Deep Capsule Networks for Accurate Microaneurysm Detection and Classification in Retinal Images

Diabetic retinopathy (DR), is a leading cause of vision. Early detection and intervention are crucial for preventing irreversible vision impairment. Microaneurysms (MAs), small balloon-like dilatations in blood vessels, are often the earliest detectable signs of DR. This study explores an automated system using image processing and machine learning to identify microaneurysms (MAs) in five stages and early DR signs, in retinal images. Leveraging the Kaggle dataset, the system extracts key features from MAs using image processing techniques. These features are fed into a machine learning algorithm, specifically a Gaussian Naive Bayes classifier, trained to differentiate true MAs from non-MAs. The system achieves Enhanced accuracy, demonstrating its potential to assist ophthalmologists in early DR detection. Further optimization and validation on larger datasets are planned to refine the system for real-world implementation. This study highlights the promise of image processing and AI in automating DR screening, potentially improving accessibility and reducing vision loss from DR.