Few-Shot Magnet Localization Using Sim-to-Real Transfer Learning.

Area of Science:

• Medical instrumentation
• Machine learning in healthcare
• Biomedical engineering

Background:

• Magnetic localization is a key radiation-free technology for real-time medical instrument tracking.
• Current physics-model-based methods face limitations in accuracy due to imperfect assumptions and computational demands.
• Need for more efficient and accurate localization techniques in medical procedures.

Purpose of the Study:

• To develop an efficient and accurate learning-based magnetic localization method.
• To achieve low computation cost and few-shot generalization for magnetic tracking.
• To demonstrate the practical efficacy of the proposed method in medical applications.

Main Methods:

• A single, compact neural network was designed for magnetic localization.
• Transfer learning was employed using a simulation dataset for few-shot generalization.
• The method was evaluated using four 3-channel sensors within a 100×100×100mm workspace.

Main Results:

• Achieved mean position error of 1.66 mm and orientation error of 2.15°.
• Demonstrated a 2x improvement in accuracy compared to traditional magnetic model-based methods.
• Successfully validated in a real-time catheter tracking experiment.

Conclusions:

• The proposed learning-based method offers efficient and accurate magnetic localization.
• This technique provides a significant advancement over existing physics-model-based approaches.
• The method shows strong potential for practical applications in real-time medical instrument tracking.