Study And Implementation Of Vision Based Pulmonary Nodules Filtering Algorithm

In China,lung cancer is the first major cancer,lung cancer mortality accounts for more than 20%of cancer mortality.Lung cancer is usually characterized as lung nodules in the early stages,the cell in lung nodules turning cancerous may cause lung cancer.Therefore,the detection of early lung nodules is essential.However,early detection of pulmonary nodules is very difficult,the workload is high.And it need high degree of professionalism and is susceptible to the subjective judgement of doctors.Therefore,missed diagnosis and misdiagnosis often occur,especially in the areas with poor medical conditions,that early detection rate of lung nodules is low.Based on the thinking of computer-aided detection and through the analysis of the domestic and foreign lung nodule detection algorithm research status,an algorithm of early lung nodules detection based on machine vision was proposed in this paper.The algorithm is mainly divided into three parts,the extraction of lung parenchyma,the establishment of suspected nodules and false nodules discrimination.Among them,the extraction of lung parenchyma adopts the traditional image processing method and unsupervised learning method K-means clustering algorithm.The establishment of the suspected nodule sets used U-net network to segment the suspected nodules,and carries out the data enhancement method based on distortion,in order to improve the robustness of the algorithm.And VGG16 was used for binary classification,thus the false nodules were judged.There are three main points of innovation in this paper:1)Adopt the method of combining traditional image processing process and unsupervised learning K-means clustering algorithm to divide the lung parenchyma,and improve accuracy.2)Aiming at the different shapes of lung nodules,the data enhancement method based on distorted deformation is adopted to improve the accuracy and robustness of the algorithm.3)The identification of pulmonary nodules using the image semantic segmentation algorithm based on the U-net and the VGG16 classification algorithm is used to improve the accuracy of segmentation.In general,the algorithm can realize the detection of early pulmonary nodules,and has good accuracy and robustness.