Deep Unfolding RPCA for High Resolution Flow Estimation

Numerous techniques have been proposed to produce high precision blood flow or contrast-enhanced ultrasound estimates from fast ultrasound sequences. Among them, robust principal component analysis (RPCA)-based methods are known as superior to most state-of-the-art techniques. In particular, these techniques may include a deconvolution step which allows for further improvement of the resolution of the estimated blood flow images. However, they rely on many hyperparameters that have to be manually adjusted to obtain the optimal solution. To overcome this limitation, we propose a new deep unfolding neural network based on the DRPCA iterative algorithm, which enables the reconstruction of high-resolution and high-sensitivity blood flow components. Compared to other state-of-the-art methods, the proposed algorithm showed interesting performances, in terms of PSNR and SSIM on simulation data.