Research On Lung Tissue Segmentation Method For Pulmonary Function Evaluation

Lung diseases,such as chronic obstructive pulmonary disease,emphysema and lung cancer,seriously endanger human life and health.Quantitative assessment of pulmonary function is of great significance for the diagnosis and treatment of lung disease.The pulmonary function evaluation system based on lung tissue segmentation can evaluate the lung function status of the whole lung,single lung,and single lung lobe.And it can clearly show the scope and location of the lesion area.It is of great significance for the accurate assessment of the degree of lung tissue damage,the development of lung volume reduction surgery plan,and the evaluation of the efficacy after surgery.Due to various factors such as motion,volume effect and offset field effect in the lung CT imaging process,the problems of gray overlap,blurred boundaries,and difficulty in separation between lung tissues are caused.Most of the existing lung tissue segmentation methods either ignore the effective anatomical structure information and result in low segmentation accuracy,or have the high calculation complexity that result in limited clinical application.To solve these problems,this paper deeply studies the lung tissue segmentation method for pulmonary function evaluation,and designs a pulmonary function quantitative evaluation system based on lung tissue segmentation,which verifies the effectiveness of the lung tissue segmentation method.The research content of this article is mainly divided into the following aspects:(1)In order to accurately extract relevant lung tissue for quantitatively assessing lung function,this paper studies the segmentation of lung parenchyma,bronchus,and blood vessels based on threshold segmentation and region growing algorithm.First,coarse segmentation of lung tissue is performed based on global threshold algorithm and three-dimensional region growing algorithm.Next,an improved region growing method for bronchus segmentation is proposed to solve the problem that bronchiole are difficult to extract and leakage is easy to happen,.Then,left lung and right lung are separated based on integral projection method,and the lung parenchyma is repaired based on the morphological closed operation.Finally,to solve the problem that high density tissue in the lung parenchyma is easy to interfere with the extraction of blood vessels,a pulmonary vessel segmentation method is proposed based on threshold algorithm and connected domain algorithm.(2)To solve the problem that the pulmonary fissure is not easy to detect due to low resolution and fracture,a fissure segmentation method is proposed based on Hessian filter.First,the suspected lung fissure voxels are extracted based on the Hessian filter.Then,the lung fissure points are finely extracted based on the connected domain algorithm,which is based on the inner product of the normal vector.Finally,the fracture points of the lung fissure points are connected based on the morphological closed operation.(3)To solve the problem of lobe bronchus segmentation failure,which due to false branches and multiple bifurcations in the bronchial tree skeleton,a lobe bronchus segmentation method is proposed based on bronchial tree model.First,bronchial tree skeleton is extracted based on the three-dimensional parallel thinning algorithm.Then,the bronchial tree model is established based on Depth First Search algorithm.Finally,the bronchial tree is labeled based on the nearest common ancestor search algorithm for multi-branch tree.(4)To solve the problem that lung lobe segmentation is difficult,which due to the deformable lung tissue and incomplete lung fissure,a lung lobe segmentation method is proposed based on supervised learning and marker-controlled watershed algorithm.First,distance features from bronchus,blood vessels and lung fissures are extracted based on Euclidean distance transformation.Next,to obtain a fusion feature map,a feature fusion model is constructed based on logistic regression.Then,lung lobe initial markers are generated based on lobe bronchus and morphological operations.Finally,lung lobes are segmented based on marker-controlled watershed algorithm.(5)To solve the problem that the cumbersome process of the traditional lung lobe segmentation method,a lung lobe segmentation method is proposed based on 3D convolutional neural network.First,the basic three-dimensional U-shaped network structure and cross-entropy loss function are applied.Then,the network structure is improved by joining residual convolution module,dilated convolution,and the Dice loss function combined with the boundary penalty term.(6)In order to verify the effectiveness of the lung lobe segmentation method in this paper,this paper independently developed a pulmonary function quantitative evaluation system based on lung tissue segmentation and verified the effectiveness of the lung lobe segmentation method by experiments.The experimental results show that the lung tissue segmentation method in this paper can effectively achieve lung tissue segmention,such as lung parenchyma segmentation,lung bronchus segmentation,pulmonary vessel segmentation,lung fissure segmentation,lobar bronchus segmentation and lung lobe segmentation.The system developed in this paper can be applied to lung function evaluation and the clinical auxiliary diagnosis of lung diseases,such as chronic obstructive pulmonary disease,emphysema and lung cancer.