Introduction
I.1 PROCESSING OF BIOMEDICAL DATA
Processing of biological and medical information has long been a dynamic field of
life science. Before the widespread use of digital computers, however, almost all
processing was performed by human experts directly. For instance, in processing
and analysis of the vital signs (such as blood pressure), physicians had to rely entirely
on their hearing and visual and heuristic experience. The accuracy and reliability
of such “manual” diagnostic processes are limited by a number of factors, including limitations of humans in extracting and detecting certain features from signals.
Moreover, such manual analysis of medical data suffers from other factors such as
human errors due to fatigue and subjectiveness of the decision-making processes.
In the last few decades, advancements of the emerging biomedical sensing and
imaging technologies such as magnetic resonance imaging (MRI), x-ray computed
tomography (CT) imaging, and ultrasound imaging have provided us with very large
amounts of biomedical data that can never be processed by medical practitioners
within a finite time span.
Biomedical information processing comprises the techniques that apply mathematical tools to extract important diagnostic information from biomedical and biological data. Due to the size and complexity of such data, computers are put to the task
of processing, visualizing, and even classifying samples. The main steps of a typical
biomedical measurement and processing system are shown in Figure I.1. As can be
seen, the first step is to identify the relevant physical properties of the biomedical
system that can be measured using suitable sensors. For example, electrocardiogram
(ECG) is a signal that records the electrical activities of the heart muscles and is used
to evaluate many functional characteristics of the heart.
Once a biomedical signal is recorded by a sensor, it has to be preprocessed and
filtered. This is necessary because the measured signal often contains some undesirable noise that is combined with the relevant biomedical signal. The usual sources of
noise include the activities of other biological systems that interfere with the desirable signal and the variations due to sensor imperfections. In the ECG example, the
electrical signals caused by the respiratory system are the main sources of noise and
interference.
The next step is to process the filtered signal and extract features that represent or describe the status and conditions of the biomedical system under study.
Such biomedical features (measures) are expected to distinguish between healthy
and deviating cases. A group of extracted features are defined based on the medical characteristics of the biomedical system (such as the heart rate calculated from
ECG). These features are often defined by physicians and biologists, and the task
of biomedical engineers is to create algorithms to extract these features from biomedical signals. Another group of extracted features is the ones defined using signal
and image processing procedures. Even though the direct biological interpretation
xxi
I.1 PROCESSING OF BIOMEDICAL DATA
Processing of biological and medical information has long been a dynamic field of
life science. Before the widespread use of digital computers, however, almost all
processing was performed by human experts directly. For instance, in processing
and analysis of the vital signs (such as blood pressure), physicians had to rely entirely
on their hearing and visual and heuristic experience. The accuracy and reliability
of such “manual” diagnostic processes are limited by a number of factors, including limitations of humans in extracting and detecting certain features from signals.
Moreover, such manual analysis of medical data suffers from other factors such as
human errors due to fatigue and subjectiveness of the decision-making processes.
In the last few decades, advancements of the emerging biomedical sensing and
imaging technologies such as magnetic resonance imaging (MRI), x-ray computed
tomography (CT) imaging, and ultrasound imaging have provided us with very large
amounts of biomedical data that can never be processed by medical practitioners
within a finite time span.
Biomedical information processing comprises the techniques that apply mathematical tools to extract important diagnostic information from biomedical and biological data. Due to the size and complexity of such data, computers are put to the task
of processing, visualizing, and even classifying samples. The main steps of a typical
biomedical measurement and processing system are shown in Figure I.1. As can be
seen, the first step is to identify the relevant physical properties of the biomedical
system that can be measured using suitable sensors. For example, electrocardiogram
(ECG) is a signal that records the electrical activities of the heart muscles and is used
to evaluate many functional characteristics of the heart.
Once a biomedical signal is recorded by a sensor, it has to be preprocessed and
filtered. This is necessary because the measured signal often contains some undesirable noise that is combined with the relevant biomedical signal. The usual sources of
noise include the activities of other biological systems that interfere with the desirable signal and the variations due to sensor imperfections. In the ECG example, the
electrical signals caused by the respiratory system are the main sources of noise and
interference.
The next step is to process the filtered signal and extract features that represent or describe the status and conditions of the biomedical system under study.
Such biomedical features (measures) are expected to distinguish between healthy
and deviating cases. A group of extracted features are defined based on the medical characteristics of the biomedical system (such as the heart rate calculated from
ECG). These features are often defined by physicians and biologists, and the task
of biomedical engineers is to create algorithms to extract these features from biomedical signals. Another group of extracted features is the ones defined using signal
and image processing procedures. Even though the direct biological interpretation
xxi