Research And Application Of Pulmonary Nodule Detection Based On Faster R-CNN

At present,the morbidity and death rate of lung cancer rank the first both in China and abroad.The incipient symptoms of lung cancer are circular or irregular pulmonary nodules with a diameter of 30 mm or less.Early diagnosis and treatment of lung cancer is the most effective way to reduce its death rate.As a result of various shapes,sizes and uneven distribution of pulmonary nodules,even professional imaging physicians find it hard to quickly and accurately diagnose the lesion area.Meanwhile,as each patient has a number of CT images,the use of computer technology to automatically extract lung nodules in CT images is a much effective way.It can remind doctors to focus attention on particular place,reduce the workload of doctors,and improve accuracy and efficiency of diagnosis.In summary,the rapid and accurate detection of lung nodule location and its symptoms from CT images is of great significance for the early diagnosis of lung cancer.Traditional pulmonary nodule detection methods based on image segmentation need a lot of assumptions but have low accuracy,which cannot meet the needs of clinical application of pulmonary nodule detection tasks.This paper presents an improved and optimized Faster RCNN algorithm for detecting pulmonary nodules.Firstly,the environment of in-depth learning is built,and then a data set which can be trained by Faster R-CNN model is produced using LIDC-IDRI database as a source to preliminarily verify the feasibility of this method,and the experimental results are analyzed and evaluated.However,there is still room for improvement in the detection accuracy of this method,which can be further optimized to improve the detection accuracy.The method of improving optimization starts with the modification of parameters and the improvement of network structure.Firstly,theoretically analyzing the feasibility of optimization,then taking comparing analysis of the result by a large amount of experiments.Finally,the two methods which are popular in object detection are compared horizontally.Compared with other popular methods,the optimized algorithm proposed in this paper improves precision of detection by more than 20%.It is verified that this method has good theoretical value and engineering application value in the field of lung nodule detection.Finally,based on the improved algorithm,a pulmonary nodule assistant detection system is built by Python’s Web framework Django.The front framework uses Bootstrap and jQuery’s CSS pattern to set up the page layout,which makes it convenient for doctors to use it in an easy way.