Solving an elastic inverse problem using Convolutional Neural Networks

TL;DR

This paper demonstrates the potential of Convolutional Neural Networks to accurately reconstruct shear modulus fields in elastic media from displacement or strain data, with promising initial results for medical imaging applications.

Contribution

The study introduces a CNN-based approach for elastic inverse problems, specifically imaging shear modulus fields from displacement or strain data, which is novel in this context.

Findings

Encouraging reconstruction accuracy observed.

6000 training examples used for CNN training.

Potential for high-contrast medical imaging applications.

Abstract

We explore the application of a Convolutional Neural Network (CNN) to image the shear modulus field of an almost incompressible, isotropic, linear elastic medium in plane strain using displacement or strain field data. This problem is important in medicine because the shear modulus of suspicious and potentially cancerous growths in soft tissue is elevated by about an order of magnitude as compared to the background of normal tissue. Imaging the shear modulus field therefore can lead to high-contrast medical images. Our imaging problem is: Given a displacement or strain field (or its components), predict the corresponding shear modulus field. Our CNN is trained using 6000 training examples consisting of a displacement or strain field and a corresponding shear modulus field. We observe encouraging results which warrant further research and show the promise of this methodology.