A multi-view assisted registration network for MRI registration pre- and post-therapy

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Biology

Background:

• Accurate image registration of pre- and post-therapy magnetic resonance imaging (MRI) is crucial for evaluating tumor treatment efficacy.
• Current MRI registration methods struggle with blurred slices, leading to inaccurate spatial information and poor alignment.
• Single-view data limitations hinder robust registration, especially in clinical settings with variable image quality.

Purpose of the Study:

• To develop an advanced MRI registration technique that overcomes the limitations of single-view data.
• To enhance the accuracy of pre- and post-therapy MRI alignment for improved tumor evaluation.
• To mitigate the impact of blurred image regions on registration outcomes.

Main Methods:

• A multi-stream fusion-assisted registration network was proposed, integrating different-view MRIs from the same patient.
• A cross-attention guided fusion module was designed to effectively leverage accurate spatial information from multi-view data.
• The network was trained and evaluated using clinical MRI data.

Main Results:

• The proposed multi-stream network demonstrated significantly improved accuracy in MRI image registration.
• Incorporating multi-view MRI data as auxiliary information enhanced alignment precision.
• The cross-attention module effectively utilized accurate spatial information, reducing errors from blurred regions.

Conclusions:

• The multi-stream fusion-assisted registration network offers a robust solution for pre- and post-therapy MRI alignment.
• Utilizing multi-view MRI data substantially improves registration accuracy in challenging clinical cases.
• This approach enhances the reliability of therapeutic effect evaluation in oncology patients.