Axial Super-Resolution Study for Optical Coherence Tomography Images Via Deep Learning

Optical coherence tomography (OCT) is a noninvasive, high resolution, and real-time imaging technology that has been used in ophthalmology and other medical fields. Limited by the point spread function of OCT system, it is difficult to optimize its spatial resolution only based on hardware. Digital image processing methods, especially deep learning, provide great potential in super-resolving images. In this paper, the matched axial low resolution (LR) and high resolution OCT image pairs from actual OCT imaging are collected to generate the dataset by our home-made spectral domain OCT (SD-OCT) system. Several methods are selected to super-resolve LR OCT images. It is shown from the experimental results that the residual-in-residual dense block network (RRDBNet) trained with different loss functions performs the best super-resolution for OCT images, and it is demonstrated from the preliminary results that deep learning methods have good generalization and robustness between OCT systems. We believe deep learning methods have broad prospects in improving the quality of OCT images.