Deep Neural Network Elasticity Imaging via Residual Force Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current elastography methods, such as strain-based elastography, assume a uniform stress distribution, leading to inaccurate elasticity predictions, especially in inhomogeneous bodies, and existing model-based methods are computationally expensive and prone to errors from noise and missing data.
Innovation Solution
A deep neural network-based method that learns the hidden elasticity distribution from strain data without labeled data, using constitutive elasticity equations and equilibrium equations to encode prior knowledge, allowing for robust predictions even with noisy and missing measurements, and generating super-resolution images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If strain-based elastography with uniform stress assumption is used, then implementation is simple and easy, but elasticity distribution prediction is inaccurate for inhomogeneous bodies
Solution Approach 1:
The patent introduces a deep learning model as an intermediary between strain measurements and elasticity distribution prediction. The model learns the complex mapping relationship from training data, serving as a mediator that translates simple strain-based inputs into accurate elasticity distributions without requiring complex physical modeling or uniform stress assumptions.
Solution Approach 2:
The patent transforms the problem from directly solving the inverse elasticity problem with physical constraints to a parameter estimation problem where the deep learning model learns optimal parameters (elasticity distribution) from training data. This parameter transformation allows the system to achieve accuracy comparable to complex model-based methods while maintaining the simplicity of strain-based approaches.
2Measurement precision
If model-based elastography is used to solve inverse problem, then elasticity distribution can be recovered in principle, but computational cost is high and prone to errors from noise and missing data
Solution Approach 1:
The patent performs preliminary action by training the deep learning model offline on a large dataset of forward elasticity problems. This pre-computed knowledge is stored in the model's weights and biases, allowing rapid inference during actual elastography applications without requiring complex real-time computations or iterative solving of inverse problems.
Solution Approach 2:
The patent creates a computational copy of the forward elasticity problem solutions through the deep learning model. Instead of solving the complex inverse problem directly, the model learns to copy the input-output relationships from training data, providing accurate elasticity predictions without requiring the computational resources needed for traditional model-based inverse solving.
3Productivity
If artificial constraints are added to reduce number of possible elasticity distributions for supervised learning, then inverse problem becomes solvable, but application scope is limited in practice
Solution Approach 1:
The patent creates a universal deep learning model that can handle diverse elasticity distribution patterns without requiring problem-specific constraints. The model is trained on a comprehensive dataset covering various tissue types and elasticity patterns, enabling it to generalize to new applications and maintain versatility across different elastography scenarios while remaining computationally solvable.
Data Source
AI summary
A method of predicting elasticity of a solid includes receiving a data set comprised of position data and corresponding strain data for points on a solid at a deep neural network (DNN), producing a predicted stress distribution, applying convolutional filters to the predicted stress distribution to produce residual force maps, and predicting an elasticity distribution of the solid by iteratively using the residual force maps and an equilibrium condition until the predicted elasticity distribution satisfies the equilibrium condition, producing the final elasticity distribution.