Improving retinal vessel segmentation with joint local loss by matting
作者：
Highlights：
• Our deep learning matting-based framework can improve the vessel segmentation performance by transforming the segmentation to matting task.
• A new loss function is proposed in vessel segmentation, in which the global pixel loss and local matting loss are combined to handle the ambiguous pixels that often reside around the boundary of the small vessels.
• Our approach is capable of improving the results of a wide range of existing methods, while it works well with images acquired by other retinal imaging instruments, such as scanning laser opthalmoscope.
摘要
•Our deep learning matting-based framework can improve the vessel segmentation performance by transforming the segmentation to matting task.•A new loss function is proposed in vessel segmentation, in which the global pixel loss and local matting loss are combined to handle the ambiguous pixels that often reside around the boundary of the small vessels.•Our approach is capable of improving the results of a wide range of existing methods, while it works well with images acquired by other retinal imaging instruments, such as scanning laser opthalmoscope.
论文关键词：Vessel segmentation,Retinal images,Deep learning,Local matting loss
论文评审过程：Received 29 January 2019, Revised 6 August 2019, Accepted 25 September 2019, Available online 27 September 2019, Version of Record 4 October 2019.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.107068
