Thickness Speed Progression Index: Machine Learning Approach for Keratoconus Detection.

Area of Science:

• Ophthalmology
• Medical Technology
• Artificial Intelligence

Background:

• Keratoconus (KC) is a progressive thinning disorder of the cornea.
• Accurate differentiation of KC, KC suspect (KCS), and normal corneas is crucial for timely intervention.
• Current diagnostic methods rely on multiple imaging modalities.

Purpose of the Study:

• To develop and validate a pachymetry-based machine learning (ML) index for differentiating keratoconus, keratoconus suspect, and normal corneas.
• To assess the diagnostic performance of the ML algorithm using corneal thickness progression data.
• To establish a novel, potentially simpler, diagnostic tool for corneal ectasias.

Main Methods:

• A retrospective study included 349 eyes from 349 patients with normal, KC, or KCS corneas.
• Six parameters from corneal thickness progression maps (Galilei system) were used to train ML algorithms.
• The Thickness Speed Progression Index was developed and validated using 5-fold cross-validation and random search over 7 ML algorithms.

Main Results:

• The Random Forest model achieved 100% accuracy and AUROC of 1.00 in distinguishing normal from KC corneas.
• In differentiating normal, KCS, and KC corneas, the model achieved 91% overall accuracy with AUROC of 0.93 (normal), 0.83 (KCS), and 0.99 (KC).
• When including sub-classifications of KCS (topographically/tomographically normal and borderline fellow eyes), accuracy was 87% with specific AUROC values for each group.

Conclusions:

• Pachymetry-based machine learning algorithms, like the Thickness Speed Progression Index, can effectively discriminate between normal, KC, and KCS corneas.
• This ML index offers a promising approach for diagnosing corneal ectasias using solely corneal thickness data.
• The findings suggest a potential for simplified, accurate screening and diagnosis of keratoconus.