Customized Prediction of Respiratory Motion - Prediction and Classification of Respiratory Motion - page 92

(a)

x 10 9

8

6

4

2

< Extended Range >

0

0 20 40 60 80 100 120

Combination number : 10 C 3

(b)

50

40

Estimated
feature
combination

30

20

10

0

40 50 60 70 80 90 100 110 120

Combination number : 10 C 3

Fig. 5.3 Dominant feature selection with feature combination vector a whole range and
b extended range. We can define 120 feature combination vectors (I) with 10 feature selection
metrics as shown in Table 5.2 and select the dominant feature combination vectors (Î) with the
minimum value of H(I), i.e. the feature combinations with Breath frequency (BRF), Principal
component coefficient (PCA), Maximum likelihood estimates (MLE), Multiple linear regression
coefficient (MLR) and Standard deviation (STD)

have the prediction parameters (neuron number for prediction and coupling param-
eters) for each class. The intraprocedure corresponds to the prediction process for each
patient in Fig. 5.1 . With the prediction parameters of the preprocedure, the intra-
procedure can operate to predict the respiratory motion of each patient. Sections 5.3.1
and 5.3.2 explain the clustering method for the group, based on breathing patterns and
how to find an optimal neuron number of the prediction process for each class,
respectively.

5.3.1 Grouping Breathing Pattern for Prediction Process

Figure 5.2 illustrates the interactive process, involved in forming a clustering
based on the breathing patterns of multiple patients. For the first step of CNN, we
need to classify the breathing patterns of multiple patients. To extract the breathing