An improved retinal blood vessel segmentation for diabetic retinopathy detection

Abstract

Diabetic Retinopathy (DR) is an eye disorder that has progressively grows towards people who suffered from diabetes. The complications in diabetes cause the damage of blood vessel at the back of the retina. In extreme cases, DR may lead to vision loss or blindness. However, this serious effect was able to be in control through timely treatment and early detection. Recently, this issue is spreading rapidly especially in working area which eventually forced the demand of diagnosis of this illness from the earliest stage. Hence, the detection of retinal blood vessel plays significant role in controlling the progressions of this illness. The advance stages of DR such as neovascularisation which leads to the growth of abnormal vessel can be controlled by extraction of retinal vessel. Therefore, the aim of this research is to develop an approach for blood vessel detection. The proposed method comprises several techniques namely contrast enhancement, background exclusion, filtration, h-maxima transform, multilevel thresholding and morphological operation. The performance of the algorithm was evaluated based on True Positive Fraction, False Positive Fraction and accuracy which achieved the result of 0.9779, 0.0408 and 0.9686, respectively.

Keywords:

• Diabetic retinopathy
• h-maxima transform
• morphological closing
• multilevel thresholding
• retinal vessel segmentation

Acknowledgement

The authors express their appreciation towards Image Science Institute for publicly provide the DRIVE database for research purposes.