SRNet: Striped Pyramid Pooling and Relational Transformer for Retinal Vessel Segmentation
Wei Yan (College of Computer Science and Engineering, Northwest Normal University),* Yun Jiang (College of Computer Science and Engineering, Northwest Normal University), Zequn Zhang (Northwest Normal University ), Yao Yan (College of Computer Science and Engineering, Northwest Normal University), Bingxi Liu (Northwest Normal University)
The 34th British Machine Vision Conference
Abstract
The morphology of retinal vessels is crucial for diagnosing and screening retinal diseases such as age-related macular degeneration and diabetic retinopathy. Retinal vessels segmentation is an indispensable part of retinal disease screening and diagnosis. However, due to the inherent complex structural features of retinal vessels, it remains a challenging visual task. Based on the type of input, retinal vessels segmentation approaches can be roughly divided into both image-level and patches-level methods, which have their respective benefits and drawbacks. To better leverage both input methods, we design a Relational Transformer Module (RTM) to effectively combine local patches-level information with image-level global contextual information. Furthermore, retinal vessels exhibit varying lengths with tree-like branching patterns, making the classical rectangular pooling inefficient in capturing accurate vessels information because they are better suited for uniformly distributed objects. To better capture contextual information, we further developed a Striped Pyramid Pooling Module (SPPM) to adapt to the tree-like distribution of retinal vessels. Based upon these foundations, we propose a retinal vessels segmentation Network with the Striped Pyramid Pooling Module and the Relational Transformer Module (SRNet). Experimental validation showed that our SRNet outperforms other advanced methods on the DRIVE and CHASE datasets.Video
Citation
@inproceedings{Yan_2023_BMVC,
author = {Wei Yan and Yun Jiang and Zequn Zhang and Yao Yan and Bingxi Liu},
title = {SRNet: Striped Pyramid Pooling and Relational Transformer for Retinal Vessel Segmentation},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0347.pdf}
}
Copyright © 2023
The British Machine Vision Conference is organised by