Leveraging side-view feature guidance for robust vascular volume analysis in OCTA
TL;DR
A side-view feature-guided network (SVF-Net) for OCTA volume analysis that integrates overlay-based feature extraction to preserve fine-grained details across slices and employs a side-view attention mechanism to capture diverse features and reduce information loss is proposed.
Abstract
Ocular diseases require accurate diagnosis and continuous monitoring to improve patient outcomes. Optical Coherence Tomography Angiography (OCTA) is a non-invasive imaging modality that provides detailed visualization of ocular vascular structures. However, existing methods face challenges, including loss of slice-direction information due to projection-based down-sampling and incomplete vascular representation from single-view images. To address these limitations, we propose a side-view feature-guided network (SVF-Net) for OCTA volume analysis. The model integrates overlay-based feature extraction to preserve fine-grained details across slices and employs a side-view attention mechanism to capture diverse features and reduce information loss. Experiments on the OCTA-500 and ROSSA datasets demonstrate that our method outperforms significant benchmarks, achieving superior segmentation accuracy and effectively delineating complex vascular structures.