Biomedical Engineering Reference
In-Depth Information
FIGURE 14.5
Effect of control definition point positions in a DNA histogram model. (a)
Probability state model for a DNA histogram where the control definition points are at their
optimal location. Notice the uniform frequency distribution (bottom, b). In (b), the second
control definition point (arrow) is moved to the left. Notice the nonuniform frequency
distribution (bottom, b).
14.4.4.4 Classifying Events into a State Of course, we still have the issue of how
to associate a particular event with a probability state to figure out. After all, we cannot
just use magic. The first step in the process is quite simple. We design a computer
algorithm that takes one event at a time and compares it with each state
s parameter
value. It asks the question, “how close is this event to the model parameter intensity
value?” Since each of these parameter values is surrounded by an uncertainty
probability distribution, it uses this distribution to calculate the probability weight.
The higher theweight, the more likely the event is associated with the state. If there are
a number of parameter values associated with the state, it simply multiplies all of these
weights together to account for how close it is to all of them.
14.4.4.5 Method of Maximum Likelihood Fails If there are a hundred states for
this axis, then we will have a hundred weights calculated for each event. This process
sounds more complicated than it really is. The next step, however, is a difficult one to
grasp immediately. How do we use these weights to figure out what state our event
belongs to? At first, you might be tempted to say, find the highest weight and associate
the event with that state. In all honesty, that was my first choice when developing
the event selection algorithm. When you use this maximum likelihood approach, the
Search WWH ::
Custom Search