Differentiable Co-Design of Ultrasound Probe Geometry and Acquisition for Plane-Wave Imaging

Abstract

Ultrasound image quality depends on both the probe geometry and the transmit and receive processing, but engineers usually design the two separately. Physics-based simulators (Field II, k-Wave, MUST/SIMUS) compute the radio-frequency channel data but give no gradients, so design studies must use grid or black-box searches that become slow as the number of parameters increases. We present a fully differentiable end-to-end pipeline for plane-wave ultrasound imaging, with synthetic speckle phantoms, an analytical frequency-domain RF simulator, delay-and-sum beamforming and an extensible image-quality objective. The element pitch, the transmit and receive apodization and the global speed of sound are differentiable parameters of one computation graph, so one reverse-mode pass gives their exact joint gradients. Studies of each parameter rank the design parameters: the optimized pitch decreases the grating lobes, and the speed-of-sound autofocus gives clear contrast and resolution gains. Over 15 test phantoms, the optimized pitch increases the cyst gCNR from 0.55 to 0.66 (about 20%), and the speed-of-sound correction adds a further 0.05. The lateral point-spread function agrees with Field II within about 10%, and on 49 real CUBDL plane-wave acquisitions from five institutions, the framework finds the global speed of sound with a mean absolute difference of 16.6 m/s from the published corrections and phantom speed of sound.