Cine MRI Image Registration for Real-Time Radiotherapy Motion Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current real-time MRI-guided radiation therapy systems face a trade-off between achieving low latency for motion tracking and preserving image quality, with existing methods struggling to accurately and efficiently register medical imagery during radiation therapy.
Innovation Solution
A deep learning model for fast deformable image registration using cine MRI scans, trained during MR-Linac treatments, predicts frame-by-frame motion and aligns images using a convolutional neural network architecture, incorporating a U-Net structure and a loss function to balance registration error and gradient magnitude, and integrates EKG signals for cardiac and respiratory motion analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image registration methods are used during real-time MRI-guided radiation therapy, then image quality is preserved, but computation time and latency are too high for real-time monitoring
Solution Approach 1:
The patent replaces conventional mechanical/image-processing-based registration methods with a deep learning neural network system. The trained DL model processes cine MRI frames to generate motion vector fields and register images, achieving both high accuracy and real-time performance (processing times of 60-100 ms per frame), thereby resolving the contradiction between registration precision and computation speed.
2Loss of time
If deep learning methods with radial k-space sampling are used to reduce latency, then processing speed improves, but radial artifacts degrade image quality
Solution Approach 1:
The patent changes the k-space sampling parameters by using fully or partially sampled Cartesian k-space grids instead of radial sampling. This parameter change eliminates radial artifacts while maintaining the fast processing capability of the deep learning model, achieving both low latency (60-100 ms) and high image quality suitable for clinical review.
3Speed
If faster cine MRI acquisition is implemented to achieve real-time monitoring, then latency is reduced, but image quality deteriorates
Solution Approach 1:
The patent performs preliminary training of the deep learning model using high-quality, fully sampled cine MRI data. This preliminary action creates a robust model that can then process faster, lower-quality cine MRI acquisitions (at 8 fps or higher) in real-time, effectively decoupling the quality requirements of training data from the speed requirements of operational data.
Solution Approach 2:
The patent replaces the direct relationship between acquisition speed and image quality with an intermediate deep learning processing step. The DL model acts as a computational bridge that can process fast-acquired images and produce high-quality registration results, breaking the direct trade-off between frame rate and image quality.
4Productivity
If automated tracking algorithms are used to monitor patient motion, then real-time monitoring capability is achieved, but registration error remains high
Solution Approach 1:
The patent replaces traditional automated tracking algorithms (such as SIFT-based methods) with a deep learning-based motion estimation network. This substitution reduces registration errors significantly while maintaining real-time monitoring capability, as the DL model learns optimal feature matching and deformation patterns from training data, achieving sub-millimeter accuracy at 60-100 ms processing speeds.
Data Source
AI summary
This invention provides a deep learning (DL) model for fast deformable image registration using 2D sagittal cine MRI acquired during radiation therapy. A DL model for fast deformable image registration is trained using cine MRI scans acquired during MR-Linac treatments of thoracic and abdominal tumors. The model uses a pair of cine MRI images as inputs and outputs a dense motion vector field (MVF) which aligns the images. The trained model is applied to predict frame by frame motion from cine MRIs in which both cardiac and respiratory motion are visible. The number of respirations and heart beats is automatically extracted by performing peak detection on high-frequency and low-frequency components of the MVF displacements corresponding to the chest wall and cardiac regions.