Does DeepSeek Provide Clinically Acceptable Intraocular Lens (IOL) Power Predictions in Cataract Surgery? A
Giovanni Ottonelli1, Giacomo De Rosa1, Jacopo Celada Ballanti1
1Department of Ophthalmology, IRCCS Humanitas Research Hospital, Rozzano, 20089 Milan, Italy.
Journal of Clinical Medicine
|December 30, 2025
Summary
The Barrett Universal II formula demonstrated superior accuracy in predicting postoperative refractive spherical equivalent compared to the generative AI model DeepSeek-R1 in cataract surgery. Established formulas remain the clinical standard for intraocular lens power calculation.
Area of Science:
- Ophthalmology
- Medical Technology
- Artificial Intelligence in Medicine
Background:
- Accurate intraocular lens (IOL) power calculation is critical for successful cataract surgery outcomes.
- Generative Artificial Intelligence (AI) shows promise in ophthalmology, but its efficacy in IOL power calculation needs evaluation.
- This study compares a novel AI model, DeepSeek, against a conventional formula for refractive prediction.
Purpose of the Study:
- To evaluate the accuracy of the open-source generative AI model DeepSeek in predicting postoperative spherical equivalent (SE).
- To compare DeepSeek's predictive accuracy against the established Barrett Universal II formula.
- To assess the suitability of general-purpose generative AI for clinical IOL power calculation.
Main Methods:
- Biometric data from 50 uncomplicated cataract surgery cases were analyzed.
- Postoperative subjective refraction was measured 30-40 days after surgery.
- Prediction error (PE), median absolute error (MedAE), and cumulative accuracy were calculated and compared using statistical tests (Wilcoxon signed-rank, McNemar).
Main Results:
- Barrett Universal II achieved a lower MedAE (0.36 D) and MAE (0.43 D) compared to DeepSeek-R1 (MedAE 0.76 D, MAE 0.77 D).
- Barrett Universal II demonstrated significantly higher cumulative accuracy at all evaluated absolute error thresholds (e.g., 71.7% vs. 25.9% at ±0.50 D).
- Statistical analysis revealed a significant paired difference favoring the Barrett Universal II formula (p < 0.0001).
Conclusions:
- The Barrett Universal II formula significantly outperformed the DeepSeek-R1 AI model in predicting postoperative SE in this cohort.
- Off-the-shelf, general-purpose generative AI models do not currently match the accuracy of validated, ophthalmology-specific formulas for IOL power calculation.
- Established formulas remain the gold standard for clinical intraocular lens power calculation.
Keywords:
DeepSeekartificial intelligence (AI)biometrycataract surgerygenerative artificial intelligenceintraocular lens (IOL)ophthalmologyspherical equivalent
