Use of a Machine Learning Method in Predicting Refraction after Cataract Surgery.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate prediction of postoperative refraction is crucial for successful cataract surgery outcomes.
• Conventional intraocular lens (IOL) power calculation formulas have limitations in precision.
• Machine learning (ML) offers a novel approach to enhance predictive accuracy.

Purpose of the Study:

• To evaluate the efficacy of ML models in predicting postoperative refraction after cataract surgery.
• To compare the predictive accuracy of ML methods against established IOL power calculation formulas.
• To determine if ML methods offer superior or non-inferior accuracy to conventional approaches.

Main Methods:

• Utilized a dataset of 3331 eyes from 2010 patients, divided into training and test sets.
• Optimized IOL formula constants and ML model parameters using training data.
• Compared predictions from SRK/T, Haigis, Holladay 1, Hoffer Q, and Barrett Universal II (BU-II) formulas against SVR, RFR, GBR, and NN ML models.

Main Results:

• The Barrett Universal II (BU-II) formula demonstrated the lowest prediction error among conventional methods.
• Several ML methods achieved lower absolute errors than BU-II.
• No statistically significant difference in accuracy was found between the best ML methods and BU-II.

Conclusions:

• ML models demonstrate potential for predicting postoperative refraction with accuracy comparable to the best conventional formulas.
• The predictive accuracy of the evaluated ML methods was not statistically inferior to the Barrett Universal II formula.
• Further research may explore advanced ML techniques for optimizing IOL power calculations.