Truace
Subscribe

Artificial intelligence for ophthalmic vessel segmentation: from multimodal imaging to clinical translation

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 narrati…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-2039· cite & verify
Artificial intelligence for ophthalmic vessel segmentation: from multimodal imaging to clinical translation
"Retinal Epithelial Cells (RPE) Grown on Scaffold" by National Institutes of Health (NIH) is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.

In brief

This structured narrative review examines how AI converts several eye-imaging types, including fundus photographs, OCT and OCTA, into quantitative vascular biomarkers. It compares segmentation tasks across the retina, choroid, cornea and conjunctiva, and assesses model architectures from Transformers to foundation models.

The central message is that good scores on a single dataset do not guarantee dependable clinical measurements. The authors argue that usefulness depends on pipelines whose structural error, biomarker uncertainty and failure states can be audited. Because devices, labels and post-processing differ, the review does not rank studies, so no model is shown to be best.

Main points

  1. The review covers color fundus photographs, scanning laser ophthalmoscopy, OCTA, structural OCT and anterior-segment images.
  2. It organizes evaluation into five levels: pixel overlap, centerline recovery, vascular topology, biomarker agreement and clinical actionability.
  3. Public fundus and OCTA datasets are catalogued, and gaps in public anterior-segment vessel benchmarks are discussed.
  4. Reported metrics are summarized without ranking studies because devices, data splits, labels and post-processing differ substantially.

The problem

AI vessel segmentation that performs well within single datasets has not consistently produced reliable clinical measurements, because errors can enter at every step from image acquisition to deployment.

The rundown

The review separates distinct measurements that are easily conflated: structural OCT choroidal vascularity index versus OCTA choriocapillaris flow-deficit mapping, and true vessel segmentation versus global hyperemia grading.

It lists requirements for clinical translation: ocular-magnification correction, quality control of retinal layers and projection artifacts, external testing across devices and populations, and out-of-distribution rejection.

Further requirements include DICOM/PACS interoperability, prospective impact studies and regulation as software as a medical device.

Sources

  1. Peer-reviewedFrontiers in Medicine2026-10-19

The debate