Xiao-Xin Guo1,2, Qun Li2, Chao Sun2, Yi-Nan Lu2. 1. Key Laboratory of Symbol Computation and Knowledge Engineering of Ministry of Education, Changchun130012, Jilin Province, China. 2. College of Computer Science and Technology, Jilin University, Changchun 130012, Jilin Province, China.
Abstract
AIM: To explore feasibility and practicability of macula localization independent of macular morphological features. METHODS: A novel method was proposed to identify macula in fundus images by using structure label transfer. Its main idea was to match a processed image with the candidate images with known structures, and then transfer the structure label representing the macular to the processed image as a result of macula localization. In this way, macula localization couldn't be influenced by lesion or other interference any more. RESULTS: The average success rate in four datasets was 98.18%. For accuracy, the average error distance in four datasets was 0.151 optic disc diameter (ODD). Even for severe lesion images, the proposed method can still maintain high success rate and high accuracy, e.g., 95.65% and 0.124 ODD in the case of STARE dataset, respectively, which indicated that the proposed method was highly robust and stable in the complicated situations. CONCLUSION: The proposed method can avoid the interference of lesion to macular morphological features in macula localization, and can locate macula with high accuracy and robustness, verifying its feasibility.
AIM: To explore feasibility and practicability of macula localization independent of macular morphological features. METHODS: A novel method was proposed to identify macula in fundus images by using structure label transfer. Its main idea was to match a processed image with the candidate images with known structures, and then transfer the structure label representing the macular to the processed image as a result of macula localization. In this way, macula localization couldn't be influenced by lesion or other interference any more. RESULTS: The average success rate in four datasets was 98.18%. For accuracy, the average error distance in four datasets was 0.151 optic disc diameter (ODD). Even for severe lesion images, the proposed method can still maintain high success rate and high accuracy, e.g., 95.65% and 0.124 ODD in the case of STARE dataset, respectively, which indicated that the proposed method was highly robust and stable in the complicated situations. CONCLUSION: The proposed method can avoid the interference of lesion to macular morphological features in macula localization, and can locate macula with high accuracy and robustness, verifying its feasibility.
Keywords:
fundus image; macula; optic disc; structure label transfer
Authors: Jian Chen; Jie Tian; Noah Lee; Jian Zheng; R Theodore Smith; Andrew F Laine Journal: IEEE Trans Biomed Eng Date: 2010-02-18 Impact factor: 4.538
Authors: Harihar Narasimha-Iyer; Ali Can; Badrinath Roysam; Howard L Tanenbaum; Anna Majerovics Journal: IEEE Trans Biomed Eng Date: 2007-08 Impact factor: 4.538