128
Biomedical Signal and Image Processing
7.3.1 BIOMEDICAL AND BIOLOGICAL FEATURES
Biomedical and biological features, as defined in this book, are the features defined
by the knowledge of biology or medicine available about the biological system under
study. As the continuation of the previous example, when detecting flu, the most
relevant feature proposed by physicians is the body temperature. In many medical
applications, a number of features are proposed by the experts (e.g., physicians and
biologists) that must be included in the classification of clustering process. In the following chapters, when discussing a number of physiological and biological systems,
some important biomedical and biological features pertinent to those systems will
be introduced.
Even though the importance of the features identified by the domain knowledge
cannot be overestimated, there are a number of other important features that may
not be defined in the medical knowledge base or even interpreted by the experts. We
group these features as a second category.
7.3.2 SIGNAL AND IMAGE PROCESSING FEATURES
In any field of study, there are many features of a quantity that may have not been identified or named by the experts but have the potential of improving the classification or
clustering significantly. In biomedical signal and image analysis, some of these features
are purely mathematical concepts that may not have a direct physiological or biomedical
meaning for the users. As an example, consider the wavelet coefficients of a certain level
of decomposition of a biomedical signal. While it may be difficult to find a direct and
specific biomedical concept for such coefficients, they are known to be extremely useful
features for clustering and classification of many important biomedical signals such as
electroencephalogram (EEG) and electrocardiogram (ECG).
It has to be mentioned that some of the apparently pure signal and image processing features can be related to the biologically meaningful features. For example, in
the detection of some diseases of the central nervous system (such as epilepsy) from
EEG, physicians often count the number of signal peaks or spikes in a given period
of time (e.g., a minute) and treat this number as an informative feature in the detection of the disease. It is clear that such a biomedical feature is very closely related
to mathematical concepts such as the power of the signal in high frequency of the
Fourier transform (FT). This example describes why one needs to study the biology
and physiology of the system under study in order to devise mathematical features
that best represent the qualitatively defined concepts quantitatively.
In this section, some of the most commonly used signal processing features often
used in biomedical signal and image processing are briefly reviewed. Many transforms and computational concepts described in the previous chapters play important
roles in extracting useful measures described in the following.
7.3.2.1 Signal Power in Frequency Bands
The high-frequency contents of a signal are often interpreted as a measure of
rapid variation in the signal. Similarly, the contents of a signal at different frequency bands quantitatively express the features that are often vital for diagnosis of
Biomedical Signal and Image Processing
7.3.1 BIOMEDICAL AND BIOLOGICAL FEATURES
Biomedical and biological features, as defined in this book, are the features defined
by the knowledge of biology or medicine available about the biological system under
study. As the continuation of the previous example, when detecting flu, the most
relevant feature proposed by physicians is the body temperature. In many medical
applications, a number of features are proposed by the experts (e.g., physicians and
biologists) that must be included in the classification of clustering process. In the following chapters, when discussing a number of physiological and biological systems,
some important biomedical and biological features pertinent to those systems will
be introduced.
Even though the importance of the features identified by the domain knowledge
cannot be overestimated, there are a number of other important features that may
not be defined in the medical knowledge base or even interpreted by the experts. We
group these features as a second category.
7.3.2 SIGNAL AND IMAGE PROCESSING FEATURES
In any field of study, there are many features of a quantity that may have not been identified or named by the experts but have the potential of improving the classification or
clustering significantly. In biomedical signal and image analysis, some of these features
are purely mathematical concepts that may not have a direct physiological or biomedical
meaning for the users. As an example, consider the wavelet coefficients of a certain level
of decomposition of a biomedical signal. While it may be difficult to find a direct and
specific biomedical concept for such coefficients, they are known to be extremely useful
features for clustering and classification of many important biomedical signals such as
electroencephalogram (EEG) and electrocardiogram (ECG).
It has to be mentioned that some of the apparently pure signal and image processing features can be related to the biologically meaningful features. For example, in
the detection of some diseases of the central nervous system (such as epilepsy) from
EEG, physicians often count the number of signal peaks or spikes in a given period
of time (e.g., a minute) and treat this number as an informative feature in the detection of the disease. It is clear that such a biomedical feature is very closely related
to mathematical concepts such as the power of the signal in high frequency of the
Fourier transform (FT). This example describes why one needs to study the biology
and physiology of the system under study in order to devise mathematical features
that best represent the qualitatively defined concepts quantitatively.
In this section, some of the most commonly used signal processing features often
used in biomedical signal and image processing are briefly reviewed. Many transforms and computational concepts described in the previous chapters play important
roles in extracting useful measures described in the following.
7.3.2.1 Signal Power in Frequency Bands
The high-frequency contents of a signal are often interpreted as a measure of
rapid variation in the signal. Similarly, the contents of a signal at different frequency bands quantitatively express the features that are often vital for diagnosis of