Research On Airway Endoscopic OCT Image Segmentation Algorithms

Respiratory disease is a worldwide epidemic disease with high morbidity and mortality.Early diagnosis and intervention are of positive significance in reducing mortality.An important indicator of early diagnosis is airway remodeling.At present,one of the main obstacles in the study of airway remodeling is inaccurate observation of airway remodeling changes.Optical Coherence Tomography(OCT)is a new imaging technique,which can observe the structure of airway surface at a depth of 2-3 mm at a resolution close to histology.Using OCT imaging system for airway imaging,segmentation algorithm can accurately quantify the degree of airway remodeling in patients,detect early precancerous lesions in smokers,track the progress of diseases longitudinally and help to evaluate the effectiveness of current and future treatment methods.In this paper,the commonly used medical image segmentation algorithms are studied,and related algorithms are used to segment airway OCT images.Graph theory and dynamic programming algorithm are used to improve the segmentation of airway OCT images by analyzing the relevant experimental results.Firstly,according to the characteristics and problems of airway OCT image,image preprocessing is carried out.Aiming at the problems of artifacts and uneven gray level in the airway OCT image,an image enhancement method based on A-scan is proposed.This method can remove artifacts in the image and enhance the contrast of the image better.In order to facilitate segmentation,the location of plastic sheath,alveoli and cartilage is used to confine the inner and outer boundary of airway wall to a smaller area of interest.Secondly,due to the problem of rugged boundary and many interference factors in airway image,the segmentation accuracy of graph theory and dynamic programming algorithm is low.In order to improve the segmentation accuracy,a boundary weighted graph theory algorithm based on gray information is proposed.The algorithm adds gray information to the weight function,and can effectively segment the rugged airway tissue boundary with many interference factors.Finally,according to the similar organizational structure of adjacent images collected by OCT,a sequential image segmentation method is proposed to obtain the region of interest of the target boundary by using the information of adjacent image segmentation.The proposed algorithm is tested on 72 airway OCT images.The experimental results show that the boundary error between the proposed algorithm and manual labeling is less than 7.2 pixels,which is better than the traditional segmentation algorithm.The result of segmenting the outer boundary of airway wall with sequential image segmentation method and manual labeling error are about 5 pixels.The result is better than that of single image segmentation.Moreover,the speed and stability of the algorithm are also better than that of single image segmentation.