Using artificial intelligence to detect keratoconus progression based on age and anterior segment optical coherence...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Keratoconus progression is a significant concern in ophthalmology.
• Accurate prediction of keratoconus progression is crucial for timely intervention.
• Current methods for assessing keratoconus progression have limitations.

Purpose of the Study:

• To establish and evaluate a novel index for defining keratoconus progression.
• To utilize deep learning on anterior segment optical coherence tomography (AS-OCT) map images and patient age.
• To assess the predictive value of an age-adjusted keratoconus progression index.

Main Methods:

• Retrospective retrieval of paired AS-OCT images and patient data.
• Development of machine learning classifiers for six AS-OCT image types.
• Establishment of a keratoconus progression index, adjusted and unadjusted for age.
• Validation of the index using accuracy, sensitivity, specificity, and AUC metrics.

Main Results:

• A dataset of 2006 paired AS-OCT images was utilized.
• The age-adjusted keratoconus progression index demonstrated superior performance.
• Key metrics significantly improved with age adjustment: accuracy (0.655 vs. 0.909), sensitivity (0.915 vs. 0.937), specificity (0.457 vs. 0.887), and AUC (0.727 vs. 0.935).

Conclusions:

• A novel, age-adjusted index for keratoconus progression was successfully developed using deep learning and AS-OCT.
• This index offers a significantly improved predictive value compared to previous methods.
• The developed index can aid in determining indications for corneal cross-linking based on initial AS-OCT imaging.