[Paper Review] Deep Learning for Medical Image Registration: A Comprehensive Review

Subrato Bharati, M. Rubaiyat Hossain Mondal+2 more|arXiv (Cornell University)|Apr 24, 2022

Radiomics and Machine Learning in Medical Imaging22 citations

TL;DR

This paper surveys deep learning-based medical image registration, covering supervised, unsupervised, GAN-based, and deep iterative approaches, across monomodal and multimodal imaging, and discusses challenges and future directions.

ABSTRACT

Image registration is a critical component in the applications of various medical image analyses. In recent years, there has been a tremendous surge in the development of deep learning (DL)-based medical image registration models. This paper provides a comprehensive review of medical image registration. Firstly, a discussion is provided for supervised registration categories, for example, fully supervised, dual supervised, and weakly supervised registration. Next, similarity-based as well as generative adversarial network (GAN)-based registration are presented as part of unsupervised registration. Deep iterative registration is then described with emphasis on deep similarity-based and reinforcement learning-based registration. Moreover, the application areas of medical image registration are reviewed. This review focuses on monomodal and multimodal registration and associated imaging, for instance, X-ray, CT scan, ultrasound, and MRI. The existing challenges are highlighted in this review, where it is shown that a major challenge is the absence of a training dataset with known transformations. Finally, a discussion is provided on the promising future research areas in the field of DL-based medical image registration.

Motivation & Objective

• Motivate the study of DL-based medical image registration and categorize existing approaches.
• Summarize supervised, dual supervised, and weakly supervised registration paradigms.
• Review similarity-based, GAN-based, and deep iterative registration methods.
• Discuss challenging data issues and diverse imaging modalities (X-ray, CT, ultrasound, MRI).
• Highlight future research directions in the field.

Proposed method

• Classify registration methods into supervised, weakly supervised, and unsupervised categories.
• Describe similarity-based and GAN-based registration approaches used in DL.
• Explain deep iterative registration with emphasis on deep similarity-based and reinforcement-learning-based strategies.
• Survey application domains and imaging modalities for monomodal and multimodal registration.
• Identify key challenges such as lack of datasets with known transformations.

Experimental results

Research questions

• RQ1 What are the main DL-based categories of medical image registration and their characteristics?
• RQ2 How do supervised, weakly supervised, and unsupervised registration methods compare in DL?
• RQ3 What are the roles of similarity-based, GAN-based, and deep iterative approaches in registration?
• RQ4 What are the primary application areas and imaging modalities of DL-based registration?
• RQ5 What are the major challenges and future directions in this field?

Key findings

• The review covers supervised, dual supervised, and weakly supervised registration frameworks.
• Unsupervised registration is discussed via similarity-based and GAN-based strategies.
• Deep iterative registration is described with emphasis on deep similarity-based and reinforcement learning methods.
• Applications span monomodal and multimodal imaging, including X-ray, CT, ultrasound, and MRI.
• A major challenge identified is the absence of training data with known ground-truth transformations.
• The paper outlines promising future research directions in DL-based medical image registration.