R&D, clinical & scientific discovery
Detecting eye disease from scans with AI
Can AI detect disease from medical scans?
Specialist capacity to read medical scans is limited, and delays in spotting disease cost outcomes. A research collaboration built AI that flags signs of eye disease from retinal scans at a level comparable to expert clinicians, helping triage which cases need urgent specialist attention. The transferable capability is AI-assisted detection on medical imaging as a triage and prioritisation aid — with clinicians making the diagnosis and the decision, never the model, and the honest caveat that clinical deployment demands a far higher evidence and regulatory bar than most AI.
The problem
Demand for expert reading of eye scans outstrips specialist capacity, delaying referral for sight-threatening conditions.
The AI approach
A deep-learning system analyses 3D optical-coherence-tomography (OCT) scans, using an interpretable two-stage design (segmentation, then referral recommendation) that can also express uncertainty.
Evidence it works
In a 2018 Nature Medicine study, the system recommended the correct referral decision for 50+ sight-threatening conditions, matching world-leading expert ophthalmologists (reported ~94% accuracy on the referral decision).
What “good” looks like
Expert-level triage that prioritises urgent cases, with transparent intermediate outputs clinicians can inspect and an explicit “refer to a human when unsure.”
Feasibility & cost shape
High regulatory and validation cost; value is greatest where specialist capacity is the bottleneck and scan volumes are high.
A benchmark for narrow, high-accuracy, interpretable clinical decision support — and for being clear-eyed about the path from a strong study to a deployed system.
Based on publicly reported information about the DeepMind with Moorfields Eye Hospital work.
This is an industry example included for illustration. It is not a Leia Intelligence project, and no client of ours is implied. Figures are as publicly reported by the original parties.
Sources: DeepMind / Moorfields · De Fauw et al., *Nature Medicine* (2018)