Application of artificial neural networks to inverse problems in biomagnetism
Kevin Kit Parker, John P. Wikswo · 2002
We have applied an artificial neural network (ANN) using the backpropagation learning algorithm to the biomagnetic inverse problem. A forward model was used to calculate the magnetic fields from propagating action potentials (APs) as would be seen in nerve or muscle bundles. This forward model depicted two design schemes of a high-resolution SQUID magnetometer, whose three pickup coils were in two configurations, a stacked array and a planar array with respect to the direction of propagation in the idealized bundle. The ANN was tasked with determining the location of the source in two dimensions, with the third being the direction of propagation which was assumed to be known. The ability of the network to determine the source location was dependent on the arrangement of the magnetometer pickup coils.