Ai-Assisted Intraocular Lens (Iol) Selection: Personalized Analysis Of Trifocal, Edof, And Monofocal Plus Lenses

This study aims to optimize the patient-specific selection process of Trifocal (TF), Extended Depth of Focus (EDOF), and Monofocal Plus (MP) intraocular lenses (IOLs) using an artificial intelligence (AI)-assisted decision-making system. By determining which lens type is most suitable for different patient profiles, a data-driven approach is proposed to support surgeons in their decision-making process.

University of Health Science, Haseki Training and Research Hospital, Istanbul, Türkiye

A simulated patient database consisting of 500 patients was created. Factors such as age, preoperative distance/intermediate/near visual acuity (LogMAR), pupil diameter, aberrometry root mean square (RMS), corneal astigmatism, night driving necessity, near vision requirements, and sensitivity to halos were included in the model. Differences in patient satisfaction among lens types were analyzed using ANOVA, and a patient-specific IOL selection prediction model was developed using the Random Forest algorithm.

The mean patient age was 67.8 ± 10.4 years, with an average pupil diameter of 4.6 ± 1.1 mm and aberrometry RMS of 0.62 ± 0.32. Among patients, 56% did not drive at night, while 41% had high near-vision demands. ANOVA analysis showed no significant difference in satisfaction among lens types (p = 0.278). TF lenses performed best for high near-vision needs (89.2/100), EDOF lenses had a balanced performance at intermediate and distance vision (85.7/100), and MP lenses were better for night drivers with low aberration (81.4/100). AI model accuracy was 56%; TF prediction performed best (Precision: 55%, Recall: 78%), EDOF had moderate predictability (Precision: 65%, Recall: 48%), while MP lenses were least accurate (Precision: 25%, Recall: 06%).

This study suggests that AI-assisted models can serve as supportive tools for patient-specific IOL selection. While the prediction accuracy for TF and EDOF lenses were satisfactory, MP lenses require additional variables for better predictability. The accuracy of AI models can be enhanced through the incorporation of broader clinical data and additional biometric parameters. The integration of AI-based personalized lens selection into clinical practice appears feasible. Further improvements in predictive accuracy may be achieved by employing more advanced models such as XGBoost or deep learning. Additionally, incorporating variables related to patient lifestyle and occupational factors could contribute to a more comprehensive model.