Displacement-based Reconstruction of Elasticity Distribution with Deep Neural Network

Reconstructing tissue elasticity distribution is an ill-posed inverse problem in ultrasound elastography. Conventional methods usually require too much iterative computation and cannot meet the real-time requirements in practice. Deep learning (DL) was recently applied to reconstruct the elasticity distribution and achieve promising results. The input of these methods is usually strain images calculated as gradients of displacement images. However, strain images are noisier than displacement images under in-vivo conditions. In this study, a displacement-based DL method is proposed for reconstructing elasticity distribution. The experimental results demonstrate that the model can learn high-dimensional mappings between displacement and elasticity distributions during training by using more readily available and more accurate displacement images as input. The proposed method can not only solve the problem of the high computational cost of traditional methods but also directly predict the elasticity distribution from the displacement image.