Comparison of accuracy of different intraocular lens power calculation methods using artificial intelligence.

Abstract

To assess the accuracy of the intraocular lens (IOL) power calculation based on three methods using artificial intelligence (AI) and one formula using no AI. During cataract surgery on 114 eyes, one type of IOL was implanted, calculated with the Hill-RBF 2.0 method. The theoretical postoperative refractions were calculated using the Kane and the Pearl-DGS methods and a vergence based formula (Barrett Universal II, BUII). The differences between the manifest and objective postoperative refractions and the predicted refractions were calculated. The percentage of eyes within ±0.5 D and ±1.0 D prediction error (PE), the mean, and the median absolute errors (MAE and MedAE) were also determined. The mean age of the patients was 69.48 years; the axial length was between 21.19 and 25.39 mm. The number of eyes within ±0.5/±1.0 D PE was 96/108 (84.21%/94.73%) using the Hill-RBF 2.0 method, 92/107 (80.70%/93.85%) with the Kane method, 91/107 (79.82%/93.85%) with the Pearl-DGS method, and 91/106 (79.82%/92.98%) with the BUII formula, using subjective refraction. With objective refractometric data, PEs were within ±0.5 D in 88 (77.19%), 83 (72.80%), 82 (71.92%), and 80 (70.17%) cases (Hill-RBF, Kane, Pearl-DGS, BUII, respectively). MAE and MedAE were also best with the Hill-RBF 2.0 method (0.3 D; 0.18 D). Better accuracy of PE might be obtained by the Hill-RBF 2.0 method compared with BUII. The Kane and Pearl-DGS methods showed similar accuracy when compared with BUII.