Validation of the Eyerobo FC Portable Fundus Camera for Diabetic Retinopathy Screening Using Public Datasets and Deep...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) screening is crucial for preventing vision loss.
• Portable fundus cameras offer potential for wider accessibility in DR screening.
• Validating new devices against established benchmarks is essential for clinical adoption.

Purpose of the Study:

• To evaluate the diagnostic performance of the portable Eyerobo FC fundus camera for DR screening.
• To compare the Eyerobo FC's performance against a desktop fundus camera benchmark.
• To assess the efficacy of transfer learning for applying AI algorithms across different fundus imaging platforms.

Main Methods:

• A prospective validation study using a three-tier experimental design.
• Training a deep learning model (EfficientNet-B4) on desktop fundus camera images.
• Validating the AI model on images from the Eyerobo FC without retraining, comparing sensitivity and specificity against a desktop benchmark.

Main Results:

• The Eyerobo FC achieved noninferior sensitivity (92.3%) and specificity (94.2%) compared to the desktop benchmark (92.7% and 94.3%).
• Area Under the Curve (AUC) for Eyerobo FC was 0.977, exceeding the desktop benchmark's 0.952.
• The AI model demonstrated 93.3% accuracy with excellent inter-grader agreement and a high image gradability rate (94.4%).

Conclusions:

• The Eyerobo FC fundus camera demonstrates comparable diagnostic performance to desktop systems for DR screening using AI.
• Transfer learning enables successful cross-domain application of AI algorithms, supporting algorithmic generalizability.
• Portable AI-assisted screening with the Eyerobo FC is suitable for resource-constrained and point-of-care settings.