Improving Retinal Vessels Segmentation via Deep Learning in Salient Region
- PDF / 1,015,411 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 54 Downloads / 212 Views
ORIGINAL RESEARCH
Improving Retinal Vessels Segmentation via Deep Learning in Salient Region Vo Thi Hong Tuyet1,2,3 · Nguyen Thanh Binh1,2 Received: 3 May 2020 / Accepted: 21 July 2020 © Springer Nature Singapore Pte Ltd 2020
Abstract Retinal vessels segmentation is one of the hard tasks. Because it may cause loss of information in the image. Therefore, the features of retinal vessel must be taken care of properly. Deep learning gives the positive results for features extraction. The retinal vessel segmentation really needs to be based on these features. Information in retina is also vital for the health status of citizen. However, the using deep learning combined with salient region will increase the accuracy of saliency. The number of parameters and size for dividing level in each input images make pixels decrease linking. This paper proposed a new method for retinal vessels segmentation based on deep learning in edge map which is created by salient map. The proposed method has three periods: edges salient map in the retinal vessel image, features extraction by convolutional neural network in salient map and segmentation based on pixel level of Sobel operator in saliency. To evaluate the results of proposed method with the other methods, we use the DRIVE dataset through the Jaccard index value. The experiments show that the JI value of the proposed method is higher than the results in other methods. Keywords Retinal vessels segmentation · Deep learning · Sobel operator · Salient region
Introduction The retina is an internal part of the eye that receives light signals. There are many different conditions that cause damage to the retina. It may partially or completely impair a person’s vision. The retina contains many vascular systems. These abnormalities in the vascular system are the main cause of retinopathy. Retinal vessels segmentation is one of the hard tasks. Because it may cause loss of information in the image. This article is part of the topical collection “Future Data and Security Engineering 2019” guest edited by Tran Khanh Dang. * Nguyen Thanh Binh [email protected]
Vo Thi Hong Tuyet [email protected]; [email protected]
1
Faculty of Computer Science and Engineering, Ho Chi Minh City University of Technology, VNU-HCM, 268 Ly Thuong Kiet Street, District 10, Ho Chi Minh City, Vietnam
2
Vietnam National University Ho Chi Minh City, Linh Trung Ward, Thu Duc District, Ho Chi Minh City, Vietnam
3
Faculty of Information Technology, Ho Chi Minh City Open University, Ho Chi Minh City, Vietnam
Many researchers in the world have proposed methods to solve this problem, such as: clustering [1], thresholding filter [2–7], domain combined with threshold [8, 9], deep learning [10], etc. The advantage of these methods is accuracy. However, the time calculating is long and complex. In recent years, researches have been used unsupervised methods, supervised methods and other methods, such as: salient region [11–15], referee and fluorescein [16], probabilistic formulation [17], classif
Data Loading...
