Evaluating Performance of Artifact Removal by Fully Dense U-Net for Microwave Induced Thermoacoustic Tomography

Abstract

Microwave induced thermoacoustic tomography (TAT) is an imaging modality based on the thermoacoustic effect. For considering the safety and comfort, the microwave radiation power and imaging time are both preferred to be less. However, it results in an issue of artifacts and noise due to sparsely spatial sampling and relative lower SNR of TA signals. Aiming to overcome this problem, deep learning-based method is an emerging technique. In this work, we evaluate the artifact removing performance of a network based on fully dense U-Net architecture. The results show that FD U-Net network could effectively remove the artifacts from the TA images and improve the image quality.