Digital Signal Processing Reference
In-Depth Information
Subject 1, MTT
Subject 1, rCBV
1.00
1.0
0.95
0.8
0.90
0.6
0.85
MFE
SOM
FVQ
FSM
NG
MFE
SOM
FVQ
FSM
NG
0.4
0.80
0.2
0.75
0.70
0.0
3
16
18
24
36
3
16
18
24
36
N
N
Subject 2, rCBV
Subject 2, MTT
1.0
1.00
0.95
0.8
0.90
0.6
0.85
MFE
SOM
FVQ
FSM
NG
MFE
SOM
FVQ
FSM
NG
0.4
0.80
0.2
0.75
0.70
0.0
3
4
6
16
36
3
4
6
16
36
N
N
Subject 3, MTT
Subject 3, rCBV
1.00
1.0
0.95
0.8
0.90
0.6
0.85
M
FE
SOM
F
V
Q
FSM
NG
MFE
SOM
FVQ
FSM
NG
0.4
0.80
0.2
0.75
0.70
0.0
3
10
16
19
36
3
10
16
19
36
N
N
Subject 4, rCBV
Subject 4, MTT
1.0
1.00
0.95
0.8
0.90
0.6
0.85
MFE
SOM
FVQ
FSM
NG
MFE
SOM
FVQ
FSM
NG
0.4
0.80
0.2
0.75
0.70
0.0
3
12
16
21
36
3
12
16
21
36
N
N
Figure 11.14
Results of the comparison between the different clustering analysis methods on
perfusion MRI data. These methods are minimal free energy VQ (MFE), Kohonen's
map (SOM), the “neural gas” network (NG), fuzzy clustering based on
deterministic annealing, fuzzy
c
-means with unsupervised codebook initialization
(FSM), and the fuzzy
c
-means algorithm (FVQ) with random codebook
initialization. The average area under the curve and its deviations are illustrated for
20 different ROC runs using the same parameters but different algorithms'
initializations. The number of chosen codebook vectors for all techniques is between
3 and 36, and results are plotted for four subjects. Subjects 1 and 2 had a subacute
stroke, while subjects 3 and 4 gave no evidence of cerebrovascular disease. The
ROC analysis is based on two performance metrics: regional cerebral blood volume
(rCBV) (left column) and mean transit time (MTT) (right column). See plate 9.
Search WWH ::
Custom Search