As the number of small pigmented choroidal lesions detected during routine imaging increases – particularly in diabetic retinopathy (DR) screenings and pre-cataract assessments – opticians and optometrists face a growing challenge: differentiating benign nevi from early-stage melanomas. A recent Swedish study has explored the current diagnostic limitations of frontline eye care providers, and evaluated the potential of a deep learning algorithm – “MelAInoma” – to enhance triage decisions.
“This study builds on our earlier publication in Ophthalmology Science… [where] we validated the algorithm rigorously and showed that its performance is at least on par with that of experienced ocular oncologists,” explains corresponding author, Gustav Stålhammar. The study was designed “not to re-evaluate diagnostic accuracy, but to explore how opticians and optometrists handle small pigmented fundus lesions and how the tool might support them.”
The research found that incorporating MelAInoma into clinical decision-making significantly increased the odds of correctly referring melanomas. Additionally, false-positive referrals were reduced by a factor of 10. Decision curve analysis further confirmed that MelAInoma consistently delivered greater net clinical benefit across all threshold probabilities from 10 percent to 90 percent, particularly in reducing unnecessary referrals without compromising sensitivity.
Importantly, 92 percent of the participants (opticians and optometrists) believed that MelAInoma would enhance their referral accuracy, and 88 percent felt that any added time was justified by the gain in diagnostic confidence. Given that nearly 70 percent of respondents encounter five or more patients annually with pigmented fundus lesions, the integration of AI tools like MelAInoma into optometric practice could hold strong potential for streamlining referral workflows and reducing pressure on ophthalmic oncology services.
“More than half of Swedish optical stores now have fundus cameras, and systematic screening programs for cataract, diabetic retinopathy, and AMD generate a growing number of incidental pigmented findings,” says Stålhammar. “Only a few ocular oncologists are available to triage these lesions, so initial evaluation often falls to opticians or general ophthalmologists who may see such cases infrequently. In this setting, MelAInoma can provide an immediate, standardized malignancy score to guide referral decisions.”
Speaking of how eye care practitioners’ responded to having this deep learning support tool, Stålhammar notes that “structured interviews conducted during the project showed near-unanimous interest in adopting the tool. Respondents cited greater diagnostic confidence, clearer referral thresholds, and reduced anxiety about missing a rare but sight- or life-threatening tumor as key advantages.”
“We are now preparing a phased roll-out of MelAInoma in Swedish optical stores to evaluate usability, turnaround time, and clinical impact under routine conditions,” says Stålhammar.
