Research And Application Of Pulmonary Nodule Detection Method Based On Deep Learning

Lung cancer is a malignant disease with high morbidity and mortality.As an early symptom of lung cancer,the appearance of lung nodules indicates that lung cells are gradually becoming cancerous.Lung nodules are generally oval or irregular spherical,and lung nodules of different sizes,shapes,density distributions,and growth rates correspond to different lung cancer incidence rates.Therefore,for the early diagnosis of lung cancer,lung nodule detection is an effective method.Methods.The design and optimization of pulmonary nodule detection methods using computers has always been the direction of continuous exploration by experts and scholars in various fields.With the advent of deep learning,deep learning-based pulmonary nodule target detection methods have also been developed.Compared with traditional methods,the deep learning-based pulmonary nodule detection method has greatly improved the accuracy and speed.However,due to the network structure,there are still problems such as low sensitivity and low accuracy for small pulmonary nodules.Especially for adhesive pulmonary nodules,it is difficult to have accurate detection results,so it needs to be improved.In response to these problems,based on Faster R-CNN,a special context-aware network is designed to detect lung nodules.Firstly,the LUNA16 dataset is used to preprocess the input data,and a new data enhancement method is proposed to solve the problem of unbalanced data distribution.Secondly,this thesis improves the network model.By optimizing parameters and optimizing network modules,several longitudinal comparison experiments are carried out,and the network parameters and modules with the best effect are selected to make them more suitable for pulmonary nodules.detection work.Thirdly,using the average precision rate as the evaluation index,by comparing the results of other single-stage detection methods commonly used in target detection with other better methods in recent years,the advanced nature of the method used in this paper is verified.The experimental results prove that the method proposed in this paper has the characteristics of low false alarm rate and high accuracy rate.Although the detection time is slightly longer,it still has high practicability,which provides a theoretical basis for the problem of low detection accuracy of small pulmonary nodules.Finally,based on the algorithm used in this thesis,an automatic detection system for pulmonary nodules is developed.The system uses the web framework Django developed based on python.Doctors can observe the location of pulmonary nodules through simple operations in the system,and the platform can manage the patient’s information in real time,which is convenient for doctors to follow up treatment.