AI Driven Functional Super-Resolution Ultrasound Imaging Guiding Focused Ultrasound Treatment

PI-Technion, Assist. Prof. Avinoam Bar-Zion, Rambam co-PI, Dr. Lior Lev Tov, MD

Clinical Problem / Unmet Need

Image-guided focused ultrasound (FUS) is an emerging therapeutic technology that enables noninvasive, targeted surgery and novel treatments, such as selective blood-brain barrier (BBB) opening and neuromodulation. Currently, most clinical FUS neurosurgeries and treatments are performed under MRI guidance. While MRI thermometry is essential for certain applications, MRI-guided procedures are costly and complex, limiting widespread accessibility. For mechanical FUS treatments in particular, MRI does not provide optimal real-time vascular feedback. There is, therefore, a critical unmet need for a more accessible, cost-effective, and functionally appropriate imaging modality to guide and monitor non-invasive focused ultrasound therapies, especially for dynamic vascular changes occurring during treatment.

TERA-sign

Perspective on Addressing the Challenge Using Artificial Intelligence

The proposed solution is to leverage Artificial Intelligence to enhance ultrasound localization microscopy (ULM), a super-resolution imaging modality that can visualize microvascular structures beyond the limits of conventional ultrasound. These methods can image blood flow down to the capillary level, achieving a functional resolution of 5µm. While early ULM methods relied on model-based localization, recent machine learning approaches have improved performance but remain limited in architectural sophistication and often fail to exploit the unique spatiotemporal and vascular properties of contrast-enhanced ultrasound data.

This research proposes advanced AI frameworks, including transformers, attention mechanisms, and graph neural networks, to better model temporal flow dynamics and vascular network structures. Additional strategies, such as transfer learning and cross-domain learning, will address the scarcity of high-quality clinical datasets. By improving ULM accuracy and adapting it to therapeutic ultrasound transducers, AI can enable real-time feedback and intelligent scanning policies, ultimately replacing or complementing MRI guidance with a more efficient and scalable solution.

Research Goals

The research is structured around three primary goals:

  1. Develop advanced AI architectures for ULM that incorporate vascular priors, temporal dynamics, and graph-based representations to significantly improve super-resolution imaging performance.
  2. Adapt ULM from imaging to therapeutic ultrasound systems, enabling deep-tissue and transcranial imaging compatible with focused ultrasound treatments.
  3. Create AI-driven guidance and feedback mechanisms for focused ultrasound therapy, improving both efficacy and safety.

Together, these goals aim to establish a clinically translatable framework for ultrasound-guided focused ultrasound treatments, improving accessibility, reducing costs, and advancing non-invasive therapeutic options worldwide.