Generative AI qMRI Reconstruction for Fast, Precise Tissue Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Quantitative MRI (qMRI) requires lengthy acquisition times for high SNR MR images, leading to inefficient and inaccurate reconstructions and approximate model fitting, which results in un-regularized estimated tissue properties maps.
Innovation Solution
Employ generative AI models, specifically diffusion models, to reconstruct contrast-weighted MR images and estimate quantitative maps by incorporating data consistency terms and a priori probability density functions for regularization, using training mechanisms that decouple prior knowledge from the reconstruction task.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple high SNR MR images are acquired for qMRI, then measurement precision of tissue properties is improved, but acquisition time increases significantly
Solution Approach 1:
The patent applies preliminary action by training generative models offline beforehand to learn the mapping between undersampled MR images and quantitative tissue properties. During actual scanning, these pre-trained models rapidly generate quantitative maps from accelerated acquisitions, eliminating the need for repeated iterative fitting and achieving both high precision and fast acquisition times.
Solution Approach 2:
The patent substitutes the traditional mechanical/iterative model fitting process with a learned generative AI system. Instead of performing time-consuming iterative optimization algorithms during scanning, the system uses pre-trained neural networks that have learned the underlying physical models and tissue property relationships, replacing the mechanical fitting process with an intelligent inference system.
2Productivity
If acquisition time is reduced for faster scanning, then productivity is improved, but measurement precision and reconstruction accuracy deteriorate
Solution Approach 1:
The system performs preliminary training of generative models using high-quality reference data before deployment. This offline preparation enables the model to compensate for undersampling effects during fast scanning, maintaining measurement precision even when acquisition time is reduced for improved productivity.
Solution Approach 2:
The patent introduces generative AI models as an intermediary between the accelerated MR imaging process and the quantitative tissue property estimation. This intermediary learns to correct artifacts and compensate for information loss from rapid scanning, bridging the gap between fast acquisition and accurate measurement.
3Device complexity
If standard least square fitting is used for quantitative map computation, then device complexity is kept simple, but reconstruction quality and measurement accuracy are insufficient
Solution Approach 1:
The patent replaces simple least-square fitting algorithms with sophisticated generative AI models that have learned complex tissue property relationships. This substitution dramatically improves reconstruction quality and measurement accuracy, trading increased algorithmic complexity for superior quantitative map quality.
Solution Approach 2:
The system transforms the processing approach by changing from deterministic iterative fitting parameters to learned probabilistic parameters from generative models. This parameter transformation enables the system to capture complex tissue variations and improve reconstruction quality beyond what traditional methods achieve.
Data Source
AI summary
Systems and methods for image reconstruction and quantitative MRI. Generative models such as diffusion models are used to reconstruct MR images and generative models and constrained mathematical models fit to estimate quantitative maps from the reconstructed MR images.