A Vessel-guided Anatomy-Aware Transformer Network for Joint Retinal Vessel and Foveal Avascular Zone Segmentation in OCT Angiography

Optical Coherence Tomography Angiography (OCT-A) is a non-invasive technique for visualizing the retinal microvasculature.Segmenting retinal vessels and the Foveal Avascular Zone (FAZ) is essential for quantitative analysis and biomarker extraction, yet remains challenging because of thin vessels, complex topology, low contrast, and noise.Existing deep learning methods typically segment vessels and FAZ separately, ignoring their anatomical relationship.We present VGAT-Net, a Vessel-Guided Anatomy-Aware Transformer Network for simultaneous segmentation of retinal vessels and the FAZ from OCT-A images.It comprises a Multi-Scale Vessel Topology Encoder, a Vessel Continuity Attention Module, a Transformer Context Aggregator, a Vessel-Guided FAZ Attention Module, and a Cross-Task Interaction Block to learn vascular topology, preserve vessel continuity, and exploit complementary anatomical information, together with a topology-preserving hybrid loss.On the OCTA-500 and ROSE-O benchmarks, using fixed subject-level splits, three random seeds, and identical training for all baselines, VGAT-Net outperforms recent CNN-, Transformer-, Mamba-, and dual-domain methods, both in a common two-head adaptation and in their published single-task formulation, achieving vessel Dice scores of 93.58% and 92.84% and FAZ Dice scores of 95.28% and 94.62%, respectively, with statistically significant gains under subject-level paired testing (Holm-corrected p < 10⁻⁹ on OCTA-500 and p < 0.05 on ROSE-O).Biomarkers derived from the segmented masks agree closely with values computed by the same script from the expert annotations (Pearson r ≥ 0.94); this quantifies the segmentationinduced biomarker error against a single expert reference and is not an independent clinical validation.The full implementation, split files, and per-baseline configurations are released publicly.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.23
Primary Topic
Retinal Imaging and Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Vessel-guided Anatomy-Aware Transformer Network for Joint Retinal Vessel and Foveal Avascular Zone Segmentation in OCT Angiography

Narayana Reddy D, M. C. Sarsamba
International journal of intelligent engineering and systems
Retinal Imaging and Analysis
article

A Vessel-guided Anatomy-Aware Transformer Network for Joint Retinal Vessel and Foveal Avascular Zone Segmentation in OCT Angiography

Narayana Reddy D, M. C. Sarsamba
article en

Abstract

Optical Coherence Tomography Angiography (OCT-A) is a non-invasive technique for visualizing the retinal microvasculature.Segmenting retinal vessels and the Foveal Avascular Zone (FAZ) is essential for quantitative analysis and biomarker extraction, yet remains challenging because of thin vessels, complex topology, low contrast, and noise.Existing deep learning methods typically segment vessels and FAZ separately, ignoring their anatomical relationship.We present VGAT-Net, a Vessel-Guided Anatomy-Aware Transformer Network for simultaneous segmentation of retinal vessels and the FAZ from OCT-A images.It comprises a Multi-Scale Vessel Topology Encoder, a Vessel Continuity Attention Module, a Transformer Context Aggregator, a Vessel-Guided FAZ Attention Module, and a Cross-Task Interaction Block to learn vascular topology, preserve vessel continuity, and exploit complementary anatomical information, together with a topology-preserving hybrid loss.On the OCTA-500 and ROSE-O benchmarks, using fixed subject-level splits, three random seeds, and identical training for all baselines, VGAT-Net outperforms recent CNN-, Transformer-, Mamba-, and dual-domain methods, both in a common two-head adaptation and in their published single-task formulation, achieving vessel Dice scores of 93.58% and 92.84% and FAZ Dice scores of 95.28% and 94.62%, respectively, with statistically significant gains under subject-level paired testing (Holm-corrected p < 10⁻⁹ on OCTA-500 and p < 0.05 on ROSE-O).Biomarkers derived from the segmented masks agree closely with values computed by the same script from the expert annotations (Pearson r ≥ 0.94); this quantifies the segmentationinduced biomarker error against a single expert reference and is not an independent clinical validation.The full implementation, split files, and per-baseline configurations are released publicly.

International journal of intelligent engineering and systemsVol. 19(12)
Jain College of Engineering and Research (IN)
Retinal Imaging and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon