Transcranial focused ultrasound (tFUS) enables the transmission of acoustic energy through the skull to converge onto minute brain regions. Numerical simulations using the K-Wave toolbox offer valuable insights into ultrasound wave propagation and focusing. However, prohibitively slow simulation speeds impede real-time visualization and analysis of tFUS. Here, in this paper, a conditional generative adversarial network (cGAN) framework has been applied to predict the acoustic field generated by tFUS phased arrays. A transcranial acoustic field dataset was generated via the K-Wave toolbox, simulating the transmission of a 128-element spherical phased array through a human skull model reconstructed from CT scans. Key parameters, including pressure distribution, focus size, and phase delay, were computed. A cGAN architecture was trained to directly map skull geometry and transducer parameters to the intracranial sound field, thereby avoiding time-consuming numerical predictions. The calculation results demonstrate that cGAN-driven simulations can predict acoustic field characteristics effectively, including focal shift-related errors stemming from skull-induced distortions.
Cheng et al. (Thu,) studied this question.