Vessel enhancement using Multi-scale Space-Intensity domain Fusion Adaptive filtering. (August 2021)

• Record Type: Journal Article
• Authors: Huang, Mingxu; Feng, Chaolu; Li, Wei; Zhao, Dazhe
• Is Part Of: Biomedical signal processing and control. Volume 69(2021)
• Publication Date: 2021-08
• Subjects: Vessel enhancement -- Hessian matrix -- Multi-scale filter -- Blood vessel indicator; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102799
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A indicating function is proposed to roughly estimate locations of vessels. The function is used to induce a local best scale in image intensity domain. The function is used to select one of the Frangi parameters adaptively. Bilateral filtering is used to preserve vessel details. Experimental results demonstrate advantages of the proposed model. Abstract: Retinal vascular disease has always been a common concern in the medical field. By performing blood vessel enhancement, fundus images help clinicians diagnosing with workload reduction and time-saving. Due to size varies of blood vessels and inhomogeneous intensities of fundus images, retinal vessel enhancement is still an open problem. In this paper, we propose a Multi-scale Space-Intensity domain Fusion Adaptive filtering model (MSIFA) to enhance retinal vessels from fundus images. Retinal vessels are roughly estimated by a proposed vessel indicating function. The indicating function is then used to define a local intensity adaptive Bilateral filtering model with spatial correlations being also considered. Finally, an improved response function is proposed to describe degrees of any a given pixel belonging to vascular regions based on eigenvalues of the Hessian matrix from images being filtered by the Bilateral filtering model. The proposed model has been evaluated with the Frangi filter and 14 representative models on STARE and DRIVE image repositories. Experimental results show advantages of the proposed models on both images repositories in terms of 6 evaluation metrics.