A quality-aware retinal screening workspace

Eyecare deserves better access.

A thoughtful screening workflow that checks image quality first, makes uncertainty visible, and helps teams decide when a specialist should take a closer look.

Quality before analysis Human review stays central Prototype, not a diagnosis
01 / Image qualityCheck before continuing
● Evidence firstNo prediction without a model
SCROLL TO EXPLORE ↓

Built for careful workflows.

Capture → Quality check → Review → Document

Designed with uncertainty in mind.

01 / Quality is the first gate

A result is only as useful as the image behind it.

Blur, poor lighting, low contrast, and a cropped retinal field can weaken any downstream analysis. RetinaReach puts image quality at the start of the workflow, not in the fine print.

Explore the states to see how the interface communicates next steps. This is an illustrative interaction, not a validated clinical quality classifier.

Potentially usable image

The image appears clear enough for the next workflow step. A real deployment would apply a validated quality model.

QUALITY GATE / PREVIEW
● Quality gate previewNo diagnosis generated
02 / A deliberate path

Less black box.
More clear next steps.

Every screening should leave a readable trail: what was checked, what remains uncertain, and what needs a human decision.

STEP 01

Capture

Record a participant code, eye, date, and the image being reviewed.

STEP 02

Check quality

Inspect basic image properties and flag issues that may need a recapture.

STEP 03

Review responsibly

Keep model output separate from quality checks. Route uncertain cases for review.

STEP 04

Keep a record

Save a local screening summary for the prototype workflow and export a CSV when needed.

03 / Designed for real constraints

A calm workspace for complicated decisions.

◉

Quality-aware from the first click

Make image limitations visible before anyone mistakes a low-quality image for a reliable result.

⌁

Uncertainty belongs in the interface

Keep the distinction between a quality flag, an AI prediction, and a specialist decision clear.

▤

Records that explain themselves.

Track quality outcomes, identify cases needing follow-up, and export basic workflow records. This prototype stores data only in the current browser on this device.

04 / The responsibility layer

Useful technology knows what it cannot do.

RetinaReach is a prototype for screening workflow design. It is not a medical device, and it does not diagnose diabetic retinopathy.

✓
No invented diagnosesNo model prediction is shown unless a real model is connected.
✓
Clear demo boundariesImage checks are heuristic demonstrations, not clinically validated thresholds.
✓
Human review mattersReal screening decisions require appropriate clinical oversight.
✓
Local prototype dataRecords stay in this browser. Do not enter identifiable patient information.

Make the next step clearer.

Open the prototype ↗
Welcome back

Sign in

Use any valid email to enter the local demo. No real account service is connected.

This is a local prototype, not secure authentication. Do not use a real password or enter patient-identifying information.

New here? Create a local profile

A fresh start

Create profile

No password needed for this prototype. Your profile stays in this browser.

Prototype only: this does not create a secure account or send your details to a server.

Already exploring? Sign in

LOCAL PROTOTYPE / WORKSPACE

Your workspace.

A clear view of the screening work recorded in this browser.

RResearch workspace
Prototype mode: no clinical AI model is connected. Quality checks are heuristic and not clinically validated. Do not use for diagnosis or patient care.
Screenings recorded0Saved in this browser
Quality follow-up0Recapture or review flags
Model analyses0No model connected

Recent screening activity

Start with a clear workflow

01

Capture

Add an image and participant code.

02

Check

Run the basic browser-side quality checks.

03

Follow up

Review any flags and export a local record.

Use fictional participant codes only. Images are processed in your browser and are not uploaded.

Image processing runs locally in this browser. The app does not upload your image. Only basic image heuristics are calculated, and no disease prediction is made.
NEW SCREENING / 01

Screening details

Use a fictional or de-identified code. Do not enter a person's name.
↑
Choose an image to inspect

Image stays in your browser and is not uploaded.

IMAGE PREVIEW
Selected retinal image preview
Your selected image will appear here.
Add the screening details and select an image. The quality report will appear here.

What this check can and cannot say

It can estimate basic image properties such as brightness, contrast, and edge detail.

It cannot reliably confirm that an image is a retinal fundus photograph, rule out disease, or determine whether an image is clinically sufficient.

Screening records

Review queue

0 items

Items are included here when the prototype's generic image heuristics flag a possible quality issue. This is not a clinical referral or diagnosis queue.