Age-related macular degeneration (AMD) is the leading cause of irreversible central vision loss in individuals over 50 in high-income countries. It affects approximately 196 million people worldwide, with this number projected to rise substantially as populations age.

The macula, located at the centre of the retina, is responsible for high-acuity central vision — essential for reading, face recognition, and fine motor tasks. AMD causes progressive degeneration of the photoreceptors and retinal pigment epithelium (RPE) in this region.

Why early detection matters

Early-stage AMD is asymptomatic or produces only subtle visual changes that patients may not notice or attribute to ageing. By the time significant symptoms emerge, irreversible structural damage has typically occurred. Early detection allows initiation of lifestyle interventions, nutritional supplementation (AREDS2 formula), and close monitoring that can meaningfully delay progression to advanced disease.

The role of fundus imaging

Colour fundus photography provides a direct view of the retinal surface. AMD hallmarks — drusen (yellow-white deposits beneath the RPE), pigmentary abnormalities, geographic atrophy, and choroidal neovascularisation — are directly visible and amenable to automated image analysis.

Severity Classification

Level Classification Key Features
0 No Macular Degeneration Normal foveal reflex. Absent or minimal small drusen (<63 µm). No RPE abnormalities. No neovascularisation.
1 Mild Macular Degeneration Medium drusen (63–124 µm) without RPE abnormality, or small drusen with pigmentary changes. No geographic atrophy in the central subfield. No late AMD features.
⚠ Screening Tool — Not a Diagnostic Device AMD severity grading is intended to flag cases for ophthalmologist review. It does not replace clinical examination with optical coherence tomography (OCT) or fluorescein angiography.

Model architecture & validation

Input
Colour fundus photograph
Standard resolution, centred on macula
Output
2-class severity score
With per-class confidence probability
Architecture
CNN (deep residual)
Transfer learning from ImageNet pre-training
Validation
Expert grader agreement
Graded by certified ophthalmologists
Accuracy
90%+ (binary)
On held-out test set
Inference time
< 3 seconds
Per image on standard GPU

How results are used

The AMD model outputs a severity level and confidence score. In the EyeMap deployment workflow, results are presented as a screening report to the optometrist, not directly to the patient. The report includes an explicit statement that findings require clinical confirmation.

Referral thresholds

Any result at Level 1 (Mild AMD) or above, or a Level 0 result with confidence below a defined threshold, triggers a recommendation for ophthalmologist referral. The referral decision remains with the qualified optometrist or clinician reviewing the report.

Longitudinal tracking

In the UK Biobank validation study, we will examine AMD classification change over time to assess the model's utility in detecting early progression — a clinically meaningful capability that static grading alone cannot provide.