Combining band-frequency separation and deep neural networks for optoacoustic imaging

In this paper we consider the problem of image reconstruction in optoacoustic tomography. particular, devise a deep neural architecture that can explicitly take into account band-frequency information contained sinogram. This is accomplished by two means. First, jointly use linear filtered back-projection method and fully dense UNet for generation images corresponding to each one frequency bands considered separation. Secondly, order train model, introduce special loss function consisting three terms: (i) separating term; (ii) sinogram-based consistency term (iii) directly measures quality which takes advantage presence ground-truth present training dataset. Numerical experiments show proposed be easily trainable standard optimization methods, presents an excellent generalization performance quantified number metrics commonly used practice. Also, testing phase, our solution has comparable (in some cases lower) computational complexity, desirable feature real-time implementation imaging.