Arbitrary Scale Super-Resolution for Medical Images

Single image super-resolution (SISR) aims to obtain a high-resolution output from one low-resolution image. Currently, deep learning-based SISR approaches have been widely discussed in medical processing, because of their potential achieve high-quality, high spatial resolution images without the cost additional scans. However, most existing methods are designed for scale-specific SR tasks and unable generalize over magnification scales. In this paper, we propose an approach arbitrary-scale (MIASSR), which couple meta-learning with generative adversarial networks (GANs) super-resolve at any scale [Formula: see text]. Compared state-of-the-art algorithms on single-modal magnetic resonance (MR) brain (OASIS-brains) multi-modal MR (BraTS), MIASSR achieves comparable fidelity performance best perceptual quality smallest model size. We also employ transfer learning enable tackle new modalities, such as cardiac (ACDC) chest computed tomography (COVID-CT). The source code our work is public. Thus, has become foundational pre-/post-processing step clinical analysis reconstruction, enhancement, segmentation.