Magnetic Resonance Fingerprinting Network for Tissue Parameter Estimation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Magnetic resonance fingerprinting (MRF) techniques face challenges in accurately estimating tissue parameters due to aliasing artifacts and the time-consuming, memory-intensive process of establishing a pre-determined dictionary, with deep learning approaches struggling with noise and convergence issues.

Innovation Solution

A machine-trained network is developed to reconstruct and estimate tissue parameters, incorporating a cascade of denoising and regression sub-problems, using a neural network architecture with separate components for reconstruction and regression, trained with reconstruction and regression losses to reduce noise and improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a pre-determined dictionary is established for tissue parameter estimation, then parameter values can be obtained through matching, but the process becomes time-consuming and memory-intensive

Engineering Contradiction:

Improvetissue parameter estimation accuracyVSAvoiddictionary establishment time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent pre-calculates and stores a dictionary of signal evolutions corresponding to different tissue parameter values before actual imaging. This preliminary action allows rapid parameter estimation during scanning by simply matching acquired signals against the pre-computed dictionary, avoiding time-consuming calculations during the actual measurement process.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces iterative reconstruction schemes that dynamically update tissue parameter estimates by alternating between dictionary matching and signal evolution calculations. This dynamic approach refines initial estimates through multiple iterations, improving accuracy while managing computational resources efficiently.

Inventive Principle:

Principle #15Dynamics

2Productivity

If under-sampled acquisition is used in MRF, then scanning time is reduced, but aliasing artifacts appear in the images

Engineering Contradiction:

Improvescanning speedVSAvoidaliasing artifacts

Core Design Contradiction:

ProductivityVSObject-affected harmful factors

Solution Approach 1:

The patent applies partial Fourier sampling techniques where only a portion of k-space is acquired, leveraging the redundancy in MRF signal evolutions to reconstruct full images. This partial action reduces scanning time while advanced reconstruction algorithms compensate for the missing data, minimizing aliasing artifacts.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The patent introduces iterative reconstruction schemes that act as intermediaries between the under-sampled raw data and the final images. These schemes use the dictionary of signal evolutions as a mediator to guide the reconstruction process, progressively eliminating aliasing artifacts while preserving anatomical details.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If deep learning is applied for tissue parameter estimation, then processing speed improves, but noise causes convergence and overfitting issues

Engineering Contradiction:

Improveparameter estimation speedVSAvoidmodel convergence stability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent incorporates feedback mechanisms in the deep learning architecture where the network's predictions are continuously refined by comparing with expected signal evolutions from the dictionary. This feedback loop allows the model to correct its own errors, improving convergence stability and reducing overfitting even in the presence of noise.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent pre-trains the deep learning network with synthetic data that includes various noise levels and conditions. This beforehand cushioning prepares the model to handle real-world noise effectively, improving its robustness and convergence behavior during actual parameter estimation tasks.

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

Data Source

PatentUS11346911B2Magnetic resonance fingerprinting image reconstruction and tissue parameter estimation

Publication Date: 2022.05.31 SIEMENS HEALTHINEERS AG

AI summary

Machine training a network for and use of the machine-trained network are provided for tissue parameter estimation for a magnetic scanner using magnetic resonance fingerprinting. The machine-trained network is trained to both reconstruct a fingerprint image or fingerprint and to estimate values for multiple tissue parameters in magnetic resonance fingerprinting. The reconstruction of the fingerprint image or fingerprint may reduce noise, such as aliasing, allowing for more accurate estimation of the values of the multiple tissue parameters from the under sampled magnetic resonance fingerprinting information.