Acoustic Simulation with Deep Learning for Low-Intensity Transcranial Focused Ultrasound Digital Twins

Transcranial focused ultrasound (tFUS) is a promising therapeutic modality capable of delivering concentrated acoustic energy to targeted brain regions. A major challenge lies in the significant distortion of the ultrasound beam caused by the skull, leading to an unpredictable shift in location and intensity of the acoustic focus. For the treatment procedure to be both safe and effective, estimating the distorted acoustic focus in real-time is essential. However, existing acoustic simulation methods to predict the acoustic field are computationally too intensive for real-time clinical use. To address this gap, we propose a deep learning-based real-time acoustic simulation method to establish a low-intensity focused ultrasound (LIFU) digital twin. Our approach rapidly estimates intracranial acoustic pressure fields during treatment by taking the acoustic free-field, skull image, and transducer placement as input using multi-modal neural networks. We evaluated model performance on both seen and unseen skull anatomies to verify generalizability. Our models achieved inference times of approximately 23 milliseconds, confirming their suitability for real-time simulation. Our method enables the construction of a digital twin framework that dynamically reflects the ongoing therapeutic state, providing a foundation for data-driven, adaptive LIFU treatment strategies. The code is available at: https://github.com/CMME-Lab/LIFUSimul-DL.git.