Deep learning-based Diabetic Retinopathy assessment on embedded system.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Diabetic Retinopathy (DR) impacts vision and is diagnosed via retinal image analysis.
• Human error in DR diagnosis can lead to misdiagnosis.
• Current diagnostic methods require expert ophthalmologist interpretation.

Purpose of the Study:

• To develop a low-cost, embedded deep learning system for DR severity grading.
• To assist ophthalmologists in diagnosing diabetic retinopathy.
• To improve the accuracy and efficiency of DR detection.

Main Methods:

• A compact deep learning algorithm, Deep-DR-Net, was designed for embedded systems.
• A cascaded encoder-classifier network with residual connections was utilized for model size reduction.
• Diverse convolutional layers were employed to enhance feature extraction for DR grading.

Main Results:

• The proposed Deep-DR-Net system demonstrated capability in detecting diabetic retinopathy.
• The system successfully graded the severity of diabetic retinopathy symptoms from retinal images.
• Experimental results validated the system's effectiveness in DR assessment.

Conclusions:

• The developed deep learning-based embedded system offers a viable solution for DR diagnosis.
• Deep-DR-Net provides an accurate and efficient method for grading diabetic retinopathy severity.
• This technology has the potential to reduce diagnostic errors and improve patient outcomes.