Medical image registration in the era of Transformers: A recent review

Abstract

Motivated by the rapid and current progress to develop intelligent image-guided intervention tools, we aim in this paper to present, a recent review of a specific family of deep learning-based medical image registration methods which is Transformer-based methods. Within the revolution of Transformers in the domain of Natural Language Processing, Vision Transformers (ViTs) have been widely investigated in many applications of medical imaging including image registration. The main goal of this work, is to survey the different models proposed in the literature to tackle the task of medical image registration using ViTs and classify them according different viewpoints by focusing on the strategy of the use of Transformer in each method. For this purpose, twenty-nine papers published between 2021 and 2024 have been reported and discussed in this study. Furthermore, to provide a comprehensive guidance to the researchers in the field, we give an overview of the different attention mechanisms and variants of Transformer used in registration architectures. A concise summary of reviewed works, datasets, evaluation metrics as well as challenges with future directions have been introduced in this paper.

Keywords

• Medical image
• Vision Transformer
• Attention
• Registration
• Affine
• Deformable