Deformable image registration with strategic integration pyramid framework for brain MRI

Yaoxin Zhang ¹, Qing Zhu ¹, Bowen Xie ²

⁰College of Computer Science, Beijing University of Technology, No. 100 Pingleyuan, Chaoyang District, Beijing 100124, China.

Magnetic Resonance Imaging +

|

March 23, 2025

View abstract on PubMed

Summary

This study introduces a new deep learning network for brain MRI registration, improving accuracy and robustness in handling large deformations by integrating features across different scales and network structures.

Area Of Science

Medical Imaging
Artificial Intelligence
Neuroscience

Background

Medical image registration is vital for clinical applications, particularly brain MRI for diagnosis and treatment planning.
Deep learning methods have advanced deformable registration, but struggle with large deformations and multi-level feature relationships.

Purpose Of The Study

To develop a novel, flexible, and efficient deep learning registration network for brain MRIs.
To address limitations in handling large deformations and complex anatomical feature relationships.

Main Methods

Proposed a strategic integration registration network utilizing a pyramid structure.
Integrated features at different scales and combined CNN encoder with Transformer decoder.
Implemented progressive optimization iterations to mitigate error accumulation in pyramid structures.

Main Results

Extensive evaluations on multiple brain MRI datasets demonstrated superior performance.
The proposed method achieved higher registration accuracy and robustness compared to existing deep learning approaches.

Conclusions

The novel network effectively handles spatial relationships and improves accuracy in brain MRI registration.
This approach offers a more efficient and robust solution for complex registration tasks.

Keywords:

Attention Coarse-to-fine Medical image registration