Extracting Blood Vessels From X-ray Coronary Angiogram Sequences

X-ray coronary angiography is important for the diagnosis of cardiovascular diseases.It can be used in the surgery navigation and therapeutic effect evaluation of percutaneous coronary intervention.Due to the imaging mechanism of X-ray,the coronary angiogram image reflects the sum of low dose X-ray attenuation for various human issues along the X-ray projection paths.Thus the contrast-filled vessels are overlapped with the noisy background layer including bones,diaphragms and lungs.These structures severely disturb the observations of vessels and thus construct the noisy complex background in the angiograms.Extracting noise-free vessel images from coronary angiograms still remains a challenging problem.Therefore,this thesis studies on the vessel extraction methods for X-ray coronary angiogram sequences.Three different processing methods for coronary angiogram sequences are proposed.First of all,based on that the coronary angiogram sequence can be regarded as the sum of a lowrank background matrix and a sparse foreground matrix,by exploiting the complex probability distribution of different structures in angiograms and the spatial-temporal continuity,this thesis presents a new graduated robust principal component analysis algorithm.This algorithm can extract the vessel layer images with high vessel visibility.Then,a vessel segmentation algorithm that combines robust principal component analysis,Radon-Like feature extraction and adaptive thresholding segmentation is proposed.This segmentation method can extract blood vessel shapes from coronary angiograms completely and accurately.Finally,based on the extracted vessel region mask,the t-TNN background complementation algorithm is performed to recover the projection images of the background structures from the original coronary angiogram sequence.Then the background layer is removed from the original angiogram sequences to get vessel layer images with accurate recovery of vessel shapes and intensities.Both the clinical X-ray coronary angiogram sequences and the synthetic sequences are used for experiments.The proposed methods are compared with other state-of-the-art algorithms.Experimental results show the superiority of the proposed algorithms.The graduated robust principal component analysis algorithm can extract complete vessel layer images with higher vessel visibility.The vessel segmentation algorithm and the background subtraction algorithm can extract precise vessel regions and accurate vessel intensities.