Denoising Retinal OCT Images Using Classical and Advanced Filtering Techniques

Digital image processing employs mathematical algorithms to enhance medical imaging. Retinal Optical Coherence Tomography (OCT), a non-invasive technique, offers high-resolution cross-sectional views of the retina, crucial for early detection of conditions such as glaucoma, diabetic retinopathy, and age-related macular degeneration. However, OCT images are often compromised by noise arising from acquisition, transmission, or hardware limitations, including Gaussian, salt-and-pepper, speckle, and Poisson noise, which can obscure fine retinal structures. This study systematically investigates the impact of these noise types on OCT images by artificially introducing noise and evaluating various denoising filters. Classical methods, including mean, median, and mode filters, were compared with advanced techniques such as Wiener filtering, wavelet-based denoising, and non-local means filtering, using both custom MATLAB implementations and built-in functions. Quantitative assessment was performed using Peak Signal-to-Noise Ratio (PSNR) and Signal-to-Noise Ratio (SNR). Results indicate that Wiener filtering excels in mitigating Gaussian, Poisson, and speckle noise, while median and mode filters are most effective against salt-and-pepper noise. For real OCT images, wavelet-based denoising provides an optimal trade-off between noise suppression and edge preservation. These insights inform the selection of appropriate filters for OCT image enhancement, supporting improved clinical analysis.