Machine Learning Algorithms to Detect Subclinical Keratoconus: Systematic Review

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Keratoconus is a progressive corneal thinning and distortion disorder.
• Early detection via corneal collagen cross-linking can prevent vision loss.
• Identifying subclinical keratoconus is challenging, necessitating advanced diagnostic tools.

Purpose of the Study:

• To systematically review and evaluate literature on algorithmic detection of subclinical keratoconus.
• To assess various machine learning algorithms and their definitions for early keratoconus detection.

Main Methods:

• A systematic review of MEDLINE, Embase, Web of Science, and Cochrane Library databases (2010-2020).
• Inclusion of full-text studies utilizing algorithms for subclinical keratoconus detection with validation.
• Adherence to PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines.

Main Results:

• Analysis of 26 eligible papers comparing measured parameters and machine learning algorithm designs.
• Detailed reporting of diagnostic criteria, demographics, sample size, acquisition systems, validation, parameter inputs, and algorithm outcomes.
• Comprehensive comparison of various approaches for subclinical keratoconus identification.

Conclusions:

• Machine learning holds potential for enhancing early keratoconus detection in clinical practice.
• Lack of consensus exists on optimal corneal parameters and machine learning algorithm design.
• Further research is recommended to refine early detection and patient stratification for timely treatment.