Research On Deep Learning Based On Pulmonary Nodule Detection Algorithm In CT Image

Lung cancer is the type of cancer with the highest morbidity and mortality in the world,which affects human health severely and pulmonary nodule is an important manifestation of it.It is very important to accurately find the location of pulmonary nodules in CT images and label them.In recent years,with the increasing clinical demands,doctors need to spend a lot of time and effort reading and analyzing CT images.Long-time and intense working can affect the accuracy of diagnosis.Therefore,the research on automatic detection of pulmonary nodules has important significance and practical application value.Compared with traditional methods,methods based on deep learning do not require manual feature extraction,this provides a new solution for the automatic detection of medical images.Based on the method of deep learning,this paper designs and implements a new detection framework for pulmonary nodules in CT medical images.The main research contents of the paper are summarized as follows:1.To solve the problem of large distribution of pulmonary nodules,a new dual channel module based on hierarchical split structure and dense connection structure is designed.It can realize multi-scale feature extraction and fusion within the network unit to enhance the transmission and reuse of multi-layer features.In view of the mutual interference between different tasks in the detection of pulmonary nodules,a spatial decoupled parallel module is proposed to make each branch focus on its specific task,avoids mutual interference.Based on the above main modules,this paper constructs a new multi-scale lung nodule detection network.2.In order to solve the problems of multiple types of pulmonary nodules,limited available characteristics,and easy to confusion of vessels and organs with similar features,based on attention mechanism,we further design a new feature extraction module and a cross stage enhanced fusion module,and combine the two modules to propose an improved pulmonary nodules detection network based on attention mechanism.Firstly,the attention mechanism is used to improve the original module to achieve adaptive learning of the importance of feature mapping patterns and optimize the feature extraction process of the network.Secondly,the enhanced fusion module based on soft attention gating is used to guide the effective aggregation of different semantic features in the encoder and decoder path.By embedding the new module into the pulmonary nodule detection network in an effective way,the performance of the proposed network is improved.3.In view of the deficiency of bounding box generation strategy in pulmonary nodules detection,the adaptive anchor generation mechanism based on k-means clustering algorithm is integrated into the algorithm framework to optimize the selection process of anchor.Experimental results show that the pulmonary nodules detection network proposed in this paper can greatly improve the effect of detection,and has achieved excellent results on the medical image dataset LUNA16.