Localization of magnetic foreign bodies using CNN and GMI magnetometer

Abstract This paper presents an algorithm based on Convolutional Neural Networks (CNN) to find the depth and angles of inclination and rotation of a foreign object inside the human body based on images of the magnetic field generated by it. The key challenge is to provide information with enough accuracy to be used in surgical procedures. We tested three distinct CNN architectures for values prediction, and our best model achieved a mean - average F1-score of 66 %, 100 %, and 98 % in the test dataset for depth and angles of inclination and rotation, respectively. We also propose an approach for converting classification values to real values and we calculate the type A uncertainty for all models, with our best model showing an uncertainty of 10.8 mm for depth, 2.0° for inclination and 7.5° for rotation values.