Diabetic retinopathy
External evaluation on Messidor-2 (n=1,744): AUC 0.9691.
Research-stage retinal AIAuditable retinal AI,
Auditable retinal AI,
ready for joint validation
RetGuard is a released research suite of three image classifiers for diabetic retinopathy, glaucoma, AMD, and DME. We are seeking imaging and clinical partners to test performance on specific fundus and OCT devices and build the evidence required for future clinical translation.
Research use only. RetGuard is not a medical device, is not cleared for clinical use, and has no current camera or OCT manufacturer affiliation.
Fundus
OCT B-scan
What is available today
3research classifiers
2imaging modalities
4studied conditions
These are separate retrospective research classifiers, not an integrated diagnostic workflow.
Reported evidence
Strong retrospective results, reported with the limits visible
The Version 1 manuscript submitted to medRxiv reports fixed-threshold transport, calibration behavior, reproducibility gates, and failed or incomplete evaluations alongside headline discrimination. That transparency is the foundation for a serious validation partnership.
Glaucoma
Strictly external evaluation on REFUGE and ORIGA; performance and threshold transport both reported.
AMD and DME
Three-class OCT classifier. External OCTID included AMD but no DME cases, so external DME performance remains unknown.
Why partner now
Turn a research result into device-specific evidence
A camera or OCT company does not need another vague AI claim. It needs a defined evaluation on its own image domain, with thresholds, calibration, failure modes, and intended use agreed before deployment is discussed.
Start with released artifacts
Research inference code, model weights, model cards, and checksums create a concrete technical starting point.
Evaluate the partner device
Lock preprocessing and run a device-specific retrospective study before making product claims.
Design prospective evidence
Co-develop gradability, subgroup, workflow, and human-factors validation for the intended setting.
Create mutual value
Generate evidence, integration learning, and a credible route to publication, licensing, or co-development if the gates pass.
Partnership pathway
A credible path from manuscript to product evidence
01
Define
Agree on modality, device, population, reference standard, intended use, and success criteria before evaluation.
02
Validate
Run a locked, device-specific study covering discrimination, thresholds, calibration, gradability, subgroups, and failure modes.
03
Translate
If the evidence gates pass, plan prospective workflow validation, quality management, regulatory strategy, and commercial terms together.
Have a fundus camera or OCT platform?
Let us define a rigorous first validation project together.