Fast acquisition and reconstruction of optical coherence tomography images via sparse representation.

Optical coherence tomography (OCT) is a powerful technology for monitoring and diagnosing eye diseases. However, speckle noise is not beneficialfor improving OCT image quality and further image analysis,such as segmentation of the retinal layer.Inspired by the rapid development of deep learning, several methods have been proposed for OCT denoising, and promising results have been obtained. Nevertheless, most methods are supervised and require paired noisy and clean images. In clinical practice, this requirement is too difficult to meet since clean images are usually obtained by averaging consecutive frames in the same location, which may lead to motion or detail blurring. To address this problem, we propose an unsupervised learning technique based on a cycle-consistent adversarial network (CycleGAN) that removes speckle noise from OCT images by learning a map from the noisy phase to the clean phase. In addition, weinterrogate our approach with extensive quantitative and qualitative metrics and compare it with several state-of-art methods. The results of the experiments indicate that the proposed method shows better speckle reduction performance than traditional methods and deep-learning methods.