Intra-operative OCT (iOCT) Image Quality Enhancement - A Super-Resolution Approach Using High Quality iOCT 3D Scans.

Effective treatment of degenerative retinal diseases will require robot-assisted intraretinal therapy delivery supported by excellent retinal layer visualisation capabilities. Intra-operative Optical Coherence Tomography (iOCT) is an imaging modality which provides real-time, cross-sectional retinal images partially allowing visualisation of the layers where the sight restoring treatments should be delivered. Unfortunately, iOCT systems sacrifice image quality for high frame rates, making the identification of pertinent layers challenging. This paper proposes a Super-Resolution pipeline to enhance the quality of iOCT images leveraging information from iOCT 3D cube scans. We first explore whether 3D iOCT cube scans can indeed be used as high-resolution images by performing Image Quality Assessment. Then, we apply non-rigid image registration to generate partially aligned pairs, and we carry out data augmentation to increase the available training data. Finally, we use CycleGAN to transfer the quality between low-resolution (LR) and high-resolution (HR) domain. Quantitative analysis demonstrates that iOCT quality increases with statistical significance, but a qualitative study with expert clinicians is inconclusive with regards to their preferences.