Model based restoration of the RF data for high resolution vascular ultrasound imaging

High resolution ultrasound imaging of vascular structures such as carotid artery is of major importance in evaluating cardiovascular diseases. However, due to the bandlimited nature of the interrogating pulse the resolution of B-scan images is often insufficient to show fine vascular details. Recently, a new class of deconvolution algorithms from the compressive sensing (CS) field has been used in ultrasound imaging for resolution improvement. Compared to the classical deconvolution methods, these CS methods (CSMs) robustly exploit the sparsity of tissue reflectivity (TR) and provide significant resolution gains. However, CSMs generally assume the ultrasound pulse is invariant throughout the propagation path which is not realistic due to frequency dependent absorption and scattering in biological tissues. As such, pulse variance severely compromises the performance of CSMs in recovering the TR with high fidelity. We propose a new technique that performs sparse deconvolution and accounts for pulse variance that occurs during wave propagation. Our method uses the Model-Based Estimation Pursuit (MBEP) to restore the RF signal in terms of Gaussian Chirplets whose delay and amplitude parameters are related to the location and strength of scatterers (i.e, TR) and its spectral parameters are sensitive to dispersion characteristics of the tissue. MBEP successively extracts GCs from RF data via partitioning, parameter estimation, and removal of the estimated echo. MBEP is highly adaptive to the varying pulses at the tissue-vessel interfaces. MBEP and a CSM (ℓ1-regularized Least Squares Deconvolution, ℓ1-LSD) were tested on ultrasound RF data acquired from a blood vessel tissue-mimicking phantom. Our analysis show that, compared to the ℓ1-LSD, MBEP offers a higher fidelity restoration of the TR as well as marked improvement in resolution gain and processing times.