Supporting data for “Magnetic Resonance-Based Positional Sensing and Learning-Driven Acoustic Trapping Toward MRI-Guided Targeted Therapy”

<p dir="ltr">Microbubble-mediated ultrasound-triggered targeted therapy presents a promising approach in chemotherapy. Specifically, microbubbles carry the therapeutic drugs, and ultrasound pulses rupture them to release payloads at tumor sites. This method circumvents systemic administration and reduces side effects. Although intermittent ultrasound exposure avoids overheating risks, its off-state allows many intact microbubbles to escape the tumor. Therefore, each trigger only acts on a limited number of microbubbles and cannot achieve a high-dose burst of drug release, resulting in insufficient drug concentration and compromising treatment outcomes.</p><p dir="ltr"><br>In light of these limitations, a proof of concept using magnetic resonance imaging (MRI)-guided acoustic trapping to localize microbubbles around tumors is proposed, preventing their dispersion due to hemodynamics. Once ultrasound trigger is actuated, it can instantaneously rupture the trapped microbubbles, achieving a burst release. This concept leverages acoustic traps, which exert acoustic radiation force to enable contact-free trapping and manipulation of subwavelength objects. Meanwhile, acoustic beams offer excellent biocompatibility, deep tissue penetration, and distinct compatibility with MRI environments. Regarding MRI, it serves as a dual-function tool, not only diagnosing tumors, but also visualizing acoustic field, thereby verifying acoustic trap’s alignment with target site. Therefore, MRI-guided acoustic trapping offers a feasible prospect for targeted therapy. However, this proof-of-concept technique currently faces several key challenges:</p><p dir="ltr"><br>(1) Low computational accuracy hinders the accurate generation of acoustic traps within body. Current digital modeling inadequately reflects real anatomical structure, weakening computation accuracy. Besides, beam<br>phase aberrations caused by tissues are often neglected.<br>(2) Computation inefficiency limits rapid phase updates required to adapt to respiratory-induced tissue motion. This prevents the maintenance of robust trapping at designated targets or may even impede trap formation entirely.<br>(3) MR environment restricts the availability of safe and cost-effective tracking methods, obstructing the sensing of array’s pose and directly challenging the targeting accuracy. When dynamic adjustment is required, the actuation system would introduce errors due to the lack of position feedback.</p><p dir="ltr"><br>To advance the proposed proof of concept, this thesis focuses on tackling these challenges through following approaches:</p><p dir="ltr"><br>(i) A learning-based inverse mapping model, from inputs (i.e., target position, elements’ positions) to outputs (i.e., phase), is designed, successfully delivering accurate generation of different acoustic traps (e.g., focal beam,<br>twin tweezer) in multi-medium, and achieving precise manipulation of a polystyrene ball with small spatial errors below 2.4 mm</p><p dir="ltr">(ii) The trained models achieve excellent computational efficiency (tens of milliseconds), supporting rapid phase updates at 15 Hz. Coupled with computer vision feedback, the designed closed-loop control effectively maintains robust trapping in time-varying multi-medium environments.<br>(iii) A compact and wireless MR-based tracking marker is engineered, producing a high-contrast spot in images with an 18-fold increase in signal intensity. Its real-time tracking (~27 Hz) enables rapid instrument localization in MRI scanner while minimizing electromagnetic interference. Moreover, it realizes accurate omnidirectional tracking, with an error of only 0.56 mm at 56 mm from the isocenter.</p><p dir="ltr"><br>In summary, these achievements, enabled by synergizing MR-based tracking with machine learning-driven acoustic trapping, substantially advance the translational potential of acoustic trapping for targeted therapy.</p>