Research platform

Three classifiers. Two modalities. A clear validation boundary.

RetGuard currently consists of separate retrospective research models. It is not a camera, not an integrated screening system, and not a clinically cleared product.

Color fundus

Diabetic retinopathy classifier

Research labels: No DR / DR
Reported evidence: External evaluation reported on Messidor-2.

Color fundus

Glaucoma classifier

Research labels: No glaucoma / glaucoma
Reported evidence: External evaluation reported on REFUGE and ORIGA.

OCT B-scan

Macular pathology classifier

Research labels: Normal / AMD / DME
Reported evidence: Internal Kermany and limited external OCTID evaluation reported.

Available today

An auditable technical package

The manuscript reports a versioned inference release with model weights, inference source, model cards, and checksums. This supports technical inspection and replication of the inference path; it does not by itself establish clinical validity.

✓Versioned research inference source
✓Research model weights
✓Model cards and checksums
✓Documented preprocessing contracts
✓Retrospective evaluation results
✓Explicit known limitations

Complete training, evaluation, and bootstrap pipelines were not public in the manuscript release and remain a reproducibility gap to address with partners.

What a first partnership can deliver

A device-specific evidence package

Data contract

Acquisition protocol, labeling, exclusions, and preprocessing locked before testing.

Performance study

Discrimination, fixed thresholds, calibration, uncertainty, and confidence intervals.

Safety analysis

Gradability, OOD behavior, failure modes, subgroups, and escalation rules.

Translation plan

Prospective workflow, human factors, quality system, regulatory, and commercial gates.

Propose a validation project