Investigating the properties of fMRI-based signature of recognizing one's own face.

Multivariate pattern analysis has revolutionized the field of neuroimaging. Many hope it will help elucidate how mental states are encoded in brain activity, though others caution that such optimism may be premature. In this study, we sought to identify an fMRI-based signature of a relatively simple but basic feeling of recognizing one's own face (SFRS), and to examine its properties. The fMRI data were acquired while participants attempted to recognize themselves in images of morphed faces. A series of binary classifications ('self' vs. 'not self') showed that the localization of most prognostic areas is consistent with published results based on univariate analysis. SFRS response classified between 'self' and 'not self' with 100 % accuracy and could accurately predict the morphing stages of presented face images. Mediation analyses showed that SFRS response acted as a mediator between the proportions of self in images and the decision to accept a given image as self. The relative insensitivity of SFRS to spatial smoothing and comparable predictive performance of a small subset of randomly selected voxels allow us to conclude that the information necessary to distinguish between the two mental states must be derived from the whole brain, and that this information is spatially smooth.