Reconstruction par acquisition compressée en imagerie ultrasonore médicale 3D et Doppler

摘要

This thesis is dedicated to the application of the novel compressed sensing theory to the acquisition and reconstruction of 3D US images and Doppler signals. In 3D US imaging, one of the major difficulties concerns the number of RF lines that has to be acquired to cover the complete volume. The acquisition of each line takes an incompressible time due to the finite velocity of the ultrasound wave. One possible solution for increasing the frame rate consists in reducing the acquisition time by skipping some RF lines. The reconstruction of the missing information in post processing is then a typical application of compressed sensing. Another excellent candidate for this theory is the Doppler duplex imaging that implies alternating two modes of emission, one for B-mode imaging and the other for flow estimation. Regarding 3D imaging, we propose a compressed sensing framework using learned overcomplete dictionaries. Such dictionaries allow for much sparser representations of the signals since they are optimized for a particular class of images such as US images.We also focus on the measurement sensing setup and propose a line-wise sampling of entire RF lines which allows to decrease the amount of data and is feasible in a relatively simple setting of the 3D US equipment. The algorithm was validated on 3D simulated and experimental data. For the Doppler application, we proposed a CS based framework for randomly interleaving Doppler and US emissions. The proposed method reconstructs the Doppler signal using a block sparse Bayesian learning algorithm that exploits the correlation structure within a signal and has the ability of recovering partially sparse signals as long as they are correlated. This method is validated on simulated and experimental Doppler data.