Optimal blind-source-separation filtering for ultrasound clutter suppression: application to ultrasound localization microscopy and speckle tracking

Ultrasound localization microscopy (ULM) exploits microbubbles to generate super-resolution images beyond the diffraction limit, and ultrasound speckle tracking (UST) allows for the estimation of tissue motion and strain. For both applications, suppression of noise and clutter is essential. This is effectively achieved using blind source separation techniques such as singular value decomposition, but given the limitations of heuristic subspace selection, useful criteria that enable automatic and adaptive selection of the desired signal components should be established. In this work, synthetic ultrasound data was used to test a comprehensive range of (proposed and novel) effective criteria based on domain knowledge for adaptive signal subspace selection for ULM and UST. For ULM, tissue clutter is most effectively suppressed by removing singular components with a mean spectral density above a frequency threshold. Also for UST, identification of signal singular components by spectral thresholding proved to be the most effective. Even though its performance for in-vivo acquisitions remains to be investigated, the proposed method shows promise for adaptive clutter suppression.