Machine-learning technology for predicting intraocular lens power: Diagnostic data generalization

Abstract

BACKGROUND: The implantation of recent intraocular lens (IOLs) allows ophthalmologists to effectively solve the surgical rehabilitation problems of patients with cataracts. The degree of improvement in the patient’s visual function is directly dependent on the accuracy of the preoperative calculation of the optical IOL power. The most famous formulas used to calculate this indicator include SRK II, SRK/T, Hoffer-Q, Holladay II, Haigis, and Barrett. All these work well for an “average patient”; however, they are not adequate at the boundaries of input variable ranges. AIM: To examine the possibility of using mathematical models obtained by deep learning of artificial neural network (ANN) models to generalize data and predict the optical power of modern IOLs. MATERIALS AND METHODS: ANN models were trained on large-scale samples, including depersonalized data for patients in the ophthalmology clinic. Data provided in 2021 by ophthalmologist K.K. Syrykh reflect the results of both preoperative and postoperative observations of patients. The source file used to build the ANN model included 455 records (26 columns of input factors and one column for the output factor) for calculating IOL (diopters). To conveniently build ANN models, a simulator program previously developed by the authors was used. RESULTS: The resulting models, in contrast to the traditionally used formulas, reflect the regional specificity of patients to a much greater extent. They also make it possible to retrain and optimize the structure based on newly received data, which allows us to consider the nonstationarity of objects. A distinctive feature of such ANN models in comparison with the well-known formulas SRK II, SRK/T, Hoffer-Q, Holladay II, Haigis, and Barrett, which are widely used in surgical cataract treatment, is their ability to consider a significant number of recorded input quantities, which reduces the mean relative error in calculating the optical IOL power from 10%–12% to 3.5%. CONCLUSION: This study reveals the fundamental possibility of generalizing a significant amount of empirical data on calculating the optical IOL power using training ANN models that have a significantly larger number of input variables than those obtained using traditional formulas and methods. The results obtained allow the construction of an intelligent expert system with a continuous flow of new data from a source and a step-by-step retraining of ANN models.