MSAF: A Multi-level Sensing Adversarial Framework for Signal Recovery in Synthetic Aperture Ultrasound

Synthetic aperture ultrasound imaging achieves two-way focusing and produces high-quality image formation by coherently combining returning echoes from successive single-element or sub-aperture transmissions across the entire transducer array. To depict transient or fast tissue dynamics, a prominent line of research efforts involves employing a limited number of sub-apertures to approximate full-aperture scans, enabling full-view imaging at high frame rates. However, existing end-to-end approaches predominantly operate in the image domain, which inherently discards phase information. Some studies use radio-frequency or in-phase and quadrature (IQ) domains, but they analyze spatial features frame by frame, thus not fully leveraging the temporal information. Motivated by high-dimensional data inherent to synthetic apertures, this study proposes a multi-level sensing adversarial framework (MSAF) to recover full IQ signals from sparse sub-aperture transmissions. MSAF utilizes sparse apertures by spatiotemporal sensing and constrains the recovered signals by global-attribute sensing. MASF was evaluated on large in vivo human echocardiography datasets and on in vivo unseen echocardiographic views and abdominal images as zero-shot testing sets. Experimental results demonstrated high-fidelity IQ signal recovery, achieving a peak signal-to-noise ratio of 32.80 dB, a structural similarity index of 0.88, a contrast-to-noise ratio of 2.92 dB, and a contrast ratio of 15.46 dB, outperforming all tailored benchmark models. Therefore, MSAF enables sparse-aperture transmissions with high signal quality and data efficiency, making it a potential approach for functional ultrasound imaging.