Networks for Joint Affine and Non-parametric Image Registration

Area of Science:

• Medical Imaging
• Deep Learning
• Computational Anatomy

Background:

• Accurate 3D medical image registration is crucial for analyzing anatomical changes over time.
• Existing methods often face challenges in speed, transformation control, and combining different registration strategies.

Purpose of the Study:

• To introduce an end-to-end deep-learning framework for rapid and regularized 3D medical image registration.
• To combine affine and non-parametric registration techniques within a unified deep learning architecture.

Main Methods:

• A three-stage framework integrating a multi-step affine network and a vector momentum-parameterized stationary velocity field (vSVF) model.
• Utilizes a U-Net-like network for momentum generation and a self-iterable map-based vSVF for refinement.
• Trained and evaluated on 3D knee MRI data from the Osteoarthritis Initiative (OAI) dataset.

Main Results:

• The framework achieves registration in a single forward pass after training, significantly improving speed.
• Demonstrates comparable performance to state-of-the-art methods.
• Offers enhanced control over transformation regularity, including the ability to produce symmetric transformations.

Conclusions:

• The proposed deep-learning framework offers a fast, efficient, and versatile solution for 3D medical image registration.
• It successfully combines affine and non-parametric registration, providing regularized and accurate transformations.
• This approach holds promise for longitudinal studies and analysis of medical imaging datasets.