Dual-stream disentangled model for microvascular extraction in five datasets from multiple OCTA instruments.

Abstract

Introduction: Optical Coherence Tomography Angiography (OCTA) is a cutting-edge imaging technique that captures retinal capillaries at micrometer resolution using optical instrument. Accurate segmentation of retinal vasculature is essential for eye related diseases measurement and diagnosis. However, noise and artifacts from different imaging instruments can interfere with segmentation, and most existing deep learning models struggle with segmenting small vessels and capturing low-dimensional structural information. These challenges typically results in less precise segmentation performance.

Methods: Therefore, we propose a novel and robust Dual-stream Disentangled Network (D2Net) for retinal OCTA microvascular segmentation. Specifically, the D2Net includes a dual-stream encoder that separately learns image artifacts and latent vascular features. By introducing vascular structure as a prior constraint and constructing auxiliary information, the network achieves disentangled representation learning, effectively minimizing the interference of noise and artifacts. The introduced vascular structure prior includes low-dimensional neighborhood energy from the Distance Correlation Energy (DCE) module, which helps to better perceive the structural information of continuous vessels.

Results and discussion: To precisely evaluate our method on small vessels, we delicately establish OCTA microvascular labels by performing comprehensive and detailed annotations on the FOCA dataset, which includes data collected from different instruments, and evaluated the proposed D2Net effectively mitigates the challenges of microvasculature region recognition caused by noise and artifacts. The method achieves more refined segmentation performance. In addition, we validated the performance of D2Net on four OCTA datasets (OCTA-500, ROSE-O, ROSE-Z, and ROSE-H) acquired using different instruments, demonstrating its robustness and generalization capabilities in retinal vessel segmentation compared to other state-of-the-art methods.