The Moorfields Dry-AMD Dataset encompasses all patients with advanced dry AMD, specifically geographic atrophy (GA), and without choroidal neovascularization, at Moorfields Eye Hospital - a leading provider of eye health services in the UK and a world-class centre of excellence for ophthalmic research and education. A detailed description of the dataset definitions can be found below.

Inclusion criteria:

• a posterior free hand draw label of ‘Geographic’ or (ii) at least one reference in their clinical notes to ‘geographic atrophy’

Exclusion criteria:

• Patients under the age of 50 at their most-recent imaging event
• Patients with at least one reference in their clinical notes to ‘Stargardt’
• Patients with a record of injections of any type
• Patients with a posterior free hand draw label of ‘Choroidal neovascularization’

The dataset will include any imaging or clinical metadata that is available for these patients. In the instance where only one eye meets the criteria above, both eyes are included in the dataset. For these reasons, the dataset will include longitudinal data from where a patient underwent monitoring during the early and intermediate stages of the disease.

Clinical metadata includes information regarding:

• patient demographics
• visual acuities (predominantly measured with Early Treatment Diabetic Retinopathy Study (ETDRS) charts)
• ocular surgeries

Additional information is provided in the ‘technical details’ tab.

The Dry-AMD dataset includes eye imaging modalities, such as:

• Optical coherence tomography (CSO, Heidelberg, Optos, Topcon, Zeiss)
• Colour fundus photographs (Topcon, Zeiss)
• Ultra-wide field photographs (Optos, Zeiss)
• Iris photographs (CSO, Zeiss)
• Keratoscope topography (CSO)
• Infrared photographs (Heidelberg, Topcon, Zeiss)
• Fluorescein angiography (Heidelberg, Optos, Topcon, Zeiss)
• Indocyanine green angiography (Heidelberg, Optos, Topcon)
• Fundus autofluorescence (Heidelberg, Optos, Zeiss)

Imaging data from CSO is subject to additional approvals.

As of July 2024, the dataset consisted of 11,168 eyes and over 688,018 ophthalmic images.