Ultrasound Computed Tomography using physical-informed Neural Network

Ultrasound computed tomography (USCT) is an important diagnostic technique for improved imaging in medical area. The reconstruction of tissue properties, such as attenuation, mass density, and speed of sound (SoS), can be used to monitor and analyze the tissue condition. In this paper, an optimized physical-based architecture called physics-informed neural network (PINN) is used to solve the inverse problems of USCT. This neural network leverages the physical information by adding the residuals of a system of Partial Differential Equations (PDE) to the loss function in the learning process. By incorporating all the information including the PDE, initial and boundary conditions, proposed nerual network can learn solutions of the wave equation. The results showed that PINN can extract the SoS value within an acceptable error.