Neural Network Inversion for Magnetic Resonance Elastography Stiffness

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current inversion algorithms for magnetic resonance elastography (MRE) and other elastography techniques face limitations in resolution and robustness due to noise sensitivity and the need for smoothing, which hinders accurate estimation of tissue mechanical properties, especially in smaller regions.

Innovation Solution

Implementing a machine learning approach using artificial neural networks (ANNs) to invert displacement data acquired during MRE, allowing for stable mechanical property estimation without relying on local homogeneity assumptions and enabling computation in the presence of noise, with the ability to process data from single or multiple frequencies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional inversion algorithms are used to compute stiffness from displacement data, then the computation can be performed with standard methods, but the resolution is limited and smoothing is required which reduces measurement precision

Engineering Contradiction:
Improvestiffness estimation precisionVSAvoidspatial resolution
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

The patent replaces traditional mechanical inversion algorithms with a neural network-based computational system. The neural network is trained on simulated data to learn the mapping from displacement data to stiffness values, substituting the conventional mathematical inversion process with a data-driven machine learning approach that achieves both high precision and fine spatial resolution without requiring smoothing operations

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary training of the neural network using extensively simulated displacement data with known stiffness values before actual measurement. This pre-training phase allows the network to learn optimal mappings and patterns, enabling it to achieve high measurement precision on real data without requiring post-processing smoothing that would otherwise reduce spatial resolution

Inventive Principle:
Principle #10Preliminary action

2Reliability

If smoothing is applied to displacement data to improve stability, then noise sensitivity is reduced, but the effective resolution is degraded

Engineering Contradiction:
Improvestability of stiffness estimateVSAvoidspatial resolution
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent replaces the traditional smoothing-based noise reduction approach with a neural network that inherently handles noisy data through its training process. The network learns to distinguish signal from noise patterns during training on simulated data, providing stable stiffness estimates without applying spatial smoothing filters that would degrade resolution

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates a computational model (neural network) that copies and learns from extensively simulated training data representing various tissue conditions and noise levels. This trained model then processes actual measurement data, providing stable results by applying patterns learned from the simulated copies rather than requiring smoothing of the original data

Inventive Principle:
Principle #26Copying

3Productivity

If algorithms assume local homogeneity to simplify computation, then processing is more efficient, but accuracy in focal diseases is reduced

Engineering Contradiction:
Improvecomputation efficiencyVSAvoidaccuracy in focal regions
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent enables the neural network to capture local variations in tissue properties by training it on simulated data that includes focal lesions and inhomogeneous regions. The network learns to process data at fine spatial scales without requiring the simplifying assumption of local homogeneity, maintaining both computational efficiency and accuracy in focal disease detection

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent substitutes traditional inversion algorithms that rely on local homogeneity assumptions with a neural network that can naturally handle heterogeneous tissue properties. The network's distributed processing architecture allows it to efficiently compute stiffness values in focal regions without requiring simplifying assumptions about tissue uniformity

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3710848B1Methods for estimating mechanical properties from magnetic resonance elastography data using artificial neural networks
Publication Date: 2024.01.03 MAYO FOUNDATION FOR MEDICAL EDUCATION & RESEARCH
  • EP3710848B1 patent drawingFigure 1~3
  • EP3710848B1 patent drawingFigure 4
  • EP3710848B1 patent drawingFigure 5

AI summary

Described here are systems and methods for magnetic resonance elastography ("MRE"), or other imaging-based elastography techniques, in which a machine learning approach, such as an artificial neural network, is implemented to perform an inversion of displacement data in order to generate estimates of the mechanical properties, such as stiffness and damping ratio, of tissues in a subject.