High-Frequency Subband Compressed Sensing with ARC Parallel Imaging

Introduction: Compressed sensing (CS) is an acquisition and reconstruction technique that can reduce the measurement size [1]. In this work, we present a novel way to efficiently combine CS and parallel imaging (PI) by separating the estimation methods in k-space. We apply CS to estimate high-frequency k-space data, and use ARC (Autocalibrated Reconstruction for Cartesian sampling) PI to estimate low-frequency k-space data. This exploits the wavelet characteristics where high-frequency wavelet subbands are typically the most sparse [2], and the ARC convolution kernel nature where regular undersampling provides more stable reconstruction. This method allows easy incorporation of the two (ARC followed by CS) and further removes possible CS failure in low-frequency region. This work has been demonstrated for high-resolution 3D breast imaging.