SVS-net: A Novel Semantic Segmentation Network in Optical Coherence Tomography Angiography Images

TL;DR

This paper introduces SVS-net, a new attention-based neural network designed for accurate and explainable vessel segmentation in OCTA images, effectively reducing artifacts and manual labeling needs.

Contribution

The study presents a novel SVS-net architecture with an attention module for improved vessel segmentation in OCTA images, addressing artifact removal and model explainability.

Findings

SVS-net outperforms existing models in accuracy, recall, F1 score, and Kappa score.

The attention module enhances vessel detection across different sizes.

The model reduces manual labeling requirements.

Abstract

Automated vascular segmentation on optical coherence tomography angiography (OCTA) is important for the quantitative analyses of retinal microvasculature in neuroretinal and systemic diseases. Despite recent improvements, artifacts continue to pose challenges in segmentation. Our study focused on removing the speckle noise artifact from OCTA images when performing segmentation. Speckle noise is common in OCTA and is particularly prominent over large non-perfusion areas. It may interfere with the proper assessment of retinal vasculature. In this study, we proposed a novel Supervision Vessel Segmentation network (SVS-net) to detect vessels of different sizes. The SVS-net includes a new attention-based module to describe vessel positions and facilitate the understanding of the network learning process. The model is efficient and explainable and could be utilized to reduce the need for manual labeling. Our SVS-net had better performance in accuracy, recall, F1 score, and Kappa score when compared to other well recognized models.