Identifying sonication regimens for safe and effective microbubble-enhanced focused ultrasound (MB-FUS) interventions is hampered by the challenges associated with predicting microbubble dynamics in tissues. This talk will present data-driven models to extract precursor signals (i.e., patterns) that are within the MB acoustic emissions (AE) collected during MB-FUS interventions in the brain to study and analyze the transition from stable to inertial MB oscillation. Through comprehensive analysis and training of the models, using more than 54 000 AE datasets collected in MB-FUS experiments in mice, we discover multi-dimensional relationships in features derived from the MB AE that can predict the onset of broadband emissions with high sensitivity. By integrating the trained models into a real-time closed-loop MB AE feedback controller it is possible to augment the MB-FUS acoustic treatment window by maximizing the BBB permeability while preventing tissue damage. Finally, we show how this data-driven feedback overcomes barriers to the delivery of nanoparticles and release of soluble biomarkers in brain tumors. Collectively, our findings demonstrate that, by expanding the treatment window, data-driven feedback can augment ultrasound nanotheranostic targeting of brain tumors and support the development of next generation AI-powered ultrasound systems for improved diagnosis, treatment, and monitoring of brain diseases.
Lee et al. (Wed,) studied this question.