Optimized Magnetic Particle Navigation Based on Data-Driven Model for Enhanced Drug Targeting

Targeted drug delivery by magnetically steering micro- and nanoparticles for specified therapy is gaining ground in the field of medical treatments. Yet fundamental challenges with regards to modeling particle movement and reaching desired regions exist. In this work, we use data-driven modeling to predict the velocities of a particle cluster from its positions and electromagnet currents in an in-vitro targeting setup. This explicit velocity prediction using a neural network, not found in previous works on this topic, is compared to a state-of-the-art physics-based model, and produced more accurate estimated particle cluster trajectories. Furthermore, for the first time, the data-driven model is integrated in model-based optimization methods designed to maximize the particle velocity in a predefined direction, or guide particles from an initial to a final position under minimized energy dissipation taking into account future time steps without requiring real-time feedback. Simulated and measured optimized particle trajectories strongly overlap in the experimental setup. With these findings, magnetic drug targeting can be made more accurate and brought closer to its clinical implementation.