Automatic multi resolution-based vascular tree extraction in fundus photography

Some diseases, e.g. Diabetic Retinopathy and retinopathy of prematurity, affect the morphology of the vascular tree, for example neovascularization in the Diabetic Retinopathy which blurs the vision. The performance of automatic extraction methods may be improved the diagnosis or evaluation of ocular or systemic diseases, and show some morphological changes such as diameter, length, branching angles for vascular or nonvascular pathology. Retinal image registration is another important application of automatic retinal vascular tree extraction. We developed a new multi resolution-based vessel segmentation algorithm that can be used in noisy fundus images, with simple and fast implementation for processing applications.Methods: Proposed schema is composed of two phases: (1) Pre-processing (2) vascular tree extraction. In first phase, first we extract the green channel of input images for the best contrast between the background tissue and vessel pixels. The second step consists of Gaussian filtering for noise reduction and applying of CLAHE operator for local contrast enhancement of fundus images. In second phase we apply our multi-resolution algorithm that which attenuates low frequencies while sharpening high frequencies for vascular tree extraction. In final step we apply morphological operation for removing the undesired objects. Evaluation of the extraction algorithm is done using the publically available databases DRIVE and STARE.Results: We compare the proposed method with the existing methods in terms of TPR, FPR and ACC. Our results for TPR =0.7632, FPR=0.0237 and ACC= 0.9536 shows that proposed method not only offers improvement in TPR values, but also reduces FPR values significantly, for both STARE and DRIVE databases.Conclusion: Automatic extraction of vascular tree is very important and necessary for many applications in retinal image Analysis. Many factors includes eye movement, small pupil size, camera misalignment may causes variation in illumination and contrast in fundus images Therefore preprocessing step is quite mandatory. In this paper, we described a comprehensive algorithm for preprocessing step. In segmentation steps using conventional morphological opening cause to remove some parts of the thin blood vessels in vascular tree that using of multi structure elements can overcome this drawback