A semi-analytical solution and AI-based reconstruction algorithms for magnetic particle tracking.

Area of Science:

• Physics
• Engineering
• Data Science

Background:

• Magnetic particle tracking (MPT) measures particle translation and rotation in opaque flows.
• Current trajectory reconstruction methods often rely on numerical optimization or filtering with artificial parameters.
• Existing analytical algorithms have limitations, particularly regarding magnetic field gradient measurement.

Purpose of the Study:

• To present a novel semi-analytical solution and reconstruction algorithm for MPT.
• To develop deep neural network (DNN)-based models for denoising MPT trajectories.
• To compare the performance of DNNs, wavelet filtering, and hybrid methods for trajectory reconstruction.

Main Methods:

• A new semi-analytical solution for MPT trajectory reconstruction is proposed.
• Deep neural network (DNN) models are developed for trajectory denoising.
• Hybrid methods combining wavelet filtering and DNNs are investigated for velocity reconstruction.

Main Results:

• The new semi-analytical method supports arbitrary sensor arrangements.
• DNN-based denoisers show superior accuracy in position reconstruction compared to traditional methods.
• Hybrid wavelet-DNN methods provide more accurate velocity reconstruction, while all methods perform well in orientation reconstruction.

Conclusions:

• The developed semi-analytical solution offers a versatile approach to MPT trajectory reconstruction.
• DNN-based denoising significantly improves position accuracy in MPT.
• Hybrid methods effectively address velocity signal over-smoothing issues, enhancing overall MPT data quality.