Enhanced photoacoustic microscopy with physics-embedded degeneration learning

Deep learning (DL) is making significant inroads into biomedical imaging as it provides novel and powerful ways of accurately and efficiently improving the image quality of photoacoustic microscopy (PAM). Off-the-shelf DL models, however, do not necessarily obey the fundamental governing laws of PAM physical systems, nor do they generalize well to scenarios on which they have not been trained. In this work, a physics-embedded degeneration learning (PEDL) approach is proposed to enhance the image quality of PAM with a self-attention enhanced U-Net network, which obtains greater physical consistency, improves data efficiency, and higher adaptability. The proposed method is demonstrated on both synthetic and real datasets, including animal experiments in vivo (blood vessels of mouse's ear and brain). And the results show that compared with previous DL methods, the PEDL algorithm exhibits good performance in recovering PAM images qualitatively and quantitatively. It overcomes the challenges related to training data, accuracy, and robustness which a typical data-driven approach encounters, whose exemplary application envisions to provide a new perspective for existing DL tools of enhanced PAM.