Related Experiment Video
Updated: Sep 18, 2025
Implantation and Evaluation of Melanoma in the Murine Choroid via Optical Coherence Tomography
Published on: December 2, 2022
Challenges for Opticians in Evaluating Small Pigmented Choroidal Lesions: Potential Support From the MelAInoma Deep
Shiva Sabazade1,2, Marco A Lumia Michalski1,2, Sara Wittskog3
1Ocular Oncology Service, St. Erik Eye Hospital, Stockholm, Sweden.
Purpose:
To evaluate the diagnostic accuracy of Swedish opticians/optometrists when triaging small pigmented choroidal lesions and determine whether the MelAInoma deep learning algorithm improves referral decisions.
Methods:
Twenty-nine opticians/optometrists graded 25 fundus photographs (5 melanomas, 20 nevi) with the Mushroom shape, Orange pigment, Large size, Enlargement, Subretinal fluid (MOLES) system and recorded referrals; 25 then used MelAInoma. Predefined referral thresholds were MOLES ≥1 or ≥3 and MelAInoma >63. Diagnostic test statistics-including sensitivity, specificity, positive (PPV) and negative (NPV) predictive values, accuracy, and standard deviation (SD)-plus odds of correct referral and decision curve net benefit were computed.
Results:
With MOLES ≥1, mean ± SD sensitivity was 98% ± 6%, specificity was 17% ± 16%, PPV was 23%, NPV was 99%, and accuracy was 33%. Raising the cutoff to MOLES ≥3 lowered sensitivity to 75% ± 29% but increased specificity to 53% ± 28%, PPV to 34%, NPV to 93%, and accuracy to 57%. MelAInoma achieved 80% sensitivity, 90% specificity, PPV 67%, NPV 95%, and 88% accuracy. Algorithm guidance quadrupled the odds of correctly referring a melanoma, reduced false-positive referrals 10-fold, and provided net clinical benefit to unaided triage across all relevant threshold probabilities.
Conclusions:
Although the findings are based on a small image set, including only five melanomas, which limits generalizability, the results suggest that opticians and optometrists detect small choroidal melanomas with high sensitivity but limited specificity. Incorporating MelAInoma at the predefined threshold reduced sensitivity, substantially improved specificity, and markedly reduced unnecessary referrals.
Translational Relevance:
MelAInoma offers practical artificial intelligence support that can streamline community referrals and facilitate earlier treatment of uveal melanoma.
