Super-resolution technology to simultaneously improve optical & digital resolution of optical coherence tomography via deep learning

Optical coherence tomography (OCT) has stimulated a wide range of medical image-based diagnosis and treatment. In cardiac imaging, OCT has been used in assessing plaques before and after stenting. While needed in many scenarios, high resolution comes at the costs of demanding optical design and data storage/transmission. In OCT, there are two types of resolutions to characterize image quality: optical and digital resolutions. Although multiple existing works have heavily emphasized on improving the digital resolution, the studies on improving optical resolution or both resolutions remain scarce. In this paper, we focus on improving both resolutions. In particular, we investigate a deep learning method to address the problem of generating a high-resolution (HR) OCT image from a low optical and low digital resolution (L
²
R) image. To this end, we have modified the existing super-resolution generative adversarial network (SR-GAN) for OCT image reconstruction. Experimental results from the human coronary OCT images have demonstrated that the reconstructed images from highly compressed data could achieve high structural similarity and accuracy in comparison with the HR images. Besides, our method has obtained better denoising performance than the block-matching and 3D filtering (BM3D) and Denoising Convolutional Neural Networks (DnCNN) denoising method.