# Learning to Address Intra-segment Misclassification in Retinal Imaging

Source: [Learning to Address Intra-segment Misclassification in Retinal Imaging](<https://www.sota2.com/research/paper/68fa6b601e209131cea2dd58>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T05:04:54.722Z

Authors: Yukun Zhou, Moucheng Xu, Yipeng Hu, Hongxiang Lin, Joseph Jacob, Pearse A. Keane, Daniel C. Alexander

Published: 2021-04-25T11:57:26.000Z

## Abstract

Accurate multi-class segmentation is a long-standing challenge in medical imaging, especially in scenarios where classes share strong similarity. Segmenting retinal blood vessels in retinal photographs is one such scenario, in which arteries and veins need to be identified and differentiated from each other and from the background. Intra-segment misclassification, i.e. veins classified as arteries or vice versa, frequently occurs when arteries and veins intersect, whereas in binary retinal vessel segmentation, error rates are much lower. We thus propose a new approach that decomposes multi-class segmentation into multiple binary, followed by a binary-to-multi-class fusion network. The network merges representations of artery, vein, and multi-class feature maps, each of which are supervised by expert vessel annotation in adversarial training. A skip-connection based merging process explicitly maintains class-specific gradients to avoid gradient vanishing in deep layers, to favor the discriminative features. The results show that, our model respectively improves F1-score by 4.4\\%, 5.1\\%, and 4.2\\% compared with three state-of-the-art deep learning based methods on DRIVE-AV, LES-AV, and HRF-AV data sets. Code: https://github.com/rmaphoh/Learning-AVSegmentation

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multi-class vessel segmentation on DRIVE-AV (test)](<https://www.sota2.com/research/sota/multi-class-vessel-segmentation-on-drive-av-test.md>): Multi-class vessel segmentation; DRIVE-AV (test); Sensitivity: 0.6987.

- [Multi-class vessel segmentation on LES-AV (test)](<https://www.sota2.com/research/sota/multi-class-vessel-segmentation-on-les-av-test.md>): Multi-class vessel segmentation; LES-AV (test); Sensitivity: 0.6294.

- [Multi-class vessel segmentation on HRF-AV (test)](<https://www.sota2.com/research/sota/multi-class-vessel-segmentation-on-hrf-av-test.md>): Multi-class vessel segmentation; HRF-AV (test); Sensitivity: 67.68.

- [Artery-vein Segmentation on UNAF external (test)](<https://www.sota2.com/research/sota/artery-vein-segmentation-on-unaf-external-test.md>): Artery-vein segmentation; UNAF external (test); Dice (Artery): 68.01.

- [Artery/Vein Segmentation on INSPIRE-AVR external (test)](<https://www.sota2.com/research/sota/artery-vein-segmentation-on-inspire-avr-external-test.md>): Artery-vein segmentation; INSPIRE-AVR external (test); Dice (Artery): 68.61.

- [Artery/Vein Segmentation on Cropped HRF external (test)](<https://www.sota2.com/research/sota/artery-vein-segmentation-on-cropped-hrf-external-test.md>): Artery-vein segmentation; Cropped HRF external (test); Dice (Artery): 0.729.

- [Artery/Vein Segmentation on LES-AV external (test)](<https://www.sota2.com/research/sota/artery-vein-segmentation-on-les-av-external-test.md>): Artery-vein segmentation; LES-AV external (test); Dice (Artery): 78.68.

- [Artery/Vein Segmentation on UZLF (test)](<https://www.sota2.com/research/sota/artery-vein-segmentation-on-uzlf-test.md>): Artery-vein segmentation; UZLF (test); Dice (Artery): 78.14.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2104.12138>)

[PDF](<https://arxiv.org/pdf/2104.12138.pdf>)

[Implementation](<https://github.com/rmaphoh/Learning-AVSegmentation>)
