AI reads eye and breast scans faster, sometimes years ahead of doctors
Two studies show AI cutting a 23-day eye-disease wait to 30 seconds in Tanzania, and spotting breast cancer clues up to six years early in Sweden.
What happened: Two new studies show AI speeding up disease detection in very different settings. In Tanzania, a randomized trial of over 2,000 diabetes patients found that AI reading eye scans gave results in 30 seconds, versus a 23-day average wait under standard care, and doubled the share of patients who followed up for treatment (36% vs 18%). Separately, Swedish researchers tested three AI tools on nearly 89,000 mammograms and found they could flag patterns linked to breast cancer up to six years before a human radiologist made the diagnosis.
Why it matters: Diabetic retinopathy can cause irreversible blindness if caught too late, and a third of Africa's 23 million diabetics have it. Tanzania has few eye specialists, so a 23-day wait for results often means patients never return for treatment. Breast cancer, meanwhile, is far more treatable when caught early. If AI can genuinely flag risk years sooner, it could open a window for closer monitoring before a tumor is detectable by standard means.
How it works, plainly: In Tanzania, patients already had photos taken of the back of their eyes; the change was having AI software analyze those images on the spot and tell patients immediately if they needed a specialist, instead of waiting for a radiologist to review the file later. In Sweden, AI tools scored old mammograms for subtle patterns; researchers then checked, years later, which patients went on to develop cancer, and found the AI's early scores were higher for those who did.
The rollout: Tanzania's government now plans to expand AI eye screening more widely after the trial results, a concrete next step. The breast cancer findings are earlier-stage: a retrospective study, not a live clinical tool. An independent radiologist cautioned that spotting a pattern in hindsight is not the same as being able to diagnose a future cancer six years ahead, and that real clinical benefit still needs to be proven.