Automated Detection of Fetal Brain Signals with Principal Component Analysis

Detection of fetal brain signals in fetal magnetoencephalographic recordings is -- due to the low signal to noise ratio -- challenging for researchers in this field. Up to now, state of the art is a manual evaluation of the signal. To make the evaluation more reproducible and less time consuming, an approach using Principal Component Analysis is introduced. Locations of the channels of most importance for the first three principal components are taken into account and their possibility of resembling brain activity evaluated. Data with auditory stimulation are taken for this analysis and trigger averaged signals from the channels selected as brain activity (manually \& automatically) compared. Comparisons are done with regard to their average baseline activity, activity during a window of interest and timing and amplitude of their highest auditory event-related peak. The number of evaluable data sets showed to be lower for the automated compared to manual approach but auditory event-related peaks did not differ significantly in amplitude or timing and in both cases there was a significant activity change following the tone event. The given results and the advantage of reproducibility make this method a valid alternative.