Using Fast Marching in Automatic Segmentation of Retinal Blood Vessels
The appearance of blood vessels in retinal images plays an important role in diagnosis of many eye diseases and system diseases. This presentation investigates a novel algorithm for automatic segmentation of blood vessels in retinal images by using vessel
- PDF / 307,511 Bytes
- 4 Pages / 595 x 791 pts Page_size
- 95 Downloads / 235 Views
School of Biomedical Engineering, College of Life Science & Biotechnology, Shanghai Jiao Tong University, Shanghai, China 2 Global R&D Center, Carestream Health Company , Shanghai, China
Abstract — The appearance of blood vessels in retinal images plays an important role in diagnosis of many eye diseases and system diseases. This presentation investigates a novel algorithm for automatic segmentation of blood vessels in retinal images by using vessel enhancement techniques and Fast Marching (FM) method. The algorithm includes the following major steps: Morlet wavelet transform, curvature estimation, matched filtering, and Fast Marching. The wavelet transform and the curvature-based method are first applied to detect the skeleton of vessels, which serve as the initial seeds of the Fast Marching algorithm. The matched filter is then used to enhance the vessels in order to extract the features used by the Fast Marching’s velocity function. Finally, the Fast Marching algorithm is applied to obtain final segmentation of retinal blood vessels. This algorithm provides effective segmentation results of retinal vessels, which can be analyzed in later processing stages leading to a complete diagnosis system. Keywords — retinal image, blood vessel segmentation, matched filter, curvature, Fast Marching
I. INTRODUCTION Retinal images have been commonly used to diagnose retinopathies and other vasculature-related diseases, such as diabetes and hypertension, because these diseases often lead to measurable abnormalities of blood vessels in color, diameter, tortuosity and length. To facilitate computeraided diagnosis, an accurate segmentation of vessels is needed as the essential input of further diagnosis. This is particularly important for remote and rural areas since physicians may be less experienced. Furthermore blood vessels can be used as landmarks for localizing other anatomical structures in retina. Previous blood vessel segmentation methods can be grouped into two major categories: thresholding based, such as edge detection and tracing based. In the thresholding methods the pixels are labeled as vessels or non-vessels by an enhancement-detection process. The vessels are enhanced by convolving mask operators with the retinal image, and are detected by thresholding the enhanced image. The Gaussian-shaped templates are the most widely used masks.
The gray-level profile of the cross section of a blood vessel can be approximated by a Gaussian-shaped curve [1]. So twelve Gaussian-shaped matched filters are designed to detect piecewise linear segments of blood vessels. Each of the twelve filters is convolved with the retinal image and only the maximum of all filter responses is preserved. The Otsu’s automatic thresholding algorithm is applied to the enhanced image to obtain the final results. Gabor filter [2], Kalman filter [3], Sobel operator [4], and morphological operator [5] are also developed in other works. The tracing algorithms start with initial seed points, and search for the next segment of blood vessels based on th
Data Loading...
