ANN-Based MRE Stiffness Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional magnetic resonance elastography (MRE) techniques face challenges in accurately estimating tissue stiffness due to low signal-to-noise ratios, leading to noisy and biased estimates, particularly when calculating derivatives from displacement data.
Innovation Solution
A system and method utilizing a trained artificial neural network (ANN) to transform MRE data, where the ANN is trained with noisy input datasets to evaluate performance and provide analytical solutions for derivatives, improving the accuracy of stiffness estimation by processing displacement data from MRE sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional finite difference methods are used to calculate derivatives from MRE displacement data, then the estimation process is simple and direct, but the results are noisy and biased due to low signal-to-noise ratio
Solution Approach 1:
The patent replaces conventional finite difference mathematical operations with a deep learning-based computational model. The neural network architecture substitutes the mechanical differentiation process, using learned transformations to estimate derivatives from noisy MRE data while suppressing noise and bias, thereby achieving higher measurement precision without the simplicity of traditional methods
Solution Approach 2:
The patent transforms the approach by changing the fundamental parameters of the estimation process. Instead of using fixed finite difference formulas, the system employs a neural network with learnable parameters that adapt to the specific characteristics of the MRE data, including the low signal-to-noise ratio conditions, enabling accurate derivative estimation through data-driven parameter optimization
2Reliability
If regularization techniques are applied to overcome noise amplification, then estimation stability improves, but the complexity of identifying appropriate regularization parameters increases
Solution Approach 1:
The neural network model performs self-service by automatically learning and incorporating appropriate regularization during its training process. The network autonomously determines the optimal balance between noise suppression and detail preservation through its learned representations, eliminating the need for external manual tuning of regularization parameters and reducing the complexity of parameter selection while maintaining estimation stability
Solution Approach 2:
The system implements feedback mechanisms during neural network training where the model continuously refines its estimates based on the input data characteristics. This feedback loop enables the network to adaptively adjust its internal parameters to optimize both stability and accuracy, avoiding the need for manual regularization parameter selection while ensuring reliable estimates
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The approach significantly enhances the accuracy of stiffness estimation by reducing noise and bias, providing more reliable mechanical property assessments compared to conventional finite difference methods.
Implementation Method 1
Magnetic resonance elastography (MRE) is a magnetic resonance imaging (MRI)-based technique that is used to estimate stiffness of tissues within the body
Implementation Method 2
the trained ANN provides one or more output datasets corresponding to an analytical solution to a derivative of a function represented in an unlabeled input dataset thereby transforming the unlabeled input dataset into one or more derivatives of the unlabeled input dataset
Implementation Method 3
estimate stiffness of the tissue based on the derivative of the displacement data
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In accordance with some embodiments, systems, methods, and media for estimating a mechanical property based on a transformation of magnetic resonance elastography (MRE) data using a trained artificial neural network are provided. In some embodiments, a system is provided, the system comprising: a hardware processor programmed to: receive displacement data of tissue in vivo; provide the displacement data to a trained ANN that was trained using noisy input datasets as training data, and derivative datasets corresponding to the noisy input datasets to evaluate performance during training, such that the trained ANN provides an output dataset corresponding to an analytical solution to a derivative of a function represented in an unlabeled input dataset thereby transforming the unlabeled input dataset into its derivative; receive, from the trained ANN, an output dataset indicative of a derivative of the displacement data; and estimate stiffness of the tissue based on the derivative.