Convolutional Neural Network for non-invasive magnetic foreign body localization in the human body

Abstract

Abstract Ferromagnetic foreign bodies accidentally inserted in patients usually need to be surgically removed. Their location can be estimated by measuring the magnetic field generated by the object and solving the magnetic inverse problem to locate the source based on the magnetic field maps configuration. Considering ferromagnetic straight needles (hypodermic or sewing, for example) and a computer simulation of the magnetic flux density based on Biot-Savart’s Law, this paper presents an algorithm based on Convolutional Neural Networks (CNN) to find the depth, the angles of inclination and rotation and the center of the needle inside the human body. The proposed model presents low RMSE values and type A measurement uncertainty for the depth, the inclination angle, the rotation angle, and for the displacement Δ between the midpoint of the distance connecting the magnetic field extreme values and the foreign body center, all of which adequate for use in therapeutic procedures.