Detection of subclinical keratoconus using an automated decision tree classification

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Keratoconus is a progressive corneal ectasia.
• Early detection of subclinical keratoconus is crucial for preventing vision loss and optimizing surgical outcomes.
• Current diagnostic methods can be subjective and time-consuming.

Purpose of the Study:

• To develop and evaluate an automated method for detecting subclinical keratoconus using a classification and regression tree algorithm.
• To assess the performance of the machine learning model in differentiating between normal corneas, forme fruste keratoconus, and keratoconus.

Main Methods:

• A retrospective case-control study involving 372 eyes from 197 patients.
• Corneal imaging using a dual Scheimpflug analyzer.
• Analysis of 55 corneal parameters using a classification and regression tree algorithm for automated classification.

Main Results:

• The automated classifier achieved 100% sensitivity and 99.5% specificity in distinguishing normal from keratoconus eyes.
• It demonstrated 93.6% sensitivity and 97.2% specificity in differentiating normal from forme fruste keratoconus.
• Key discriminant variables included posterior corneal surface asymmetry and thickness spatial distribution.

Conclusions:

• The machine learning classifier exhibits high performance in detecting forme fruste keratoconus.
• This automated tool offers a more objective approach to medical reasoning for corneal condition diagnosis.
• It has the potential to assist surgeons in identifying ectasia-susceptible corneas prior to refractive surgery.