Ray tracing intraocular lens calculation performance improved by AI-powered postoperative lens position prediction.

Area of Science:

• Ophthalmology
• Medical Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate prediction of postoperative anterior chamber depth (ACD) is crucial for optimizing refractive outcomes in cataract surgery.
• Current ray tracing power calculation suites rely on standard methods for ACD prediction, which may have limitations.

Purpose of the Study:

• To evaluate if integrating a machine learning (ML) method for precise postoperative ACD prediction enhances the refractive prediction performance of the OKULIX ray tracing suite.
• To compare the performance of standard vs. ML-based ACD predictions in cataract surgery refraction calculations.

Main Methods:

• A dataset of 4357 cataract patients' eyes was utilized.
• A validated ML method predicted postoperative ACD using preoperative biometry data.
• Refraction predictions were calculated using both standard and ML-based ACD predictions within the OKULIX suite.
• Performance was assessed using mean absolute error (MAE) and median absolute error (MedAE).

Main Results:

• The ML-predicted ACD significantly reduced both MAE (1.7%) and MedAE (2.1%) in refraction prediction compared to standard ACD.
• ML-based ACD prediction demonstrated substantial performance improvements in eyes with both short and long axial lengths (p<0.01).

Conclusions:

• Incorporating ML for postoperative ACD prediction offers a statistically significant improvement in the accuracy of the OKULIX ray tracing suite.
• The greatest benefits of ML-enhanced ACD prediction were observed in eyes with long axial lengths.