85
Discussion
For single, two-second trials of somatosensory stimulation, MVPA was able to
decode the body site of somatosensory stimulation at levels far above chance
performance. Two-way decoding performance of hand stimulation was particularly
accurate, with a mean of 92% performance, similar to performance levels reported in
two-way decoding tasks with visual stimulation (e.g. 80% accuracy for an orthogonal
orientation decoding task in (Haynes and Rees 2005a)). As in visual studies, increasing
the difficulty of the classification task decreased classifier performance (Kay, et al.
2008). However, even for closely spaced touches on the same hand, performance was
still well above chance (68% compared with 33% chance performance). Also similar to
MVPA studies of visual decoding, accuracy increased sharply as more voxels were
■ ■■ / ' . '■
included in the analysis, with the increase slowing as the number of voxels increased
beyond ten to twenty (Haynes and Rees 2005a).
Ina visual MVPA study, Vl and V2 were both able to accurately decode stimulus
orientation (Kamitani and Tong 2∞5). In the present study, Sl and S2 were able to
decode the body site Ofstimulation with high levels Ofaccuracy. Receptive fields in Sl
are small and highly somatotopic, while receptive fields in S2 are much larger and less
well-organized (Nelson, et al. 1980). This may correspond to the dissociation in
classification accuracy observed between the experiments. In experiment 1, in which
the stimuli were widely separated on the body surface, S2 and Sl classified the stimuli
with similar accuracy (both 60%). In contrast, in the 3-way finger decoding task of
experiment 2, the stimuli were closely spaced on the body surface and Sl was more