DCE MR Perfusion Mapping Under Free-Breathing Motion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing dynamic contrast-enhanced (DCE) MR images under a free-breathing protocol struggle with respiratory motion and contrast agent enhancement, leading to inaccurate lesion quantification due to spatial misalignment and intensity variations, particularly in tissues like the liver and lung.
Innovation Solution
A neural network system comprising a first sub-network for contrast enhancement state mapping and a second sub-network for registration, utilizing generative adversarial networks and cycle GANs, to align and normalize DCE MR images, enabling accurate perfusion metric assignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D elastic registration is used to compensate respiratory motion, then spatial alignment between DCE MR images is improved, but intensity variations due to contrast agent enhancement cause inaccurate registration results
Solution Approach 1:
The patent segments the registration process into two distinct stages: first performing rigid registration to correct respiratory motion and establish spatial alignment, then performing elastic registration on the rigidly registered images to handle contrast agent enhancement. This segmentation allows each registration type to optimize for its specific purpose without interference from the other challenge.
Solution Approach 2:
The patent applies rigid registration as a preliminary step before elastic registration. By first correcting the respiratory motion-induced spatial misalignment through rigid transformation, the subsequent elastic registration operates on images that are already spatially aligned, making the intensity-based similarity metric more effective at handling contrast agent enhancement.
2Ease of operation
If free-breathing protocol is used for DCE MR imaging, then patient comfort and scan feasibility are improved, but respiratory motion causes spatial misalignment between images
Solution Approach 1:
The patent replaces mechanical respiratory gating (which would require breath-holding) with a computational approach using rigid registration. The rigid registration algorithm computationally corrects respiratory motion artifacts after free-breathing acquisition, substituting mechanical control with post-processing registration to achieve spatial alignment.
3Illumination intensity
If contrast agent is administered to reduce relaxation time, then image contrast and lesion detectability are improved, but image appearance varies over time requiring complex processing
Solution Approach 1:
The patent segments the processing into distinct stages: rigid registration to handle spatial misalignment, followed by elastic registration to handle intensity variations. This segmentation simplifies the overall processing by addressing different challenges (spatial vs. intensity) in separate, specialized steps rather than attempting a single complex transformation.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Technique for assigning a perfusion metric to dynamic contrast-enhanced, DCE, magnetic resonance, MR, images, (104) the DCE MR images obtained from a MR scanner (102) and under a free-breathing protocol is provided. As to a neural network system (100) aspect, a neural network system (100) for assigning a perfusion metric to DCE MR images, comprises an input layer (112) configured to receive at least one DCE MR image (104a) representative of a first contrast enhancement state and of a first respiratory motion state and at least one further DCE MR image (104b) representative of a second contrast enhancement state and of a second respiratory motion state. The neural network (100) further comprises an output layer (116) configured to output at least one perfusion metric based on the at least one DCE MR image (104a) and the at least one further DCE MR image (104b). The neural network system (100) with interconnections between the input layer (112) and the output layer (116) is trained by a plurality of datasets, each of the datasets comprising an instance of the at least one DCE MR image (104a) and of the at least one further DCE MR image (104b) for the input layer (114) and the at least one perfusion metric for the output layer (116).