Information Technology Reference
In-Depth Information
1
step
complexity
0.9
0.8
0.7
0.6
0.5
0.4
ps01 ps02 ps03 ps04 ps05 ps06 ps08 ps09 ps11 ps13 ps14 ps15 ps16 ps17 ps18 ps19
avg
step
0.67
0.71
0.81
0.68
0.72
0.68
0.75
0.65
0.73
0.62
0.75
0.7
0.75
0.7
0.68
0.67
0.7
0.62
0.58
0.65
0.63
0.6
0.66
0.69
0.62
0.61
0.67
0.65
0.63
0.66
0.65
0.58
0.63
0.63
complexity
(a) Classification results of MVC
1
step
complexity
0.9
0.8
0.7
0.6
0.5
0.4
ps01 ps02 ps03 ps04 ps05 ps06 ps08 ps09 ps11 ps13 ps14 ps15 ps16 ps17 ps18 ps19
avg
0.82
0.67
0.77
0.81
0.85
0.9
0.79
0.73
0.8
0.77
0.91
0.71
0.81
0.8
0.76
0.8
0.79
step
complexity
0.88
0.63
0.75
0.8
0.83
0.89
0.83
0.79
0.85
0.76
0.87
0.79
0.81
0.86
0.85
0.81
0.81
(b) Classification results of SVVC
Fig. 6. Classification results
columns, the highest accuracy reaches high to 91.49% in individual participants,
and the average accuracy across 16 participants is 79.4% [SD = 0.06]. As to the
complexity classifiers, denoted in dark columns, the highest accuracy is 89% in
individual participants, and the average accuracy across 16 participants is high
to 81.3% [SD = 0.06]. The accuracy of SVVC is higher than MVC significantly
[
F
(1, 33) = 99.651,
p<
0. 001].
4
Discussion
It is well known, classification performance depends on features selection, and
suitable features can improve the classifier performance effectively. The high ac-
curacy in our classification demonstrat that selecting PPC, PFC, ACC, FG, cau-
date as features is suitable for classifying high-level cognitive states. Significant
BOLD pattern differences as shown in Figure 7 can explain the high accuracy
in our classification. Because these regions are related to solving the 4*4 Su-
doku, significant task differences in BOLD signal can improve the accuracy of
classification when these regions are selected as features. The high classification
accuracy might indicate that these regions take part in problem solving. Hence,
the classification method might be used to analyze activated brain regions in cog-
nitive tasks. Researchers have already used this pattern classification in fMRI
data analysis to explore brain activity [1, 2]. It is also interesting to see that
although the same features are selected in two methods mentioned above, the
accuracy of SVVC is higher than MVC significantly. This is due to information
lossless by feature extracting from BOLD effect time course in SVVC.
Search WWH ::




Custom Search