K-ERA · Infectious Keratitis AI Research Network
K-ERA
EvidenceWhy It MattersHow It WorksWhy Federated
Google login
Infectious keratitis · AI research platform

Federated validation for
infectious keratitis AI.

K-ERA is a research platform for evaluating infectious keratitis AI across hospitals while raw patient data remain on local systems.

Sites apply on the web, continue in the desktop app on a hospital PC, and contribute local cases under institutional approval.

Google login
Read more ↓
1. Sign in on the web → 2. Get approved → 3. Install the desktop app → 4. Start local case work
Platform type
Clinician-facing research platform with a federated validation workflow.
Workflow
Web approval first, then desktop case work and local training on a hospital PC.
Focus disease
Infectious keratitis, with bacterial-versus-fungal differentiation as the current public benchmark task.
Data boundary
Raw images and patient identifiers stay inside each hospital.
Principal Investigator
Jinho Jeong, M.D., Ph.D.
Dept. of Ophthalmology · Jeju National University Hospital
Current Evidence

Founding benchmark results
from the initial single-site cohort.

The public benchmark is intentionally strict: white-light images only, patient-disjoint evaluation, visit-level prediction, and leakage-aware controls.

The app can still register multimodal visits, including fluorescein, but the current default AI path remains anchored to that white-light benchmark. A single-center feasibility result tells you what is possible. Multi-site validation is what makes that evidence credible across institutions.

101
Patients in the founding single-center feasibility cohort.
258
Culture-confirmed visits evaluated with patient-disjoint 5-fold validation.
658
White-light slit-lamp images in the current published benchmark.
0.677
Best visit-level AUROC in the current leakage-aware benchmark.
Manuscript under peer review · Founding cohort: Dept. of Ophthalmology, Jeju National University Hospital
Why This Matters

Early treatment decisions are difficult,
and external validation remains limited.

Infectious keratitis remains one of the leading causes of preventable corneal blindness worldwide. Bacterial and fungal keratitis can present with overlapping features, making early differential diagnosis difficult even for experienced clinicians.

What remains limited is not only model performance, but the ability to build broader validation cohorts under real privacy and governance constraints.

Clinical reality
The decision is time-sensitive
Bacterial and fungal keratitis often overlap visually early on, yet treatment direction can diverge quickly and delay can cost vision.
Evidence gap
Internal benchmarks do not solve external validation
Most keratitis AI work still depends on single-site evidence. The real bottleneck is building broader validation under real hospital variation.
Operational barrier
Multi-site validation requires governance
Hospitals need a practical way to review, validate, and contribute cases without exporting raw patient data. K-ERA is designed around that operational requirement.
How It Works

Approval on the web.
Case work on the desktop.
Review stays in the loop.

Case authoring, image upload, and AI assessment all run from the K-ERA desktop app installed on a hospital PC. The web portal handles account approval only — patient images never reach a web server. This is a deliberate security boundary, not a technical limitation.

Step 01
Sign in. Request access.
Sign in with Google on this website to request institutional approval. The web portal is for account management and hospital access only — patient images never reach a web server.
Step 02
Install the desktop app. Upload images.
After approval, install the K-ERA desktop app on a hospital PC. Upload white-light, fluorescein, and slit images per visit. The app supports multimodal visits, but the current default AI path still follows the white-light benchmark used in the submitted study.
Step 03
Draw the lesion box.
Draw a loose box around the lesion. K-ERA runs MedSAM to refine it into ROI previews and lesion crops. No manual annotation pipeline required.
Step 04
Run a local training round.
Approved sites trigger image-level or visit-level training from the desktop app. Weight deltas go to central review before aggregation — raw data stays on-site.
Platform capabilities
Lesion preparation
MedSAM-assisted ROI workflow
Draw a box around the lesion. K-ERA runs MedSAM to refine it into a cornea mask and lesion crop — reducing manual preprocessing while keeping the clinician in the loop.
The current benchmark used manual lesion prompts. The desktop app operationalizes the same workflow for routine clinical cases.
Case assessment
AI inference and similar case retrieval
The desktop app returns a visit-level prediction with confidence percentage, GradCAM activation, and multi-model ensemble breakdown. A separate DINO retrieval rail surfaces the most similar cases from the research corpus — both are working features today.
Inference runs locally on the hospital PC. Retrieval queries the central embedding index. The app already stores white-light, fluorescein, and slit views, but the current default AI path remains white-light-based because that is the benchmark most fully validated in the submitted study.
Federated training
Review-gated aggregation across hospitals
Approved sites trigger local image-level or visit-level training rounds from the desktop app. Signed weight deltas go to central review before aggregation, and privacy budget reporting with budget guardrails is tracked alongside each round.
The full training pipeline is implemented and tested. Aggregation remains review-gated rather than blind auto-merge, and privacy budget guardrails can warn or block a round before rollout.
Why Federated

Raw data stays on-site.
Validation reaches across hospitals.

Multi-site validation is difficult when institutions cannot transfer raw images and patient identifiers. A federated workflow allows each hospital to keep patient data on-site while still contributing to a shared research network.

In K-ERA, participating sites contribute reviewed weight updates, de-identified metadata, and limited review assets rather than raw clinical data. This keeps the privacy boundary intact while making broader validation possible.

Hospital A → local training → Δ weights
Hospital B → local training → Δ weights → Central review
Hospital C → local training → Δ weights

Reviewed aggregation → redistributed to approved nodes
K-ERA Desktop App

All patient-facing case work and local training remain on the K-ERA desktop app installed on a hospital PC. The browser is used only for approval and access management.

Pilot enrollment open
We are preparing the first participating hospital for the clinical validation network.

After institutional approval, sites can install the desktop app, register cases locally, and join the review and aggregation workflow.

Contact us to join →
What participating sites receive
  • →Multi-site publications follow ICMJE authorship criteria and written contribution agreements established before submission.
  • →Access to the K-ERA AI inference model and confidence outputs for enrolled cases.
  • →Similar-case retrieval across the shared research corpus via the DINO embedding index.
  • →Federated weight updates distributed to approved nodes after central review.
Governance
Site governance
The founding site is already operating under its own approved research workflow. Participating hospitals can follow their own institutional review and governance process before case enrollment, and K-ERA can provide protocol documentation on request.
Authorship
Multi-site publications follow ICMJE authorship criteria and written contribution agreements established before submission.
Data flow
Raw images and identifiers never leave the site. The central server receives only reviewed weight deltas, de-identified metadata, and low-resolution thumbnails.
A founding-site result is a starting point.
Multi-site validation is the next step.
K-ERA is open to ophthalmology departments interested in prospective case registration, federated validation, and participation in multi-site publications under their own institutional governance process.
Google login
Read the evidence
Open to ophthalmology departments worldwide
Jinho Jeong, M.D., Ph.D. · Dept. of Ophthalmology, Jeju National University Hospital
dr.jinho.jeong@gmail.com
K-ERA
© 2026 K-ERA Research Network · TinyStar Labs
PrivacyTermsContact