Lens-Aware Differentiable Beamforming for In Vivo Distributed Aberration Correction with Curvilinear Transducers
We previously introduced ultrasound autofocusing, an iterative model-based aberration correction technique that estimates local sound speed and incorporates it into beamforming to correct image distortion from heterogeneous media. In this work, we extend ultrasound autofocusing to curvilinear arrays and introduce advancements to the underlying model. A differentiable bent-ray tracing approach accounts for refraction through the transducer lens, while a new adaptive-grid initialization accounts for changes in speckle position with sound speed. The method is validated in silico and in calibrated sound speed phantoms. Our distributed aberration-correction method is then applied to a first large-scale in vivo evaluation comprising 313 liver acquisitions from 76 high-BMI human subjects. In images containing anechoic regions, contrast and CNR improved by $1.38 \pm 1.60$ dB (+18.0%) and $0.09 \pm 0.14$ (+10.2%), respectively. Improvements were also observed in speckle brightness (+20.3%), coherence factor (+13.1%), lag-one coherence (+2.7%), common-midpoint correlation coefficient (+0.7%), and common-midpoint phase error (-9.1%; lower is better), with all metric improvements statistically significant. Target structure and visibility also improved significantly. These results demonstrate the potential of ultrasound autofocusing for clinically applicable distributed aberration correction.