Artificial intelligence as a medical device for ophthalmic image analysis: a scoping review of regulated devices

Abstract

<title>Abstract</title> This scoping review aims to identify regulator-approved ophthalmic image analysis AIaMDs in three jurisdictions, examine their characteristics and regulatory approvals, and evaluate the available evidence underpinning them, as a step towards identifying best practice and areas for improvement. 36 AIaMDs from 28 manufacturers were identified − 97% (35/36) approved in the EU, 22% (8/36) in Australia, and 8% (3/36) in the USA. Most targeted diabetic retinopathy detection. 19% (7/36) did not have published evidence describing performance. For the remainder, 131 clinical evaluation studies (range 1–22/AIaMD) describing 192 datasets/cohorts were identified. Demographics were poorly reported (age recorded in 52%, sex 51%, ethnicity 21%). On a study-level, few included head-to-head comparisons against other AIaMDs (8%,10/131) or humans (22%, 29/131), and 37% (49/131) were conducted independently of the manufacturer. Only 11 studies (8%) were interventional. There is scope for expanding AIaMD applications to other ophthalmic imaging modalities, conditions, and use cases. Facilitating greater transparency from manufacturers, better dataset reporting, validation across diverse populations, and high-quality interventional studies with implementation-focused outcomes are key steps towards building user confidence and supporting clinical integration.

Ähnliche Arbeiten