The Missing Faces of AI: Challenging Health Inequity in Eye Care

When it comes to scaling AI, health inequity remains a persistent issue. Some countries have the means to quickly roll out new technology; but when others do not, healthcare standards fracture on a global scale. And nowhere is this issue more acute than in eye care.

An estimated 1 billion people worldwide have a visual impairment that could have been prevented if the right action had been taken at the right time. This "action", however, hinges on long-term disease monitoring to spot any potential red flags — in the UK, diabetic retinopathy screening programs are fit for this exact purpose.

But those in low-resource settings are less fortunate: ophthalmologists are even more stretched, and with such limited capacity, patients slip through the cracks.

What's more, AI, a so-called potential "breakthrough" in diagnostics, falls short when applied to non-European populations. They’re typically trained on data that lacks representation from these groups, and in many cases, the necessary data simply doesn’t exist. If AI is not deployed responsibly, such prominent barriers could very-well balloon the health equity gap.

In this episode, Dr Charles Cleland, ophthalmologist and clinical research fellow at the London School of Hygiene and Tropical Medicine, considers the tumultuous journey ahead for AI in eye care.

At the intersection of digital innovation and public health, much of Dr Cleland's research is anchored in Tanzania, where he tested an AI system for the detection of diabetic retinopathy in local communities. Now, he brings his expertise to shed light on some of the biggest unanswered questions for AI in ophthalmology — and what those questions mean through a global lens.

Dr Cleland’s work and this conversation focus on three main areas:

1. The role of AI in diabetic retinopathy screening in low-resource settings, including early results from field studies in Tanzania
2. The challenges of algorithm generalizability, especially when AI models are trained predominantly on European datasets and applied to African populations
3. The risk of perpetuating bias in medical AI, and the importance of building equitable, context-specific tools that serve diverse patient groups

This episode explores how digital innovation could help reduce disparities in eye care, and the structural, ethical, and technical hurdles that must be addressed to make that possible. As AI continues to evolve, this discussion asks: can it truly serve the people most at risk of being left behind?