Particle Swarm Optimization Applied to EEG

Source Localization of Somato sensory

Evoked Potentials

Abstract
One of the most important steps in presurgical diagnosis ofmedically intractable epilepsy is to find
the precise location of the epileptogenic foci. Electroencephalography (EEG) is a noninvasive tool
commonly used at epilepsy surgery centers for presurgical diagnosis. In this paper, a modified
particle swarm optimization (MPSO) method is used to solve the EEG source localization problem.
The method is applied to noninvasive EEG recording of somatosensory evoked potentials (SEPs)
for a healthy subject. A 1 mm hexahedra finite element volume conductor model of the subjects
head was generated using T1-weighted magnetic resonance imaging data. Special consideration
was made to accurately model the skull and cerebrospinal fluid. An exhaustive search pattern and
the MPSO method were then applied to the peak of the averaged SEP data and both identified the
same region of the somatosensory cortex as the location of the SEP source. A clinical expert
independently identified the expected source location, further corroborating the source analysis
methods. The MPSO converged to the global minima with significantly lower computational
complexity compared to the exhaustive search method that required almost 3700 times more
evaluations.

CHAPTER-1
INTRODUCTION

Modern medicine applies variety of imaging techniques of the human body. The group of
electrobiological
electromyography

measurements
(EMG,

comprises

muscular

items

contractions),

as

electrocardiography

(ECG,

heart),

electroencephalography

(EEG,

brain),

magnetoencephalography (MEG, brain), electrogastrography (EGG, stomach), electrooptigraphy

(EOG, eye dipole field). Imaging techniques based on different physical principles include computer
tomography (CT), magnetic resonance imaging (MRI), functional MRI (fMRI), positron emission
tomography

(PET),

and

single

photon

emission

computed

tomography

(SPECT).

Electroencephalography is a medical imaging technique that reads scalp electrical activity generated by
brain structures. The electroencephalogram (EEG) is defined as electrical activity of an alternating type
recorded from the scalp surface after being picked up by metal electrodes and conductive media.
The EEG measured directly from the cortical surface is called electrocortiogram while when using
depth probes it is called electrogram. In this article, we will refer only to EEG measured from the head
surface. Thus electroencephalographic reading is a completely non-invasive procedure that can be
applied repeatedly to patients, normal adults, and children with virtually no risk or limitation. When
brain cells (neurons) are activated, local current flows are produced. EEG measures mostly the currents
that flow during synaptic excitations of the dendrites of many pyramidal neurons in the cerebral cortex.
Differences of electrical potentials are caused by summed postsynaptic graded potentials from
pyramidal cells that create electrical dipoles between soma (body of neuron) and apical dendrites
(neural branches). Brain electrical current consists mostly of Na+, K+, Ca++, and Cl- ions that are
pumped through channels in neuron membranes in the direction governed by membrane potential [2].
The detailed microscopic picture is more sophisticated, including different types of synapses involving
variety of neurotransmitters. Only large populations of active neurons can generate electrical activity
recordable on the head surface. Between electrode and neuronal layers current penetrates through skin,
skull and several other layers. Weak electrical signals detected by the scalp electrodes are massively
amplified, and then displayed on paper or stored to computer memory . Due to capability to reflect
both the normal and abnormal electrical activity of the brain, EEG has been found to be a very
powerful tool in the field of neurology and clinical neurophysiology. The human brain electric activity
starts around the 17-23 week of prenatal development. It is assumed that at birth the full number of
neural cells is already developed, roughly 10 11 neurons . This makes an average density of 10 4

neurons per cubic mm. Neurons are mutually connected into neural nets through synapses. Adults have
about 500 trillion (5.1014) synapses. The number of synapses per one neuron with age increases,
however the number of neurons with age decreases, thus the total number of synapses decreases with
age too. From the anatomical point of view, the brain can be divided into three sections: cerebrum,
cerebellum, and brain stem. The cerebrum consists of left and right hemisphere with highly convoluted
surface layer called cerebral cortex.
1. EEG: History, brain waves, applications
During more than 100 years of its history, encephalography has undergone massive progress. The
existence of electrical currents in the brain was discovered in 1875 by an English physician Richard
Caton. Caton observed the EEG from the exposed brains of rabbits and monkeys. In 1924 Hans Berger,
a German neurologist, used his ordinary radio equipment to amplify the brain's electrical activity
measured on the human scalp. He announced that weak electric currents generated in the brain can be
recorded without opening the skull, and depicted graphically on a strip of paper. The activity that he
observed changed according to the functional status of the brain, such as in sleep, anaesthesia, lack of
oxygen and in certain neural diseases, such as in epilepsy. Berger laid the foundations for many of the
present applications of electroencephalography. He also used the word electroencephalogram as the
first for describing brain electric potentials in humans. He was right with his suggestion, that brain
activity changes in a consistent and recognizable way when the general status of the subject changes,
as from relaxation to alertness . Later in 1934 Adrian and Matthews published the paper verifying
concept of human brain waves and identified regular oscillations around 10 to 12 Hz which they
termed alpha rhythm.
1.1Brain waves classification
For obtaining basic brain patterns of individuals, subjects are instructed to close their eyes and relax.
Brain patterns form wave shapes that are commonly sinusoidal. Usually, they are measured from peak
to peak and normally range from 0.5 to 100 V in amplitude, which is about 100 times lower than
ECG signals. By means of Fourier transform power spectrum from the raw EEG signal is derived. In
power spectrum contribution of sine waves with different frequencies are visible. Although the
spectrum is continuous, ranging from 0 Hz up to one half of sampling frequency, the brain state of the

individual may make certain frequencies more dominant. Brain waves have been categorized into four
basic groups :
- beta (>13 Hz),
- alpha (8-13 Hz),
- theta (4-8 Hz),
- delta (0.5-4 Hz).

Figure 1.1 Brain wave samples with dominant frequencies belonging to beta, alpha, theta,
and delta band.
The best-known and most extensively studied rhythm of the human brain is the normal alpha rhythm.
Alpha can be usually observed better in the posterior and occipital regions with typical amplitude about
50 V (peak-peak). According to our experiences alpha was also significant between posterior and
central regions in comparison to other regions. Alpha activity is induced by closing the eyes and by
relaxation, and abolished by eye opening or alerting by any mechanism (thinking, calculating). Most of
the people are remarkably sensitive to the phenomenon of eye closing, i.e. when they close their eyes
their wave pattern significantly changes from beta into alpha waves. The precise origin of the alpha
rhythm is still not known. Alpha waves are usually attributed to summated dendrite potentials. Evoked
potentials (e.g. generated in brain stem) often consist of fibre potentials (axonal) and synaptic
components.
EEG is sensitive to a continuum of states ranging from stress state, alertness to resting state, hypnosis,
and sleep. During normal state of wakefulness with open eyes beta waves are dominant. In relaxation

or drowsiness alpha activity rises and if sleep appears power of lower frequency bands increase. Sleep
is generally divided into two broad types: non rapid eye movement sleep (NREM) and REM sleep.
NREM and REM occur in alternating cycles. NREM is further divided into stage I, stage II, stage III,
stage IV. The last two stages corresponds to deeper sleep, where slow delta waves show higher
proportions. With slower dominant frequencies responsiveness to stimuli decreases.
Various regions of the brain do not emit the same brain wave frequency simultaneously. An EEG signal
between electrodes placed on the scalp consists of many waves with different characteristics. A large
amount of data received from even one single EEG recording presents a difficulty for interpretation.
Individuals brain wave patterns are unique. In some cases, it is possible to distinguish persons only
according to their typical brain activity. For example, subjects who regard themselves as rational types
or as holistic/intuitive types may demonstrate certain higher activity in their frontal left and frontal
right hemisphere respectively.
1.1.1Applications
The greatest advantage of EEG is speed. Complex patterns of neural activity can be recorded occurring
within fractions of a second after a stimulus has been administered. EEG provides less spatial
resolution compared to MRI and PET. Thus for better allocation within the brain, EEG images are
often combined with MRI scans. EEG can determine the relative strengths and positions of electrical
activity in different brain regions.
According to R. Bickford research and clinical applications of the EEG in humans and animals are
used to:
(1) monitor alertness, coma and brain death;
(2) locate areas of damage following head injury, stroke, tumour, etc.;
(3) test afferent pathways (by evoked potentials);
(4) monitor cognitive engagement (alpha rhythm);
(5) produce biofeedback situations, alpha, etc.;
(6) control anaesthesia depth (servo anaesthesia);
(7) investigate epilepsy and locate seizure origin;
(8) test epilepsy drug effects;
(9) assist in experimental cortical excision of epileptic focus;

(10) monitor human and animal brain development;

(11) test drugs for convulsive effects;
(12) investigate sleep disorder and physiology.
Symmetry of alpha activity within hemispheres can be monitored. In cases of restricted lesions
such as tumour, hemorhage, and trombosis, it is usual for the cortex to generate lower frequencies.
EEG signal distortion can be manifested by reduction in amplitude; decrease of dominant
frequencies beyond the normal limit; production of spikes or special patterns. Epileptic conditions
produce stimulation of the cortex and the appearance of high-voltage waves (up to 1000 V)
referred to as spikes or spike and wave . EEG patterns have been shown to be modified by a
wide range of variables, including biochemical, metabolic, circulatory, hormonal, neuroelectric,
and behavioural factors. By tracking Electroencephalography (EEG) signal is the recording of
spontaneous electrical activity of the brain over a small period of time.
The term EEG refers that the brain activity emits the signal from head and being drawn. It is
produced by bombardment of neurons within the brain. It is measured for a short duration of 20-40
minutes with the help of placing multiple electrodes over the scalp. The researchers have found that
these signals indicate the brain function and status of whole body. Therefore EEG signal provides
valuable information of the brain function and neurobiological disorders as it provides a visual
display of the recorded waveform and allows computer aided signal processing techniques
to characterize them. This gives a prime motivation to apply the advanced digital signal processing
techniques for analysis of EEG signals. EEG signal acquisition follows a noninvasive procedure
and proper signal processing and pattern recognition tools can emerge an important device for
automated system to recognize electroencephalographic changes. Real time automated EEG signal
analysis in clinical settings is of great assistance to clinicians in detecting neurobiological diseases
and prevention of the possibility of missing (or misreading) information. However, automated
classification of EEG signals is a challenging problem as the morphological and temporal
characteristics of EEG signals show significant variations for different patients and under different
temporal and physical conditions. This ability motivates us to apply the advanced digital signal
processing techniques on the EEG signals. The study of neuronal function and neurophysiological
properties with EEG signals
generation and their detection plays the vital role in detection, diagnosis of brain disorders

and brain diseases.

1.1.2 Neural Activities
The Central Nervous System (CNS) consists of two types of cells, nerve cells and glia cells. All the
nerve cell consists of axons, dendrites and cell bodies. The proteins developed in the cell body are
transmitting information to other parts of the body. The long cylindrical shaped axon transmits the
electrical impulse. Dendrites are connected either to the axons or dendrites of other inside cells and
receive the electrical impulse from other nerves cells. The study of brain found that each nerve of
human is approximately connected to 10000 other nerves. The electrical activity is mainly due the
current flow between the tip of dendrites and axons, dendrites and dendrites of cells. The level of
these signals is in Mv

Figure 1.2: Structure of neuron

range and its frequency is less than 100Hz .
1.1.3 EEG signal generation
The EEG signal is the current measured between the dendrites of nerve cells in the cerebrum region
of the brain during their synaptic excitation. This current consists of electric field detected by
electroencephalography (EEG) equipment and the magnetic field quantified by electromyogram

(EMG) devices . The human head consists of many parts some of which is scalp, skull, and brain.
The noise is either to be produced internally in the brain or externally over the scalp due to the
system used for recording. The degree of attenuation caused by skull is nearly hundred times larger
than the soft tissues present in the head. As the level of signal amplitude is very low, the more
number of neuron in excitation state can only produce the recordable potential by the scalp

electrode

system.

Thus

the

amplifiers

are

Figure 1.4: Conventional 10-20 EEG electrode positions for the placement of 21 electrodes
used to increase the level of signals for later processing . The brain structure is divided into three
regions, cerebrum, cerebellum and brain stem. The cerebrum region defines the initiation of

movement, conscious sensation and state of mind. The cerebellum region plays a role in voluntary
actions like muscle movements. The brain stem region controls the respiration functioning, heart
regulation, biorhythms and neural hormones. It is clear that the EEG signals generated from brain
can determine the status of whole body and brain disorders .
1.1.4 Recording of EEG signals
EEG is most often recorded from many electrodes in a arranged in a particular pattern or montage.
A common standard for describing these position is the International 10/20 System. These methods
are cheap and give a continuous record of brain activity with better than millisecond resolution. No
other tool can achieve this high temporal resolution and for this reasons many of the detailed
discoveries of dynamic cognitive processes have been reported using EEG and ERP (Event Related
Potentials) methods. The first electrical variation was noted by using a simple galvanometer. But in
recent generation they are recorded by using EEG systems which consist of multiple electrodes,
amplifiers single for each channel to amplify the attenuated signals and followed by filters to
remove the system noise and registers . Later, EEG systems are equipped with computerized
system to store the large data, for easy and correct analysis of EEG signals by using high sampling
rate and more number of quantization levels with the help of advanced signal processing tools.
The Analog to Digital Converter (ADC) circuits are used to convert the analog EEG signal to
digital form. The bandwidth for EEG signal is 100 Hz. So, the minimum 200 samples/sec are
required for sampling the EEG signal to satisfy the Nyquist criterion. Sometimes, the higher
sampling rate of 2000 samples/sec is used for getting high resolution of EEG signals. The capacity
of connected memory units depends upon the amount of data recorded. The different types of
electrodes used for recording of high quality data. 1.1.4 Frequency bands of EEG Signals Most of
EEG waves range from 0.5-500 Hz, howev r following five frequency bands are clinically relevant:
(i) delta, (ii) theta, (iii) alpha, (iv) beta and (v) gamma. Delta waves: Delta waves frequency is up
to 3 Hz. It is slowest wave having highest amplitude. It is dominant rhythm in infants up to one
year and adults in deep sleep. Theta waves: it is a slow wave having frequency range from 4 HZ to
7 Hz. It emerges with closing of the eyes and with relaxation. It is normally seen in young children
or arousal in older children and adults. Alpha waves: Alpha activity has frequency range from 7 HZ
to 12 Hz. it most commonly seen in adults. Alpha activity occur rhythmically on both sides of the

head but are often slightly higher in amplitude on the non dominant side, especially in right-handed
individuals. Alpha wave appears with closing eyes (relaxation state) and disappears normally with
opening eyes/stress. it is regarded as a normal waveform. Beta waves: Beta activity is fast having
small amplitude. It has frequency range from 14 HZ to 30 Hz. It is the dominant rhythm in patients
who are alert or anxious or who have their eyes open. Beta waves usually seen on both sides in
symmetrical distribution and is most evident frontally. It may be absent or reduced in areas of
cortical damage. It is generally regarded as a normal rhythm and observed in all age groups. These
are mostly appeared in frontal and central portion of the brain. The amplitude of beta wave is less
than 30 V.
Gamma waves: It is fastest waves of brain having frequency range falls from 30-45 Hz. It is also
called as fast beta waves. These waves have very low amplitude and present rarely. But, the
detection of these rhythms plays an important role in finding the neurological diseases. These
waves are occurred in front central part of the brain. It suggests the event-related synchronization
(ERS) of the brain .
1.1.5 Advantages of EEG
There are various advantage of EEG signals some of them can be stated as follows:
Temporal resolution of EEG signal is high.
EEG measure the electrical activity directly.
EEG is a non-invasive procedure.
EEG has ability to analyze brain activity.
Major disadvantage is EEG that it is difficult to find out the source of electrical activity
from where it is coming out
1.1.6 Uses of EEG
Diagnose epilepsy and see what type of seizures are occurring. EEG is the most useful
and important test in confirming a diagnosis of epilepsy. Check for problems with loss of
consciousness or dementia.
Help find out a persons chance of recovery after a change in consciousness.
Find out if a person who is in a coma is brain-dead.
Study sleep disorders, such as narcolepsy.

 Watch brain activity while a person is receiving general anesthesia during brain surgery.
Help find out if a person has a physical problem (problems in the brain, spinal cord, or
nervous system) or a mental health problem.
1.1.7 Normal and abnormal EEG signals
Normal EEG [7]: In adults who are awake, the EEG shows generally alpha waves and
beta waves. The two sides of the brain show similar patterns of electrical activity. There
are no abnormal bursts of electrical activity and no slow brain waves on the EEG tracing.
If flashing lights (photic stimulation) are used during the test, one area of the brain (the
occipital region) have a brief response after each flash of light, but the brain waves are
normal.
Abnormal EEG [7]: Different patterns of electrical activity are obtained from the two
sides of brain. This represent a problem in one area or side of the brain is present. The
EEG shows sudden bursts of electrical activity (spikes). These sudden changes may be
caused by a brain tumor, injury, epilepsy. When a person has epilepsy, the location and
exact pattern of the abnormal brain waves may help show what type of epilepsy or seizures
the person has. In many people with epilepsy, the EEG may appear completely normal
between seizures. The EEG records changes in the brain waves that may not be in just one
area of the brain.
1.2 Data Base
The raw EEG signal is obtained from repository source as mentioned in [30] which consists
of total 5 sets (classes) of data (SET A, SET B, SET C, SET D, and SET E) corresponding
to five different pathological and normal cases. Three data sets are selected from 5 data
sets in this work. These three types of data represent three classes of EEG signals (SET
A contains recordings from healthy volunteers with open eyes, SET D contains recording
of epilepsy patients in the epileptogenic zone during the seizure free interval, and SET E
contains the recordings of epilepsy patients during epileptic seizures) as shown in Fig. 1.3.
All recordings were measured using Standard Electrode placement scheme also called as
International 10-20 system. Each data set contains the 100 single channel recordings. The
length of each single channel recording was of 26.3 sec. The 128 channel amplifier had
been used for each channel [37]. The data were sampled at a rate of 173.61 samples per
second using the 12 bit ADC. So the total samples present in single channel recording are

nearly equal to 4097 samples (173.6123.6). The band pass filter was fixed at 0.53-40 Hz
(12dB/octave) [38].
1.3 Objective
The main objective of our research is to analyze the acquired EEG signals using signal
processing tools and classify them into different classes. The secondary goal is to improve

Figure 1.3: Raw EEG signal of class A, class D, and class E

the accuracy of classification. In order to achieve this: (i) first a feature extraction technique
is proposed, which combines extracted coefficients of Daubechies-2 wavelet at fourth level
decomposition and Burgs algorithm individually, (ii) a two-level committee neural network
(CNN-2) is proposed for better classification of EEG signals. The final level decision i.e.
the recognition accuracy of the classifier will be made by CNN-2,(iii) select a set of input
features that are effective for identification of EEG signal using genetic algorithm,(iv) make
certain optimum selection of nodes in the hidden layer using genetic algorithm for each ANN
structure of two-level CNN to get effective classification of EEG signal.
1.4 Thesis outline
Chapter 1 of the thesis explains the basic theory behind the EEG signals, this chapter also
discuss about the used data in the experiment.

CHAPTER-2
PROPOSED METHOD

Feature extraction is the collection of relevant information from the signal. Due to high
transitions the signals are unable to distinct and insufficient to answer the status of individual.

So, the DSP tools have been used for extracting the desired characteristics of
different EEG signals and provided the knowledge of human state. A number of feature
extraction methods have been present for study the EEG signal. The wavelet transform is
used for extracting the feature vectors from EEG signal because of its ability to characterize
time-frequency information which is important in this context [3], [12]- [13]. The model
based approach give spectral analysis technique using Burg autoregressive (AR) method
is also used to extract the features representing EEG signals [13]. The research work [13]
demonstrated that the Burgs AR coefficients based features well represent the EEG signals.
The benefits of the model based spectra are that it has good statistical stability for short
segments of signal and has good spectral resolution which is less dependent on the length
of the signal [13], [14], [15].
AR method is frequently used as model based method since AR parameters can be easily
estimate by solving linear equations. [1], [8]- [9]. N. Hazarika et. al. reported a classification
technique based on wavelet transform and artificial neural network [10]. U belli [11]
presented wavelet based feature set and a mixture of experts network for classification of
three different pathological cases of 300 EEG recordings with accuracy 93.17 percent. In
her another reported technique [12], a combined neural network with wavelet based feature
set is used for classification of the same EEG data set and a performance

Figure 2.4: Flow diagram for classification of EEG signal

2.3.1 Normalization
Each recorded signals of all the 3 classes are normalized to -1 and +1 value. This is done by
dividing all the samples with the maximum absolute value. The normalized plot of 3 class
is shown in Fig. 2.5.
Normalize(k) =
Signal(k)

||(Signal(k))||max

Figure 2.5: Raw EEG Normalization of class A, class D and class E

Segment Detection
A rectangular window of length 256 discrete samples is selected from each channel to form
a single EEG segment. The total numbers of time series present in each class are 100 and
each single channel time series contained 16 EEG signal segments. Therefore total 1600
EEG segments are produced from each class. Hence, total 4800 EEG segments are obtained
from three classes. The 4800 EEG segments are divided into training and testing data sets.
The 2400 EEG signal segments (800 vectors from each class) are used for testing and 2400
EEG signal segments (800 segments from each class) are used for training [11].The plot of
EEG segment from three classes (A, D, E) is shown in Fig. 2.6.

AR features: Autoregressive (AR) coefficients describe the important features of EEG

signals, the correct choice of model order is important. Too low and too high order gives the
poor estimation of power spectral density (PSD) [13]. In the present work, Burgs method
is used for calculating the AR coefficients. The model order of 10th is set to the input
signals by minimizing the forward and backward prediction errors for the AR parameters to
satisfy the Levinson-Durbin recursion. The vector of 11 AR coefficients is obtained for each
The essence of magnetic resonance imaging (MRI) is spectroscopy, specifically nuclear magnetic
resonance (NMR) spectroscopy. The technique as a whole has had enormous influence and has
generated more than one Nobel prize. From the first development in NMR awarded to Felix Bloch
(Bloch, 1946) and Edward Mills Purcell (Bloembergen, 1948) in the 1952 Physics prize to the
latest recognition of Paul C. Lauterbur (Lauterbur, 1986) and Sir Peter Mansfield (Mansfield and
Maudsley, 1977) in the 2003 prize for Physiology or Medicine. At its core, NMR utilizes the
absorption and emission of radiofrequency electromagnetic radiation (RF) by nuclei made
susceptible by placement in a magnetic field. If the nuclei of interest has a non-zero quantum of
nuclear spin, its placement into a magnetic field will cause it to align either with or against the
applied field, as it approximates a rotating charge. This applied magnetic field then forces the
particular nuclei of interest into one of several energy levels, the proton, by way of example, falls
into one of two energy levels. The difference in energy levels, termed Zeeman splitting (after Pieter
Zeeman, 1902 Nobel Laureate in Physics), provides for the possibility of radiative absorption or
emission by the nuclei as they are passed from one energy level to another. By slight modulation of
the applied field using the magnetic field gradient coils present in the scanner, it is possible to
measure a different specific absorption and emission frequency and phase for many small volumes
within the bore of the scanner. These small volumes are commonly termed voxels, to specify a
three-dimensional unit of resolution similarly as the 2 term pixel specifies a two-dimensional unit
of resolution. While the precise physics of nuclear spin resonance are deserving of a treatise in and
of themselves (Lambert and Mazzola, 2004), it is useful to have some specific understanding of the
concepts central to MRI. When a suitable RF pulse is applied to the nuclei within the magnetic
field, the populations at the susceptible energy levels are altered. Due to the slight predominance of
nuclei in the lower energy state, as predicted by the Boltzmann distribution, a net absorption of
radiation occurs. Depending on the length of the RF pulse, the precession frequencies of the nuclei
are brought into phase and deflected all or partially into the magnetization plane transverse to the

applied field. As time passes, various effects contribute to the eventual relaxation of this magnetic
moment back to equilibrium. The two types of relaxation commonly measured by MRI are termed
T1 and T2 relaxation. T1, or longitudinal, relaxation is governed by so-called spin-lattice
interactions. That is, the inverted susceptible nuclei transfer their energy to the surrounding milieu,
or lattice, thereby relaxing back into a lower energy state. Classically, this may be conceived of as
the angles of the spin axis of the nuclei gradually realigning with the direction of the applied
magnetic field. The energy of the nuclei is absorbed by the lattice often in the form of increased
molecular rotational or vibrational energy, which, if sustained for an excessive period, can be
observed as an increase in temperature. T2, or transverse, relaxation is governed by so-called spinspin interactions. In spin-spin interactions, energy is transferred directly between two susceptible
nuclei with similar precession frequencies. As the spins simply reverse orientation relative to the
external magnetic field, no net change in T1 occurs. The drop in T2 is due to the loss of phase
coherence when the transfer occurs.
In discussion of MRI pulse sequences, particularly the gradient echo which finds common use in
fMRI, the term T2* is often encountered. In actual measurement, the signal decays much faster
than would normally be predicted by T2 relaxation. The source of this additional speed is
relaxation from the spins of the individual nuclei dephasing as they encounter local fluctuations in
the magnetic field, and is designated as the T2* rate of relaxation.
1.1 From NMR to MRI
Why was NMR used for so long in analytical chemistry before being applied to imaging The
problem was that an NMR signal could provide information about the type of material in a sample,
but the exact location of the signal source was unknown. With no spatial information it is
impossible to form an image. Eventually a solution was devised. In the 1970s, Paul Lauterbur
(American), Richard Ernst (Swiss) and Peter Mansfield (British) all independently demonstrated
the imaging potential of NMR, and MRI was born. In 1978 the first image of a human head was
taken. The key to this sudden burst of success was something known as spatial encoding. Spatial
encoding is described in detail in Section. Ernst received the Nobel Prize in Chemistry in 1991, and
Lauterbur and Mansfield were awarded the Nobel Prize in Medicine in 2003 for their discoveries
concerning magnetic resonance imaging. The press release went on to say: Paul Lauterbur . . .

discovered the possibility to create a two-dimensional picture by introducing gradients in the

magnetic field. Peter Mansfield . . . showed how the signals could be mathematically analysed,
which made it possible to develop a useful imaging technique. The decision to award Lauterbur
and Mansfield the Nobel prize was the source of much controversy. The cause was not because
Lauterbur and Mansfield were undeserving recipients, but because Damadian thought the prize
should also have been awarded to him. Believing that he had been treated unfairly, he put full-page
advertisements in a number of international newspapers protesting the decision. This controversy is
recalled by many even those who remember nothing else about MRI. The first scanners reached
the market by 1982 but only became widely available in the 1990s. By 2003 there were around 75
million scans performed per year using 10 000 MRI machines worldwide.

Figure .1 Modern MRI: (a) a scanner at St Georges Radiology (Christchurch, New Zealand),
and(b) an image of the authors brain acquired using it.
1.2 The effects of patient motion
The data acquisition time in MRI varies depending on the particular sequence used. It is, however,
often of the order of minutes, meaning that there is a high chance the patient will move during the
acquisition. A number of types of motion can adversely affect the final image:
Large-scale/bulk motion: motion of the entire object being imaged, such as the head or torso of a
patient.
Respiration: motion caused by the breathing of the patient.
Cardiac motion: motion of the heart muscle and major blood vessels.
Blood flow: blood moving through an otherwise stationary slice.

 Gastrointestinal peristalsis: motion of the intestine and its contents due to peristalsis.
Of these five, the last four have somewhat satisfactory solutions. Motion from respiration can
sometimes be prevented by breath-holding; however, for longer scans techniques such as
respiratory gating or respiratory ordered phase encoding (ROPE) are used. Similarly, the problem
of cardiac motion is commonly mitigated by cardiac gating. Blood flow effects can be reduced
using techniques such as gradient moment nulling. Peristalsis can be temporarily frozen for a
short time during imaging using antiperistaltic drugs such as hyoscine butylbromide. Most of the
above techniques have been available for many years: a summary is available in a 1986 review
article by Bellon et al. Bulk motion is, however, still a major problem in MRI. This thesis focuses
on bulk motion that occurs during a single acquisition sequence, not motion occurring between
sequences; the latter can be corrected for using image registration techniques. The author
performed a simple experiment to quantify the head motion of healthy subjects trying to remain
motionless. Results of this experiment are given in Appendix.
For simplicity of analysis, attention is restricted to motion consisting of rigid-body translation and
rotation within the slice being imaged and therefore involving the x- and y dimensions only. If a
body is translated a distance (x0, y0) from a reference position during data acquisition, a phase
error is generated in the data. Similarly, if the body is rotated from a reference position, acquired
lines in k-space are correspondingly rotated from their nominal orientation. Data acquisition is
extremely rapid in the frequency-encode (FE) direction so motion is effectively frozen for this
time. Thus, inconsistencies which exist in k-space due to motion only occur between lines in the
phase-encode (PE) direction. Inconsistencies in the PE direcion result in ghosting or blurring in the
same direction in the image.Here, ghosting is defined as repeated versions of the image or part of
the image while blurring is similar to motion blur in a photograph. Blurring is caused by a wide
central peak in the point spread function (PSF).
1.3 Structural MRI
Structural MRI makes use of the small differences between voxels to differentiate neighboring
types of tissue. For instance, the juxtaposition of a water-rich and water-poor tissue will lead to the
eventual resolution of distinct voxel intensities as the T1 relaxation rates differ. In the particular

example of a T1 image, it is illustrative to examine the different relaxation rates of nuclei in each of
three environments commonly encountered in a scan of a neuroscience subject: bone, brain, and
cerebral-spinal fluid (CSF.) In bone, the surrounding lattice is a solid with few protons to begin
with, and those that are present have a long T1 relaxation time due to the scarcity of spin-lattice
interactions in the rigid material. In brain, the mobility and molecular absorption frequencies of the
tissue now provides for such relaxation events, leading to a shortening of the T1 relaxation time,
and a presence of signal on the image. In CSF, the milieu is so mobile, suitable absorption
frequencies are thinly distributed through a broad range causing the probability of an appropriate
transfer event to fall, once again lengthening T1 relaxation time. Due to this relationship, CSF
appears with the same low image intensity as bone. While imaging of the T1 susceptibility 4 of
tissue is common in anatomical studies, it is not unusual to acquire a T2 image as well, due to
certain types of pathology being more readily discernable in a T2 image. Acquisition time in an
anatomical study is usually only constrained by fatigue of the patient within the bore or clinical
concerns. Essentially, only one measurement of each voxel volume needs to be made, although
scans are often repeated for purposes of signal averaging. The resolution of such scans can
therefore be very high to accommodate clinical discrimination of tissue features, with any drop in
signal being overcome by lengthening the acquisition time of the scan. Contrast this method with
the need for rapid whole volume acquisition in functional imaging, as discussed below.
1.4 Functional MRI
Differing from anatomical imaging, functional imaging requires speed and sensitivity of
acquisition, usually at the expense of resolution. There is nothing fundamentally different about the
signal which is acquired in a functional imaging scan, theoretically any type of anatomical imaging
protocol could be used, however the main concern is now not anatomical discriminability, but
change in tissue properties over time. T2 weighted images are almost universally used due to the
reliance on imaging a changing local magnetic field due to various effects arising from functional
activation. Although still a much debated topic, the foundation of functional MRI rests on the
observation that activated neural tissue experiences a vascular change (Ogawa et al., 1990). The
volume and flow of blood to an activated area of cortex can increase dynamically, with an
accompanying change in oxygen Whether this activation truly represents an increase in neuronal
activity is a current subject of investigation by means of combined electrophysiology and fMRI

(Silva et al., 2000; Attwell and Iadecola, 2002; Logothetis, 2002, 2003; Logothetis and Wandell,
2004). Such comparative studies have shown a correlation between the increase of Fmri signal and
spiking activity during a variable-coherence motion task (Rees et al., 2000) and the direct
correlation of the fMRI response to the local field of simultaneously recorded neurons (Logothetis,
2002). Fundamentally, multiple low-resolution images of the area of interest are acquired in rapid
succession, with a volume being completed typically on the order of two seconds. These images
can then be compared in series to produce a time series of intensities for each voxel. The nature of
the signal thus produced is dependent on
the method of investigation, mainly differing in method of contrast used, as discussed below. This
time series may then be used for experimental evaluation of response to an applied stimulus, in
either a block design or event related paradigm. In a block design experiment, the subject performs
two or more differing tasks, each individually and constantly for an extended period of time. The
relative levels of activation during each task are then compared to gain insight into the differential
activation of a cortical region. For an event related experiment, a single task is performed multiple
times, and the ensuing responses are aligned upon the onset of the task and averaged in order to
measure the time course of activation in a given cortical region. Hence, the strength of a block
design paradigm is localization of functionally responsive areas, while the strength of an
eventrelated experiment is insight into the nature of the ensuing response.
1.5 Blood Oxygen Level Dependent (BOLD) Contrast
It is the magnetic properties of hemoglobin itself which make a truly noninvasive functional
imaging technique possible with MRI. Ever since the local blood oxygenation state was linked to
signal intensity changes (Ogawa et al., 1990), the BOLD contrast method has been nothing short of
a revolution in functional imaging. The technique depends on the fact that de oxyhemoglobin is
weakly paramagnetic, and thus causes a local decrease in signal strength in vasculature and
surrounding tissue. Specifically, as blood oxygenation decreases, its T2* shortens as well.
However, when an area of cortex becomes active, cerebral blood flow and blood volume increase
leading to an eventual lowering of the oxygen extraction fraction of the blood that is, the blood
supply increases beyond the demand, causing an eventual decrease in the level of de
oxyhemoglobin. As the amount of de oxyhemoglobin diminishes, the paramagnetic effects are

removed, and a rise in signal intensity is measured. In BOLD imaging, therefore, an activated area
of cortex would demonstrate an elevation in signal intensity.

I.5.1 Exogenous Contrast

Much like the BOLD method described above, the use of an exogenous contrast agent in functional
imaging depends on the eventual change in local concentration of a magnetically active substance.
A wide variety of contrast agents are used clinically for diagnostic purposes, with gadolinium a
compound long known to affect magnetic relaxation rates(Bernheim et al., 1959) perhaps being
the most common (Gries and Miklautz, 1984; Villringer et al., 1988). Superparamagnetic iron
oxide agents, such as Ferridex, also have use in the imaging of organ tissue, but can be readily
adapted as a functional contrast as well. Recently, the agent MION has been developed specifically
for fMRI and demonstrates improvements over other iron oxide contrasts in terms of spatial and
temporal resolution (Shen et al., 1993; Mandeville et al., 1998). In this case, when an areas of
cortex becomes active, the local vasculature expands with an increase in cerebral blood volume,
resulting in a given volume of brain having relatively more of the iron oxide agent. This local
concentration of iron oxide produces a focal inhomogeneity in the magnetic field of the scanner.
Nuclei affected by this inhomogeneity dephase quickly producing a drop in T2*. Much like the
effects then of deoxygenated blood, a drop in signal intensity is observed. It is important to note
that this is the opposite effect of the BOLD signal: with an iron oxide contrast agent, an increase in
activation yields a decrease in signal intensity.
1.6 fMRI Investigation of the Visual Motion Aftereffect
Similar to the electrophysiology, investigation of the time course of fMRI activation during a MAE
task shows a correlation between perceived strength of motion and the MR signal strength (Tootell
et al., 1995). Furthermore, this technique allowed the anatomical localization of an area in human
cortex which responded similarly to the macaque motion sensitive areas MT and MST (Tanaka et
al., 1986). This early study did not have the spatial resolution to separate the two regions
functionally, and the area was therefore termed the MT+ complex in human. More recent
investigations involving hemifield stimuli similar to those used here were able to functionally
dissect human MT+ into putative MT (pMT) and MST (pMST) regions posteriorly and anteriorly,

respectively (Dukelow et al., 2001). A separate study has linked the effects of selective attention to
an upregulation of the response during the MAE (Huk et al., 2001).

As the technique begins to proliferate, macaque cortex beyond the primary visual areas is currently
being investigated. Anterior regions in the ventral stream of object and shape recognition are being
described (Denys et al., 2004) and are demonstrated to show selective activation for the
presentation of intact versus scrambled objects. Similarly, these areas have also been found to
respond to images of faces (Tsao et al., 2003). Beyond passive viewing experiments, areas of
cortex responsive to the execution and direction of saccades have been functionally imaged in
frontal eye fields and posterior parietal cortex (Koyamaet al., 2004).
I.7 SURGICAL PLANNING
Secondary only to primary diagnostic imaging, MRI finds clinical use as a surgical planning tool.
Whether the procedure demands the targeting of a structure for repair or stimulation, or the
demarcation of the extent of a region to be excised, as in the case of a 12 tumor, the high-resolution
of MRI clearly provides a major aid. Indeed, beyond investigation in humans, anatomical imaging
has been used for localization and differentiation in species useful in experimentation, such as the
rhesus monkey (Price et al., 1997). Adapted to the neuroscience laboratory, medical uses may still
be prevalent, but in the case of single and multiple cell electrophysiology a prime concern is the
targeting of implanted electrodes or recording chambers (Asahi et al., 2003; Scherberger et al.,
2003). Precise drug delivery, necessary in lesioning, also benefits from image-based targeting
(Blaizot et al., 1999). While use of a magnet-compatible and visible stereotaxic device has been
developed for surgical planning in the macaque (Saunders et al., 1990), such a technique requires
specialized equipment that may not be readily available. Fortunately, anatomical landmarks
commonly used in stereotaxic positioning, such as the internal auditory meati, can be visualized by
imaging. The acquired image can be modified by computational methods (Cox, 1996a) to any
alignment, in this case orthogonal to a stereotaxic setup. Using anatomical or functional landmarks
to identify the structures of interest on the MRI image, the surgeon is then able to position
chambers and/or electrodes by using the coordinate frame of the stereotaxic device inside the
operating theatre (Dubowitz, 2002a). Such precise targeting helps to ensure that neurons of interest

fall near subsequent electrode penetrations, paring the time and effort of the experimenter and
subject.

Visual motion is an integral and vital facet of sensory perception. While information may certainly
be conveyed by a static scene, it is changing or moving scenes which are most prevalent in daily
life. A simple, yet striking, illustration of the cortical processing of motion is the visual motion
aftereffect (MAE), an early subject of psychophysics (Addams, 1834; Wohlgemuth, 1911). After
viewing a scene with a constant direction of motion for several seconds, the opposite of that motion
is perceived upon subsequent viewing of a static image. The percept of the aftereffect can be
extended by the placement of an intervening dark, or storage, period between the two phases of the
stimulus. Investigation of motion and its aftereffect has been pursued in many species, with
fundamental studies of neuronal activity performed using single unit electrode recordings (Barlow
and Hill, 1963; Maffei et al., 1973; Petersen et al., 1985). The MAE has also served as the subject
of previous fMRI investigations (Tootell et al., 1995; He et al., 1998; Culham et al., 1999; Culham
et al., 2000; Seiffert et al., 2003). A human homologue of macaque cortex area MT and MST,
fundamental in motion processing, has been identified and is commonly referred to as V5, MT+, or
the MT+ complex. Both areas respond to motion in the visual field with MT responding
preferentially to simple motion, while MST has a role in more 16 complex processing such as
object motion (Tanaka et al., 1993; Vanduffel et al., 2002) and the perception of heading and selfmotion (Andersen et al., 2000). When functionally
identified, this area typically is found on the lateral aspect of the occipital lobe, within orabout the
posterior reaches of the inferior temporal sulcus. Like macaque MT and MST, the area responds to
stimuli such as first- or second-order motion, and shares anatomical similarities with the macaque
region (Seiffert et al., 2003). Recently, further functional subdivision of this area has been possible
using fMRI, such that putative homologues of the specific areas MT and MST, dubbed pMT and
pMST in the human, can be identified based on their response to partial visual field stimuli
(Dukelow et al., 2001). The areas differ inthe important aspect that receptive fields of area MST are
much larger than those of area MT, reaching 30 degrees or more across the vertical visual meridian
into the contralateral visual field (Desimone and Ungerleider, 1986; Tanaka et al., 1986; Raiguel et
al., 1997). Tootell et al. (1995) demonstrated that human visual areas from V2 to MT+ have an

increasing activation during a MAE stimulus but did not differentiate between pMT and pMST.
Towards that end, this study combines the investigation of the MAE with the use of a hemifield
stimulus to attempt the analysis of the aftereffect in both subdivisions of the human MT+ complex.
Storage of the MAE is known to occur when a blank field ispresented between the moving and
static stimuli (Thompson and Wright, 1994; Verstraten
et al., 1994), and has been shown to prolong and postpone the functional activation during fMRI
investigation (Culham et al., 1999). A storage period is used in this study to aid in the separation of
activity resulting from real motion vs. perceived motion.
All data was collected on a 1.5 T Magnetom Vision clinical MRI scanner (Siemens, Erlangen,
Germany) with 25 mT/m (300 s rise time) gradients. A conventional birdcage head coil was
used for all measurements. Anatomical imaging was performed using a magnetization prepared
rapid acquisition gradient echo sequence (TR 14 ms, TE 4 ms, 30 degree flip angle) with a 256 x
256 mm field of view on a 256 x 256 matrix yielding a native in plane resolution of 1 x 1 mm. 150
slices were acquired sagittally at 1 mm spacing to preserve voxel isotropy. Functional imaging was
performed using a custom low-bandwidth (833 Hz/Px) echo-planar gradient echo sequence (TE 50
ms, TR 2 s effective, 90 degree flip angle) with a 256 x 256 mm field of view on a 64 x 64 matrix
yielding an in plane resolution of 4 x 4 mm. 16 slices were acquired coronally at 4 mm spacing
giving a final isotropic voxel resolution of 4 mm.
The advances in imaging technology, diagnostic imaging has become an indispensable tool in
medicine today. X-ray angiography (XRA), magnetic resonance angiography (MRA), magnetic
resonance imaging (MRI), computed tomography (CT), and other imaging modalities are heavily
used in clinical practice. Such images provide complementary information about the patient. While
increased size and volume in medical images required the automation of the diagnosis process, the
latest advances in computer technology and reduced costs have made it possible to develop such
systems. Blood vessel delineation on medical images forms an essential step in solving several
practical applications such as diagnosis of the vessels (e.g. stenosis or malformations) and
registration of patient images obtained at different times. Segmentation algorithms form the
essence of medical image applications such as radiological diagnostic systems, multimodal image
registration, creating anatomical atlases, visualization, and computer-aided surgery Vessel
segmentation algorithms are the key components of automated radiological diagnostic

systems. Segmentation methods vary depending on the imaging modality, application domain,
method being automatic or semi-automatic, and other specific factors. There is no single
segmentation method that can extract vasculature from every medical image modality. While some
methods employ pure intensity based pattern recognition techniques such as thresholding followed
by connected component
analysis , some other methods apply explicit vessel models to extract the vessel contours.
Depending on
the image quality and the general image artifacts such as noise, some segmentation methods may
require
image preprocessing prior to the segmentation algorithm. On the other hand, some methods apply
post-processing to overcome the problems arising from over segmentation. vessel segmentation
algorithms and techniques can be divided into six main categories, pattern recognition techniques,
model-based approaches, tracking-based approaches, artificial intelligence-based approaches,
neural network-based approaches, and miscellaneous tube-like object detection approaches. Pattern
recognition techniques are further divided into seven categories, multi-scale approaches, skeletonbased approaches, region growing approaches, ridge-based approaches, differential.
2.1 FUNCTIONAL IMAGING IN THE MACAQUE
While functional imaging in human subjects has become commonplace in the neurosciences, the
use of non-human primates, such as the macaque, is a novel development. It is an aspect of
functional imaging with many unique concerns and requirements. The maintenance of a colony of
macaques in proximity to an available research magnet prevents this technique from being
investigated currently at more than a few sites worldwide. Additionally, all the difficulties of
maintaining the animal and many10 of the concerns of electrophysiology in the macaque now must
be dealt with in the cramped, isolated and potentially dangerous bore of an imaging system.
However, this has not prevented investigation and development of methods for acquiring fMRI
data in both the awake and anesthetized monkey.
The earliest functional imaging experiments conducted in the macaque used positron emission
tomography (Takechi et al., 1994), but it was not long before the use of fMRI was proposed
(Westergaard et al., 1997) and ultimately accomplished (Dubowitz et al., 1998; Stefanacci et al.,

1998). Initial results in visual cortex demonstrated that fMRI and the BOLD contrast method of
imaging were viable tools for neurophysiologic investigation of the macaque. Early work was done
entirely in clinical and research magnets all with a common horizontal bore. This confined the
animal to be positioned in the sphinx position prone with the head raised a technique which is
still used today. Development of vertical bore magnets at a handful of facilities now enables natural
upright positioning of the animal.
The next step from observing the BOLD response evoked in visual cortex was comparing the
results to human data (Dubowitz et al., 2001b) and mapping these effects in an ordered fashion
such as retinotopy or motion selectivity. By presenting a stimulus in subregions of the visual field,
the retinotopy of macaque visual cortex was determined and found to be Imaging is an essential
aspect of medical science to visualize the anatomical structures of the human body. Several new
complex multidimensional digital images of physiological structures can be processed and
manipulated to help visualize hidden diagnostic features that are otherwise difficult or impossible
to identify using planar imaging methods. Magnetic Resonance Imaging (MRI) is a significant
technique for examining the human body, it helps to clarify and distinguish the neural architecture
of the human brain. It is unharmed method of obtaining images of the specific structure in the
human body. MRI scanner employs magnetic field and radio waves to generate exhaustive images
of the human brain, its data is most relevant in the studies of a head, specifically, for tracking the
size of brain tumor and other brain related problems. It helps for early detection of intracranial
tumors and precise estimation of tumor boundaries. Analytical, MRI scan can also been used to
assess the maturity of the central nervous system and diagnose malformations. The resonance is
also crucial for imaging of vascular changes. Using this method of diagnostic, imaging allows
obtaining information about aneurysms and accompanying symptoms. It is also helps in showing
seditious changes of the central nervous system and gives
accurate assessment of the degree of brain atrophy . The automatic classification of brain's MRI
can thus be used to identify regions having various brain diseases like cerebro-vascular, Alzheimer,
brain tumor, inflammatory, etc. The segmented MR images used in the medical diagnostic process
depends on a combination of two, often conflicting, requirements; i.e. the removal of the
unnecessary information present in the original MR images and the maintenance of the significant
details in the resulting segmented images. MRI segmentation methods are usually evaluate based

on their ability to differentiate: i) between cerebro-spinal fluid (CSF), white matter, and gray
matter, and ii) between normal tissues and abnormalities. Many segmentation techniques have been
proposed in the recent years, which were used for segmentation of brain tissues from MRI, are
classical pattern recognition methods, rule-based systems, image analysis methods, crisp and fuzzy
clustering procedures, feed-forward neural networks, fuzzy reasoning, geometric models to
determine lesion boundaries, connected component analysis, deterministic annealing, atlas based
methods and contouring approaches. Lots of researches have been performed for the segmentation
of MR brain images to detect and extract tumor regions from these images. Some of these related
works regarding the segmentation of brain tissues using clustering and other methods can be found.
2.2 Image segmentation
Image segmentation methods can be classified into three categories: edge-based methods, regionbased methods, and pixel-based methods. For Brain segmentation, two types of segmentation
techniques have been adopted in theliterature; i.e. region detection methods and boundary
detection methods. Mostly, the existing methods are dedicated for specific objects. The K-means
clustering technique is a pixel-based method, it is one of the most simple techniques, it's
complexity is relatively lower than other region-based or edge based methods. Furthermore, Kmeans clustering is suitable for biomedical image segmentation as the number of clusters is usually
known for images of particular regions of the human anatomy. Combined with the existing
methods and aiming to get better results, it is useful to take soft segmentation methods
into account. In soft segmentation, pixels are classified into different classes with various degrees
of uncertainty which are specified by functions. The larger the value of function for a specific
pixel, the larger the possibility that this pixel belongs to that cluster. The fuzzy C-means (FCM)
clustering algorithm is soft segmentation method, and has aroused comprehensive attention. There
have been many different families of fuzzy clustering algorithms proposed, for instance see .
2.3 Clustering
Clustering is the process of grouping feature vectors into classes in the selforganizing mode. Let
{x(q): q = 1,,Q} be a set of Q feature vectors. Each feature vector x(q)= (x1(q), , xN (q)) has
N components. The process of clustering is to assign the Q feature vectors into K clusters {c(k): k
= 1, , K}, usually by the minimum distance assignment principle. Choosing the representation of

cluster centers (or prototypes) is crucial to the clustering. Feature vectors that are farther away from
the cluster center should not have as much weight to those are close. These more distant feature
vectors are outliers usually caused by errors in one or more measurements or a deviation in the
processes that formed the object. The simplest weighting method is arithmetic averaging; it adds all
feature vectors in a cluster and takes the average as prototype. Because of its simplicity, it is still
widely used in the clustering initialization. The arithmetic averaging gives the central located
feature vectors the same weights as outliers. To lower the influence of the outliers, median vectors
are used in some proposed algorithms. To be more immune to outliers and more representatives,
the fuzzy weighted average is introduced to represent prototypes The earlier fuzzy clustering
algorithms; when the distance between the feature vector and the prototype is large, the weight is
small, and it is large when the distance is small. Using Gaussian functions to generate fuzzy
weights is the most natural way for clustering. It is not only immune to outliers but also provides
appropriate weighting for more centrally and densely located vectors. It is used in the fuzzy
clustering and fuzzy merging (FCFM) algorithm.

CHAPTER-3
RESULTS

CHAPTER-4

MATLAB

INTRODUCTION TO MATLAB
What Is MATLAB?
MATLAB is a high-performance language for technical computing. It
integrates computation, visualization, and programming in an easy-to-use
environment

where

problems

and

solutions

mathematical notation. Typical uses include

are expressed in

familiar

Math and computation

Algorithm development
Data acquisition
Modeling, simulation, and prototyping
Data analysis, exploration, and visualization
Scientific and engineering graphics
Application development, including graphical user interface building.
MATLAB is an interactive system whose basic data element is an array that does not require
dimensioning. This allows you to solve many technical computing problems, especially those with
matrix and vector formulations, in a fraction of the time it would take to write a program in a scalar
non interactive language such as C or FORTRAN.
The name MATLAB stands for matrix laboratory. MATLAB was originally
written to provide easy access to matrix software developed by the LINPACK
and EISPACK projects. Today, MATLAB engines incorporate the LAPACK and
BLAS libraries, embedding the state of the art in software for matrix
computation.
MATLAB has evolved over a period of years with input from many users.
In university environments, it is the standard instructional tool for introductory
and advanced courses in mathematics, engineering, and science. In industry,
MATLAB is the tool of choice for high-productivity research, development, and
analysis.
MATLAB features a family of add-on application-specific solutions called
toolboxes. Very important to most users of MATLAB, toolboxes allow you to
learn

and

apply

specialized

technology.

Toolboxes

are

comprehensive

collections of MATLAB functions (M-files) that extend the MATLAB environment

to solve particular classes of problems. Areas in which toolboxes are available

include signal processing, control systems, neural networks, fuzzy logic,

wavelets, simulation, and many others.
The MATLAB System:
The MATLAB system consists of five main parts:
Development Environment:
This is the set of tools and facilities that help you use MATLAB functions
and files. Many of these tools are graphical user interfaces. It includes the
MATLAB desktop and Command Window, a command history, an editor and
debugger, and browsers for viewing help, the workspace, files, and the search
path.
The MATLAB Mathematical Function:
This is a vast collection of computational algorithms ranging from
elementary functions like sum, sine, cosine, and complex arithmetic, to more
sophisticated functions like matrix inverse, matrix eigen values, Bessel
functions, and fast Fourier transforms.
The MATLAB Language:
This is a high-level matrix/array language with control flow statements,
functions, data structures, input/output, and object-oriented programming
features. It allows both "programming in the small" to rapidly create quick and
dirty throw-away programs, and "programming in the large" to create
complete large and complex application programs.
Graphics:
MATLAB has extensive facilities for displaying vectors and matrices as
graphs, as well as annotating and printing these graphs. It includes high-level
functions for two-dimensional and three-dimensional data visualization, image
processing, animation, and presentation graphics. It also includes low-level
functions that allow you to fully customize the appearance of graphics as well
as to build complete graphical user interfaces on your MATLAB applications.
The MATLAB Application Program Interface (API):
This is a library that allows you to write C and Fortran programs that
interact with MATLAB. It includes facilities for calling routines from MATLAB

(dynamic linking), calling MATLAB as a computational engine, and for reading

and writing MAT-files.
MATLAB WORKING ENVIRONMENT:
MATLAB DESKTOP:Matlab Desktop is the main Matlab application window. The desktop contains five sub
windows, the command window, the workspace browser, the current directory window, the
command history window, and one or more figure windows, which are shown only when the user
displays a graphic.
The command window is where the user types MATLAB commands and expressions at the
prompt (>>) and where the output of those commands is displayed. MATLAB defines the
workspace as the set of variables that the user creates in a work session. The workspace browser
shows these variables and some information about them. Double clicking on a variable in the
workspace browser launches the Array Editor, which can be used to obtain information and income
instances edit certain properties of the variable.
The current Directory tab above the workspace tab shows the contents of the current
directory, whose path is shown in the current directory window. For example, in the windows
operating system the path might be as follows: C:\MATLAB\Work, indicating that directory
work is a subdirectory of the main directory MATLAB; WHICH IS INSTALLED IN DRIVE
C. clicking on the arrow in the current directory window shows a list of recently used paths.
Clicking on the button to the right of the window allows the user to change the current directory.
MATLAB uses a search path to find M-files and other MATLAB related files, which are
organize in directories in the computer file system. Any file run in MATLAB must reside in the
current directory or in a directory that is on search path. By default, the files supplied with
MATLAB and math works toolboxes are included in the search path. The easiest way to see which
directories are on the search path. The easiest way to see which directories are soon the search path,
or to add or modify a search path, is to select set path from the File menu the desktop, and then use
the set path dialog box. It is good practice to add any commonly used directories to the search path
to avoid repeatedly having the change the current directory.

The Command History Window contains a record of the commands a user has entered in the
command window, including both current and previous MATLAB sessions. Previously entered
MATLAB commands can be selected and re-executed from the command history window by right
clicking on a command or sequence of commands. This action launches a menu from which to
select various options in addition to executing the commands. This is useful to select various
options in addition to executing the commands. This is a useful feature when experimenting with
various commands in a work session.
Using the MATLAB Editor to create M-Files:
The MATLAB editor is both a text editor specialized for creating M-files and a graphical
MATLAB debugger. The editor can appear in a window by itself, or it can be a sub window in the
desktop. M-files are denoted by the extension .m, as in pixelup.m. The MATLAB editor window
has numerous pull-down menus for tasks such as saving, viewing, and debugging files. Because it
performs some simple checks and also uses color to differentiate between various elements of code,
this text editor is recommended as the tool of choice for writing and editing M-functions. To open
the editor , type edit at the prompt opens the M-file filename.m in an editor window, ready for
editing. As noted earlier, the file must be in the current directory, or in a directory in the search
path.
Getting Help:
The principal way to get help online is to use the MATLAB help browser, opened as a
separate window either by clicking on the question mark symbol (?) on the desktop toolbar, or by
typing help browser at the prompt in the command window. The help Browser is a web browser
integrated into the MATLAB desktop that displays a Hypertext Markup Language(HTML)
documents. The Help Browser consists of two panes, the help navigator pane, used to find
information, and the display pane, used to view the information. Self-explanatory tabs other than
navigator pane are used to perform a search.

CHAPTER-5

SIGNAL PROCESSING

What Is the Signal Processing Toolbox?

The Signal Processing Toolbox is a collection of tools built on the

MATLAB numeric computing environment. The toolbox supports a
wide range of signal processing operations, from waveform generation to
filter design and implementation, parametric modeling, and spectral
analysis. The toolbox provides two categories of tools:
Command line functions in the following categories:

Analog and digital filter analysis

Digital filter implementation

FIR and IIR digital filter design

Analog filter design

Filter discretization

Spectral Windows Transforms

Cepstral analysis

Statistical signal processing and spectral analysis

Parametric modeling

Linear Prediction

Waveform generation

A suite of interactive graphical user interfaces for

Filter design and analysis

Window design and analysis

Signal plotting and analysis

Spectral analysis

Filtering signals

Related Products
The MathWorks provides several products that are especially relevant to
the kinds of tasks you can perform with the Signal Processing Toolbox.
For more information about any of these products, see either

The online documentation for that product if it is installed or if you are reading the
documentation from the CD

The MathWorks Web site, at http://www.mathworks.com; see the products section

The toolboxes listed below all include functions that extend the capabilities of
MATLAB. The blocksets all include blocks that extend the capabilities of
Simulink.

Table 0-1:
Product

Description

Communications
Blockset

Design and simulate communications systems

Communications Toolbox Design and analyze communication systems

Data Acquisition Toolbox Acquire and send out data from plug-in data
acquisition boards
Database Toolbox

Exchange data with relational databases

DSP Blockset

Design and simulate DSP systems

Embedded Target for the Deploy and validate DSP designs on Texas
TI
TMS320C6000
DSP
Platform

Instruments C6000 digital signal processors

Fuzzy Logic Toolbox

Design and simulate fuzzy logic systems

Image Processing
Toolbox

Perform image processing, analysis, and

algorithm development

Table 0-1:
Product

Description

MATLAB Link for Code Use MATLAB with RTDX-enabled Texas

Composer Studio
Instruments digital signal processors
Development Tools
Neural Network Toolbox Design and simulate neural networks
Optimization Toolbox

Solve standard and large-scale optimization

problems

Simulink

Design and simulate continuous- and

discrete-time systems

Statistics Toolbox

Apply statistical algorithms and probability

models

System Identification
Toolbox

Create linear dynamic models from measured

input-output data

Wavelet Toolbox

Analyze, compress, and denoise signals and

images using wavelet techniques

All
Toolb
ox
Users
Use the Function
Reference

for

locating
information

on

specific functions.
Reference
descriptions
include a synopsis
of the functions
syntax, as well as a
complete
explanation
options
operations.

of
and
Many

reference
descriptions
include
examples,

also
helpful
a

description of the
functions
algorithm,

and

references

to

additional reading
material.

Use this manual in

conjunction
the

with

software

to

about

the

learn

powerful
that

features
MATLAB

provides.
chapter

Each
provides

numerous
examples

that

apply the toolbox

to

representative

signal

processing

tasks.
Some

examples

use the MATLAB

random

number

generation function
randn.
In these cases, to
duplicate
results

the
in

example, type
randn
('state
',0)

the

befor
e
runni
ng the
exam
ple.
Instal
ling
the
Signa
l
Proce
ssing
Toolb
ox
To determine
if the Signal
Processing
Toolbox

is

installed

on

your

system,

type

this

command

at

the MATLAB
prompt.
When

you

enter

this

command,

MATLAB
displays
information
about

the

version

of

MATLAB
you

are

running,
including
list

of

all

toolboxes
installed
your

on

system

and

their

version
numbers.
For
information
about
installing the
toolbox,

see

the MATLAB
Installation
documentatio
n

for

platform.
Tech
nical
Conv

your

entio
ns
This manual and
the

Signal

Processing
Toolbox functions
use the following
technical notations.
Nyquist frequency

One-half the sampling frequency. Some

toolbox functions normalize this value to 1.

x(1)

The first element of a data sequence or

filter, corresponding to zero lag.

or w

Analog frequency in radians per second.

or w

Digital frequency in radians per sample.

Digital frequency in hertz.

The interval from x to y, including x but

Typo
grap
hical
Conv
entio

[x, y)

not
including y.

...

Ellipses in the argument list for a given

syntax on a function reference page
indicate all possible argument lists for that
function appearing prior to the given
syntax are valid.

ns
This

manual

uses some or
all of these
conventions.

Exampl
Item

Convention

Example code

Monospace font

To assign the value 5 to A,

enter
A=5

Function names, syntax,

Monospace font
filenames, directory/folder
names, and user input

The cos function finds the

cosine of each array element.
Syntax line example is
MLGetVar ML_var_name

Buttons and keys

Boldface with book title caps

Press the Enter key.

Literal strings (in syntax

descriptions in reference
chapters)

Monospace bold for literals

f = freqspace(n,'whole')

Mathematical

Italics for variables

This vector represents the

polynomial p = x2 + 2x + 3.

expressions
Standard text font for
functions, operators, and
constants
MATLAB output

Monospace font

Menu and dialog box titles Boldface with book title caps

MATLAB responds with

Choose the File Options

menu.

New terms and for

emphasis
Signa
l
Proce
ssing
Basic
s

The
follo
wing
topics
descri
be
how
to
begin
using
MAT
LAB
and
the
Signa
l
Proce
ssing
Toolb
ox for

Italics

An array is an ordered
collection of information.

your
signal
proce
ssing
applic
ations
. It is
assum
ed
that
you
have
basic
knowl
edge
and
under
standi
ng of
signal
s and
syste
ms,
includ
ing
such
topics
as
filter
and
linear

syste
m
theor
y and
basic
Fouri
er
analy
sis.

Signa
l
Proce
ssing
Toolb
ox
Centr
al
Major
featur
es
and
key
areas
of the
toolb
ox
Featu
res (p.

1-2)
Repre
sentin
g
Signa
ls (p.
1-4)
Vecto
r and
matri
x
repres
tation
of
signal
s
Wave
form
Gener
ation:
Time
Vecto
rs
Perio
dic
and
aperio
dic
wavef

orms,
seque
nces
(impu
lse,
and
Sinus
oids
(p. 16)
step,
ramp)
,
multi
chann
el
signal
s,
pulse
trains,
sinc
and
Dirichlet functions
Worki
ng
with
Data
(p. 1-

13)
Meth
ods of
inputt
ing
and
impor
ting
data
Filter
Imple
menta
tion
and
Analy
sis
Filteri
ng
discre
te
signal
s
(p. 114)
The
filter
Funct
ion

(p. 117)
Math
emeti
cal
infor
matio
n

on

the
filter
functi
on
Other
Funct
ions
for
Filteri
ng (p.
1-19)
Other
types
of
filter
functi
ons
availa
ble in
the
toolb
ox

Impul
se
Respo
nse
(p. 123)
Impul
se
respo
nse
detail
s
Frequ
ency
Respo
nse
(p. 125)
Frequ
ency
respo
nse
detail
s
ZeroPole
Analy
sis (p.

1-33)
Zplane
poles
and
zeros
Linea
r
Syste
m
Mode
ls (p.
1-36)
Discr
etetime
and
contin
uoustime
linear
syste
m
model
s
and transformations
Discr
ete

Fouri
er
Trans
form
(p. 148)
DFT
detail
s
Select
ed
Biblio
graph
y (p.
1-51)
Sourc
es for
additi
onal
infor
matio
n

Signal
Toolbox

Processing
Central

Features
T
he
Si
gn
al
Pr
oc
es
si
ng
T
oo
lb
ox
fu
nc
ti
on
s
ar

e
al
go
rit
h
m
s,
ex
pr
es
se
d
m
os
tl
y
in
M
fil
es
,
th
at
i
m
pl
e
m
en
ta

va
ri
et
y
of
si
gn
al
pr
oc
es
si
ng
ta
sk
s.
T
he
se
to
ol
bo
x
fu
nc
ti
on
s
ar
e
a

sp
ec
ial
iz
ed
ex
te
ns
io
n
of
th
e
M
A
T
L
A
B
co
m
pu
tat
io
na
l
an
d
gr
ap
hi

ca
l
en
vi
ro
n
m
en
t.

Fil
ter
ing
an
d
FF
Ts
Tw
o
of
the
mo
st
im
por
tan
t
fun
cti

on
s
for
sig
nal
pro
ces
sin
g
are
not
in
the
Si
gn
al
Pr
oc
ess
ing
To
olb
ox
at
all,
but
are
bui
ltin
M

AT
LA
B
fun
cti
on
s:

f
il
t
e
r
a
p
p
li
e
s
a
d
i
g
it
a
l
f
il
t
e
r

t
o
a
d
a
t
a
s
e
q
u
e
n
c
e
.

f
f
t
c
a
l
c
u
l
a
t
e
s
t

h
e
d
i
s
c
r
e
t
e
F
o
u
r
i
e
r
t
r
a
n
s
f
o
r
m
o
f
a
s

e
q
u
e
n
c
e
.
Th
e
op
era
tio
ns
the
se
fun
cti
on
s
per
for
m
are
the
ma
in
co
mp
uta

tio
nal
wo
rkh
ors
es
of
cla
ssi
cal
sig
nal
pro
ces
sin
g.
Bo
th
are
des
cri
be
d
in
thi
s
ch
apt
er.
Th
e

Si
gn
al
Pr
oc
ess
ing
To
olb
ox
use
s
ma
ny
oth
er
sta
nd
ard
M
AT
LA
B
fun
cti
on
s
an
d
lan
gu

ag
e
fea
tur
es,
inc
lud
ing
pol
yn
om
ial
roo
t
fin
din
g,
co
mp
lex
arit
hm
eti
c,
ma
tri
x
inv
ers
ion
an

d
ma
nip
ula
tio
n,
an
d
gra
phi
cs
too
ls.
Sig
nal
s
an
d
Sy
ste
ms
Th
e
bas
ic
ent
itie
s
tha

t
too
lbo
x
fun
cti
on
s
wo
rk
wit
h
are
sig
nal
s
an
d
sys
te
ms
.
Th
e
fun
cti
on
s
em
ph
asi

ze
dig
ital
, or
dis
cre
te,
sig
nal
s
an
d
filt
ers
, as
op
po
sed
to
an
alo
g,
or
co
nti
nu
ou
s,
sig
nal
s.

Th
e
pri
nci
pal
filt
er
typ
e
the
too
lbo
x
su
pp
ort
s is
the
lin
ear
,
tim
einv
ari
ant
dig
ital
filt
er
wit

ha
sin
gle
inp
ut
an
da
sin
gle
out
put
.
Yo
u
ca
n
rep
res
ent
lin
ear
tim
einv
ari
ant
sys
te
ms
usi
ng

on
e
of
sev
era
l
mo
del
s
(su
ch
as
tra
nsf
er
fun
cti
on,
sta
tespa
ce,
zer
opol
egai
n,
an
d
sec

on
dord
er
sec
tio
n)
an
d
co
nv
ert
bet
we
en
rep
res
ent
ati
on
s.

Ke
y
Ar
eas
:
Fil
ter
De

sig
n
an
d
Sp
ect
ral
An
aly
sis
In
ad
diti
on
to
its
cor
e
fun
cti
ons
,
the
too
lbo
x
pro
vid
es
ric

h,
cus
to
mi
zab
le
sup
por
t
for
the
ke
y
are
as
of
filt
er
des
ign
an
d
spe
ctr
al
ana
lys
is.
It
is
eas

y
to
im
ple
me
nt
a
des
ign
tec
hni
qu
e
tha
t
sui
ts
yo
ur
ap
pli
cat
ion
,
des
ign
dig
ital
filt
ers
dir

ect
ly,
or
cre
ate
ana
log
pro
tot
yp
es
an
d
dis
cre
tiz
e
the
m.
To
olb
ox
fun
cti
ons
als
o
esti
ma
te
po

we
r
spe
ctr
al
de
nsi
ty
an
d
cro
ss
spe
ctr
al
de
nsi
ty,
usi
ng
eit
her
par
am
etri
c
or
no
np
ara
me

tric
tec
hni
qu
es.
Ch
apt
er
2,
Fi
lter
De
sig
n
an
d
Im
ple
me
nta
tio
n
an
d
Ch
apt
er
3,
St
atis
tic

al
Sig
nal
Pro
ces
sin
g,
res
pec
tiv
ely
det
ail
too
lbo
x
fun
cti
ons
for
filt
er
des
ign
an
d
spe
ctr
al
ana
lys

is.

Some filter design and spectral analysis functions included in the toolbox are

Computation and graphical display of frequency response

System identification

Generating signals

Discrete cosine, chirp-z, and Hilbert transforms

Lattice filters

Resampling

Time-frequency analysis

Basic communication systems simulation

Interactive Tools
The power of the Signal Processing Toolbox is greatly enhanced by its easy-to-use
interactive tools. SPTool provides a rich graphical environment for signal viewing, filter
design, and spectral analysis. The Filter Design and Analysis Tool (FDATool) provides a
more comprehensive collection of features for addressing the problem of filter design. The
FDATool also offers seamless access to the additional filter design methods and
quantization features of the Filter Design Toolbox when that product is installed. The
Window Design and Analysis Tool (WinTool) provides an environment for designing and

comparing spectral windows.

Extensibility
Perhaps the most important feature of the MATLAB environment is that it is extensible.
MATLAB lets you create your own M-files to meet numeric computation needs for
research, design, or engineering of signal processing systems. Simply copy the M-files
provided with the Signal Processing Toolbox and modify them as needed, or create new
functions to expand the functionality of the toolbox.
Representing Signals
The central data construct in MATLAB is the numeric array, an ordered
collection of real or complex numeric data with two or more dimensions.
The basic data objects of signal processing (one-dimensional signals or
sequences, multichannel signals, and two-dimensional signals) are all
naturally suited to array representation.

Vector Representation
MATLAB represents ordinary one-dimensional sampled data signals, or
sequences, as vectors. Vectors are 1-by-n or n-by-1 arrays, where n is the
number of samples in the sequence. One way to introduce a sequence
into MATLAB is to enter it as a list of elements at the command prompt.
The statement
x = [4 3 7 -9 1]
creates a simple five-element real sequence in a row vector.
Transposition turns the sequence into a column vector

x = x'
resulting in
x =4 3 7 -9 1
Column orientation is preferable for single channel signals because it
extends naturally to the multichannel case. For multichannel data, each
column of a matrix represents one channel. Each row of such a matrix then
corresponds to a sample point. A three-channel signal that consists of x, 2x,
and x/ is
y = [x 2*x x/pi]

CHAPTER-6
CONCLUSION

Electroencephalography (EEG) signals provide valuable information to study the brain function
and neurobiological disorders. Digital signal processing gives the important tools for the analysis
of EEG signals. The EEG signal was collected from the standard data base. These EEG signals
were not distinguishable with human eyes. We used the signal processing tools to distinct them
and Provide the status of the individual. DWT and AR method are used for the feature extraction.
Performance of proposed mixture of features is compared to the DWT features and AR features
using ANN classifier. Experiment result shows that the mixture of features classify the EEG
signals more accurately. Therefore, the two feature extraction schemes are partially
complementary in nature whose combination helps to represent and characterize the signal more
effectively than using any of them alone. Performance of proposed Two-level committee neural
network shows that the accuracy increases from 960.04 percent to 97.29 percent.

CHAPTER-7
REFERENCES

[1] E. D. ..U beyli, Statistics over features: EEG signals analysis, Computers in Biology
and Medicine, vol. 39, Issue 8, pp. 733-741, 2009.
[2] Ali B. Usakli, Improvement of EEG Signal Acquisition: An Electrical Aspect for State
of the Art of Front End, Computational Intelligence and Neuroscience, vol. 2010, 2009.
[3] E. D. ..U beyli, Measuring saliency of features representing EEG signals using signaltonoise ratios, Expert Systems with Applications, vol. 36, pp. 501-509, 2009.
[4] Saeid Sanei and J.A. Chambers, EEG Signal Processing, John Wiley and Sons Ltd, England,
2007
[5] H.L. Attwood and W. A. MacKay, Essentials of Neurophysiology, B. C. Decker, Hamilton,
Canada, 1989
[6] M. Teplan,Fundamentals of EEG measurements,Measmt Sci. Rev., vol. 2, no. 2, 2002.
[7] http://www.everydayhealth.com/health-center/electroencephalogram-eeg-results.aspx
[8] H. Adeli a, Z. Zhou b, and N. Dadmehr c, Analysis of EEG records in an epileptic
patient using wavelet transform, Journal of Neuroscience Methods, vol. 123, pp. 69-87,
2003.
[9] M. Unser and A. Aldroubi, A review of wavelets in biomedical applications, Proceedings
of the IEEE, vol. 84, Issue 4, pp. 626-638, 1996.
[10] N. Hazarika, J. Z . Chen, and A. Chung Tsoi, Classification of EEG signals using the
wavelet transform, Signal Processing, vol. 59, pp. 61-72, 1997.
[11] E. D. ..U beyli, Wavelet/mixture of experts network structure for EEG signals
classification, Expert Systems with Applications, vol. 34, pp. 1954-1962, 2008.
[12] E. D. ..U beyli, Combined neural network model employing wavelet coefficients for EEG
signals classification, Digital Signal Processing, vol. 19, pp. 297-308, 2009.
[13] E. D. ..U beyli, Least squares support vector machine employing model-based methods
coefficients for analysis of EEG signals, Expert Systems with Applications, vol. 37, pp. 233-239,
2010.
[14] Steven M. Kay and S. Lawrence Marple, Spectrum Analysis-A Modern Perspective,
Proceedings of the IEEE, vol. 69, no. 11, 1981.
[15] P. Stoica and R. L. Moses, Introduction to spectral Analysis, Upper Saddle River, NJ:
Prentice-Hall, 1997.

[16] J. P. Burg, Maximum entropy spectral analysis, 37th Meeting Society of Exploration
Geophysicists, Oklahoma City, OKla, 1967.
[17] http://www.dtic.upf.edu/ xserra/cursos/TDP/referencies/Park-DWT.pdf.
[18] M. Unser and A. Aldroubi, A review of wavelets in biomedical applications, Proceedings
of the IEEE, vol. 84, no. 4, pp. 626638, 1996.
[19] S. Haykin, Neural Networks: A Comprehensive Foundation, New York: Macmillan,
1994.
[20] A. J. C. Sharkey, On combining artificial neural nets, Connection Science, vol. 8, no.
3/4, pp. 299-314, 1996.
[21] N. P. Reddy and O. A. Buch, Speaker verification using committee neural networks,
Computer Methods and Programs in Biomedicine, vol. 72, Issue 2, pp. 109-115, 2003.
[22] S. Ari and G. Saha, In search of an optimization technique for Artificial Neural Network
to classify abnormal heart sounds, Applied Soft Computing, vol. 9, Issue 1, pp. 109-115, 2009.
[23] M. L. Raymer, W. F. Punch, E. D. Goodman, L. A. Kuhn, and A. K. Jain, Dimensionality
Reduction Using Genetic Algorithms, IEEE transactions on evolutionary computation, vol. 4,
no. 2, 2000.