EMKLAS: A New Automatic Scoring System for Early and Mild Keratoconus Detection

Area of Science:

• Ophthalmology
• Medical Informatics
• Biomedical Engineering

Background:

• Keratoconus (KC) detection and classification, especially in early and preclinical stages, lack a definitive gold standard.
• Accurate diagnosis is crucial for timely intervention and management of KC progression.

Purpose of the Study:

• To develop a unique predictive model for classifying early and mild keratoconus (KC).
• To establish the probability of correct classification for each case using demographic, optical, and geometric variables.
• To create a web application for real-time predictions based on new patient data.

Main Methods:

• A dataset of 178 eyes (74 controls, 104 KC) was analyzed, including demographic, clinical, pachymetric, and geometric parameters.
• An ordinal logistic regression model was implemented and programmed as a web application.
• Key variables identified: age, gender, corrected distance visual acuity, 8-mm corneal diameter, and posterior minimum thickness point deviation.

Main Results:

• The EMKLAS classifier demonstrated 73% global accuracy during training (95% CI: 65%-79%).
• Validation showed high accuracy for control (79%) and mild KC (80%) groups, with 69% accuracy for the early KC group.
• The model identified specific variables contributing to KC classification.

Conclusions:

• The developed web application and EMKLAS score offer a fast, objective, and quantitative method for assessing early and mild KC.
• This tool can assist ophthalmology professionals in the diagnostic decision-making process for keratoconus.
• The model aids in classifying preclinical KC, addressing the absence of a single gold standard.