Full Wave Inversion for Ultrasound tomography Using Physics Based Deep Neural Network

Ultrasound Computed Tomography (USCT) is a promising imaging modality for reconstructing tissue properties. Full-wave inversion (FWI) provides an enhanced contrast image. In this paper, we introduce an optimized Physics-Informed Neural Network (PINN) architecture to solve the inverse problems of FWI. PINNs are neural networks that can approximate universal functions using partial differential equations that govern the physics of the system. When training the network, the deep neural network can converge to learn the solutions of the wave equation by minimizing the loss function. The results showed that our proposed networks can retrieve the Speed of Sound (SoS) information and significantly decrease the computational cost via the transfer learning technique.