A Dipole Imaging Method Based on Azimuthal Equidistant Projection

Abstract

The temporal and spatial information of dipoles has been demonstrated to be very important in the decoding of MI-tasks. So, how to make full use of this information is very meaningful. In traditional dipole decoding methods, the spatial filters are applied to extract the spatial feature, which may not utilize the real position of the dipoles, resulting in a low recognition accuracy. In order to solve the problem, a dipole imaging method based on azimuthal equidistant projection (AEP) is proposed, named as AEPDI. Firstly, the dipoles coordinates and amplitudes within region of interest (ROI) and the region of time (TOI) are extracted; Then the 3D dipoles coordinates are projected onto XOY, XOZ and YOZ planes by AEP algorithm respectively. After a series of processing of the dipoles coordinates, the two-dimensional images of the dipoles are constructed. Finally, a multi-branch 3DCNN is designed to recognize the 3D feature data of the three planes. The abundant experiments show that dipole imaging method with the AEP algorithm can effectively improve the classification accuracy, and the multi-directional dipoles feature fusion achieves the highest classification accuracy.