Research On Image Segmentation Of Lung Nodules Based On U-Net

Lung medical images are an important auxiliary tool for lung cancer diagnosis and treatment.The main manifestation of lung cancer in early medical images is pulmonary nodules.Therefore,screening and segmentation of pulmonary nodules is an effective way to diagnose lung cancer.Countless pulmonary medical images increase the workload of radiologists.The development of artificial intelligence technology provides a new direction for assisted artificial diagnosis and makes important progress in the diagnosis and treatment of pulmonary nodules.In order to further improve the efficiency and accuracy of pulmonary nodule diagnosis by using artificial intelligence technology,a segmentation algorithm of pulmonary nodule image based on U-Net is proposed.The main work is as follows:First,to solve the problem of incomplete lung nodule images,cracks at image mosaics and long image mosaics in the process of image mosaics,an image mosaics algorithm based on the combination of oriented fast and rotated brief(ORB)and random sample consensus(RANSAC)is proposed to eliminate the cracks at the mosaics of lung nodule images,The accuracy of pulmonary nodule image mosaic is improved.Second,the size and shape of the pulmonary nodules in the pulmonary nodule image are different,and the pulmonary nodules in the pulmonary nodule image may have similar shape characteristics with the lung related tissues,so it is difficult to ensure the accuracy of the pulmonary nodule image segmentation.In order to solve the above problems,three methods,namely,maximum variance between classes,Canny edge detection and k-means clustering,are used to extract the lung image contour from the lung image,and the lung parenchyma extraction algorithm is used to segment the lung image to reduce the error in the segmentation of the lung nodule image.Thirdly,aiming at the inaccurate segmentation of the U-Net network structure in the process of lung nodule image segmentation,a U-Net lung nodule image segmentation algorithm with dual attention module is proposed.The algorithm adds spatial attention module and channel attention module on the U-Net network structure,uses self attention mechanism to analyze the global context information from the perspective of time and space,strengthens the characteristics of the pulmonary nodule image,overcomes the interference of different factors on the segmentation results,and improves the segmentation effect of the pulmonary nodule image.