Diabetic retinopathy detection through convolutional neural networks with synaptic metaplasticity

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Diabetic retinopathy, a diabetes complication, causes irreversible vision loss.
• Early detection of diabetic retinopathy is vital for preventing blindness.
• Automated detection methods using fundus images are essential for timely diagnosis.

Purpose of the Study:

• To develop an automated method for early diabetic retinopathy detection.
• To leverage bio-inspired synaptic metaplasticity in convolutional neural networks.
• To improve the accuracy and efficiency of diabetic retinopathy screening.

Main Methods:

• A bio-inspired approach incorporating synaptic metaplasticity into convolutional neural networks.
• Synaptic metaplasticity was integrated into the backpropagation stage of convolutional operations.
• The method was evaluated using a public diabetic retinopathy dataset and four CNN architectures.

Main Results:

• Convolutional neural networks enhanced with synaptic metaplasticity demonstrated improved learning rates and accuracy.
• The proposed method outperformed existing approaches, even with smaller training datasets.
• The InceptionV3 architecture with synaptic metaplasticity achieved 95.56% accuracy, 94.24% F1-score, 98.9% precision, and 90% recall.

Conclusions:

• Convolutional neural networks incorporating synaptic metaplasticity offer a high-performance solution for early diabetic retinopathy detection.
• The approach exhibits fast convergence rates and simplified training.
• This method shows significant potential for clinical application in screening for diabetic retinopathy.