We have reported that perceived body postures could be decoded/classified from cortical EEG signals (Kitazaki et al, 2008, ECVP). The classification of upright and inverse postures was performed at about 80% either with natural postures or unnatural postures that we cannot take. The classification of natural and unnatural postures was more accurate with upright postures (63%) than inverse postures (55%). The purpose of the present study was to compare performance of neural decoding with performance of behavioral/perceptual experiments to see correlation between neural signals and perception. We conducted two behavioral experiments: discrimination of orientations (upright and inverse postures), and discrimination of biomechanical naturalness (natural and unnatural postures). One of human body postures (256 gray-scale computer graphics) was presented on a CRT display and 10 participants were asked to discriminate orientation or naturalness as quickly and accurately as possible. Independent variables were the orientation (upright and inverse postures), the naturalness (natural and unnatural postures), and the viewpoints (0:front, 45, 90:side, 135, 180:back deg). In results, behavioral/perceptual performance was higher both in correct rate and reaction time for the orientation discrimination than the naturalness discrimination. The discrimination of naturalness was easier with upright postures than inverse postures, and deteriorated with accidental views (0 and 180 deg). The EEG data during observation of the identical postures were re-analyzed to decode the orientation and the naturalness of postures for different viewpoints. The accuracy of neural decoding was higher for generic views (45 and 135 deg) than the accidental views. These results indicate good correlations between the performance of neural decoding and behavioral/perceptual experiments for the effects of generic views, inversion, and biomechanical constraint. It is suggested that neural decoding of EEG signal can be a useful tool for quantitatively predicting perceptual processing in brain.
Supported by Nissan Science Foundation to MK, Grant-in-Aid for Scientific research (B) MEXT Japan to SN, and The Global COE program ‘Frontiers of Intelligent Sensing’ to YI