| 摘要: |
| 光学相关断层血管成像（OCTA）在眼科的研究和临床应用中，可对视网膜微血管进行非侵入性可视化，被广泛应用于眼科等研究领域.本研究致力于比较、分析基于不同基础模型的深度学习分割算法和传统分割算法在视网膜OCTA图像上的血管分割效果.研究结果表明，深度学习分割算法的效果明显优于传统分割算法，尤其以特征重建网络（FRNet）分割算法表现最好.在OCTA-3 mm，OCTA-6 mm和ROSSA数据集上，相较于传统模糊均值算法，FRNet分割算法的精确度分别高出4.78%，3.10%和3.43%，能有效克服噪声干扰和血管断裂问题.研究结果揭示了在OCTA图像分割任务中，采用深度学习算法，特别是FRNet算法，具有显著的性能优势，为OCTA数据相关医学研究提供了有力的支持. |
| 关键词: 光学相关断层血管成像（OCTA） 视网膜图像血管分割 深度学习 |
| DOI：10.20192/j.cnki.JSHNU（NS）.2025.02.003 |
| 分类号:TP391.4 |
| 基金项目: |
|
| Research on retinal segmentation algorithm based on OCTA |
|
ZHANG Xinxin, WANG Xiaomei, HU Xiangyu, ZHU Yuanyuan
|
|
College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| Optically correlated tomography angiography （OCTA） is used in the research and clinical application of non-invasive visualization of retinal microvasculature， which is widely used in ophthalmology and other research fields. A variety of methods have been applied to vascular segmentation on retinal OCTA images. This paper is dedicated to compare and analyze the vascular segmentation effects of traditional segmentation algorithms and deep learning segmentation algorithms based on different basic models for OCTA images. The results showed that the deep learning segmentation algorithms are obviously superior to traditional segmentation algorithms， especially the feature reconstruction net （FRNet） segmentation algorithm. On the OCTA-3 mm， OCTA-6 mm， and ROSSA datasets， the FRNet segmentation algorithm accuracy is 4.78%， 3.10% and 3.43% higher than that of traditional fuzzy mean algorithms respectively. At the same time， it can effectively overcome the problems of noise interference and vascular fracture of traditional algorithms. The results reveal that the deep learning algorithms， especially FRNet， have significant performance advantages in OCTA image segmentation tasks， which provid the strong support for OCTA data-related medical research. |
| Key words: optically correlated tomography angiography（OCTA） retinal image vascular segmentation deep learning |
