Machine Learning-based Identification of Eyes With a History of Previous Myopic Laser Vision Correction

Area of Science:

• Ophthalmology
• Medical Artificial Intelligence
• Biomedical Data Science

Background:

• Myopic laser vision correction (LVC) is a common procedure, but identifying eyes that have undergone LVC can be challenging.
• Accurate identification of post-LVC eyes is crucial for appropriate ophthalmic examination and treatment planning.

Purpose of the Study:

• To develop and validate a machine learning model for the accurate identification of eyes with previous myopic laser vision correction (LVC).

Main Methods:

• A machine learning model (ML-LVC) was developed using a dataset of 35,269 eyes from Seoul National University Bundang Hospital (SNUBH).
• The ML-LVC model was internally and externally validated using datasets from SNUBH (41,508 examinations) and Seoul National University Hospital (SNUH) (5,517 examinations).
• Model performance was compared against the Cooke-Riaz-Wendelstein (CRW1) index.

Main Results:

• The ML-LVC model achieved high performance with areas under the receiver operating curve of 0.9970 (internal) and 0.9960 (external).
• Accuracies were 0.9908 (internal) and 0.9927 (external), with sensitivities of 0.9528 (internal) and 0.9222 (external).
• The ML-LVC model demonstrated significantly superior prediction compared to the CRW1 index in both validation sets (P = .004, P = .005).

Conclusions:

• The developed ML-LVC model significantly outperforms the CRW1 index in identifying eyes with prior myopic LVC.
• This AI-driven tool offers a valuable advancement for ophthalmologists, enhancing diagnostic capabilities.
• The ML-LVC model is accessible via a web application, facilitating widespread clinical adoption.