Artificial intelligence to unlock the potential of acoustic holography: Case study on transcranial focused ultrasound

Optimal beam shaping in holographic Focused Ultrasound (FUS) is essential for maximizing therapeutic outcomes, particularly through heterogeneous structures such as bone and soft tissue. Traditional methods often lack precision, while advanced techniques rely on case-specific constraints, making treatment planning time-consuming and less accessible. Although Artificial Intelligence (AI) is gaining attention in FUS, it is frequently viewed as a novel concept rather than a fully integrated tool for precise clinical applications. This study aims to shift that perspective by demonstrating how AI-powered methodologies can streamline transcranial holographic FUS treatment planning, offering improved efficiency and adaptability. Our AI-driven framework uses skull and target geometries extracted from large, age-diverse patient datasets, segmented using FreeSurfer from CT and MRI scans. Physics-informed datasets are generated using full-wave time-domain and mixed-domain simulations. A conditional generative adversarial network with a U-Net-based generator is trained using a composite loss function. Performance is validated using hydrophone pressure measurements with an ex vivo skull. Comparative analysis shows that the framework accelerates computation, improves robustness, and simplifies planning where conventional theoretical models often face limitations. [Work supported by the U.S. National Science Foundation (NSF) under Award No. CAREER CMMI 2143788, and CMMI 2016474, which are gratefully acknowledged.]