北京理工大学 增强现实智能医疗实验室

Abstract—Current centerline detection methods for vascular trees usually require a large mount of human interactions, which introduces considerable amounts of computation and may lead to large errors. This paper presents a fully automatic vascular centerline extraction method for X-ray angiographic images. First, an adaptive vascular enhancement algorithm is proposed to enhance different scale vascular branches by the Frobenius norm of Hessian matrix. Second, a discriminant function defined on the derivative distribution of the image is utilized to estimate the initial positions and direction vectors of seed points. Then, new ridge points and direction vectors from the initial point are optimized by iteratively calculating the local arc lengths. Third, fake tiny vessels are removed by predefining the number of connecting components. Experiments show that vascular centerline can be extracted automatically and accurately from the X-ray angiographic images. As the developed method is fast and doesn’t require any human intervention, it can be applied to the clinical application for the diagnosis and treatment of cardiovascular diseases.