Retinal Optical Coherence Tomography Image Denoising using Convolutional Autoencoder

Optical Coherence Tomography (OCT) is an imaging modality that is frequently used to monitor and diagnose retinal diseases. However, speckle noise in the acquired OCT images makes them difficult to interpret and analyse. In this paper, we present convolutional autoencoder (CAE) architecture for denoising retinal OCT images that incorporates an encoder and a decoder. Encoder extract relevant features of OCT images and subsequently decoder reconstruct the denoised images using the extracted features. To assess the efficacy of the proposed CAE, we used Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Equal Number of Looks (ENL) as evaluation metrics. Furthermore, the denoising efficacy of the presented model was compared against conventional image denoising methodologies, namely Block-matching and 3D filtering (BM3D), Non-local Means (NLM), Median, and Bilateral filtering. The superiority of the proposed denoising model on OCT image is indicated by the evaluation metrics.