Customized Prediction of Respiratory Motion - Prediction and Classification of Respiratory Motion - page 91

For the interactive process of multiple markers, we use the coupling degree lij
representing the degree to which component (i) depend on one another (j), as
shown at the coupling matrix block in Fig. 5.1 . The coupling degree lij and the
coupling matrix P with p 9 p matrix including all components of coupling degree
can be defined using Eq. ( 4.6 ). We may expect combined relationships between
marker i and j if the coupling degree lij is close to one, i.e., tight coupling,
whereas we may expect released relationships if the coupling degree is far from
one, i.e., loose coupling. With these coupling effects in mind, the prediction system
for multiple patients should organize the whole respiratory motion datasets into
some specific breathing motions that associate together in a group based on the
respiratory patterns. For such associate processes of the multiple patient interac-
tions, we would like to analyze respiratory patterns and extract usable prediction
parameters which are repeatedly utilized in the training data of a group in a manner
going back to the learning process of the respiratory prediction.

5.3 Proposed Filter Design for Multiple Patients

This chapter explains the detailed modeling prediction process based on the
breathing patterns of multiple patients. The procedure for the interactive prediction
consists of the preprocedure (interactive process for multiple patients) and the
intraprocedure (prediction and correction process). We show the interactive pro-
cess for multiple patients in Fig. 5.2 .
In the preprocedure we would like to get the clustering of respiratory motion based
on the breathing patterns of the multiple patients. After the clustering, each class can

< Group based on Breathing Features < Find Optimal

Neuron Number >

Multiple
Markers

Breathing
Pattern (BP)

# Group

BP

Discriminant
Criterion
J =argmin( S W / S B

Multiple
Markers

Breathing
Pattern (BP)

BP

Prediction
Parameters

Grouping

Multiple
Markers

Breathing
Pattern (BP)

# Group

BP

Discriminant
Criterion
J =argmin( S W / S B

Multiple
Markers

Breathing
Pattern (BP)

BP

Fig. 5.2 Interactive process for multiple patients. Here, a multiple markers input in Fig. 5.3
corresponds to three markers in Fig. 5.1 . The preprocedure (interactive process for multiple
patients) can provide the clustering of breathing pattern based on multiple patients and the
prediction parameters for each class