Vision-Based Eye Image Classification for Ophthalmic Measurement Systems.

Area of Science:

• Ophthalmic instrumentation
• Medical imaging analysis
• Computer vision in healthcare

Background:

• Ophthalmic instrument accuracy is challenged by invalid eye image frames from non-collaborative patients and operators.
• Effective pupillary light reflex measurement requires robust image analysis to overcome these challenges.

Purpose of the Study:

• To investigate and compare machine learning, deep learning, and expert systems for improving ophthalmic instrument performance.
• To enhance the accuracy of pupillary light reflex measurements using vision-based classification algorithms.

Main Methods:

• Development and comparison of machine learning, deep learning, and expert systems for eye image classification.
• Creation and release of the PopEYE dataset (15,000 images from 22 subjects) for testing.
• Analysis of classification accuracy and computational load for embedded system implementation.

Main Results:

• Evaluation of different vision-based algorithms for classifying eye status in ophthalmic images.
• Assessment of computational efficiency for potential deployment on resource-constrained embedded boards.
• Demonstration of improved performance metrics for the ophthalmic instrument using the developed methods.

Conclusions:

• Vision-based classification algorithms can significantly improve ophthalmic instrument performance for pupillary light reflex measurements.
• The PopEYE dataset provides a valuable resource for research in ophthalmic image analysis.
• The study offers insights into selecting appropriate algorithms for embedded ophthalmic devices based on accuracy and computational load.