Accuracy of a new intraocular lens power calculation method based on artificial intelligence

The purpose of this study is to develop and assess the accuracy of a new intraocular lens (IOL) power calculation method based on machine learning techniques.

The following data were retrieved for 260 eyes of 260 patients undergoing cataract surgery: preoperative simulated keratometry, mean keratometry of posterior surface, axial length, anterior chamber depth, lens thickness, and white-to-white diameter; model and power of implanted IOL; and subjective refraction at 3 months post surgery. These data were used to train different machine learning models (k-Nearest Neighbor, Artificial Neural Networks, Support Vector Machine, Random Forest, etc). Implanted lens characteristics and biometric data were used as input to predict IOL power and refractive outcomes. For external validation, a dataset of 52 eyes was used. The accuracy of the trained models was compared with that of the power formulas Holladay 2, Haigis, Barrett Universal II, and Hill-RBF v2.0.

The SD of the prediction error in order of lowest to highest was the new method (designated Karmona) (0.30), Haigis (0.36), Holladay 2 (0.38), Barrett Universal II (0.38), and Hill-RBF v2.0 (0.40). Using the Karmona method, 90.38% and 100% of eyes were within ±0.50 and ±1.00 D respectively.

The method proposed emerged as the most accurate to predict IOL power.