Chapter 13 - Medical image super-resolution

Abstract

Super-resolution (SR) algorithms have been widely developed in the last two decades in order to enhance the clarity of image details by obtaining more high-resolution images from one or more low-resolution ones. Such algorithms have been extensively applied in different fields, including primarily remote sensing, analysis of human face images, biometrics, and automated driving. In particular, SR methods have been recently developed for medical image data of different modalities, such as magnetic resonance imaging (MRI) and computed tomography (CT). Medical image SR is a challenging problem for several reasons. In fact, visualizing the details of living tissues in medical imaging is hindered by constraints of acquisition time, sampling frequencies, low radiation dose, and hardware limitations. In this work, we address these challenges and introduce a taxonomy of the SR methods based on four different criteria, namely, the number of input low-resolution images (multiple images or a single image), the SR method (reconstruction-based or learning-based methods), the SR domain (image intensity or feature domains), and the resolution type (spatial, temporal, or spectral SR). Then, we provide a detailed review of the literature on the SR methods for different medical imaging modalities, including MRI, CT, ultrasound, and microscopy. Finally, current limitations and challenges are discussed, and thus several future research directions are pointed out.