Artificial intelligence driven intraocular lens power calculation in extreme axial myopia

Abstract

Abstract Accurate intraocular lens (IOL) power calculation is critical in cataract surgery, especially in patients with extreme axial myopia where traditional formulas often yield inaccurate results. This study retrospectively evaluated the accuracy of two AI-driven IOL formulas (Hill-RBF, Kane), the Barrett Universal II formula, and the traditional SRK/T formula in patients with axial lengths ≥ 30.0 mm. Data from 80 eyes of 51 patients treated at the Institute of Science Tokyo were analyzed. Postoperative refractive errors were recalculated, and accuracy was assessed using mean error (ME), mean absolute error (MAE), and median absolute error (MedAE). Statistical analyses included the Wilcoxon signed-rank test and chi-square test. The Kane and Hill-RBF formulas demonstrated significantly lower MAE (0.51 D and 0.52 D, respectively) compared to SRK/T ( P  < 0.05). MAE of the Barrett Universal II formula was 0.66D, which was not significantly different from SRK/T. In eyes with axial lengths ≥ 32.0 mm, Kane achieved the lowest MAE and MedAE (0.44 D and 0.40 D). Both Kane and Hill-RBF showed lower refractive errors > ± 1.0 D (7.5%) compared to SRK/T (42.5%). AI-driven formulas, particularly Kane and Hill-RBF, significantly improve refractive accuracy in extreme axial myopia. Their clinical adoption may enhance postoperative visual outcomes and reduce the need for corrective interventions.