Abstract
The detection of the optic disc (OD) and fovea is essential to many automatic diagnosis systems for retinal diseases. The single shot multibox detector (SSD) can generate predictions from feature maps of various resolutions, which has not been introduced into the OD and fovea detection. To enhance the detection performance, we propose an improved SSD network, which has strengthened information flow enabled by the dense connections. The proposed method can achieve multiscale detection of the OD and fovea with strengthened feature propagation. Extensive experiments on the publicly available Messidor database and local fundus images are performed to evaluate the performance of the proposed method. Compared with seven types of representative solutions in the Messidor database, the proposed method can achieve competitive performance compared to state-of-the-art algorithms. Furthermore, the proposed method is applied to the excyclotropia screening. The screening results demonstrate promising application prospects for the proposed method in medical practice.
© 2020 Optical Society of America
Full Article | PDF ArticleMore Like This
Qianlong Zhu, Xinjian Chen, Qingquan Meng, Jiahuan Song, Gaohui Luo, Meng Wang, Fei Shi, Zhongyue Chen, Dehui Xiang, Lingjiao Pan, Zuoyong Li, and Weifang Zhu
Biomed. Opt. Express 12(10) 6529-6544 (2021)
Zhiqiang Tian, Yaoyue Zheng, Xiaojian Li, Shaoyi Du, and Xiayu Xu
Biomed. Opt. Express 11(6) 3043-3057 (2020)
Pedro J. Navarro, Diego Alonso, and Kostas Stathis
J. Opt. Soc. Am. A 33(1) 74-83 (2016)