Magnetic Resonance Imaging (MRI)

Republic of Iraq
Ministry of Higher Education And Scientific Research

University of Technology
Applied Science Department
Applied Physics Branch
2010 - 2011

A project Submitted to Department of Applied
Science, Applied Physics Branch
University of Technology
In partial Fulfillment of the Requirement for the
Degree of the B.Sc
In Applied physics
Percent By:
1. Ali Jafar Hadi
2. Mohammed waleed moaid

Supervised by :

Dr. AbdulHadi Kadhim Al-Ogaily

2011

Magnetic Resonance Imaging
( MRI )

2011

Magnetic Resonance Imaging

(
(
)
)

( ( 8 88 8 ) )

Dedication
i

Dedication

To my Mother & father
To my Brothers & Sisters
To my friends

Acknowledgment
ii

Acknowledgment

All the praises and thanks are to Allah, the most beneficent, and the
most merciful and his continuous help over all the period of my life.
I would like to express my deep appreciation and gratitude to my
supervisor , AbdulHadi Kadhim Al-Ogaily who gave me great
support and valuable advice though the progress of this project.
I would like to thank to the Head of the Applied Science department
and the stuff and sub-stuff at department of the Applied Science
specially the Examining Committees Certification, Dr. Uday Muhsin
Nayef and assist.lecturer Abbas Fadhil Sabbar for their continuous
support and help.
Finally, I would like to thank all kind, helpful and lovely people who
helped me in the project.

Ali & Mohammad

iii
Contents

Contents

Dedication
Acknowledgment
Contents
i
ii
iii
CHPTER 1 Introduction 1
1.1 History of MRI
1.2 NMR phenomenon
1.3 Aim of this work

2
3
5

CHPTER 2 The Magnetic Resonance Signal 6
2.1 Nuclear Magnetism
2.2 Precession
2.3 Polarization and Coherence
2.4 Radio Frequency Pulses
2.4.1 90 Pulse Excitation and FID
2.4.2 180 Pulse Refocusing
2.5 Relaxation and Image Contrast
2.5.1 Overview
2.5.2 The Bloch Equations
2.5.2.a Longitudinal Bloch Equation
2.5.2.b Transverse Bloch Equation
2.6 T
2
- weighted Imaging
2.7 T
2
*
- weighted Imaging
2.8 T
1
-weighted Imaging
2.9 Summary

7
8
9
10
11
14
17
17
18
18
19
20
23
23
27

CHPTER 3 Brain Tumors Identification in Magnetic Resonance Images 29
3.1 Abstract
3.2 Magnetic Resonance Imaging
3.3 overview about imaging
3.4 Brain Tissues
3.5 Image Analysis System
30
30
32
33
34
iv
Contents

3.6 Image Segmentation Methods
3.7 Segmentation Methods
3.7.1 Computer Aided Segmentation Method
3.7.2 Manual Segmentation Methods
3.8 Texture Feature Extraction
3.8.1 Statistical Features for Texture Analysis
3.8.1.a First-Order Statistics
3.8.1.b The mean
3.8.1.c The standard deviation
3.8.1.d The entropy
3.9 Methodology
3.9.1 Unsupervised Segmentation Method

36
36
36
37
38
39
39
40
40
40
40
40

CHPTER 4 Result and Discussion 43
4.1 Result and Discussion
4.2 Conclusion

44
45
References 47
1

C H A P T E R
1

Introduction
RI has revolutionized diagnostic
dddddimaging in medical science. An
important advantage of diagnostic MRI as
compared to CT is that the former does not
use ionizing radiation , The development of
magnetic resonance imaging (MRI) began with
discoveries in nuclear magnetic resonance
(NMR) in the early 1900s , Magnetic
Resonance Imaging (MRI) has become an
important noninvasive imaging modality. MRI
has found a number of applications in the
fields of biology, engineering, and material
science. Because it provides unique contrast
between soft tissues .
M

2
CHPTER 1 : Introduction
CHPTER ONE :
Introduction

1.1 History of MRI

Since its first implementation by Lauterbur
(1)
, MRI has found a number of
applications in the fields of biology, engineering, and material science. Because it
provides unique contrast between soft tissues (which is generally superior to that of
CT) and high spatial resolution, MRI has revolutionized diagnostic imaging in
medical science. MRI is based on the phenomenon of Nuclear Magnetic Resonance
(NMR), independently discovered by Bloch
(2)
and Purcell
(3)
.
The development of magnetic resonance imaging (MRI) began with discoveries in
nuclear magnetic resonance (NMR) in the early 1900s. At this time, scientists had
just started to figure out the structure of the atom and the nature of visible light and
ultraviolet radiation emitted by certain substances. The magnetic properties of an
atom's nucleus, which is the basis for NMR, were demonstrated by Wolfgang Pauli
in 1924. The first basic NMR device was developed by I. I. Rabi in 1938. This
device was able to provide data related to the magnetic properties of certain
substances. However, it suffered from two major limitations. Firstly, the device
could analyze only gaseous materials, and secondly, it could only provide indirect
measurements of these materials. These limitations were overcome in 1945, when
two groups of scientists led by Felix Bloch and Edward Purcell independently
developed improved NMR devices. These new devices proved useful to many
researchers, allowing them to collect data on many different types of systems. After
further technological improvements, scientists were able to use this technology to
investigate biological tissues in the mid 1960s.
(1) Lauterbur was a 1962 graduate of the University of Pittsburgh and credits the idea of the MRI to a brainstorm one day at
a suburban Pittsburgh Big Boy, with the MRI's first model scribbled on a table napkin. The further research that led to
the Nobel Prize was performed at Stony Brook University in the 1970s.
(2) Felix Bloch was a Swiss physicist who was awarded the 1952 Nobel Prize for "their development of new ways and
methods for nuclear magnetic precision measurements
(3) Edward Mills Purcell was an American physicist who shared the 1952 Nobel Prize for Physics for his independent
discovery (published 1946) of nuclear magnetic resonance in liquids and in solids. Nuclear magnetic resonance (NMR)
has become widely used to study the molecular structure of pure materials and the composition of mixtures.

3
An important advantage of diagnostic MRI as compared to CT is that the former
does not use ionizing radiation, the use of NMR in medicine soon followed. The
earliest experiments showed that NMR could distinguish between normal and
cancerous tissue. Later experiments showed that many different body tissues could
be distinguished by NMR scans. In 1973, an imaging method using NMR data and
computer calculations of tomography was developed. It provided the first magnetic
resonance image (MRI). This method was consequently used to examine a mouse
and, while the testing time required was more than an hour, an image of the internal
organs of the mouse resulted. Human imaging followed a few years later. Various
technological improvements have been made since to reduce the scanning time
required and improve the resolution of the images. Most notable improvements
have been made in the three-dimensional application of MRI.
Although the potential of NMR spectroscopy was recognized almost
immediately and its development started soon after the discovery of NMR, it took
more than two decades to implement NMR-based imaging. Even after the
pioneering work by Lauterbur and the development of basic imaging techniques by
Kumar and Mansfield, several more years were required to design and develop
imaging hardware at the level necessary to produce high-quality diagnostic images
of the human body. Despite its relatively slow beginning, MRI has become an
indispensable diagnostic tool since the early 1980s.

1.2 NMR phenomenon
The foundation of the NMR phenomenon is the interaction between an
external magnetic field and nuclei which have a nonzero magnetic moment.
According to the classical theory of electromagnetism, the motion of individual
nuclear moments in a static magnetic field B
0
is a precession about B
0
at an angular
frequency
0
, known as the Larmor frequency, which is proportional to the strength
of the magnetic field. Another well-known fact is that the energy of interaction with
B
o
depends on the direction of nuclear moments in such a way that the minimum
energy corresponds to the state in which the moments are parallel to B
o
. As a result,
in thermal equilibrium the majority of nuclear magnetic moments are aligned along
the external field. The alignment of magnetic moments gives rise to nonzero
magnetization in macroscopic samples of solids, liquids, or gases containing a large
number of nuclei (e.g., ~10
23
cm
-3
of hydrogen nuclei in water). The NMR
phenomenon is observed when a macroscopic sample in a static magnetic field is

4
irradiated by an oscillating magnetic field of frequency that equals the frequency
of precession,
o
. The NMR phenomenon can be best explained by using the
arguments put forward by Bloch . Suppose that a macroscopic sample is placed
between the poles of a magnet that produces a static magnetic field, B
o
. Under the
influence of B
o
the sample becomes magnetized. In thermal equilibrium the nuclear
magnetization in the sample can be expressed as
M= B
o
.. (1(
Where is the nuclear susceptibility. Suppose that the sample experiences an
oscillating magnetic field B1 produced by an alternating current in a coil
encompassing the sample, and assume that the oscillating field is perpendicular to
B
o
.It can be shown that at resonance, achieved when the frequency of B1 equals
o

, even a weak oscillating field applied as a pulse can rotate the magnetization in the
sample and place it in the transverse (i.e., perpendicular to B
o
) plane. After the
excitation pulse ends, the transverse magnetization in the sample processes about
B
o
. During the precession the transverse magnetization decays because of nuclear
interactions and no uniformity of B
o
. From Faraday's law of induction it follows
that the time-varying magnetization induces voltage in the coil. The induced
voltage, which identifies the presence of the transverse magnetization in the sample,
can be Fourier transformed to obtain a NMR spectrum. For many applications of
NMR spectroscopy it is critically important that the observed signal includes
contributions from nuclei in different chemical environments (e.g., hydrogen nuclei
in water and fat). Such nuclei normally have slightly different frequencies of
precession depending on the chemical composition of a sample. NMR spectra can
be used to identify chemically different populations of nuclei and to determine their
relative amounts in a sample. We are indebted to F. Bloch for a set of widely used
phenomenological equations that describe the dynamics of nuclear magnetization in
media. In particular, Bloch has predicted that thermal fluctuations would cause
exponential relaxation of the longitudinal (i.e., parallel to B
0
) magnetization, M
lon
,
to the equilibrium state in equation (1):
M
lon
- Bo exp (
1
) (2)
Where T
1
is a characteristic relaxation time. On the other hand, Bloch has suggested
that magnetic interactions between neighboring nuclei cause exponential decay of
the transverse (i.e., perpendicular to B
0
) magnetization, M
tr
, with a characteristic
time constant T
2
:
M
tr
exp (-t/T2).. (3)

5
Although the Bloch model has important limitations (e.g., it fails to describe
NMR in solids), it has proven to work well for liquids and biological systems in
general. Throughout the Bloch equations are used as the main tool for analysis of
the NMR phenomenon and associated effects important for MRI.
The pioneering work of Lauterbur has demonstrated that the NMR signal acquired
in the presence of external magnetic field gradients can be used to obtain MR
images of the transverse magnetization in the object. The basic effect that makes
MRI possible is that in the presence of magnetic field gradients the frequency of
precession, w0, becomes spatially dependent. The observed NMR signal is a sum of
many signals produced by nuclei at different locations in the sample.
Hydrogen nuclei (
1
H) appear to be the best target for in vivo MRI. They have
the following advantages: a) among all nuclei present in tissues hydrogen nuclei
produce the greatest NMR signal; b)
1
H MRI in vivo achieves excellent contrast
between different tissues. The remarkable feature of
1
H MRI is that it provides a
variety of ways to manipulate image contrast. Since the inception of MRI it has
been apparent that the major source of contrast in MR images is the dependence of
T
1
and T
2
relaxation times, and proton density on the biochemical composition of
different tissues. The studies of Damadian and other investigators have
demonstrated that in certain instances malignant tissues have longer T
1
and T
2

relaxation times than normal tissues. Because of the exponential dependence on the
tissue relaxation times, the NMR signal is extremely sensitive to small variations in
T
1
and T
2
. Moreover, by adjusting acquisition parameters, the dependence of the
signal on the relaxation Introduction 5 times can easily be varied in order to
improve image contrast. [1]

1.3 Aim of this work
The aim of this work is to produce a method or algorithm for analyzing MRI in
easiest way. The method or algorithm is based on semi-automatic unsupervised
segmentation method and then identify the tumor region using 1
st
order statistical
features, The method has several steps, starting from preliminary segmentation to
getting new MR segment image to improve the result segmentation, by assigning
each classified point to its nearest mean value, and then isolate the tumor region. An
important task of this work, the program in visual basic language have been used
.this paper has given the results visually, since the result in visual has more power
than numerical result.
6

C H A P T E R
2

The Magnetic
Resonance Signal
he hydrogen atom is most commonly used
fffffff when performing MRI scanning, since the
hydrogen nucleus gives the strongest signal, and
there is plenty of hydrogen around in the human
body in molecules of water, fat and proteins.
Potentially, other nuclei could be used, but the poor
sensitivity would make for very poor quality
images, and excessively long scan times
T

7
CHPTER 2 : The Magnetic Resonance Signal
CHPTER TWO:
The Magnetic Resonance Signal

2.1 Nuclear Magnetism

In the materials that we traditionally think of as being magnetic (such as iron
and steel), the magnetism arises from two properties of electrons: charge (all
electrons are negatively charged) and motion (electrons rotate or spin about their
axis). A moving charge will always generate a magnetic field, and in this way an
electrical current flowing through a wire generates a magnetic field around the wire.
While a single rotating electron generates a magnetic field, electrons tend to form
themselves into pairs, with one electron rotating in one direction and the other
electron rotating in the opposite direction.

The magnetic field of one electron cancels out the field from the second, and
the net magnetism of the pair is zero. In materials with unpaired electron spin, the
magnetic field from individual electrons may be randomly aligned, so that under
these circumstances the net field from the material will be zero. However, if such a
material is brought into a magnetic field (the applied magnetic field), this causes a
partial alignment of the electronsmagnetic fields and the material becomes
slightly, and temporarily magnetized.
These types of material are called paramagnetic.

We do not usually think of people as being magnetic, and indeed some of the
principal constituents of the human body (e.g., water, lipids, proteins) do not show
any magnetic effects from their electronic structure. However, another, weaker form
of magnetism (nuclear paramagnetism) arises from the charge and spin of nuclear
particles in just a few types of nuclei, where the spin of individual protons and
neutrons is not cancelled out within a nucleus. Table 1 shows a list of some of the
most interesting nuclei (from a biological point of view) that exhibit nuclear
magnetism.

8

a
(Relative sensitivity indicates the amount of signal (compared to hydrogen) and is for equal numbers of atoms per
unit volume of tissue. The relative sensitivity takes into account both the Larmor frequency and the natural
abundance of the given isotope.)

The hydrogen atom is most commonly used when performing MRI scanning,
since the hydrogen nucleus gives the strongest signal, and there is plenty of
hydrogen around in the human body in molecules of water, fat and proteins.
Potentially, other nuclei could be used, but the poor sensitivity would make for very
poor quality images, and excessively long scan times.

2.2 Precession

The hydrogen nucleus spins about its axis, generating the nuclear magnetism,
but when the nucleus is placed in a magnetic field, it undergoes another form of
rotary motion called precession. This type of motion can also be seen with a child's
spinning top if it is tilted at an angle to the gravitational field.

Figure 2.1
The angular momentum of
a spinning top (left)
interacts with the
gravitational field to
produce a rotary
precessional motion. In a
similar way, the magnetism
of the nucleus (right)
interacts with the applied
B0 field to cause precession
about the field.

1

9

The frequency of this precessional motion is called the Larmor frequency, and is
proportional to the applied magnetic field strength:
0
= B0 ..(1)
In the Larmor equation, above, is the gyromagnetic ratio and is constant for all
nuclei of the same type, but varies between different types, as shown in Table 1

2.3 Polarization and Coherence

Since nuclear magnetism is a form of paramagnetism, the individual nuclear
spins , do not have any preferred alignment until they are placed in the applied field
(the B
0
field). A large component of any MRI scanner is the magnet in which the
patient lies, which causes the nuclear spins to become aligned, or polarized, such
that some spins point in the general direction of the applied field, and some point
in a direction opposed to the applied field.

Figure 2.2
When no magnetic field is applied (left), nuclear spins have no preferred alignment so that the net
magnetization within the patient is zero. When a magnetic field is applied (right), polarization occurs
with some spins being aligned with the field, and some opposed to it. There are slightly more spins
aligned with the field than opposed, giving rise to magnetism within the patient the net
magnetization.

10
There are slightly more spins aligned with the applied magnetic field than
there are opposed to it, although the difference in the numbers is very small. This
difference is the excess population of spins. Now, when the magnetism of the
individual spins is added together, the net result is a small amount of
magnetization in the same direction as the applied field that arises within the
patient. The amount of this net magnetization is proportional to the strength of the
applied field, hence the need for a powerful magnet at the heart of the MRI scanner.
However, because the net magnetization points in the same direction as the applied
field and is much, much smaller than it, it is very difficult to measure directly. We
need to tilt the patients magnetization away from the applied field in order make it
measurable.

2.4 Radio Frequency Pulses

The patients net magnetization is made measurable by applying a burst
(pulse) of electromagnetic energy. The transfer of energy from the electromagnetic
wave to the nuclear magnetization is a resonance phenomenon, meaning that the
frequency of the pulse must be close the spinsnatural frequency (the Larmor
frequency) in order to have the desired effect. Since the Larmor frequency is in the
radio frequency range (MHz), then the pulse is called a radio frequency (RF) pulse.
Before the RF pulse is applied, the patients net magnetization is usually shown
schematically as in Figure 2.3. Note that a) only the excess population of spins is
shown, and b) although no individual spin is aligned with the B0 field, the net result
of adding all these spins together is a net magnetization that points in the direction
of that field.
Since the patients magnetization has a direction as well as a magnitude, it is a
vector quantity called the net magnetization vector, and given the symbol M. The
magnitude of M when the patient has become fully magnetized is given the symbol
M0, and depends on the strength of the B0 field and the number of hydrogen atoms
per unit volume of tissue that contribute to the magnetization vector. An RF pulse
tilts this vector out of alignment with the B0 field, and the angle of tilt (or the flip
angle) depends on the strength and duration of the RF pulse. When one or more RF
pulses is applied, and the MRI signal is measured (see fig 2.3 ), then this is called a
pulse sequence.

11

2.4.1 90 Pulse Excitation and FID

A 90 pulse will tilt the net magnetization vector by 90 . If M starts off aligned
with the B0 field (by definition, the z direction), then a 90 pulse flips the
magnetization into the x-y plane (see Figure 2.4).

Figure 2.3
Three different ways of representing the patients magnetization. a) shows the excess population of
spins that (by definition) tend to be aligned with the applied field. b) is a schematic way of drawing
these spins that indicates that the component of magnetization perpendicular to B0 is random such that
any transverse components of magnetization cancel to zero. c) is the resultant magnetization vector
after adding the individual nuclear magnets together. This net magnetization vector also points in the
same direction as the applied field, and is therefore difficult to detect directly.

Figure 2.4
Individual spins (left) and the net
magnetization vector (right) a)
before and b) just after a 90 RF
pulse. The vector has been tilted out
of alignment with the B0 field and
into the x-y plane.

12
M then precesses in the x-y plane so that, in effect, we now have the patients
magnetization rotating in a plane perpendicular to the applied field. The method of
detecting M is analogous to a motor cars alternator which generates electricity to
power the ignition, lights etc. An alternator consists of a magnet inside some coils
of wire, with the magnet being driven by the motor to that it rotates inside the coils.
This generates an electromotive force (voltage) within the coils.
Similarly, we wrap a coil of wire around the patient (the receiver coil) and
can measure an oscillating voltage across the coil, with the frequency of oscillation
being equal to the Larmor frequency, and the size of the voltage being dependenton
the number of hydrogen nuclei that are within the coil.

While the magnetization remains in the x-y plane, and the individual nuclear
spins are oriented in the same direction as shown in Fig. 2.4 (i.e., they are
coherent), a voltage will be measured in the coil. However, there are slight, natural
variations in the Larmor frequency due to the interactions between magnetic
particles (such as the nuclei) within the patient, and these variations in frequency
cause the signal to die away (the magnetization loses coherence) normally within
the course of a few hundred milliseconds or so. The initial voltage and its
subsequent decay are known as the free induction decay (FID), since the voltage is
produced by electromagnetic induction, and once M is in the x-y plane it can rotate
freely without further influence from RF pulses.
Figure 2.5
A voltage will be induced across the ends of a coil of wire placed around a rotating magnet. In an MRI
scanner, this rotating magnetism originates in the patient after a 90 pulse has been applied.
An oscilloscope connected to the coil would show a sinusoidally-oscillating voltage, with the frequency
of oscillation equal to the Larmor frequency, and the peak-to-peak voltage being proportional to the
number of spins within the sensitive region of the coil.

13

After application of the 90 RF pulse, the voltage measured in the receiver coil
that is wrapped around the patient takes the form of a sinusoidal oscillation with
decaying amplitude. The envelope that encloses the FID, shown as a dotted line
in Fig. 2.6, indicates the rate at which the sinusoid decays, and the time constant
that characterizes the decay is called T2
*
(T2 star). T2
*
depends on both the
physico-chemical properties of the tissue and on the uniformity, or homogeneity, of
the applied magnetic field.

Tissues that have a high free water content tend to have longer T2
*
values than
tissues with a dense matrix of cell membranes and cellular structures. Thus, T2
*
can
be used as the basis for imaging many pathological tissues such as MS lesions,
since the density of these cell membranes and structures is reduced by processes
such as demyelination and axonal loss, leading to increased T2
*
.
As mentioned above, T2
*
decay contains contributions from both the
physicochemical properties of the tissue, and magnetic field inhomogeneities within
the patient. These inhomogeneities may result from either imperfections in the
design and manufacture of the magnet, or from the magnetic properties of the
patient.

Any variations in the tissues magnetic properties within the patient will result in
small but important perturbations in the magnetic field uniformity; these sorts of
variation occur, for example, around the interface between the brain tissue and
frontal sinuses.

It can be seen from the Larmor equation that any variation in B0 field strength will
lead to different precessional frequencies throughout the patient, so that the initial
Figure 2.6
A 90 pulse of RF energy
tilts the magnetization
vector into the x-y plane
where it generates an
oscillating voltage in the
receiver coil. The
amplitude of this voltage
decays with a time
constant
T2
*

14
magnetization vector, being composed of magnetization from different parts of the
patient, will begin to spread out (or dephase) in the x-y plane.

This is mechanism by which the FID signal decays after it is initially generated by
the 90 pulse (Fig. 2.7).

2.4.2 180 Pulse Refocusing

The initial 90 pulse rotates the magnetization which is initially along the z-axis
(longitudinal magnetization) down into the x-y plane and generates the measurable
or transverse magnetization. Similarly, a 180 pulse rotates the magnetization
vectors by 180 , and one use of these pulses is to reverse the dephasing that occurs
after a 90 pulse because of the variations in the B
0
field.
Figure 2.7
After the initial 90 pulse, dephasing of transverse magnetization occurs, with the magnetization from
two locations within the patient, labeled a and b, being shown. A 180 pulse is applied and rotates the
magnetization about the y-axis so that the magnetization is flipped over, reversing the phase of a and b.
The magnetization continues to precess and comes back into phase at a time TE called the echo time.
Deviations from the average motion are shown, so that superimposed on this is a precession at the
average Larmor frequency.

15

Shown in Fig. 1.8 are two packets of magnetization, labeled a and b, with a
rotating clockwise (relative to the average) because it experiences a higher B0 field
than average (and hence has a higher precession frequency), and b rotating
anticlockwise because it experiences a lower B0 field than average. Deviations from
the average motion are shown, so that superimposed on this dephasing is precession
at the average Larmor frequency. A 180 pulse will rotate the magnetization about
the y-axis, which has the effect of reversing the phase that accumulated before this
pulse. After the 180 pulse, a continues to rotate clockwise and b anticlockwise and
eventually at a time called the echo time, or
TE, the individual magnetization vectors come back into phase generating again a
high voltage in the receiver coil. This spin-echo is what is commonly measured
during an MRI scanning sequence.
Notice that the voltage in the receiver coil is somewhat reduced in amplitude
compared to the initial amplitude after the 90 pulse. That is because the loss of
signal due to T2
*
relaxation has two causes, and only one of these (the magneticfield
inhomogeneities) can be reversed by the 180 pulse. A second cause of signal
Figure 2.8
After the 90 pulse creates magnetization in the x-y plane, variations in the Larmor frequency within the
patient will cause a spreading out, since some magnetization precesses slower, and some faster than the
average. This spreading out in the x-y plane results in a reduction in the size of the net magnetization
vector and is the cause of the decay in the voltage measured in the receiver coil. Deviations from the
average motion are shown, so that superimposed on this spreading out is a precession at the average
Larmor frequency.

16
Figure 2.9
Multiple spin-echoes are formed
by applying 180 pulses in
succession. The echoes form in the
spaces between the pulses, and the
amplitude of echoes in the train
decreases exponentially with a time
constant T2.

dephasing is the interactions that occur on a microscopic scale between magnetic
entities such as unpaired electrons and other nuclear magnetism within the patient.
Notice that the echo forms at a time exactly twice that between the 90 pulse and the
180 pulse. If the time between the 90 and 180 pulses is varied, then TE will also
vary and the amplitude of the echo will change, as shown in Fig. 2.9. Because the
effects of magnetic field inhomogeneity are removed in the process of forming the
echo, then the decrease in echo amplitude as TE increases is due only to the
microscopic interactions between magnetic particles within the tissue. These
interactions are fundamentally related to the physico-chemical properties of the
tissue, and not to the vagaries of the uniformity of the magnetic field. The decay of
the echo amplitude shown in Fig. 2.9 is exponential with TE and has a time constant
called T2, the transverse relaxation time. T2 is also called the spin-spin relaxation
time, since the loss of transverse magnetization in this way results largely from
interactions between the magnetism of individual spins.

After the echo is formed, the magnetization again begins to decay because of
the magnetic field inhomogeneities. Multiple echoes can be created by applying a
series of 180 pulses. The echoes form between the pulses, and these multiple
echoes are called the echo train. In this way, the MRI signal can be measured more
than once, since each echo gives us a measurement. The envelope that encloses the
peaks of the echoes in the train is also an exponential decay with time constant T2
(Fig. 2.10).

17
Figure 2.10
Increasing the time between the 90 and 180 pulses causes the spin-echo to form later. While the effects
of magnetic field inhomogeneity are reversed by the 180 pulse, some reduction in the echo amplitude
with increasing echo time remains; the time constant for echo amplitude decay is T2, the transverse
relaxation time.

2.5 Relaxation and Image Contrast

2.5.1 Overview

The above has given the background to understand the nature of the MRI
signal, how that signal is generated and how it decays due to transverse relaxation
(T2
*
).
The following will give more detail about how these processes can be used to create
MR images with contrast that helps in the diagnosis and characterization of disease.
The brightness of any tissue in an MRI image is fundamentally related to the
number of hydrogen atoms (or nuclear spins) per unit volume of that tissue since
this determines the voltage that is generated in the receiver coil by magnetic
induction. However, we can exploit the relaxation characteristics to modulate that
brightness according to the physico-chemical properties of the tissue; these
properties are different for different tissue types, and for tissue affected by disease.

18

2.5.2 The Bloch Equations

The nuclear spins exist in equilibrium with their surroundings. At
equilibrium, the magnetization within the patient is aligned with B0 field (defined as
the longitudinal direction, or z direction) as illustrated in Fig. 2.3. An RF pulse
disturbs the spins from this equilibrium, for example by tilting the net
magnetization vector into the x-y, or transverse, plane as shown in Fig. 2.4. The
Bloch equations, formulated by Felix Bloch in 1946, describe the return of
magnetization to equilibrium after such a disturbance. [2]

2.5.2.a LONGITUDINAL BLOCH EQUATION

This is a differential equation that describes the return of the longitudinal
component of magnetization (Mz) to equilibrium:

The solution to this equation depends on the nature of the disturbance from
equilibrium. For example, if a 90 pulse is applied at time t=0, then this tilts the
longitudinal magnetization completely into the x-y plane so that Mz = 0. Then the
solution, illustrated in Fig. 2.11a, is:

If, instead, a 180 pulse is applied at time t=0, the longitudinal magnetization is
inverted (tilted upside down), so that Mz = -M0. Then the solution, as shown in Fig.
2.11b, is:

Each of these particular forms is of interest in MRI, since both 90 and 180 pulses
are commonly applied, and contrast is affected via the longitudinal magnetization.

19
Figure 2.11
Two particular solutions to the
longitudinal Bloch equation.
Before an RF pulse is applied,
Mz is at its equilibrium value
M0.
(a) A 90 pulse tilts the z
magnetization into the x-y plane
so that Mz = 0 at t = 0 before it
recovers exponentially back to
equilibrium.

(b) A 180 pulse inverts the z
magnetization so that Mz = -M0
at t = 0 before again it recovers
exponentially to equilibrium.

2.5.2.b TRANSVERSE BLOCH EQUATION

This is a differential equation that describes the return of the transverse
component of magnetization (Mxy) to equilibrium:

Of course, at equilibrium, the transverse component of magnetization is zero (see
Fig. 2.4a), so this differential equation simply describes how the transverse
magnetization returns to zero after it has been disturbed. Again, the solution
depends on the nature of the disturbance but if a 90 pulse is applied at time t=0,
then this tilts the longitudinal magnetization completely into the x-y plane so that
Mxy = M0. Then the solution, illustrated in Fig. 2.12, is:

20

We have seen in Section 2.1.3 that in the real world the magnetic field
inhomogeneities cause more rapid decay of transverse magnetization, so that
thedecay constant should really be T2
*
in Eq. [6]. However, if we measure the
transverse magnetization in a spin-echo, as shown in Fig. 2.9, then the decay
constant T2 applies.

2.6 T2-weighted Imaging

T2-weighted imaging creates image contrast (differences in tissue brightness)
that depends on variations in T2. T2-weighted imaging is simpler to understand than
T1-weighting, and is therefore discussed first.
After a 90 pulse, the magnetization lies in the x-y plane and generates the maximum
voltage in the receiver coil. This voltage first dies away and then, if a 180 pulse is
applied, reforms as an echo at a time TE (Fig. 2.8). The amplitude of the echo
depends only on the T2 of the tissue, which in turn reflects the physicochemical
properties of that tissue.
Fig. 2.13 shows transverse relaxation curves for two tissues with different T2
values: tissue 'a' has the longer T2 and could represent an MS lesion, while tissue 'b'
could represent normal tissue. An echo time can be chosen so that the difference in
Figure 2.12
The solution to the transverse Bloch equation when a 90 pulse is applied, converting all z
magnetization to transverse magnetization so that Mxy = M0 at t = 0. Mxy decays exponentially to
its equilibrium value of zero.

21
Figure 2.13
Transverse relaxation curves for two tissues aand b, with the difference between aand b
shown as the lower dotted curve. This difference is the contrast between the two tissues, and is
maximized by judicious choice of TE in T2-weighted imaging.

signal generated by the two tissues, and therefore the difference in brightness, is
maximized. When TE is chosen in this way to give contrast, this is known as T2-
weighted imaging.

Figure 2.14 shows an axial section of a normal brain, cutting through the
lateral ventricles. The same section is shown with three different echo times: 15 ms,
50 ms and 100 ms. Notice that in the top row of images, where the display
brightness settings have been kept fixed, the brightness of all tissues decreases with
increasing TE, as expected from Fig. 2.12.

In the bottom row of images, adjustments have been made to the display
brightness setting, as would normally be done, giving the illusion that the signal
from some tissues increases with increasing TE.

22
Figure 2.14
Axial section of a normal brain with increasing echo time (TE) from left to right. In the top row of
images, the display brightness settings have been kept fixed, showing that all tissues decline in
brightness, with some declining more quickly than others. CSF has the longest T2 and declines most
slowly, while gray matter has a slightly longer T2 than white matter and appears brighter in all images
largely because of increased proton density. In the bottom row, the image display
brightness has been adjusted as would be done for filming.

Figure 2.15 shows a typical application of T2-weighted imaging in multiple
sclerosis. The echo time has been chosen to give good contrast between normal
white matter and the MS lesions, which have an elevated T2 and therefore appear
hyperintense on this image.

23

2.7 T2
*
-weighted Imaging

T2*-weighted imaging employs the same concept as T2 weighting in that there
is a delay between the generation of transverse magnetization and its measurement.
The difference is that in a T2
*
-weighted sequence, no 180 refocusing pulse is used,
so that the signal decays partly because of magnetic field inhomogeneities with a
time constant T2
*
.This type of scan is sometimes known as a gradient-echo
sequence, or sometimes as a FLASH sequence (an acronym for Fast Low-Angle
SHot).

2.8 T1-weighted Imaging

Longitudinal magnetization is not directly measured by MRI: to generate a
voltage in the receiver coil, this longitudinal magnetization must first be tilted into
the x-y plane. However, if multiple RF pulses are to be applied, then longitudinal
relaxation affects the amount of longitudinal magnetization that is available for
tilting when the next pulse is applied.
Figure 1.15
Axial T2-weighted spin-echo image (TE = 90 ms) of a patient with multiple sclerosis. Multiple lesions
can be seen, particularly around the lateral ventricles and they are hyperintense because of both their
longer T2 and increased proton density.

24
Figure 2.16
To collect data for an MR image, a series of 90 RF pulses must be applied, with a separation between
each pulse of TR. The amount of longitudinal magnetization before each pulse determines how much
signal that pulse will generate. Hence, T1 influences the signal intensity in the image as
TR is shortened.

Consider a sequence of 90 pulses applied in succession, with a time between
pulses of TR (the repetition time) as shown in Fig. 2.16. Each pulse tilts the
zmagnetization down into the x-y plane, generating a measurable signal. Between
the pulses, Mz recovers according to the curve described by Eq. 3, so at a given
TR, different degrees of recovery of the longitudinal magnetization in different
tissues produce contrast according to the T1 of these tissues, as illustrated in Fig.
2.17. Remember that the next 90 pulse converts this recovered longitudinal
magnetization into measurable signal, so the T1 indirectly affects the amplitude of
the signal and therefore the brightness of that tissue in the MR image. Notice that
this is only true for signal generated by the second and subsequent pulses, since the
amplitude of the signal after the first pulse is independent of the T1; this first signal
is usually discarded.
A repetition time can be chosen so that the difference in signal generated by
the two tissues, and therefore the difference in brightness, is maximized (Fig. 2.7).
When TR is chosen in this way to give contrast, this is known as T1-weighted
imaging.

25

Figure 2.18 shows a normal brain, with three different repetition times: 600 ms,
1500 ms and 4000 ms. In the top row of images, where the display brightness
settings have been kept fixed, the brightness of all tissues decreases as TR is
reduced, in line with Fig. 2.11. In the bottom row of images, where adjustments
have been made to the display brightness settings, there is the illusion that the
signal from some tissues increases with decreasing TR. [5]

Figure 2.17
Longitudinal relaxation curves for two tissues aand b, with the difference between aand b
shown as the lower dotted curve. This difference is the contrast between the two tissues, and is
maximized by judicious choice of TR in T1-weighted imaging.

26

T1-weighted imaging is typically used in the head to delineate parenchyma from
CSF, and is most frequently used to show tissue hypoplasia or atrophy. Fig. 2.19
shows a routine T1-weighted scan typical of those used to answer many clinical
questions. The TR on this type of scan is chosen to give high contrast between gray
matter and CSF, giving good definition of the outline of the brain structures.

.

Figure 2.18
Axial section of a normal brain with increasing repetition time (TR) from left to right. In the top row of
images, the display brightness settings have been kept fixed, showing that all tissues increase in
brightness as TR is increased, with some increasing more quickly than others. CSF has the longest T1
and increases most slowly, while white matter has a slightly shorter T1 than gray and recovers more
quickly. In the bottom row, the image display brightness has been adjusted as would be done for filming.
Figure 2.19
A T1-weighted coronal section through the temporal
lobes and lateral ventricles of a female (age 79 yrs)
presenting to memory clinic, showing general atrophy
of the brain

27
1.2.6 Summary

The basic factor determining the brightness of a tissue is the number of hydrogen
atoms, or protons, per unit volume of tissue, since the signal intensity after an RF
pulse is proportional to the proton density in a scan with a long TR and a short TE.
Weighting by the relaxation times can be introduced either by shorter repetition
time (TR) to give T1 weighting, or by longer echo time (TE) to give T2 weighting.
This is summarized and illustrated in Fig. 2.20.

Using a combination of short TR and long TE is not generally useful, since
the short TR serves to darken tissues with long relaxation times, while long TE
darkens tissues with short relaxation times. Thus, the brightness of all tissues is
generally depressed and the resulting image has low intensity and poor contrast (see
Fig. 2.20).
In biological tissues, it is a general rule that T1 and T2 vary in concert: if T1 is long
for a particular tissue, then T2 will also be long, and vice versa. However, there is no
strict relationship between them, and the only approximate rule is that the more

Figure 2.20
Summary of relaxation time weighting of MR images. The fundamental brightness is determined by the
number of hydrogen atoms per unit volume of tissue (proton density). Decreasing the TR introduces T1
weighting, and increasing the TE introduces T2 weighting. A combination of short
TR and long TE causes a decrease in intensity for all tissues, giving low signal intensity and little
contrast.

28

solid a tissue, or the higher the content of large molecules such as proteins, and
lipids in cell membranes then the lower will be both T1 and T2.
An MRI examination will normally comprise more than one imaging pulse
sequence. In the CNS, T1-weighted sequences are typically used to show gross
anatomy, since the nervous system tissue is shown in relief against a background of
dark CSF. Because of this, T1-weighted imaging is sometimes referred to as
anatomical scanning. T2-weighted sequences are normally used to show changes
in tissue that are brought about by disease, since the relaxation times are invariably
lengthened in damaged tissue and this shows as a bright region. T2- weighted
scanning is sometimes referred to as imaging pathology.

29

C H A P T E R
3

Brain Tumors Identification in
Magnetic Resonance Images
rain tumors have become a global problem
ccccccparticularly in Iraq because of the existed
environmental polluted sources. From the
classification point of view, brain tumors show the
same attributes as other bodys tumors, as there is
benign and malignant tumors. Nevertheless, the
brain tumors has it's particular from being to handle
B

30
CHPTER 3 : Brain Tumors Identification in Magnetic Resonance Images
CHPTER THREE :
Brain Tumors Identification in Magnetic Resonance Images

3.1 Abstract
Brain tumors have become a global problem particularly in Iraq because of the
existed environmental polluted sources. From the classification point of view, brain
tumors show the same attributes as other bodys tumors, as there is benign and
malignant tumors. Nevertheless, the brain tumors has it's particular from being to
handle. But, with a great advances in the early detection of the tumors by medical
imaging techniques such as Computed Tomography (CT) scan, Magnetic
Resonance Imaging (MRI) and Ultrasound (US) give easy access for excision of
brain tumors. Medical Image analysis and processing has great significance in the
field of medicine, especially in non-invasive treatment and clinical study. In this
study, MR images are utilized using image-processing techniques, to differentiating
infected tissues from normal brains tissues.
A semi-automatic unsupervised successive segmentation technique, using 1
st

order statistical features, has been proposed to isolating tumor from normal tissues.
The introduced technique includes two stages; firstly, the mean, standard deviation,
and image entropy have been adopted as a 1
st
order statistical features. Because of
the unsupervised segmentation technique is based on the intensity information in an
image. For this reason, coloring interference was the problem faced us in isolating
the tumor area from the Gray Matter (GM) brains tissue. To overcome this color-
overlapping problem, an adaptive multi-stage segmentation by using the mean value
of each pre-segmented classes has been introduced to extract the tumor tissue from
those showed similar behaviors. [4]

3.2 Magnetic Resonance Imaging
Magnetic resonance (MR) imaging (MRI), in conjunction with computerized
stereotactic systems, has become an indispensable tool for guiding brain tumor
resections. During the last decade, image guidance has led to improvement in both
lesion localization and definition of tumor margins, thus increasing the precision

31
and reducing the invasiveness of brain tumor surgery. Data obtained by means of
newly developed MRI techniques, such as diffusion tensor MRI and functional
MRI, allow for visualization of white-matter fiber tracts and cortical activity,
respectively.
Along with the exponential spread of intraoperative navigational tools, the
role of computer-assisted surgical planning has expanded. Surgical planning takes
advantage of both two-dimensional (2D) and three-dimensional (3D) models, which
integrate data from multiple imaging modalities, each highlighting one or more
aspects of morphology or function. Thus, multimodality fusion-based models are
used increasingly for optimizing lesion targeting and simulating different surgical
approaches.
MRI is based on the principles of nuclear magnetic resonance (NMR). The
NMR phenomenon relies on the fundamental property that protons and neutrons
that make up a nucleus possess an intrinsic angular momentum called spin. When
protons combine to form nucleus, they combine with oppositely oriented spins.
Thus, nuclei with an even number of protons have no net spin, whereas nuclei with
an odd number of protons possess a net spin. Hydrogen nuclei have an NMR signal
since its nucleus is made up of only a single proton and possess a net spin. The
human body is primarily fat and water, which have many hydrogen atoms. Medical
MRI primarily images the NMR signal from the hydrogen nuclei in the body
tissues. The net spin of the nucleus around its axis gives it an angular moment.
Since the proton is a positive charge, a current loop perpendicular to the rotation
axis is also created, and as a result, the proton generates a magnetic field.

The joint effect of the angular moment and the self-generated magnetic field
gives the proton a magnetic dipole moment parallel to the rotation axis. Under
normal condition, one will not experience any net magnetic field from the volume
since the magnetic dipole moments are oriented randomly and on average equalize
one another. When placed in a magnetic field, a proton with its magnetic dipole
moment processes around the field axis. The frequency of this precession, eo, is the
resonant frequency of NMR and is called the Larmor frequency. The precession
frequency is directly proportional to the strength of the magnetic field, i.e.
) 1 .........(
O o
B e =

32
Where B0 is the main magnetic field strength, and is a constant called gyro
magnetic ratio, which is different for each nucleus (42.56 MHz/Tesla for protons).

Given a specimen, the application of a magnetic field B0 would create a net
equilibrium magnetization M0 per cubic centimeter, which is aligned to the B0
field. The M0 is the net result of summing up the magnetic fields due to each of the
H nuclei and is directly proportional to the local proton density (or spin density).
However, M0 is many orders of magnitude weaker than B0 and is not directly
observable. By tipping M0 away from the B0 field axis with an appropriate RF
pulse having a frequency equals to the Larmor frequency, a longitudinal
magnetization component ML and a transverse magnetization component MT is
produced. When the RF pulse is turned off, the longitudinal magnetization
component ML recovers to M0 with a relaxation time T1, and the transverse
magnetization component MT dephases and decays to zero with a relaxation time
T2. During relaxation, the protons lose energy by emitting their own RF signal with
the amplitude proportional to MT. This signal is referred to as the free-induction
decay (FID) response signal. T2 indicates the time constant required for the FID
response signal from a given tissue type to decay. An RF coil placed around the
object being imaged measures the FID response signal.

3.3 overview about imaging
Image processing and computer vision have become important within the past
three decades. It is a rapidly growing area of computer science. Its growth has been
fueled by technological advances in digital imaging, computer processors and mass
storage devices fields that traditionally used analog imaging are now switching to
digital systems, for their flexibility and affordability. Important examples are
medicine, film and video production, photography, remote sensing, and security
monitoring etc. These and other sources produce huge volumes of digital image
data every day, more than could ever be examined manually. Image processing
techniques aim to improve the pictorial information for human interpretation and
the processing of image data for autonomous machine perception.
Medical image analysis and processing has great significance in the field of
medicine, especially in non-invasive treatment and clinical study. It has emerged as
one of the most important tools to identify as well as diagnose various disorders.
This tool assists the physicians and Radiologists to improve their opinions on

33
specific diagnosis. In addition, imaging analysis helps the physicians to visualize
and analyze the image for better understanding of abnormalities in internal
structures. The various scanning techniques have incarnated as an inevitable tool for
the physicians to help them in diagnosis.
There are multiple medical imaging techniques such as Magnetic Resonance
Imaging (MRI), Computed Tomography(CT), Ultrasound, and other more. This
paper has concentrated on (MRI), because it is an imaging technique used primarily
in medical settings to produce high quality images inside the human brain. As well
as MRI is a non-invasive technique to view biological tissues, using the
phenomenon of nuclear magnetic resonance.
Image segmentation is one of the most important tasks in computer vision and
image analysis. The applications of segmentation techniques range from medicine
(e.g., locating a lesion) to industry (e.g., robotic vision) and the military (e.g., target
detection). With the increasingly prominent role of medical imaging, the diagnosis
and treatment of disease, segmentation techniques have been extended for
extracting clinically useful information about anatomic structures through
modalities. The segmentation of the images can be performed either manually or
using image processing and computer vision techniques. Trained radiologists
perform segmentation of MRI manually, but now there are many recent
developments are employing to segment the MRI since manual of images is a time
consuming process and is susceptible to human errors.

3.4 Brain Tissues
It has been demonstrated in fig.(3.1-a) that; the normal brain is generally
divided into three mainly different tissues; i.e. GM, WM, and CSF, as well as four
ventricles of the brain are connected cavities within the brain, where cerebrospinal
fluid CSF is produced , each has the following function:
Gray matter forms the outer layer of the cortex, encasing the inner white matter
almost completely. Regions are gray because of the high concentration of cell
bodies, and are therefore sites of high neuronal connections and computational
activity.
White matter is made up of nerve fibers that connect different parts of the cortex, as
well as the cortex with other parts of the brain. It regions are white because of the
high concentration of axons, and are therefore major signal pathways.

34
Cerebrospinal Fluid, also called CSF, is a clear substance that circulates through
the brain and spinal cord. It provides nutrients and serves to cushion the brain and,
therefore protect it from injury. As this fluid gets absorbed, more is produced from
the choroids plexus, a structure located in the ventricles. A brain tumor can cause a
build-up or blockage of CSF.
Tumor and Edema for abnormal brain fig.(3.1-b), in addition to the three tissue
classes as mentioned in above we add a two new classes which are, tumor and
edema tissue. A tumor is an abnormal growth of tissue .Unlike tumor structures,
there is no spatial prior for the edema. As a consequence, the probability density
function for edema cannot be initialized automatically. We approach this problem
as follows: first, we have found that edema, when present is most evident in white
matter. Also, we noticed from tests with supervised classification that the edema
probability density appears to be roughly between CSF and white matter. Thus,
white matter and edema would result in similar probability density functions. The
bimodal distribution is then initialized by modifying the mean value for edema to be
between white matter and CSF, using prior knowledge about properties of edema .

a

b

3.5 Image Analysis System
Image analysis has great significance in the field of image processing,
especially in non-invasive treatment and clinical study. Image analysis, as
illustrated in figure (3.2), involves the following operations. [6]
Tumor
Edema

Ventricle

Body of
caudate

Internal
capsule

Gray matter

White matter

CSF

Figure 3.1
Show brain tissues in MR image (upper) normal image and (lower) is a pathologic image

35

At the outset, the segmentation process can be split into three major steps:
- Pixel Classification: In this step, pixels are classified into a certain number of
classes. In general, the number of classes is equal to the number of tissues in the
brain, so that there exists a one-to-one correspondence between the classes and the
tissues. Methods such as FCM, Finite Gaussian Mixture Model, Neural Networks ,
etc. accomplish task
- Correction: In this step, we remove any misclassification caused due to
inhomogeneities of the radio frequency and magnetic fields, imaging noise, etc. n x
n majority filters and Markov Random fields are used for this purpose.

- Tissue labeling: During this step, a class of pixels is assigned a unique tissue
name. This is done using a priori knowledge about density-tissue relationship or by
using an anatomical atlas knowledge base. Image segmentation simply refers to the
decomposition of an image into regions, but does not label the regions with the
tissue types. Classification is the step that takes care of this. In supervised
segmentation, this is done by an operator when labeling training data, where pixels
are associated with anatomical tissues. Unsupervised methods return segmentation
with different clusters; however the association of clusters with tissue types must
still be performed. Often, this is done by an operator, or implicitly understood by an
investigating physician. This is the simplest and the most effective method for
classification in unsupervised segmentation, though many other schemes also exist.
All of our above discussion, vindicate our choice of an unsupervised method
(namely FCM and FGMM) to segment the brain MRI.

Preprocessing

Segmentation
Feature Extraction
For ROI

Classification

Raw Image
Figure 3.2
Major steps of image segmentation and classification system

36
3.6 Image Segmentation Methods
Segmentation is based on homogeneity of the image regions; i.e. intensity,
color, texture, or the combination of all them. Ideally, segmentation method finds
those sets that correspond to distinct anatomical structures or regions of interest
(ROI) within the image. When the constraint that regions be connected is removed,
then determining the sets R
i
is called pixel classification and the sets themselves are
called classes. Pixel classification is often a desirable goal in medical images,
particularly when disconnected regions belonging to the same tissue need to be
identified. Determination of the total number of classes i in pixel classification is a
difficult task. Medical image segmentation has two sorts of objectives, either to find
out regions containing certain body structures or to extract the contours of body
structures. For the first objective, segmentation is conducted based on some
regional similarity criteria, such as intensity, gradient, texture and so on. ROIs are
normally treated as foreground regions R
f
and the remaining are treated as
background regions R
b
. The problem formulation under this objective is modified
accordingly. The problem is to determine regions R
f
and R
b
, such that:
| = =
b b f
R I R R
f
R ,
For the second objective, segmentation is conducted based on some discontinuity
criteria, such as edges. The problem is to find a set of points C = {C
i
}, which
tightly enclose the ROI.

3.7 Segmentation Methods:

3.7.1 Manual Segmentation Methods:
Medical images are traditionally, segmented; i.e. the region of interest ROI are
delineated by some expert peoples (as experienced radiologists and physicians). In
manual segmentation, a skilled operator, using a computer mouse or track ball,
traces the structures of interest on each slice of the image dataset. Manual
segmentation benefits from anatomical knowledge that the user employs for
boundary identification of the objects . Manual segmentation method has at least
four disadvantages:
1. Generally requires a high level of expertise.

37
2. It is time and labor consuming. Segmentation of a series of 1500-2000
transverse images of 512512 pixels usually takes 2 to 4 hours.
3. It is subjective and therefore not reproducible. For example, a disagreement ratio
of 14 ~ 22% by experts in brain tumor segmentation was reported in.
4. The brightness and contrast of the display screen affects the segmentation process
and the subsequent analysis.

3.7.2 Computer Aided Segmentation Method
The advantages of automating segmentation include increasing reliability,
consistency, and reproducibility. However, the most practical benefit is that while
the computer may work longer at the task, it takes less human time. However,
medical image segmentation, which is usually effortless and swift for the human
visual system, is a considerable challenge in computer based automatic program.
This will either decrease the cost of a volumetric study, allow more subjects to be
analyzed, or since the time commitment of manual methods is prohibitive, it can
also allow even more comprehensive analyses. Also, there are numerous problems
faced in automatic MRI segmentation, some of them are mentioned below . [7]
It is hard to represent domain knowledge in a computer system. The shapes
of soft tissues inside the human body are not only complex but also highly variable.
Large variability in medical images may happen between two different people.
Difficulty comes from the medical image itself due to the intrinsic properties of the
medical imaging systems.
As a tradeoff between manual and automatic segmentation, semi-automatic
segmentation algorithms which require certain human interactions; e.g. defining a
seed in the center of the ROI, or roughly initialize a close contour which contains
the ROI etc . Semi-automatic segmentation methods, to certain extent, alleviate
some labor of experts, and ROIs are directly selected in manual interactions.
However, it can not solve the problems in manual segmentation substantially, since
the interactions such as initialization may still be quite subjective, and the
interactions may still be affected by the brightness and contrast of the display
screens .

38

3.8. Texture Feature Extraction
Although there is no strict definition of the image texture, it is easily perceived
by humans and is believed to be a rich source of visual information about the nature
and three-dimensional shape of physical objects. A different approach to texture
analysis has been investigated in literature, like statistical-based approach,
structural-based approach, model-based approach, and transform-based approach. In
this paper, we investigate statistical-based approach of texture analysis using first-
order statistics of T
2
weighted images in normal and abnormal patients, to identify
good" metrics that would aid as markers in detecting early abnormal cases.

Image Segmentation Methods
Manual Segmentation Computer Aided (CAD) Segmentation
Grayscale
Region Based segment.

Edge Based segment.

Multi-Spectral

Unsupervised
(Clustering)

Isolate
Fuzzy c-means
The c-means
Supervised
Minimum distance
Maximum likelihood
Parallelepiped
Figure 3.3
Overview of the image segmentation methods.

39
3.8.1 Statistical Features for Texture Analysis
Statistical features for texture analysis is one of the simplest methods, so far
used to measure the texture features behaviors. The recognition performance of the
1
st
order measures of this kind of test has found to be not vigorous (often utilizes the
1
st
order histogram measures; mean, standard deviation, entropy and moment).
3.8.1.a First-Order Statistics
Let us assume an image of dimensions N M, to be a function f(x, y) of two
space variables x and y, where x = 0, 1, N - 1 and y = 0, 1 M -1. The function
f(x,y) can take discrete values i = 0,1,.,G -1,where G is the total number of
intensity levels within the image. The intensity-level histogram is a function
showing the number of pixels in the whole image, are given as :
) 2 .........( ) ), , ( ( ) (
1
0
1
0
=
=
N
x
M
y
i y x f i h o

Where o(j,i) the Kronecker delta function
=
=
=
i j
i j
i j
, 0
, 1
) , ( o

The image histogram plots the relative frequency of each pixel value that
occurs in a grayscale image. The histogram of intensity levels is obviously a concise
and simple summary of the statistical information contained in the image. Since, the
calculation of the gray level histogram involves single pixels; it contains the first-
order statistical information about the image (or its fragment). The histogram of an
image, a first order statistical method, provides the simplest method for describing
texture. The probability mass function of occurrence of the intensity levels is
obtained by dividing the values h (i) by the total number of pixels in the image.
1 - ....G 0,1,...... z
(3) .......... ,.........
) (
) (
=

=
i
i
N M
i h
z

The shape of the histogram provides many clues as to the image characteristics; it
shows the dominant image brightness level, number of objects within the image etc.
The histogram of an image is (z
i
), for z
i
=0, 1, 2... G-1, as shown in fig. (3.4).
Different parameters can be extracted from the histogram, e.g. the image mean, the
image entropy, and the image moments, as mentioned :

40
3.8.1.b The mean: In terms of image distribution probability values, the image
mean is given by :
) 4 .......( .......... ) (
1
0
=
=
G
z
i i
i
z z

3.8.1.c The standard deviation: provides useful information about the
variation of intensity through the image, the standard deviation, given by :
) 5 ......( .......... ) ( ) (
1
2
i i
G
z
i
o

=
3.8.1.d The entropy: The entropy is a measure of the uniformity or randomness
of image grey distribution, given by :
) 6 .......( )......... ( log ) (
1
0
i
G
z
i
z z H
i

=
=

3.9. Methodology
3.9.1 Unsupervised Segmentation Method
In this paper, a semi-automatic unsupervised successive segmentation method
has been introduced to isolating brains tumor tissue from healthy tissues, based on
certain 1
st
order desired features. The introduced technique includes two stages;
T2-WEIGHTED-TSE-AXIAL
0.00E+00
2.00E-02
4.00E-02
6.00E-02
8.00E-02
1 20 39 58 77 96 115 134 153 172 191 210 229 248
Grayvalue
P
r
o
b
a
b
i
l
t
y
Series1
Figure 3.4
The MR grey scale image and its histogram

41
firstly, for each image window, compute the 1st order statistical features (i.e. mean,
standard deviation, and entropy).

Because of the unsupervised segmentation, technique is based on the
intensity information in an image. For this reason, coloring interference was the
problem faced us in isolating the tumor area from the Gray Matter (GM) brains
tissue. To overcome this color-overlapping problem, an adaptation process as a
second stage was proposed to refining the result of preliminary segmentation by
assigning each classified point to its nearest mean value. [8]

This has been done by computing the mean values of the preliminary classes,
the pixels value of the raw image. The refining result is demonstrated as shown in
fig.(4), which is concerned as the proposed supervised-unsupervised segmentation
technique. The closeness between the counted means obtained by the supervised
operation with the image pixel values is performed using the minimum distance
creation method. Different samples of MR images for normal and abnormal brains
have been tested, for different patients
The following steps can summarize the operations involved two-Stage
Unsupervised Segmentation Algorithm

Unsupervised Segmentation Algorithm
First stage

Step1: Input MR Image for chosen patient;
Step2:Define Windows size(M), number of desired Classes and Minimum
allowable deviated error between classes;
Step3: For each image window, compute the 1
st
order statistical features (i.e.
mean, standard deviation, and Entropy (see eqs. 4,5 & 6);
Step4: The classification search is performed on the test image, pixel to pixel,
from the upper left image corner; left to right, up to down.
Step5: A new set of features is counted, compared with previous belongs to
past windows. For the first tested window, the pixel center classified as

42
Class-1. Successively, counted features are compared with those stored
in vector to determine the closet class.
Step6: If the difference between the newly counted features with the stored was
larger than a decided minimum deviation error, the pixel is assigned as
new class number. Other it takes the class number belong to the closet
absolute minimum features.
Step 7: if the number of the assigned class is found to be larger than the
selected number of classes, then the pixel given a number out of the
classs range, referring as to be undefined class. Thus, normally,
numbers of classified regions are equal to the number of the decided
classes plus one.
Step8: The preliminary segmentation process is terminated when the
classification operation reach the last down-corner image point.

Second stage

Step 9: Compute the mean values of each class;
Step 10: Reclassify the MR image, using the counted means (step 9), and stop.

Original slice

Preliminary segmentation

Before correction process

Final segmentation

Figure 3.5
Two stages unsupervised segmentation method, applied on T
1
-weight MR image.
(a) The first 3 images represent stage-1; (b) The 4
th
image represents the final
43

C H A P T E R
4

Result and Discussion

44
CHPTER 4 : Result and Discussion
CHPTER FORE :
Result and Discussion

4.1 Result and Discussion
As it has been mentioned in above, the adopted unsupervised segmentation method
requires to define priory only number of classes, windows size and minimum deviated error.
Figs.(14-a and b) represents different segmented images obtained by 33 pixels windows size,
minimum deviated error distance equal 12, and for varying number of classes. The method has
been applied on both normal and abnormal brain images.
a-Normal Case

Original image

Segmented by 5
classes

Segmented by 7
classes

Segmented by 9
classes
B-Abnormal Cases

Original image

Segmented by 5
classes

Segmented by 7
classes

Segmented by 9
classes

45

Original image

Segmented by 5
classes

Segmented by 7
classes

Segmented by 9
classes

As can be seen, the preferable number of classes ranged between (5-9), which represent
the main brains tissues. The minimum deviated distance MDD has been chosen as to be "12" due
to the gained results through our presence course of work. One of the main advantages acquired
from the implementation of the unsupervised classification is that; numbers of regions are defined
priory by the operators. Definitely, for abnormal brains, number of regions should be, at least,
being increased by one of the normal brain tissues. Cases number of tissues class should be
increased by two referring to the tumor and edam tissues. For best differentiation between normal
and abnormal MRI brains, number of regions adopted in this project was found as to be 7 classes.

4.2. Conclusion

The problems usually encountered when trying to design an automatic system for tumors
detection system is that; a large number of tumors types are existed, which differs in size, shape,
location, tissue composition and homogeneity. Consequently, preliminary work on tumor
segmentation relied imperfectness in differentiation between normal and infected tissues.

A semi-automated segmentation technique was introduced to isolate the tumor tissues from the
rest of brains tissues, by selecting number of points of interest from the tumor region. Many
different orders statistics features have been used in performing these segmentation techniques;
i.e. mean, standard deviation, and entropy. The minimum distance criterion has utilized to test the
similarity or dissimilarity between brain components. A better result we obtained in comparison
with edge detection methods, by assigning different colors for different tissues.
Figure 2.2
Unsupervised segmentation results for normal and abnormal T2-Weighted MR images of , for
different classes Mask Size = 33 , and MDD = 12.

46
Because of the image, segmentation technique is based on the intensity information in an
image and has been used for the segmentation tissue classes. For this reason, coloring interference
was the problem faced us in isolating the tumor area from the grey matter brains tissue .To
overcome this color-overlapping problem, an adaptive multi-stage segmentation has been
introduced to extract the tumor tissue from those showed similar behaviors.
The normal and abnormal(i.e. tumor ) tissue classes have the same or overlapping grey value
distributions; such methods have been successfully applied to identify normal anatomical
structures, but may fail in the presence of highly variable brain tumors.

References
47

References

[1] Peter, A. R. An introduction to Magnetic Resonance in Medicine Edited
by, Oxford Blackwell scientific publications, Third edition,1993.

[2] Rosenfeld, A., Kak, A.C, " Digital Picture Processing " 2
nd
Edition
Academic Press, 1982.

[3] Bizais, Y., Barillot C and Paola RD , Information Processing in Medical
Imaging, Proceedings of 14
th
Int. Conf., Kluwer Academic,1995.

[4] Seemann T., Digital Image Processing using Local Segmentation Ph.D.
Thesis Submit for ,School of Computer Science and Software Engineering,
Monash University, Australia,2002.
[5] Warfield SK, Talos F, Tei A, et al. Real-time registration of volumetric
brain MRI by biomechanical simulation of deformation during image
guided neurosurgery. Comput Visual Sci. 2002;5:311.

[6] Curry, T. S., E. Dowdey, J. & Murry, R. C., Christensens of Physics of
Diagnostic Radiology Lea & Fibiger ,Philadelphia, London 4
th

Edition,1990.
[7] Kesavamurthy T., Subha Rani S.,Pattern Classification using Imaging
Techniques for Infarct and Hemorrhage Identification in the Human
Brain,Calicut Medical Journal Vol.4,No.3,2006.

[8] Rane S. D. Parallel Magnetic Resonance Imaging: Characterization and
Comparison , MS.C. Thesis, Office of Graduate Studies of Texas

Magnetic Resonance Imaging (MRI)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Magnetic Resonance Imaging (MRI)

Uploaded by

Copyright:

Available Formats

Republic of Iraq

Ministry of Higher Education And Scientific Research

You might also like