Trust the image.
Verify the AI.
Protect the patient.
RetinaX is an AI-assisted screening workflow for diabetic retinopathy designed around image quality, explainability, retinal evidence, confidence-aware triage and human validation.
Bring the fundus image into RetinaX.
Select a patient fundus image to begin the screening workflow.
Upload fundus image
PNG, JPG or JPEG
Your fundus image will appear here
Checking whether the image is safe to analyze.
Focus, illumination, retinal visibility and gradability are verified first.
Original fundus vs. CLAHE-enhanced image.
RetinaX preserves the original image and uses CLAHE enhancement for preprocessing and prediction-stability checking.
Multiple evidence streams, one screening workflow.
DR severity, vessel structure and lesion evidence are shown as parallel outputs.
Preprocessing
CLAHE, normalization, denoising and resizing.
COMPLETEDR classification
Five-level severity assessment.
READYVessel analysis
Retinal vessel segmentation and structural evidence.
READYLesion evidence
IDRiD-based lesion evidence where available.
READYSee what the model is looking at.
Grad-CAM is shown as model-focused visual evidence, not as proof of a lesion diagnosis.
Estimated optic disc and experimental fovea candidate. Not clinically validated.
Not an NV detector. Dense normal vessels and segmentation artifacts may be highlighted. Specialist review is required.
NV assessment: unavailable — no NV-specific model.
AI assists. A human makes the final review decision.
Referable, uncertain or safety-triggered cases can be escalated.
Priority human review
Review the fundus image, Grad-CAM and retinal evidence before finalizing the case.
RetinaX screening report.
A compact summary for the screening team and specialist referral workflow.