Retina and Choroid Segmentation Method Based on Multi-scale Feature Extraction and Fusion

The change of layer thickness of retina and choroid is closely related to eye diseases. In clinical applications, optical coherence tomography (OCT) is utilized to visualize retina and choroid. Precise segmentation of the retina and choroid in fundus OCT images is essential for the early detection and diagnosis of ocular diseases. Effective feature extraction and feature fusion have not been fully explored in current studies, which results in poor segmentation results. This paper presents a novel end-to-end segmentation framework for simultaneous segmentation of the retina and choroid. To extract more effective features, a multi-scale channel feature extraction module is designed, which increases the network's receptive field and allows for the acquisition of effective channel information. To fuse deep features with shallow features, a cross fusion module is proposed, capturing the relationship between high-level features and low-level features. Experiments demonstrate that our method outperforms others in segmentation performance.