results verify the advantages and disadvantages of the proposed method in various aspects.By comparing other algorithms,good detection sensitivity and classification accuracy are finally obtained,which provides an effective method for doctors to detect and classify lung nodules,so as to realize the adjuvant treatment of lung cancer.