FPGA-based Edge Detection for Diabetic Retinopathy Screening

Early detection of Diabetic Retinopathy (DR) is critical to prevent vision loss caused by diabetes. This paper presents an FPGA-based implementation of edge detection algorithms to identify retinal abnormalities such as aneurysms, hard exudates (yellow spots) and cotton wool spots associated with DR. The present study focuses on preprocessing retinal images, applying Gaussian smoothing to reduce noise, and utilizing edge detection filters such as Sobel, Prewitt, Scharr, Laplacian, and Roberts. The proposed methodology computes gradient magnitudes to detect edges, followed by thresholding to identify significant features. Performance metrics, including Peak Signal-to-Noise Ratio (PSNR), Mean Square Error (MSE), Structural Similarity Index (SSIM), precision, recall, and entropy, are utilized to assess the efficiency of each filter. Experimental results demonstrate that Gaussian smoothing significantly improves edge detection accuracy and offers reduction in noise, while the presented hybrid filters offer improved noise reduction in comparison to conventional filters.