Computational Reconstruction of 3D Stomach Geometry using Magnetic Field Source Localization

In this study, we investigated the feasibility of computationally reconstructing the 3D geometry of the stomach by performing source localization of the magnetic field (MF) induced from the stomach surface. Anatomically realistic stomach and torso models of a human participant, reconstructed from the CT images, were used in the computations. First, 128 coils with a radius of 5 mm were positioned on different locations on the stomach model. Next, MF at the sensor positions were computed using Bio-Savart law for the currents of 10 and 100 mA. Then, three noise levels were defined using the biomagnetic data recorded from the same participant and two additional sets of generated white-noise resulting in mean signal to noise ratios (SNR) of 20 and 10 dB. Finally, for each combination of the current and noise level, the magnetic dipole (MDP) approximation was performed to estimate coil positions. The performance of the source localization was assessed by computing the goodness of fit (GOF) values and the distance between the coil and the estimated MDP positions. We obtained GOF values over 98% for all coils and a mean localization error of 0.69 +/- 0.08 mm was achieved when 100 mA current was used to induce MF and only biomagnetic data was added. When additional white-noise was added, the GOF values decreased to 95% and the mean localization error increased to around 4 mm. A current of 10 mA was enough to localize the coil positions with a mean error around 8 mm even for the highest noise level we tested but for the few coils furthest from the body surface, the error was around 10 cm. The results indicate that source localization using the MDP approximation can successfully extract spatial information of the stomach.