129
Clustering and Classification
biomedical signals. For instance, in EEG (as will be described later), there are four
important “waves” called alpha, beta, gamma, and delta that play central roles in the
analysis of an EEG. These waves are nothing but variations at different frequencies
and therefore can be easily extracted and measured using a filter designed in the
frequency domain. Once a signal is filtered at a prespecified frequency range (using
a band-pass filter), the power of the components passing through the filter describes
how strong those particular frequencies are in the signal, for example, whether or not the
prespecified waves exist in the recorded EEG.
7.3.2.2 Wavelet Measures
Wavelet transform (WT) provides a number of coefficients that decompose a signal
at different scales (as discussed in Chapter 5). These features that are commonly used
for classification and clustering of biomedical signals and images were discussed in
Chapter 6.
7.3.2.3 Complexity Measures
As discussed in Chapter 6, complexity measures describe the sophisticated structure
of biological systems quantitatively. For example, fractal dimension, as described in
Chapter 6, is one of the most important complexity measures that expresses the complexity of a signal and is heavily used in biomedical signal processing techniques.
Many biological systems are known to become less complex as they get older. For
example, the study of the fractal dimension of EEGs taken from people belonging
to different age groups indicates that as people get older, their fractal dimension
decreases. This feature together with other complexity measures such as mobility,
complexity, and entropy was described in Chapter 6.
7.3.2.4 Geometric Measures
Geometric features play an important role in image classification. Some of the main
geometric features are described in the following.
Area: In almost all medical image classification applications, one needs to measure
the size of the objects in an image. The main feature typically used for this is the
area of the object. In image processing, the area of an object is often measured as
the number of pixels inside the object. This means that after segmentation of the
objects, the number of the pixels inside each closed region (object) indicates the size
or area of the region. In processing of tumors, the area of the tumor in a 2-D image
is considered as one of the most informative measures.
Perimeter: The evaluation of the perimeter of an object is often an essential part of
any algorithm designed for biomedical image processing. Contour features such as
perimeter allow distinguishing among the objects having the same area but different
perimeter. For instance, two cell nuclei with the same area can have a very different
contour shape; one can have a circular counter while the other might be a very long
and narrow oval. Since the perimeter of the second cell is much more than that of
the first one (while having the same area), the value of perimeter can be used as an
informative image processing feature.