Appropriate Null Models for Testing the Effect of the Head Model on MEG Functional Connectivity Fingerprinting

Functional connectivity (FC)-based neural fingerprinting is an approach that promises to identify and/or differentiate subjects within a cohort on the basis of the patterns of statistical dependencies between time series recorded mostly if not always noninvasively, with electroencephalography (EEG), magnetoencephalography (MEG), or functional magnetic resonance imaging (fMRI). The message is that brain activity is what differentiates subjects, or what makes a neural fingerprint "unique". In EEG- and MEG-derived FC fingerprinting, the activity recorded at the sensors is usually projected back into cortical sources by means of an inverse model depending on head and brain shapes, sensor locations, and tissue conductivity, and further reduced in dimension to obtain time series of regional activity, used to compute FC. We argue that the role of the head model in fingerprinting has been so far tested by means of suboptimal or incomplete null models. Here we employed a set of experiments aimed to decouple recorded sensor data and head model for each subject, constituting what we consider more appropriate tests of the null hypothesis of no head model effect on fingerprinting. When the same source-level data are used for all subjects, so that only the head model varies across subjects, identification is perfect (EER = 0) across all frequency bands. When the same head model is applied to all subjects, so that only the neural data vary, identifiability and differentiability are substantially degraded compared to the standard matched condition. These results jointly show that the null hypothesis of no head model effect on fingerprinting cannot be sustained. We make no claim about the relative size of neural versus anatomical contributions, nor about the specific mechanism through which the head model affects fingerprinting metrics.