The name is absent

78

trial of 25%). For the same subject, five additional classifiers were trained and tested,
producing a total classification across six scan series of 462 correct and 138 incorrect
(77%); the chance likelihood of this performance was vanishingly small {p < lθ"). In
every one of 7 subjects, decoding performance was much greater than chance, with a
mean of 68% ± 3% SEM across subjects. Changing the ratio of training to testing data
(from five scan series for training and one scan series for testing to three scan series for
training and three scan series for testing) did not change classification accuracy. '
We also examined the ability of separate sets of classifiers to perform two-way
discriminations between the left and right hand of stimulation, and the left and right
foot of stimulation. Across subjects, the mean classification performance was 91% ± 2%
SEM for two-way hand decoding and 85% ± 1% for two-way foot decoding (both p < 10’
⁹⁹ under the binomial distribution with success probability per trial of 50%).

Having shown that MVPA across all active areas could successfully decode the
body site of stimulation, we wished to determine Ifdifferent brain areas differed in their
decoding ability. Classifiers were separately trained and tested using only the voxels in
each of three ROIs: SI, S2 and MST∕STP. The four-way decoding performance across
subjects was 60% ± 1% for the Sl ROI; 60% ± 1% for the S2 ROI and 30 ± 0.4% for the
MST∕STP ROI (Fig. ЗА). The scores of the SI, S2 and MST∕STP ROIs were entered into a
one-factor ANOVA, which revealed a significant effect of area (F(2,18)=51, p<10^⁷).

To study distributed representations in somatosensory cortex, we created two
additional ROIs consisting of voxels in the foot region of Sl (Slfoot) and voxels in the
hand region of Sl (Slhand), as determined by their anatomical location and preference

<<   86   87   88   89   90   91   92   >>

More intriguing information