Dilation-Supervised Learning: A Novel Strategy for Scale Difference in Retinal Vessel Segmentation | IEEE Journals & Magazine | IEEE Xplore

Dilation-Supervised Learning: A Novel Strategy for Scale Difference in Retinal Vessel Segmentation


Impact Statement:Retinal vessel segmentation is an important auxiliary method to assist clinicians in diagnosing ophthalmic diseases. It makes the doctor's diagnosis fast and efficient, a...Show More

Abstract:

Retinal fundus image segmentation based on deep learning is an important method for auxiliary diagnosis of ophthalmic diseases. Due to the scale difference of the blood v...Show More
Impact Statement:
Retinal vessel segmentation is an important auxiliary method to assist clinicians in diagnosing ophthalmic diseases. It makes the doctor's diagnosis fast and efficient, and reduces the doctor's workload. However, retinal blood vessels have serious scale differences, and the existence of many thin blood vessels greatly affects the performance of segmentation. To solve this problem, researchers have proposed some multiscale feature extraction modules or models, but the existing methods still have the problems of insufficient performance or high model complexity. The dilation-supervised learning method is proposed in this article, which uses the dilated labels to alleviate the scale difference for training, and provides a new research idea to solve the problem of scale difference and small object segmentation.

Abstract:

Retinal fundus image segmentation based on deep learning is an important method for auxiliary diagnosis of ophthalmic diseases. Due to the scale difference of the blood vessels and the imbalance between foreground and background pixels in the fundus image, the deep learning network will inevitably ignore thin vessels when downsampling and feature learning. For the scale difference problem, this article aims to tackle its limitation from two perspectives: changing the supervised approach and adapting the feature learning. Correspondingly, a dilation-supervised learning method and an adaptive scale dimensional attention mechanism which are used to construct a two-stage segmentation model is proposed. Moreover, we introduce a quantitative approach to evaluate the scale difference of the blood vessels. With the help of the proposed weighted loss function, the segmentation results are refined, and the class imbalance problem between foreground and background pixels is resolved. Finally, the...
Published in: IEEE Transactions on Artificial Intelligence ( Volume: 5, Issue: 4, April 2024)
Page(s): 1693 - 1707
Date of Publication: 18 July 2023
Electronic ISSN: 2691-4581

Funding Agency:


References

References is not available for this document.