Development of an embedded diagnostic tool for visual misalignment screening

Area of Science:

• Biomedical Engineering
• Computer Science
• Ophthalmology

Background:

• Strabismus, or eye misalignment, requires early detection for effective treatment.
• Existing screening methods can be costly and inaccessible, especially in low-resource settings.
• Computer vision and deep learning offer potential for automated, low-cost diagnostic tools.

Purpose of the Study:

• To design, implement, and validate an affordable embedded system for preliminary strabismus screening.
• To leverage computer vision and deep learning for accurate and real-time eye condition analysis.
• To create a portable, user-friendly prototype for accessible health screening and education.

Main Methods:

• Developed a hardware system using Raspberry Pi 4, USB camera, and 3D-printed chin rest.
• Implemented software in Python with OpenCV and a NASNetLarge deep learning model (TensorFlow Lite).
• Validated the system using proprietary and balanced datasets, including 10-fold cross-validation.

Main Results:

• Achieved 96.30% classification accuracy on a proprietary dataset (27 images).
• Demonstrated 95.6% average accuracy with strong generalization (F1-score, precision, recall >94%) on a 1000-image dataset.
• Validated a novel treatment mechanism for reliable eye tracking and microstrabismus detection.

Conclusions:

• The developed open-source system offers a low-cost, accurate solution for preliminary strabismus screening.
• Its portability and ease of use make it suitable for community health initiatives and educational purposes.
• The system shows significant potential for improving accessibility to eye care, particularly in resource-limited environments.