Reference Standard for Validation of Age-Related Macular Degeneration Screening Algorithms

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) shows promise for screening undiagnosed age-related macular degeneration (AMD).
• A standardized image labeling framework is crucial for consistent AI model development, validation, and deployment in non-specialist settings.
• Current reference standards and imaging modalities require expert consensus for optimal AI application.

Purpose of the Study:

• Establish expert consensus on an image-based reference standard for labeling AMD.
• Define a framework to support the consistent training and evaluation of AI-based screening algorithms for AMD.
• Evaluate existing reference scales (AREDS, Beckman) and imaging techniques (color, OCT, autofluorescence) for AMD detection.

Main Methods:

• A modified Delphi consensus study involving fellowship-trained retina specialists, ophthalmologists, and AI/imaging specialists.
• Two rounds of structured surveys assessing opinions on reference standards, imaging modalities, AMD features, and referral criteria.
• Consensus defined using a 9-point Likert scale with predefined statistical thresholds for agreement.

Main Results:

• Consensus reached on adopting the Beckman Classification as the level 1 reference standard for AMD labeling.
• Strong agreement on using Optical Coherence Tomography (OCT) for identifying key AMD features like drusen, geographic atrophy (GA), and choroidal neovascularization (CNV).
• Consensus achieved on referral thresholds: urgent for neovascular AMD, non-urgent for GA and intermediate AMD.

Conclusions:

• A consensus-based reference standard for AI-driven AMD image labeling has been established.
• These recommendations aim to standardize AI model development and evaluation for AMD screening.
• The defined standard is intended to facilitate AI deployment while remaining distinct from clinical practice guidelines.