Retinal Vessel Segmentation Using Multi-Scale Line Detection

Abstract

Inspection of the retinal vasculature may reveal precursors of serious diseases such as hypertension, diabetes, cardiovascular disease and stroke. In this paper an effective method for automatically extracting the vascular network in retinal images is presented. The proposed method is based on a multi-scale line detection, which is the line responses at varying scales. Linearly combining these line responses produces the final segmentation for each retinal image. The multi-scale line detection is applied on a vessel enhanced image whose noise and optic disc is removed and the contrast of blood vessels (including thin vessels) is enhanced by top-hat transformation and line detector filter. The preprocessing with retinal image results in the response of thin vessels with multi-scale detection is more sensitive and free from the influence of the optic disc, so the proposed method can get the very detail vascular tree in segmentation results. The performance of the proposed method was evaluated on two publicly available DRIVE, STARE databases. Experimental results have also shown the proposed method achieves high local accuracy (a measure to assess the accuracy at regions around the vessels) and approximates the average accuracy of a human observer. Moreover, the method is simple, fast, and robust to noise, so suitable for being integrated into a computer-assisted diagnostic system for ophthalmic disorders.
Copyright © 2013 Praise Worthy Prize - All rights reserved.