Modeling the propagation of (shear) waves in tissues or tissue like media can give further insight and a better understanding of the chances and challenges involved in shear wave elastography.
We are working on tissue models ranging from basic test cases to complex simulations involving nonlinear inhomogeneous materials and reflections while focusing on efficient computations at the same time.
These computational models are compared to experimental data obtained using high-frame-rate ultrasound imaging and optical coherence tomography.
Furthermore the models can be used to train machine learning frameworks, for more efficient and accurate processing of raw input data.
Simulation of wave propagations in soft tissue
Displacement field of a tissue like medium after a periodic excitation in x-direction. The shear waves can be observed in the vicinity of the excited region.
Capturing shear waves using high-frame-rate ultrasound imaging.
Since classical numerical methods depend heavily on assumptions regarding material types and parameters, it is difficult to make accurate predictions for waves propagating in inhomogeneous media when there are uncertainties in the material properties. Therefore, we investigate a generic and purely data-driven approach to predict wave propagation using a fully convolutional neural network, which is based on the well-known U-Net architecture.
Evolutions and predictions of wave field dynamics in complex and heterogeneous domains. The upper left panels show the simulated ground truth, the upper right panels the predictions of our neural network. The bottom row shows the absolute prediction error. On the left is the location-dependent error evolution, while the graph on the right shows the error averaged over the entire domain.