Medical image super-resolution (SR) reconstruction is an active area of research, with the potential to make medical imaging safer and more accessible by enhancing the quality of images produced with low-cost equipment or reduced radiation doses. This paper provides a comprehensive overview of the deep learning (DL) approaches utilized in the medical image SR field, systematically categorizing recent developments since 2018.

The paper explains how SR functions in medical imaging and its potential to impact areas such as cardiac imaging, cancer screening, and brain diagnostics. SR methods are organized into three main types: 2D image SR (further divided into single-image and reference-based methods), video SR, and 3D volume SR, with each category further subdivided based on the DL architectures used. The authors discuss selected designs, including convolutional neural networks, generative adversarial networks, transformers, and so on, expertly summarizing the strengths and weaknesses of each approach. This review situates itself within the broader context of medical image enhancement and aims to clarify the current state, methods, and future directions for SR technology.

For a general audience, this review offers an engaging introduction to SR in medical imaging. However, it is particularly valuable for researchers, medical imaging professionals, and engineers working on artificial intelligence (AI) solutions for healthcare. The content is well constructed and grounded in real-world applications, and addresses practical challenges such as limited or incomplete datasets, the need for structural integrity, and factors like computational intensity, among others. Diagrams throughout the text are highly useful in navigating the detailed information presented.

What is missing, however, are sample images before and after SR processing, which would greatly enhance the paper. To support this, readers may refer to [1,2,3].