Fuzzy Clustering and Genetic Algorithms - Biomedical Signal Analysis: Contemporary Methods and Applications - page 217

7

Fuzzy Clustering and Genetic Algorithms

Besides artificial neural networks, fuzzy clustering and genetic algo-
rithms represent an important class of processing algorithms for biosig-
nals.
Biosignals are characterized by uncertainties resulting from incom-
plete or imprecise input information, ambiguity, ill-defined or overlap-
ping boundaries among the disease classes or regions, and indefiniteness
in extracting features and relations among them. Any decision taken at a
particular point will heavily influence the following stages. Therefore, an
automatic diagnosis system must have sucient possibilities to capture
the uncertainties involved at every stage, such that the system's out-
put results should reflect minimal uncertainty. In other words, a pattern
can belong to more than one class. Translated to clinical diagnosis, this
means that a patient can exhibit multiple symptoms belonging to several
disease categories. The symptoms do not have to be strictly numerical.
Thus, fuzzy variables can be both linguistic and/or set variables. An ex-
ample of a fuzzy variable is the heart-beat of a person ranging from 40
to 150 beats per minute, which can be described as slow, normal, or fast.
The main difference between fuzzy and neural paradigms is that neural
networks have the ability to learn from data, while fuzzy systems (1)
quantify linguistic inputs and (2) provide an approximation of unknown
and complex input-output rules.
Genetic algorithms are usually employed as optimization procedures
in biosignal processing, such as determining the optimal weights for
neural networks when applied, for example, to the segmentation of
ultrasound images or to the classification of voxels.
This chapter reviews the basics of fuzzy clustering and of genetic
algorithms. Several well-known fuzzy clustering algorithms and fuzzy
learning vector quantization are presented.

7.1 Fuzzy Sets

Fuzzy sets are an important tool for the description of imprecision and
uncertainty.
A classical set is usually represented as a set with a crisp boundary.
For example,