Postoperative data leakage in AI-based IOL power prediction formulas: a hidden source of bias
Experimental Eye Research · 2026-01 · 262 · 110700
DOI: 10.1016/j.exer.2025.110700 ↗Kurz gefasst
Fachartikel in der Zeitschrift Experimental Eye Research, 2026-01. Verfasst von Guillaume Debellemanière und 9 weiteren Autor:innen. Beteiligte Einrichtung: Department of Ophthalmology, Rothschild Foundation Hospital, Paris, France. Hauptthemen laut MEDLINE: Lenses, Intraocular, Refraction, Ocular, Optics and Photonics, Artificial Intelligence. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This study addresses a critical bias in non-optical formulas purely based on AI models and designed to predict the implanted power using the postoperative spherical equivalent (SE) as an input. Two predictive models were developed using the XGBoost algorithm on a training set of 4588 eyes implanted with Finevision IOLs. One model (AI Pred. SE) was trained to predict the postoperative SE from the implanted power and biometric data, while the second model (AI Pred. Power) was designed to predict the implanted IOL power using the achieved SE and biometric data. Both models were evaluated on an independent testing set using standard performance metrics. The logic of the optical behavior was then assessed for each model, by varying the relevant input (SE for AI Pred. Power, Power for AI Pred. SE) across a broad range of values. On standard evaluation, the AI Pred. Power model showed the lowest prediction error (Friedman test p < 0.0001), outperforming conventional formula. However, despite its accuracy in predicting the implanted IOL power, this architecture returned a predicted power that was largely invariant and non-physiologically responsive to changes in the target refraction regardless of the target refraction entered as an input, for a given eye. This finding underscores the importance of excluding real postoperative refractive outcomes from any evaluation processes or formula design. We hypothesize that this common evaluation paradigm for pure-AI formulas creates a fundamental and undetectable bias, producing models that are optimized for retrospective prediction but are incapable of prospective clinical application.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Ophthalmology, Rothschild Foundation Hospital, Paris, France.
- Department of Ophthalmology and Visual Sciences, McGill University, Montreal, QC, Canada; LASIK MD, 1250 Rene-Levesque Blvd W, MD Level, Montreal, QC, H3B 4W8, Canada.
- Faculté de Médecine Université de Montréal, QC, Canada.
- University of Toronto, Toronto, Canada.
- Memira Eye Center, Oslo, Norway.
- Department of Ophthalmology, Rothschild Foundation Hospital, Paris, France. Electronic address: dgatinel@for.paris.
Zitieren
APA
Debellemanière, G., Mechleb, N., Gauvin, M., Wallerstein, A., Abboud, L., Dai, H. Y. L., Tabunar, L., Brenner, L., Saad, A., & Gatinel, D. (2026). Postoperative data leakage in AI-based IOL power prediction formulas: a hidden source of bias. Experimental Eye Research, 262, 110700. https://doi.org/10.1016/j.exer.2025.110700
Vancouver
Debellemanière G, Mechleb N, Gauvin M, Wallerstein A, Abboud L, Dai HYL, et al. Postoperative data leakage in AI-based IOL power prediction formulas: a hidden source of bias. Experimental Eye Research. 2026;262:110700. doi:10.1016/j.exer.2025.110700
BibTeX
@article{Debellemanire2026,
author = {Debellemanière, Guillaume and Mechleb, Nicole and Gauvin, Mathieu and Wallerstein, Avi and Abboud, Lara and Dai, Huan Yu Lily and Tabunar, Lauren and Brenner, Luis and Saad, Alain and Gatinel, Damien},
title = {Postoperative data leakage in AI-based IOL power prediction formulas: a hidden source of bias},
journal = {Experimental Eye Research},
year = {2026},
volume = {262},
pages = {110700},
doi = {10.1016/j.exer.2025.110700}
}
EUVIMED