Research On Lung Parenchyma Segmentation And Lung Nodule Detection Based On CT Images

Lung cancer is currently one of the most common malignant tumors due to the highest rate of incidence and mortality all around the globe,the main form of early lung cancer is lung nodules.The best way to improve the cure rate and increase survival time of lung cancer patients is early detection and timely diagnosis.With the continuous development of computed tomography and especially the application of multi-detector-row computed tomography technology,every tester has 400 to 500 scanned images.Although these thin lung slice scan images can improve the detection rate of lung nodules and reduce the missed diagnosis of small nodules,the review of a large number of CT images may cause subjective misdiagnosis of radiologists,which leads to an increase in the cost of later treatment.In order to reduce the rate of small nodules misdiagnosis,computer-aided diagnosis systems are used to detect lung nodules to help radiologists diagnose.Two key technologies in computer-aided diagnostic systems are lung parenchymal segmentation and lung nodule detection.In the medical field,CT images have ultra-high resolution so that various organs of the body can be clearly observed.However,since CT images are gray-scale images,it is also difficult to distinguish blurred gray-scale between tissue edges.Factors such as the differences in samples can be influenced by testers,test methods and test environment.Noise and artifact also have some effects.These all make the generalization of the algorithm in the auxiliary diagnosis system hard and the accuracy is difficult to improve.And the nodules are unconventional organizations,its characteristics are complex and variable.It is necessary to design more complex algorithms for noudle detection.In this paper,experiments and researches were performed on lung parenchymal segmentation and lung nodule detection,based on the image features of chest CT images and the medical imaging findings of lung nodules.The main work of this paper includes:(1)The extraction of lung parenchyma using convolutional neural network.Through the establishment of the Mask-RCNN model for training,learning and testing based on sample datas,it is verified that the deep learning method can be practically applied to the lung segmentation task.By comparing the lung parenchymal segmentation method based on convex hull algorithm and the anthor method based on fuzzy modeling idea and iterative relative fuzzy connection algorithm,it is verified that the advantage of deep learning method over traditional algorithm is faster segmentation speed.And deep learning method has better robustness and greater space for improvement.(2)Detection of pulmonary nodules using convolutional neural networks.Because of the diverse forms of pulmonary nodules,traditional algorithms for the detection of pulmonary nodules are often based on specific morphological features of the lung nodules,such as solitary nodules,ground-glass nodules.The traditional algorithm is difficult to generalize the detection of nodules,and it is often directly misdetected for larger noudles.Therefore,the deep learning model is used to extract the characteristics of lung nodules.In this paper,the R-FCN model is used to detect the lung nodules.The position-sensitive score maps of the R-FCN is used to express the position information of the nodules to improve the accuracy of nodule detection.And by comparing with other network models,the experimental results show that the model used in this paper has a better effect in the lung nodule detection task.In this paper,the researches based on conventional medical images are carried out.The task of lung parenchymal segmentation and lung nodule detection is completed.It has good theoretical value and engineering application value in practical application.