Statistical comparison of classifiers applied to the interferential tear film lipid layer automatic classification

Area of Science:

• Ophthalmology and Computer Vision

Background:

• The tear film lipid layer's heterogeneity necessitates classification based on thickness.
• Guillon's interference pattern categories provide a framework for classifying lipid layer thickness.
• Interference patterns manifest as color texture, amenable to automated analysis.

Purpose of the Study:

• To develop and evaluate automated methods for classifying tear film lipid layer interference patterns.
• To compare various texture analysis techniques and machine learning algorithms for this classification task.

Main Methods:

• Image acquisition of the eye to capture tear film interference patterns.
• Region of interest detection and extraction of low-level features to form a feature vector.
• Classification of feature vectors using diverse texture analysis methods across three color spaces and multiple machine learning algorithms.

Main Results:

• Exhaustive study conducted on a dataset of 105 images from healthy subjects.
• Statistical analysis of results from various texture analysis and machine learning approaches.
• Maximum classification accuracy exceeding 95% achieved.

Conclusions:

• Automated classification of tear film lipid layer interference patterns is feasible and highly accurate.
• The developed method offers a faster and more objective alternative to manual expert classification.
• Potential for improved diagnosis and management of dry eye disease through objective tear film assessment.