Oleg Michailovich (University of Waterloo)
The coherent nature of ultrasound imaging along with the bandlimitedness of its beamforming mechanism impose limitations on the resolution and contrast of ultrasound scans. While rarely corrigible via refinement of hardware design, the above limitations can be effectively mitigated by means of post-processing. In this case, the problem of image restoration is usually cast in the form of an inverse problem, which is subsequently solved using numerical optimization. One of the most well-known examples of such problems aims at recovering the reflectivity function of insonified tissues from raw ultrasound data which is considered to be a blurred and noisy version of the former. While common to imaging science in general, the above problem is known to be particularly challenging in the case of ultrasound imaging for a number of reasons. Chief among those are both inter-subject and intra-scan variability of the blur operator as well as the spatial inhomogeneity of the statistical properties of tissue reflectivity functions. Under the above conditions, the inverse problem at hand acquires the form of a non-stationary blind deconvolution (NSBD) problem, a solution to which is generally hard to find. Accordingly, the first part of the talk will cover some recent approaches to solving the above problem, followed by demonstration of its practical significance. Thereafter, we will take a brief tour to the realm of diffusion Magnetic Resonance Imaging (dMRI), overviewing its principal challenges as well as some ways of overcoming thereof. We will also discuss the possibility of processing dMRI data by means of deep neural networks (DNN), with a particular emphasis on the importance of adaptation of network architecture to the exclusive properties of dMRI signals. Finally, we will attempt to make a connection between the two imaging modalities, posing several intriguing questions to be addressed in future research.