Digital Subtraction Angiography Based On Multi-Scale Spatial Filtering

Digital Subtraction Angiography originated in1980’s as a medical imaging technology.The technique of DSA, an important part of Digital Radiography, has been applied on a largescale to the clinical practice across the world. However, artifacts are produced clinically in thesubtraction images due to patients’ various unconscious motion. In this paper, a vesselenhancement algorithm is proposed for the removal of artifacts. The major job list is asfollows:1. In common image registration algorithms, regular mesh nodes are regarded as CPs(control points) without considering the image feature. In this paper, the points on the vascularstructures are used on the image registration. A multi-scale vessel enhancement algorithmwith content-adaptive filter parameter on scale factor is proposed. By employing thisapproach, many fine vascular structures can be visualized and the contrast between thevascular structures and the background structures becomes obvious. It also avoids distortionsof the vessels in the subsequent processes of image re-stretching. These enhancementscontribute to locating the CPs in the blood vessels.The eigenvalues of Hessian matrix are applied to remove the noises caused by themulti-scale adaption.An adaptive algorithm on an initial parameter used in the vessel similarity filter ispresented. The algorithm can initialize this initial parameter automatically on the basis of thedistribution histogram of the input image.2. An approach used to choose points on the contrast image as CPs is proposed. In orderto extract CPs and place them on the vessels, the error diffusion algorithm is employed on theenhanced vascular image and the mean-shift algorithm is applied to decrease the amount ofCPs. By applying the approach, the CPs are simplified to a great extent and placed in thevascular regions where intensive changes occur. The proposed approach has preferablepracticality.3. An algorithm is presented which matches the CPs belonging to the contrast image andthe mask image respectively. The neighborhood similarity measurement algorithm is appliedto search the CPs in the mask image for the points corresponding to the CPs belonging to thecontrast image. This algorithm, removing a mass of points without correspondence, makes thematching more efficient and the subsequent image registration more accurate.