Predicting Early Keratoconus Progression Using Biomechanics via Multi-Machine Learning: A Multicenter 2-Year...

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Data Science

Background:

• Keratoconus is a progressive corneal disease.
• Early detection of keratoconus progression is crucial for timely intervention.
• Current methods may not reliably predict early disease progression.

Purpose of the Study:

• To develop and validate a machine learning-based index for predicting early keratoconus progression.
• To utilize corneal biomechanical and tomographic data for prediction.
• To establish a reliable tool for initial patient assessment.

Main Methods:

• A multicenter prospective cohort study involving 247 eyes.
• Collected biomechanical (Corvis ST) and tomographic (Pentacam) data.
• Applied machine learning algorithms (Random Forest, XGBoost, etc.) to predict progression.

Main Results:

• The Random Forest model, termed BEPI, showed the highest predictive performance.
• BEPI achieved high sensitivity (0.989) and specificity (0.985) in the training cohort.
• In external validation, BEPI demonstrated 0.941 accuracy and 0.947 F1-score.

Conclusions:

• The BEPI accurately predicts early keratoconus progression using corneal biomechanics.
• BEPI offers a novel index for evaluating disease progression at the first visit.
• This tool facilitates personalized clinical decisions to improve patient prognosis and preserve vision.