Artificial intelligence for ophthalmic vessel segmentation: from multimodal imaging to clinical translation
- 1Hangzhou Lin'an Traditional Chinese Medicine Hospital, Affiliated Hospital of Hangzhou City University , ,
- 2School of Ophthalmology and Optometry, Wenzhou Medical University , ,
Abstract
Artificial intelligence (AI)-based vessel segmentation converts color fundus photographs, scanning laser ophthalmoscopy (SLO), optical coherence tomography angiography (OCTA), structural OCT, and anterior-segment images into quantitative vascular biomarkers. Strong performance within individual datasets has not consistently yielded reliable clinical measurements because errors can arise during acquisition, anatomical preprocessing, vessel extraction, biomarker calculation, and deployment. This structured narrative review compares segmentation tasks involving retinal vascular trees, OCTA capillary plexuses, the structural choroid and choriocapillaris, macular neovascularization, corneal neovascularization, and conjunctival microvasculature. We assess modern architectures in relation to vessel geometry, severe foreground-background imbalance, scale variation, crossings, and susceptibility to artifacts. The methods considered include multiscale feature pyramids, deformable and directional operators, Transformers, state-space models, diffusion approaches, and foundation models. We organize the synthesis around five evaluation levels: pixel overlap, centerline recovery, vascular topology, biomarker agreement, and clinical actionability. Public fundus and OCTA datasets are catalogued with their acquisition and annotation characteristics, and gaps in public anterior-segment vessel benchmarks are discussed. Because devices, data splits, labels, and post-processing differ substantially, reported metrics are summarized without ranking studies against one another. The review distinguishes structural OCT choroidal vascularity index from OCTA choriocapillaris flow-deficit mapping, as well as true vessel segmentation from global hyperemia grading. Clinical translation will require ocular-magnification correction, quality control of retinal layers and projection artifacts, external testing across devices and populations, out-of-distribution rejection, DICOM/PACS interoperability, prospective impact studies, and regulation as software as a medical device. We conclude that clinical utility depends on an auditable pipeline that measures structural error, biomarker uncertainty, and failure states throughout the workflow.
Related
- Artificial intelligence across the gastric cancer care continuum: Clinical evidence, multimodal integration, and translation
- AdaSwitch3D: Reliability-Aware Human-in-the-Loop Artificial Intelligence for Breast Tumor Segmentation in Dynamic Contrast-Enhanced MRI Across Unseen Centers
- Clinical Translation of Artificial Intelligence in Oncology: Real-World Validation, Workflow Integration, and Precision Medicine Applications
- The integration of artificial intelligence into computed tomography: from segmentation to the standardisation of clinical decisions
- Myocardial Stiffness Tracking & Assessment Toolkit Driven by Artificial Intelligence For Reproducible Benchmarking of Temporal Segmentation and Shear-Wave Velocity Stabilization on Synthetic Data
BibTeX
@article{wu2026artificial,
title = {{Artificial intelligence for ophthalmic vessel segmentation: from multimodal imaging to clinical translation}},
author = {Lingyan Wu and Jiahao Tang and Shanping Zhao},
journal = {Frontiers in medicine},
year = {2026},
month = oct,
doi = {10.3389/fmed.2026.1912838},
url = {https://europepmc.org/article/PMC/PMC13724346},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial intelligence for ophthalmic vessel segmentation: from multimodal imaging to clinical translation
AU - Lingyan Wu
AU - Jiahao Tang
AU - Shanping Zhao
PY - 2026
DA - 2026/10/19
JO - Frontiers in medicine
DO - 10.3389/fmed.2026.1912838
UR - https://europepmc.org/article/PMC/PMC13724346
AB - Artificial intelligence (AI)-based vessel segmentation converts color fundus photographs, scanning laser ophthalmoscopy (SLO), optical coherence tomography angiography (OCTA), structural OCT, and anterior-segment images into quantitative vascular biomarkers. Strong performance within individual datasets has not consistently yielded reliable clinical measurements because errors can arise during acquisition, anatomical preprocessing, vessel extraction, biomarker calculation, and deployment. This structured narrative review compares segmentation tasks involving retinal vascular trees, OCTA capillary plexuses, the structural choroid and choriocapillaris, macular neovascularization, corneal neovascularization, and conjunctival microvasculature. We assess modern architectures in relation to vessel geometry, severe foreground-background imbalance, scale variation, crossings, and susceptibility to artifacts. The methods considered include multiscale feature pyramids, deformable and directional operators, Transformers, state-space models, diffusion approaches, and foundation models. We organize the synthesis around five evaluation levels: pixel overlap, centerline recovery, vascular topology, biomarker agreement, and clinical actionability. Public fundus and OCTA datasets are catalogued with their acquisition and annotation characteristics, and gaps in public anterior-segment vessel benchmarks are discussed. Because devices, data splits, labels, and post-processing differ substantially, reported metrics are summarized without ranking studies against one another. The review distinguishes structural OCT choroidal vascularity index from OCTA choriocapillaris flow-deficit mapping, as well as true vessel segmentation from global hyperemia grading. Clinical translation will require ocular-magnification correction, quality control of retinal layers and projection artifacts, external testing across devices and populations, out-of-distribution rejection, DICOM/PACS interoperability, prospective impact studies, and regulation as software as a medical device. We conclude that clinical utility depends on an auditable pipeline that measures structural error, biomarker uncertainty, and failure states throughout the workflow.
SN - 2296-858X
ER -