Biomedical Engineering Reference
In-Depth Information
Ta b l e 4 .
mean bout duration quartiles of different clusters, in epochs
wake
N1
N2
SWS
REM
Q1
Q2
Q3 Q1 Q2 Q3 Q1 Q2 Q3 Q1
Q2
Q3 Q1
Q2
Q3
k
=2
cluster 1 (n=211) 1.4
2.9
8.4 1.1 1.4 2.2 5.9
12
24
17
28
50
19
27
44
k
=2
cluster 2 (n=33)
16
21
51 1.5 2.5 4.6 4.6
11
20
15
24
40
15
18
33
k
=3
cluster 1 (n=148) 1.5
3.1
8.6 1.0 1.3 2.0 5.8
12
23 9.4
20
36
20
28
43
k
=3
cluster 2 (n=19)
26
32
71 1.5 2.5 4.1 4.8
11
19 9.7
17
34
12
12
21
k
=3
cluster 3 (n=77)
1.4
3
10 1.2 1.8 3.3 5.8
13
25
34
45
76
17
25
47
k
=4
cluster 1 (n=157) 1.4
3.0
8.2 1.1 1.4 2.1 5.8 12.0 24
11
22
41
16
24
39
k
=4
cluster 2 (n=15)
32
39
73 1.6 2.7 4.6 4.1
10
18
11
16
30 5.3
5.6
9.9
k
=4
cluster 3 (n=48)
1.3
2.8
10 1.2 1.9 3.7 5.1
13
23
39
48
77
13
22
46
k
=4
cluster 4 (n=24)
2.2
5.4
23 1.1 1.5 2.3 7.2
15
24
17
29
53
52
59
80
not found to be significant for
k
=2
,
3
. Additionally, both Kruskal-Wallis and ANOVA
find highly significant
(
p<
10
−
6
)
differences among clusters in the SWS bout duration
quartile variables for
k
=3
,
4
. In contrast, the differences in the stage NREM2 bout
duration quartiles among clusters are not found to be significant for any of the clustering
families,
k
=2
,
3
,
4
. Pairwise statistical comparisons provide additional information,
andarediscussedinsection3.3below.
Pairwise Comparisons. Bout Duration Characteristics of Individual Clusters.
The
following are a few noteworthy statistically significant differences in bout durations.
The reader is also referred to Table 4, and Fig. 5, 6, 7 in conjunction with this discussion.
Below, the precise family (value of
k
) is omitted when bout duration characteristics of
a given cluster number are qualitatively similar for different values of
k
.
Cluster
1
.
Clusters
1
and
3
share the property that their median wake bout duration
quartiles are significantly lower than for clusters
2
and
4
(Wilcoxon
p<
0
.
05
). On
the other hand, cluster
1
has significantly lower SWS bout duration quartiles than clus-
ter
3
. See Fig. 5, Fig. 6, and Fig. 7. The bout duration characteristics of cluster
1
are
remarkably stable across values of
k
.
Cluster
2
.
Cluster
2
consistently has significantly higher wake bout duration quartiles
than any other cluster, for
k
=2
,
3
,
4
(Wilcoxon
p<
0
.
02
). The single exception is
the variable wake.Q1 in the case
k
=4
. Low sample sizes for
k
=4
clusters
2
and
4
(
15
and
24
, respectively) likely contribute to the latter isolated nonsignificance finding.
One also observes that, in the progression from
k
=2
to
k
=3
to
k
=4
,cluster
2
has
monotonically decreasing REM bout duration quartiles.
Cluster
3
.
As observed in section 3.3,
145
of the
148
instances (approximately
98%
)in
the
k
=3
version of cluster
3
belong to the
k
=2
version of cluster
1
. The remainder of
the
k
=2
cluster
1
instances form the majority of the
k
=3
cluster
3
. Therefore, it is not
surprising that many of the bout duration quartiles for the
k
=3
version of cluster
3
are
similar to those for cluster
1
. See Fig. 6 and Table 4. However, there is an immediately
noticeable difference between clusters
1
and
3
for
k
=3
, namely the fact that cluster
3
has visibly higher SWS bout duration quartiles than all other clusters, including cluster
Search WWH ::
Custom Search