Optimization of intraocular lens optical power calculation using artificial intelligence capabilities

Purpose. Development of a technology for optimal formula determination of the IOL required optical power, based on the artificial intelligence (AI) and the individual clinical characteristics of the patient’s eye. Material and methods. A retrospective analysis of 1337 uncomplicated phacoemulsification with implantation of a monofocal IOL RAO100C (Rayner) was performed. For artificial intelligence training and models testing, a database was created containing 1080 clinical cases (1080 eyes); the process of filling the database was fully automated. The accuracy criterion for IOL prediction was the range of spheroequivalent (SE) values in the postoperative period <|0.5|. Results. A retrospective analysis showed that in uncomplicated cataract surgery, the accuracy of the target refraction achievement is high and does not go beyond ±0.5 D. 16 artificial intelligence models were developed and tested, of which 4 were selected based on accuracy metrics – 1 model for each formula (Barrett, Haigis, Holladay 2, HofferQ). The best models turned out to be DecisionTreeClassifier. Validation of AI models was carried out by comparing the choice of IOL diopter between the AI and an expert. All models showed an advantage over the expert. A software interface for this service was developed and implemented. Conclusion. A selective database has been created for training and testing artificial intelligence models that determine IOL optical power. A system has been developed to help a doctor make a decision to determine the required IOL optical power based on artificial intelligence, depending on the individual clinical characteristics of the patient’s eye. The high efficiency of the developed technology has been proven on a test model (90–95%). The program interface was developed and implemented. Key words: IOL calculation, artificial intelligence