Diagnosis of Forme Fruste Keratoconus Using Corvis ST Sequences with Digital Image Correlation and Machine Learning

Area of Science:

• Ophthalmology and Biomedical Engineering
• Corneal Biomechanics and Imaging Analysis

Background:

• Early detection of forme fruste keratoconus (FFKC) is crucial for preventing vision loss.
• Corvis ST (CVS) tonometry provides dynamic corneal response data, but distinguishing FFKC remains challenging.
• Incremental digital image correlation (DIC) offers detailed biomechanical insights into corneal deformation.

Purpose of the Study:

• To apply incremental digital image correlation (DIC) to Corvis ST (CVS) data for corneal displacement and strain analysis.
• To evaluate the performance of machine learning (ML) models integrating these biomechanical data for FFKC detection.
• To compare the diagnostic accuracy of ML models with existing CVS parameters.

Main Methods:

• 100 subjects (50 normal, 50 FFKC) underwent Corvis ST (CVS) imaging.
• Incremental DIC reconstructed full-field corneal displacement, strain, velocity, and strain rate evolution.
• Naïve Bayes (NB) and Random Forest (RF) models, including a voting classifier, were trained on eight biomechanical evolution curves.

Main Results:

• The voting ensemble ML model achieved an Area Under the Curve (AUC) of 1.00, and the RF model achieved an AUC of 0.99.
• Existing CVS parameters, Radius and A2 Time, showed AUCs of 0.948 and 0.938, respectively, but were outperformed by ML models.
• ML model performance improved with incremental time points during corneal deformation, indicating temporal pattern significance.

Conclusions:

• Integrating incremental DIC-derived biomechanical data with ML models offers superior accuracy in differentiating FFKC from normal corneas.
• This approach surpasses the diagnostic performance of current Corvis ST (CVS) parameters.
• Analyzing the biomechanical responses and temporal patterns of the inner cornea holds significant potential for early keratoconus detection.