Adopting machine learning to automatically identify candidate patients for corneal refractive surgery

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Screening candidates for corneal refractive surgery is crucial to prevent complications.
• Current screening methods lack definitive accuracy, leading to potential misdiagnoses.
• Machine learning (ML) offers a potential solution for improved clinical decision support.

Purpose of the Study:

• To evaluate the efficacy of machine learning as a clinical decision support tool for identifying suitable corneal refractive surgery candidates.
• To develop and validate an ML architecture that integrates multi-instrument patient data and expert clinical decisions.

Main Methods:

• An ML architecture was developed using five heterogeneous algorithms.
• An ensemble classifier was created to enhance predictive performance.
• The model was trained and validated on a large dataset (10,561 subjects for training, 2640 for internal validation, and 5279 for external validation).

Main Results:

• The ensemble classifier demonstrated high prediction performance with an area under the receiver operating characteristic curves of 0.983 (internal) and 0.972 (external).
• ML models significantly outperformed traditional methods like percentage of tissue ablated and Randleman ectatic score.
• The model successfully reclassified a patient with postoperative ectasia into the ectasia-risk group.

Conclusions:

• Machine learning algorithms utilizing comprehensive preoperative data achieve high performance in screening corneal refractive surgery candidates.
• Automated ML analysis of preoperative data provides a safe and reliable clinical decision-making tool.
• This approach enhances the accuracy of identifying suitable candidates and mitigating risks associated with corneal refractive surgery.