A review of convolutional neural networks-based medical image registration methods

Read full text Summarise with VIRAT
Authors: Hussein Abdulkhalek Midhat, Asim Majeed Murshid
Cited by: 0
References: 42

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Registration of medical images is an integral part of the modern imaging processes, which allows spatially aligning anatomical structures in various modalities or longitudinal acquisitions. Although they have been used long, the traditional registration algorithms can be limited due to large anatomical variation, complicated deformations, and uneven intensity distribution. Recent developments in Convolutional Neural Networks (CNNs) have led to offering data-driven methods that can learn data-driven informative representations in space and predict deformation fields with enhanced accuracy and stability. This review focuses on the key types of CNN-based supervised, unsupervised and weakly supervised registration and describes how novel architecture-based innovations of encoder-decoder networks, attention systems, and multi-resolution systems have contributed to robustness in multimodal and clinically challenging ones. The datasets and evaluation methodologies that are widely used in the literature are also mentioned in the discussion, which gives a consistent perspective on the modern trends in benchmarking practices. Through the synthesis of these advances, the review outlines the advancement in the registering of medical images using CNN and establishes directions that are emerging in the research that can define the future generation of the registration methods.

Fields of Study & Subjects

Medical Image Segmentation Techniques Medical Imaging and Analysis Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
International Journal of Communication and Information Technology
Volume
7
Issue
1
Pages
165–175
Publication date
1 January 2026
Language
EN