Related Experiment Video
Updated: Feb 25, 2026
05:05
Four-Dimensional CT Analysis Using Sequential 3D-3D Registration
Published on: November 23, 2019
8.5K
Biomechanics-Informed Non-Rigid Medical Image Registration With Elasticity Theories
IEEE Transactions on Medical Imaging
|February 23, 2026
Summary
Physics-informed neural networks (PINNs) enable efficient and accurate soft tissue registration by enforcing biomechanical constraints. This approach improves medical image alignment, outperforming existing methods in computational speed and accuracy.
Area of Science:
- Medical Imaging
- Computational Biomechanics
- Machine Learning
Background:
- Biomechanical modeling enhances medical image registration by ensuring biophysically plausible spatial transformations.
- Current methods for biomechanical-constrained registration are computationally intensive or require cumbersome data generation.
- A need exists for efficient and accurate registration techniques that incorporate biomechanical principles.
Purpose of the Study:
- To apply physics-informed neural networks (PINNs) for biomechanically constrained soft tissue registration.
- To formulate and develop algorithms for both forward (registration) and inverse (parameter estimation) problems using PINNs.
- To compare the performance of linear and nonlinear elasticity theories within the PINN framework for registration and parameter identification.
Main Methods:
- Physics-informed neural networks (PINNs) were utilized to model 3D elastic soft tissues and enforce biomechanical constraints via partial differential equations (PDEs).
- Forward problem: Registration algorithm aligning point sets using PINN-imposed biomechanics.
- Inverse problem: Algorithm for estimating physical parameters (material properties) concurrently with registration.
- Comparison of network architectures (single vs. dual branch) and elasticity theories (linear vs. nonlinear).
Main Results:
- PINNs-based registration methods achieved state-of-the-art performance, surpassing existing biomechanical-model-based and learning-based approaches.
- The proposed methods successfully ensured biomechanical constraints of soft tissues post-registration.
- Evaluations on MRI-US registration and prostate cancer biopsy data demonstrated the efficacy of the PINN approach.
Conclusions:
- PINNs offer an efficient and effective framework for biomechanically constrained medical image registration.
- The developed PINN-based methods provide accurate registration and reliable physical parameter estimation.
- This work presents a significant advancement in applying physics-informed machine learning to soft tissue biomechanics and medical imaging.
