Electric Field Models of Transcranial Magnetic Stimulation Coils with Arbitrary Geometries: Reconstruction from Incomplete Magnetic Field Measurements

TL;DR

This paper presents a method to reconstruct accurate coil models for transcranial magnetic stimulation from incomplete magnetic field measurements, enabling flexible and efficient modeling of arbitrary coil geometries.

Contribution

It introduces a dipole approximation technique with minimum norm estimation and orthogonal matching pursuit for sparse models, applicable to arbitrary coil geometries from limited data.

Findings

Accurate magnetic field representation from sparse data.

Sparsification of coil models without losing accuracy.

Enhanced flexibility in coil model construction.

Abstract

Background: Calculation of the electric field induced by transcranial magnetic stimulation (TMS) in the brain requires accurate models of the stimulation coils. Reconstructing models from the measured magnetic fields of coils so far worked only for flat coil geometries and required data of their full magnetic field distribution. Objective: To reconstruct models of coils with arbitrary winding geometries from spatially incomplete magnetic field measurements. Methods: Dipole approximation via minimum norm estimation with a cross validation procedure simultaneously assessing predictability and reproducibility of the field approximation. The methods were validated on both simulated and acquired magnetic flux density data. Furthermore, reconstruction of coil models from sparsely sampled data is investigated and a procedure for obtaining sparse dipole coil models based on orthogonal matching pursuit is suggested. Results: Given that measured data from regions around the coil are available, the magnetic vector potential can accurately be represented by fitted dipole models even from sparsely sampled data. Considerable sparsification of dipole models is possible while retaining accurate representation of the field also close to the coil. Conclusion: Our versatile approach removes an important hurdle for constructing coil models from measurement data and enables a flexible trade-off between the accuracy and computational efficiency of the reconstructed dipole models.