Learning hidden elasticity with deep neural networks

TL;DR

This paper presents a novel deep learning-based elastography method that learns the hidden elasticity of solids from strain measurements without labeled data, enabling high-resolution and robust elasticity imaging.

Contribution

It introduces a physics-supervised neural network approach that accurately estimates elasticity, reconstructs distributions in unmeasured areas, and achieves super-resolution imaging.

Findings

Accurately learns hidden elasticity from measured strains.

Robust to noisy and missing data.

Enables super-resolution elasticity imaging.

Abstract

We introduce a de novo elastography method to learn the elasticity of solids from measured strains. The deep neural network in our new method is supervised by the theory of elasticity and does not require labeled data for training. Results show that the proposed method can learn the hidden elasticity of solids accurately and is robust when it comes to noisy and missing measurements. A probable elasticity distribution for areas without measurements may also be reconstructed by the neural network based on the elasticity distribution in nearby regions. The neural network learns the hidden elasticity of solids as a function of positions and thus it can generate elasticity images with an arbitrary resolution. This feature is applied to create super-resolution elasticity images in this study. We demonstrate that the neural network can also learn the hidden physics when strain and elasticity distributions are both given. The proposed method has various unique features and can be applied to a broad range of elastography applications.