Pulmonary Nodule Detection Algorithm Based On Adaboost And Faster-RCNN

As the environment worsens,lung cancer is the leading cause of cancer-related deaths worldwide.Pulmonary nodules are early signs of lung cancer.To make the diagnosis of pulmonary nodules contained in CT images remind doctors to focus on examination can reduce the doctor’s workload and make the diagnosis of digital and engineering,can effectively improve the diagnostic accuracy and efficiency of the work.Rapid and accurate detection of pulmonary nodules and their signs from CT images is of great significance and research value for the early diagnosis of lung cancer.Due to the shape,size and distribution of pulmonary nodules,even experienced doctors cannot make high-precision judgments.On the other hand,each patient.has hundreds of CT images.Therefore,the detection of pulmonary nodules has become particularly important and challenging.Based on the traditional method of pulmonary nodules detection and the hypothesis conditions are more complex computations cannot meet the needs of real-life pulmonary nodule detection technology.After analysis,we chose Adaboost-based pulmonary nodule detection algorithm under a small sample.The method allow learning in a supervised manner,only based on pulmonary nodule image block extracted from the original CT images without access to around-information annotations.We have written a GUI interface for doctors to edit and annotate so that doctors can label the CT images containing pulmonary nodules to solve the problem of lack of medical image labeling.Then set up a network of pulmonary nodules detection by selecting the optimal classifier through multiple network structures and parameter settings.Apply LBP features involving rotation invariance of Image to Training Classifier.In order to validate our approach,we use a synthetic database to mimic the task of detecting pulmonary nodule automatically from CT images-as commonly encountered in automatic detection of medical images applications and show that classifier can automatically detect pulmonary nodules from the lungs CT images accurately.The algorithm optimizes the traditional pulmonary nodule extraction algorithm and improves the detection accuracy.The algorithm of pulmonary nodule detection based on Adaboost is very good for the small sample detection and the failure of big data.Therefore,this paper proposes the Faster-RCNN-based pulmonary nodule detection algorithm under big data.First set up a deep learning environment,and then set the data interface so that it matches the network interface of Faster-RCNN.Secondly,set up the single-class classification network of Faster-RCNN and parameters to train different models(including ZF and VGG models).Using different models to test different databases.Pulmonary nodule detection based on Faster-RCNN is highly accurate.The two methods presented in this paper evaluate the algorithm in terms of detection accuracy and robustness to noise.The proposed pulmonary nodule detection technology has good engineering application value.