Comparing The Accuracy Of Artificial Intelligence- Based Formulas In Calculation Of Intraocular Lens Power

To compare the accuracy of artificial intelligence- based formulas Kane, Hill-RBF and Barrette Universal II in calculation intraocular lens power.

We investigated patients who had undergone phacoemulsification surgery with 9 kinds of intraocular inplantation including monofocal, EDOF and multifocal from January 2025 to February 2025 at Vietnam-Russia International Eye Hospital-HCMC. Patients must be older than 18 years old to be selected. We eliminated patients who had corneal refractive surgery or other ocular surgery before; post operation VA less than 20/30 and had no refractive data after 3 weeks postoperatively.

This retrospective study included 123 eyes of 91 patients. Biometry was perfomed using IOL Master 700 which was included Barrette Universal II formula. Kane and Hill -RBF 3.0 were caculated online by inputting biometry data from IOL master 700. After surgery they were also collected VA, refractive errors. Mean numerical error (MNE) and mean absolute error (MAE) and proportion of eyes within ±0.25D ,±0.5D and ±0.75D were assessed for the accuracy of formulas. Subgroup analyses were performed for axial length and type of implants.

Kane formula demonstrated the lowest MAE (0,21±0,19D), MedAE (0,14D) and highest percentages of eyes with prediction errors ±0.25D and ±0,5D with 68.3% and 92.7% respectively; although there was no statistically significant difference between Barrett Universal II, Hill-RBF and Kane. Subgroup analysis by axial length (AL) and types of implant were also investigated. In short-eye subgroup, Kane also demonstrated lowest MAE while Hill-RBF took this placement in long-eye subgroup. There was statistically significant difference between formulas in eyes which had multifocal lens implantation where Kane exhibited the lowest MAE (0,19±0,15D), MedAE (0,14D).

Kane was proved the most accurate formula in terms of MAE and MedAE in whole group and also in subgroup short-eye, normal-eye. In subgroup by types of implants, Kane was also the most accurate, especially in multifocal implantation with significant difference.