Dense Dilated Network With Probability Regularized Walk for Vessel Detection

The detection of retinal vessel is of great importance in the diagnosis and treatment of many ocular diseases. Many methods have been proposed for vessel detection. However, most of the algorithms neglect the connectivity of the vessels, which plays an important role in the diagnosis. In this paper, we propose a novel method for retinal vessel detection. The proposed method includes a dense dilated network to get an initial detection of the vessels and a probability regularized walk algorithm to address the fracture issue in the initial detection. The dense dilated network integrates newly proposed dense dilated feature extraction blocks into an encoder-decoder structure to extract and accumulate features at different scales. A multi-scale Dice loss function is adopted to train the network. To improve the connectivity of the segmented vessels, we also introduce a probability regularized walk algorithm to connect the broken vessels. The proposed method has been applied on three public data sets: DRIVE, STARE and CHASE_DB1. The results show that the proposed method outperforms the state-of-the-art methods in accuracy, sensitivity, specificity and also area under receiver operating characteristic curve.

I. Introduction

Retinal vessels are commonly analyzed in the diagnosis and treatment of various ocular diseases such as age-related macular degeneration, diabetic retinopathy, glaucoma, hypertension, arteriosclerosis and multiple sclerosis [1]–[3]. Clinicians have found that the structure of retinal blood vessels is associated with diabetes [4], [5] which affects the lining of the blood vessels in eyes. Poor circulation in the retinal vessels often compounds these problems by causing the generation of fragile new vessels. Further, vessel segmentation is also the basis for subsequent retinal artery/venous vessel classification task, which is of direct assistance to the ophthalmologist in terms of diagnosis and treatment of eye disease [6], [7]. Therefore, it is important for the clinicians to segment and analyze the retinal vessels, especially the tiny ones. However, manual demarcation of retinal vessels is a costly and time-consuming task. In recent years, extensive work has been done to segment the vessel automatically. Li et al. used the Hessian matrix and random walks to segment retinal vessel [8]. Nguyen et al. utilized a multi-scale line detection scheme for vessel segmentation [9]. Bankhead et al. developed a fast retinal vessel detection method by using wavelets [10] and edge location refinement [11]. Soares et al. used supervised classification to apply wavelets to 2-D retinal blood vessel segmentation [12]. Läthén et al. proposed a quadrature filtering method with multi-scale analysis for retinal vessel segmentation [13]. Zhao et al. [14] further proposed a vessel segmentation model based on saliency guided infinite perimeter active contour. A more detailed review of traditional vessel detection methods can be found in [15], [16]. A common limitation of these traditional vessel detection algorithms is that they often require some manually determined parameters tuned based on local data set, which makes the algorithms less scalable.