Transcranial focused ultrasound (tFUS) is a promising therapeutic modality with the potential to transform treatment of neurological disorders. Intravascular microbubbles are precisely excited with ultrasound to safely and reversibly open the blood-brain barrier, enabling non-invasive drug delivery, targeted in space and time. Yet a major bottleneck is the fidelity of control, as the physics is complex, non-linear, and slow to model. This means treatment often needs to be done in open-loop or based on low-dimensional heuristics.

Our approach is to make the entire tFUS pipeline (acoustic propagation, microbubble dynamics, sensing modalities) fast and differentiable, to unlock the possibility of accurate, real-time, closed-loop feedback. To do so, we employ neural surrogates to speed up physics simulations, apply machine learning to localise and classify cavitation behaviour from measured emissions, and leverage advances in differentiable programming to solve a range of problems across inverse design, state estimation, and treatment optimisation. The research reframes the blood-brain barrier opening as a highly steerable and controllable interface to the brain.

ARIA's Scalable Neural Interfaces space calls for new tools to interface with the brain at scale. Many axes exist to pursue this vision, often focusing on the essential biological or hardware dimensions of the problem. Ours is a complementary bet: using advanced computational science to accelerate experiments and maximise precision, making treatments safe and effective. If successful, the clinical reality of targeted drug delivery comes much closer. Beyond our own results, though, the work is catalytic by design; the differentiable physics and simulation tools form an open-source infrastructure that the wider community can build on, lowering the barrier to a wide range of related technologies.

Callum is a PhD candidate at Imperial College London, where he uses modern computational paradigms to improve the safety and efficacy of focused ultrasound-driven targeted drug delivery in the brain. Previously, he was a reinforcement learning research engineer at Oxa Autonomy and InstaDeep. He holds an MSc in Artificial Intelligence from the University of Edinburgh and a BSc in Electrical & Computer Engineering from the University of Cape Town. He is also the founder of the Centuriae writing retreat project.