Guideline on Initial Point Finding and Search Bounds Setting for Biomedical Magnetic Localization

Permanent magnet tracking technology provides a feasible way to localize noninvasive in vivo biomedical devices such as wireless capsule endoscopes. However, current permanent magnet tracking technology that typically relies on an optimization algorithm suffers from a trade-off between the accuracy, consistency, and speed of the results. In particular, in practical applications such as the magnetically actuated capsule, poor localization results are undesirable. In other words, high consistency and accuracy must be achieved concurrently. In this study, we propose an initial point finding method for optimization algorithms based on a semi-soft classifier that adopts a fully connected network and a convolution neural network as base classifiers. Furthermore, we present search bound setting guidelines for the optimization algorithms based on the estimated initial point. By setting tight bounds, the consistency and speed of the process can be improved while maintaining high accuracy. The proposed method is validated via simulations and experiments, and outperforms the previous methods under the same system configuration. Within an effective working space of 42 cm
$\times30$
cm
$\times24$
cm, it achieves an update rate of approximately 9-Hz, higher localization consistency, zero occurrence rate of poor localization results, an average error of 1.3 mm for position estimation, and an average error of 2.7° for orientation estimation.