Research On Key Algorithms For Quantitative Evaluation Of Pulmonary Function Based On MSCT

With the aging of the population,the atmospheric pollution increasingly serious,the morbidity and mortality of chronic obstructive pulmonary disease(COPD)increased year by year,which not only seriously affect the patient's quality of life,but also caused great economic burden to family and society.Because of its slowly developing process,the early diagnosis,treatment and evaluation is of great significance to prevent and control COPD.Multi-slice CT(MSCT)can reveal early stages of emphysema without evident clinical symptoms,and can find abnormalities of pulmonary anatomy before pulmonary function test(PFT),which has obvious advantages for the early diagnosis of lung disease.Quantitative evaluation of pulmonary function based on MSCT assists radiologists to evaluate the severity of airway remodeling and the limited degree of pulmonary function through morphological analysis of pulmonary airway and pulmonary functional parameter,such as lung volume,lung density,emphysema index.However,the performance of quantitative evaluation system of pulmonary function has large development space,and still need to solve a series of key problems,such as accurate segmentation of various pulmonary structure,pulmonary structure positioning based on skeleton extraction,automatic detection of pulmonary lesions and functional evaluation,realistic visualization of pulmonary tissue or organ.This research focuses on the further study on various strategies of increasing accuracy of quantitative evaluation,including high precision segmentation of pulmonary structure,such as pulmonary parenchyma,pulmonary airway and pulmonary vascular,accurate extraction of pulmonary airway skeleton,accurate calculation of pulmonary function parameters.The contributions of this dissertation are as follows:(1)In order to solve the problem that blurring organization's boundaries caused by motion artifacts and exisiting diseased tissue such as inflammation and emphysema,lungs adhesions near around the mediastinum,high precision pulmonary parenchyma segmentation algorithm suitable for pulmonary emphysema is proposed.The lung's region of interest(ROI)is chosen from a cross-section image who is in the middle part of the lung,then the iterative threshold which segment the ROI is used to the entire scanning slice.As a result,the efficiency of threshold selection is improved,and the robustness of the threshold selection is enhanced by the lungs' ROI error detection mechanism.Pulmonary airways segmentation step is added to the process of lung parenchyma segmentation,which eliminates small airways inside of pulmonary,and improves the accuracy of quantitative parameters,for example lung volume.A parallel three-dimensional region growing method is utilized to label the left and right lung.To separate the adhesion of left and right lungs,we design a method combing iterative multi-adhesion positioning with scan lines,boundary detection and dynamic planning.The method solves the issue of multi-adhesions positioning,and reduces the loss of pixels during separation of left and right lungs with deep adhesion.Experimental results show that the precision and efficiency of the algorithm can meet the needs of the clinical application through a large number of clinical datasets and VESSEL 12 dataset.(2)To solve the problem that terminal airways are difficult to trace out completely because that partial volume effect,noise,luminal mucus cause to inhomogeneous density,and the contrast between airway wall and lumen is changing continuously,a two stage lung airway segmentation method including coarse airway segmentation stage and fine airway segmentation stage is proposed.The main bronchus and thicker bronchus have homogeneous density,and little partial volume effect.In the coarse segmentation stage,a conservative threshold is used to extract them to avoid the leakage.Then,in the fine segmentation stage,the light projected sampling method is employed to tubular morphology analysis of airway.Because gray contrast reduced lung airway lumen and airway wall in the local scope are distinguished,it can extract more terminal airway branch and prevent leakage simultaneously.Experimental results show that by comprehensively testing and comparison with the state-of-the-art methods through EXTRA09 dataset,the proposed method has substantial advantages over other methods.(3)To solve the problem of that the complex morphology,criss-crossing structure with the surrounding tissue,inhomogeneous gray distribution,blurring boundary between vascular and airway wall,and fine terminal branch leads to accurate segmentation of pulmonary vascular is difficult,a fully automatic 3D multi-scale pulmonary vascular segmentation is proposed.The method using multi-label MRF optimization method optimize tubular filter based on characteristic value analysis of the Hessian matrix.And the problem of vascular scale selection is transformed to the problem of energy minimization,which can solve by the graph cut method.The proposed method makes the scale distribution of pulmonary vascular is in accordance with the regularity of morphology,and improves the accuracy of the segmentation of pulmonary vascular.Additionally,it overcomes the shortcomings that the traditional vascular filter is significantly influenced by noise,and is easy to break at bifurcation.Experimental results demonstrate that the proposed method,different from other similar algorithms,can effectively improve the consistency of local vascular scale distribution and the connectivity of terminal vascular,and have a better advantage in terms of accuracy through a large number of VESSEL 12 dataset.(4)To solve the problem of that the diameter of airway changes greatly,the airway's surface got by airway segmentation is uneven and rough,the existing skeleton extraction method is sensitive to the boundary noise,resulting to burrs which need subsequent pruning,a novel 3D skeleton extraction algorithm based on fast marching method(FMM)is proposed.The method uses an improved SUSAN algorithm to detect local morphological characteristics of terminal branches.The advance of this step lies in the inhibition of boundary noise while preserving the endpoint of terminal branch(the starting point to the terminal branch),by which it improves the accuracy of endpoint detection.At the same time,due to using FMM to construct double distance fields,which increase the precision of the distance fields to sub pixel level,the skeleton that subsequently got by minimum cost path planning has more accurate topology structure and more smooth path.Experimental results show that the algorithm effectively improves the performance of skeleton extraction of terminal brach,and the skeleton is anti-noise,centered,and smoother.In order to satisfy the need for clinical diagnosis,all of the core algorithms are integrated into the medical image computing(MIC)platform for forming a prototype system of quantitative evaluation of pulmonary function.By means of a hierarchical structure and a plug-in application development model of MIC platform,a clinical application system with interaction friendly and strong expansibility is realized.The prototype system provides intuitive diagnostic information assisted doctors,for example chart show of quantitative analysis parameters,and user interaction function of aided diagnosis,for example 3D multi-model visualization analysis.Take advantage of the research environment,all the key algorithm in this paper has been applied to Neusoft 128 slice CT workstation,and achieves industrial applications.