You are on page 1of 66

CHAPTER 1

INTRODUCTION

1.1 OVERVIEW

In medical image processing the detection of new Multiple Sclerosis (MS) lesions
or an inflammatory diseases that is an inflammatory disease on Magnetic Resonance Imaging
(MRI) is important as a disease activity and surgical purpose.
Brain Magnetic Resonance Imaging (MRI) is widely used in surgical and diagnosis purpose in
that the image processing is used to give the result.
Brain tissues classification
Brain tissues classified as three ways that is,
1. White Matter(WM).
2. Gray Matter(GM).
3. Ceribrospial Fluid(CSF).
Multiple Sclerosis(MS)
Multiple sclerosis (MS) is a chronic autoimmune disorder affecting movement,
ensation, and bodily functions. It is caused by destruction of the myelin insulation covering nerve
fibers (neurons) in the central nervous system (brain and spinal cord).
MS is a nerve disorder caused by destruction of the insulating layer surrounding
neurons in the brain and spinal cord. This insulation, called myelin, helps electrical signals pass
quickly and smoothly between the brain and the rest of the body.When the myelin is destroyed,
nerve messages are sent more slowly and less efficiently. Patches of scar tissue, called plaques,
form over the affected areas, further disrupting nerve communication. The symptoms of MS
occur when the brain and spinal cord nerves no longer communicate properly with other parts of
the body. MS causes a wide variety of symptoms and can affect vision, balance, strength,
sensation, coordination, and bodily functions.

1
Multiple sclerosis affects more than a quarter of a million people in the United States.
Most people have their first symptoms between the ages of 20 and 40; symptoms rarely begin
before 15 or after 60. Women are almost twice as likely to get MS as men, especially in their
early years. People of northern European heritage are more likely to be affected than people of
other racial backgrounds, and MS rates are higher in the United States, Canada, and Northern
Europe than in other parts of the world. MS is very rare among Asians, North and South
American Indians, and Eskimos.
Causes and symptoms
Multiple sclerosis is an autoimmune disease, meaning its cause is an attack by the
body's own immune system. For unknown reasons, immune cells attack and destroy the myelin
sheath that insulates neurons in the brain.
The symptoms of multiple sclerosis may occur in one of three patterns:

 The most common pattern is the "relapsing-remitting" pattern, in which there are
clearly defined symptomatic attacks lasting 24 hours or more, followed by complete or
almost complete improvement.
 The period between attacks may be a year or more at the beginning of the disease, but
may shrink to several months later on. This pattern is especially common in younger
people who develop MS.
 In the "primary progressive" pattern, the disease progresses without remission or with
occasional plateaus or slight improvements. This pattern is more common in older
people.
 In the "secondary progressive" pattern, the person with MS begins with relapses and
remissions, followed by more steady progression of symptoms.

Between 10-20% of people have a benign type of MS, meaning their symptoms progress very
little over the course of their lives.
Because plaques may form in any part of the central nervous system, the symptoms of MS vary
widely from person-to-person and from stage-to-stage of the disease. Initial symptoms often
include:Muscle weakness, causing difficulty walking

 Loss of coordination or balance

2
 Numbness, "pins and needles," or other abnormal sensations
 Visual disturbances, including blurred or double vision.

Voxel-wise Bayesian classifier

The Bayesian Classification represents a supervised learning method as


well as a statistical method for classification. Assumes an underlying probabilistic
model and it allows us to capture uncertainty about the model in a principled way
by determining probabilities of the outcomes. It can solve diagnostic and predictive
problems.
Bayesian Classification provides a useful perspective for understanding
and evaluating many learning algorithms. It calculates explicit probabilities for
hypothesis and it is robust to noise in input data
Assumption: training set consists of instances of different classes described
conjunctions of attributes values
Task: Classify a new instance d based on a tuple of attribute values into one of
the classes.The Bayesian classifier that uses the Naïve Bayes assumption and
computes the MAP hypothesis is called Naïve Bayes classifier. One of the most
practical learning methods

Properties of Bayesian classifier

 Incrementality: with each training example, the prior and the


likelihood can be updated dynamically: flexible and robust to errors.

3
 Combines prior knowledge and observed data: prior probability of
a hypothesis multiplied with probability of the hypothesis given the
training data
 Probabilistic hypothesis: outputs not only a classification, but a
probability distribution over all classes

Applications

 Medical Diagnosis
 Text classification

Lesion level Random classifier

In brain MRI this classifier is used to detecting new lesions that is


injuries, extra cells ,also called as inflammatory diseases.Here, Markov Random
Field (MRF)is going to use.

Markov Random Field?(MRF)

A set of points is a clique if all its members are neighbours. A potential


is a nearest neighbor potential if VA(z)=0 whenever A is not a clique.

Brain lesions distribution

4
Figure 1.1: Multiple sclerosis in Brain MRI

Multiple Sclerosis(MS) is considered a disease of the white matter


because normally lesions appear in this area but,it is also possible to find some of
them in the gray matter.[ It is a progressive disease characterized by disseminated
demyelination of nerve fibers of the brain and spinal cord. It begins slowly, usually
in young adulthood, and continues throughout life with periods of exacerbation and
remission. The first signs are often paresthesias, or abnormal sensations in the
extremities or on one side of the face. Other early signs are muscle weakness,
vertigo, and visual disturbances, such as nystagmus, diplopia (double vision), and
partial blindness. Later in the course of the disease there may be extreme emotional
lability, ataxia, abnormal reflexes, and difficulty in urinating.

A history of exacerbation and remission of symptoms and the presence


of greater than normal amounts of protein in cerebrospinal fluid are characteristic.
Most of the brain and spinal cord will show characteristic lesions. As the disease
progresses, the intervals between exacerbations grow shorter and disability
becomes greater. Treatment involves drugs that affect the function of the immune
system; acute episodes, also called exacerbations, are often treated with

5
corticosteroids. Physical therapy may help postpone or prevent specific disabilities.
The patient is encouraged to live as normal and active a life as possible.

1.2 LITERATURE SURVEY

Cohn Elliott, Dougles L.Arnold, D.LouisCollius,Tal Arbel gives brief


information about Automated segmentation of multiple sclerosis lesions in brain
MRI by using I)Shochastic model.(ii)Markov model to increase the load
correlation but, less accuracy [1].

Petronella Anbek, H.c.Bisschops and Jeroen Vander Grand suggested that


the probabilistic segmentation of white matter lesions in MR imaging by using K-
Nearest Neighbor method to increase the accuracy but, the computational time is
increased [2].

Baseem A.Abdullah, Mel-Ling Shyu, Terri A.Scanduna, Nigel John


gives brief details about the segmentation of Multiple Sclerosis Lesions in brain
MRI is used to provide accurate edges but, the design is complex [3].

Anjum Hayet, Gandal, Ahmad Khan gives brief information about a


review of automated techniques for brain tumor detection from MR images for an
Identification of brain tissues and detect the diseases.It used to provide high
sensitivity [4].

M.Eifadal, J.Zhang, H.Denn suggested that the abnormality detection of


brain MR image segmentation by using Iterative Conditional Mode Algorithm
but,it gives less accuracy[5].

6
Muhamad Abdel.Mottaleb, Baseem A.Scandura contributed that a
topology-preserving approach to the segmentation of brain images with multiple
sclerosis lesions by using T2-Weighted techniques but, it have computational
complexity [6].

V.B.Padole, D.S.Chaudhari, K.K.Mektha gives brief description about


brain MAI Segmentation by using T1 and T2 weighted spatially constrained Fuzzy
C-means clustering to provide Effective performance and robust to noise [7] .

Michel J.Berg,Sken Ebholm,Edward A.Ashton suggested that the


accuracy and reproducability of manual and semiautomated quantification of Ms
lesions by MRI Geometrically Constrained Region Growth Method. It required
large time for computation [8].

Mareel Bose,Fabrico Hertz,Jem-Paul gives brief idea about automated


change detection in muiltimodal serial MRI:application to MS lesions evolution
by using joint histogram equalization method. It used to reduce the fault but,
sensitivity is less [9].

Reza Forghanl,Alan C.Evans gives brief information about the


automation “Pipeline” Analysis of 3-D MRI Data for clinical Trials:Application to
MS is used enhance to detect small treatment effects [10].

A.Criminist,J.Sholton,A.Blake Los Angeles suggested that the


discriminative,Semantic Segmentation of brain tissue in MR images by using
random decision forest classifier. It gives accurate classification but, CSF is moat
difficult to label accurate [11].

7
Guido Gorig, Daniel Wolti,Alan Colchester gives brief information about
exploring the discrimination power of the time domain for segmentation and
characterization of lesions in MRI data by using spatial analysis .It provide high
sensitive to detect fluctuation structures [12].

Stevan L.Hartmann, Mitchell H.Parks,Benoit.M.Dswart gives idea about


the automatic 3-D segmentation of internal structures in brain MR Images is
difficult to find the regions [13].

Fernand.S.Cohen, Zhengweiyang, Jonathan Nissanov suggested that the


automatic Matching of homologous histological Sections for identification and
curve matching, robust to non- uniform sampling noise [14].

1.3 PROBLEM FORMULATION

 Need both reference and follow up scans to an identify inflammations and to


gives difference images.
 Fault detection rate is reduced in T1 and T2-weighted techniques
 The single point and ground truth approaches are to provide less
performance and less accuracy.
 It required several time points to detect an inflammations.

Phase 1

In the phase 1 two stage of classification for detecting an inflammatory


disease in brain MRI.In medical image field that is useful for surgical purpose and
disease activity.
8
1.4 OBJECTIVE

The main objective of this project is the classification of tissues for


detecting an inflammatory disease in brain MRI by using two stage of classifier
that is,

i) Voxel wise Bayesian classifier.


ii) Lesion level Random forest classifier.

It used to increase the accuracy, performance, sensitivity and also reduce


the computational time.

1.5 ORGANIZATION OF THE REPORT

Chapter 1 gives the introduction about the project report and also gives
the objective of the project with literature survey and organization of the project.

The chapter deals with design methodology and its implementation using
MATLAB.

Background information relevant to understanding the problem, its


context, and the proposed solution is presented in Chapter 3.

First, an overview of the thesis is explained using the detailed study on


literature survey. Second, the various comparative performance of both Bayesian
and Lesion –level classification has been explained. Finally, conclusion ad future
work content are presented.

9
CHAPTER 2

SOFTWARE DESDRIPTION

2. MATLAB

2.1 INTRODUCTION

MATLAB is Short form of "MATrix LABoratory", MATLAB was


invented in the late 1970s by Cleve Molar, then chairman of the computer science
department at the University of New Mexico. He designed it to give his students
access to LINPACK and EISPACK without having to learn Fortran. It soon spread
to other universities and found a strong audience within the applied mathematics
community. Jack Little, an engineer, was exposed to it during a visit Moler made
to Stanford University in 1983. Recognizing its commercial potential, he joined
with Moler and Steve Bangert. They rewrote MATLAB in C and founded The
MathWorks in 1984 to continue its development. These rewritten libraries were
known as JACKPAC.

MATLAB was first adopted by control design engineers, Little


specialty, but quickly spread to many other domains. It is now also used in
10
education, in particular the teaching of linear algebra and numerical analysis, and
is the defacto choice for scientists involved with image processing.

2.2 SYNTAX

MATLAB is built around the MATLAB language, sometimes called M-


code or simply M. The simplest way to execute M-code is to type it in at the

prompt, >> in the Command Window, one of the elements of the MATLAB
Desktop. In this way, MATLAB can be used as an interactive mathematical shell.

Sequences of commands can be saved in a text file, typically using the


MATLAB Editor, as a script or encapsulated into a function, extending the
commands available.In many other languages, the semicolon is required to
terminate commands. In MATLAB the semicolon is optional. If a statement is not
terminated with a semicolon, then the result of the statement is displayed.

Variables

Variables are defined with the assignment operator, =. MATLAB is


dynamically typed, meaning that variables can be assigned without declaring their
type, and that their type can change. Values can come from constants, from
computation involving values of other variables, or from the output of a function.

Vectors / matrices

11
MATLAB is the "Matrix Laboratory", and so provides many convenient
ways for creating matrices of various dimensions. In the MATLAB vernacular, a
vector refers to a one dimensional (1×N or N×1) matrix, commonly referred to as
an array in other programming languages. A matrix generally refers to a multi-
dimensional matrix, that is, a matrix with more than one dimension, for instance,
an N×M, an N×M×L, etc., where N, M, and L are greater than 1. In other
languages, such a matrix might be referred to as an array of arrays, or array of
arrays of arrays, etc. Matrices can be defined by separating the elements of a row

with blank space or comma and using a semicolon to terminate each row. The list
of elements should be surrounded by square brackets [].

2.3 MATLAB IMAGE PROCESSING

MATLAB provides a suitable environment for image processing.


Although MATLAB is slower than some languages (such as C), its built in
functions and syntax makes it a more versatile and faster programming
environment for image processing. Once an algorithm is finalized in MATLAB,
the programmer can change it to C (or another faster language) to make the
program run faster.MATLAB does not have the easy to use interfaces of Adobe
Photoshop. MATLAB is used to test and tweak new image processing techniques
and algorithms. Almost everything in MATLAB is done through programming and
manipulation of raw image data and not a user interface. The effects and filters in

12
Photoshop (or any other image editing software) are actually algorithms. With
MATLAB, the user can create these complex algorithms that are applied in
Photoshop.

Image matrices

MATLAB handles images as matrices. This involves breaking each pixel


of an image down into the elements of a matrix. MATLAB distinguishes between
color and grayscale images and therefore their resulting image matrices differ
slightly.

Color images

A color is a composite of some basic colors. MATLAB therefore breaks


each individual pixel of a color image (termed ‘true color’) down into Red, Green
and Blue (RGB) values. What we get as a result, for the entire image, are 3
matrices, one representing each color.

The three matrices are stacked next to each other creating a 3 dimensional
m by n by 3 matrix. For an image which has a height of 5 pixels and width of 10
pixels the resulting in MATLAB would be a 5 by 10 by 3 matrix for a true color
image.

Gray scale images

A grayscale image is a mixture of black and white colors. These colors,


or as some may term as ‘shades’, are not composed of Red, Green or Blue colors.
But instead they contain various increments of colors between white and black.

13
Therefore to represent this one range, only one color channel is needed.Thus we
only need a 2 dimensional matrix, m by n by 1. MATLAB terms this type of
matrix as an Intensity Matrix, because the values of such a matrix represent
intensities of one color.

Color maps

MATLAB also allows the use of colormaps. Basically a colormap is a m


by 3 matrix representation of all the possible colors in the image/matrix.

Modifying, creating or applying one of MATLAB’s predefined colormaps will


effectively change the color range of an image.

Pixel values

MATLAB, by default, will use integer values (which MATLAB terms as


uint8) that have a range of integers from 0 to 255 to represent a pixel value. 0
stands for the lightest color and 255 stands for the darkest color possible. This
applies to the RGB (Red Green Blue) color channels. In the case of the Red
channel, lower numbers will produce lighter (pink) values for red, and higher
numbers near 255 will produce darker (maroon) values of red.

For the intensity matrix (m by n by 1) this scale applies for colors


between white and black (or depends on the colormap being used).The second type
of pixel values used for images are called double which are floating point
(decimal) numbers between 0 and 1. This range is proportional to the uint8 range

14
and therefore multiplying each double pixel value by 255 will yield a uint8 pixel
value. Similarly conversion from uint8 to double is done by dividing the uint8
value by 255.

MATLAB does have casting functions unit 8 () and double (). But these
only change the data type and do not scale the values. Scaling must be done
manually. The reason MATLAB has two formats is because uint8 values take less
storage. But in many older versions of MATLAB (version 6.0) direct arithmetic
operations on uint8 values is not possible because of accuracy issues. Therefore to
perform arithmetic operations, the pixel values must be converted

to double first. In version 2006a, this is not an issue as MATLAB simply changes
uint8 to double first, does the operations, and then changes the values back to
uint8.

Image processing toolbox

MATLAB does have extensions which contain functions geared towards


image processing. They are grouped under the ‘Image Processing Toolbox’. In
some versions it is an optional extra which users have to pay for, while in others it
comes packaged with the software.

This toolbox is especially helpful for applying numerous filters (such as


linear and deblur) and also includes algorithms which can detect lines and edges in
an image. A programmer can write almost all the features in this extension.
Otherwise the command ‘edit comamndName” usually allows a user to see/modify

15
lines of the built-in functions that MATLAB and its Image Processing Toolbox
provide.

Basic tasks

Converting a color image to graycale

1. Since we know that a RGB is a m-n-3 matrix and a grayscale image is a m-


n-1 matrix we can take label our new grayscale intensity matrix as either one
of the red, green or blue channel. In this way we are accounting for only one
channel values for our grayscale image.

2. Another common way is to combine elements of the red, green and blue
channel with a weight by using, weightR * RedChannel +
weightG*GreenChannel + weightB*BlueChannel. Where, the weights,
weightR + weightG + weightB = 1. This method allows us to mix each
channel selectively to get a grayscale image.
3. More recent versions of MATLAB have a function called rgb2gray() -which
turns an RGB matrix to a grayscale matrix. This function is an
implementation of the option listed above.

Working formats in matlab

If an image is stored as a JPEG-image on your disc we first read it into


MATLAB. However, in order to start working with an image, for example perform

16
a wavelet transform on the image, we must convert it into a different format. This
section explains four common formats.

Intensity image
This is the equivalent to a “gray scale image” and this is the image we will
mostly work with in this course. It represents an image as a matrix where every
element has a value corresponding to how bright or dark the pixel at the
corresponding position should be colored. There are two ways to represent the
number that represents the brightness of the pixel.The double class.This assigns a
floating number between 0 and 1 to each pixel. The value 0 corresponds to black
and the value 1 corresponds to white. The other class is called uint8 which assigns
an integer between 0 and 255 to represent the brightness of a pixel.

The value 0 corresponds to black and 255 to white. The class uint8 only
requires roughly 1/8 of the storage compared to the class double. On the other
hand, many mathematical functions can only be applied to the double class. We
will see later how to convert between double and uint8.

Binary image

This image format also stores an image as a matrix but can only color a
pixel black or white . It assigns a 0 for black and a 1 for white.

Indexed image

17
This is a practical way of representing color images. In this course we
will mostly work with gray scale images but once you have learned how to work
with a gray scale image you will also know the principle how to work with color
images. An indexed image stores an image as two matrices. The first matrix has
the same size as the image and one number for each pixel.

The second matrix is called the color map and its size may be different
from the image. The numbers in the first matrix is an instruction of what number to
use in the color map matrix.
RGB image

This is another format for color images. It represents an image with three
matrices of sizes matching the image format. Each matrix corresponds to one of
the colors red, green or blue and gives an instruction of how much of each of these
colors a certain pixel should use.

Multiframe image

In some applications we want to study a sequence of images. This is


very common in biological and medical imaging where you might study a
sequence of slices of a cell. For these cases, the multiframe format is a convenient
way of working with a sequence of images. In case you choose to work with
biological imaging later on in this course, you may use this format.

Fundamentals

A digital image is composed of pixels which can be thought of as small dots


on the screen. A digital image is an instruction of how to color each pixel. We will
18
see in detail later on how this is done in practice. A typical size of an image is 512-
by-512 pixels. Later on in the course you will see that it is convenient to let the
dimensions of the image to be a power of 2. For example, 2 9=512. In the general
case we say that an image is of size m-by-n if it is composed of m pixels in the
vertical direction and n pixels in the horizontal direction. Let us say that we have
an image on the format 512-by-1024 pixels. This means that the data for the image
must contain information about 524288 pixels, which requires a lot of memory!
Hence, compressing images is essential for efficient image processing. You will
later on see how Fourier analysis and Wavelet analysis can help us to compress an
image significantly. There are also a few “computer scientific” tricks to reduce the
amount of data required to store an image. There are many different data types, or
classes, that you can work with in the MATLAB software.

You can build matrices and arrays of floating-point and integer data,
characters and strings, and logical true and false states. Function handles connect
your code with any MATLAB function regardless of the current scope. Structures
and cell arrays, provide a way to store dissimilar types of data in the same array.
There are 15 fundamental classes in MATLAB. Each of these classes is in the form
of a matrix or array. With the exception of function handles, this matrix or array is
a minimum of 0-by-0 in size and can grow to an n-dimensional array of any size. A
function handle is always scalar one by one.Here are a couple of basic MATLAB
commands which do not require any tool box for displaying an image.

Displaying an image given on matrix form

19
MATLAB
Operation:
command:

Display an image represented as the imagesc(X)


matrix X.

Adjust the brightness. s is a parameter


such that
brighten(s)
-1<s<0 gives a darker image, 0<s<1 gives
a brighter image.

Change the colors to gray. colormap(gray)

Table-2.1: Displaying an image in given matrix form

Sometimes your image may not be displayed in gray scale even though
you might have converted it into a gray scale image. You can then use the
command colormap to “force” MATLAB to use a gray scale when displaying an
image.

20
CHAPTER 3

1.2 IMAGE PROCESSING

Vision allows humans to perceive and understand the world surrounding us.
Computer vision aims to duplicate the effect of human vision by electronically perceiving and
understanding an image. Giving computers the ability to see is not an easy task - we live in a
three dimensional (3D) world, and when computers try to analyze objects in 3D space, available
visual sensors (e.g., TV cameras) usually give two dimensional (2D) images, and this projection
to a lower number of dimensions incurs an enormous loss of information. In order to simplify the
task of computer vision understanding, two levels are usually distinguished; low level image
processing and high level image understanding. Low level methods usually use very little
knowledge about the content of images. High level processing is based on knowledge, goals, and

21
plans of how to achieve those goals. Artificial intelligence (AI) methods are used in many cases.
High level computer vision tries to imitate human cognition and the ability to make decisions
according to the information contained in the image.

3.2 IMAGE REPRESENTATION

 GRAY
 RGB
 BINARY
 2D
 3D

3.3 IMAGE FILE FORMATS:

 BMP – Bit map file format


 JPG – jpeg
 TIF – Tagged image file format
 PNG -portable network graphics
 PGM -portable gray map

3.4 CATEGORIES OF IMAGE PROCESSING:

Image processing operations can be roughly divided into three major


categories

 Image Compression
 Image Enhancement and Restoration
 Measurement Extraction.

Image compression

22
Image compression is the application of data compression on digital
images. In effect, the objective is to reduce redundancy of the image data in order
to be able to store or transmit data in an efficient form. Image compression can be
lossy or lossless. Lossless compression is sometimes preferred for artificial images
such as technical drawings, icons or comics and also for medical images.

This is because lossy compression methods, especially when used at low


bit rates, introduce compression artifacts. Lossless compression methods may also
be preferred for high value content, such as medical imagery or image scans made
for archival purposes. Lossy methods are especially suitable for natural images
such as photos in applications where minor (sometimes imperceptible) loss of
fidelity is acceptable to achieve a substantial reduction in bit rate.

Methods for lossless image compression are:

 Run-length encoding - used as default method in PCX and as one of possible


in BMP, TGA, TIFF
 Entropy coding
 Adaptive dictionary algorithms such as LZW - used in GIF

Methods for lossy compression:

 Reducing the color space to the most common colors in the image. The
selected colors are specified in the color palette in the header of the
compressed image. Each pixel just references the index of a color in the
color palette. This method can be combined with dithering to avoid
posterization.

23
 Chroma subsampling. This takes advantage of the fact that the eye perceives
brightness more sharply than color, by dropping half or more of the
chrominance information in the image.

 Transform coding. This is the most commonly used method. A Fourier-


related transform such as DCT or the wavelet transform are applied,
followed by quantization and entropy coding.Fractal compression. The best
image quality at a given bit-rate (or compression rate) is the main goal of
image compression.

However, there are other important properties of image compression schemes:

Scalability generally refers to a quality reduction achieved by


manipulation of the bitstream or file (without decompression and re-compression).
Other names for scalability are progressive coding or embedded bitstreams.
Despite its contrary nature, scalability can also be found in lossless codecs, usually
in form of coarse-to-fine pixel scans. Scalability is especially useful for previewing
images while downloading them (e.g. in a web browser) or for providing variable
quality access to e.g. databases. There are several types of scalability:

 Resolution progressive: First encode a lower image resolution; then encode


the difference to higher resolutions.
 Component progressive: First encode grey; then color.

24
Region of interest coding: Certain parts of the image are encoded with higher
quality than others. This can be combined with scalability (encode these parts first,
others later).

Meta information: Compressed data can contain information about the image
which can be used to categorize, search or browse images. Such information can

include color and texture statistics, small preview images and author/copyright
information.

The quality of a compression method is often measured by the Peak


signal-to-noise ratio. It measures the amount of noise introduced through a lossy
compression of the image. However, the subjective judgement of the viewer is also
regarded as an important, perhaps the most important measure.

Image enhancement and restoration

Enhancement of a compressed image may reveal artifacts of the


compression process. Evidence that information has been discarded from the image
may limit its usefulness in a criminal investigation. Image enhancement
techniques can be used to modify the brightness and contrast of an image, to
remove blurriness, and to filter out some of the noise. Using mathematical
equations called algorithms, the computer applies each change to either the whole
image or targets a particular portion of the image.

For example, global contrast enhancement would affect the entire image,
whereas local contrast enhancement would improve the contrast of small details,
such as a face or a license plate on a vehicle. Some algorithms can remove
25
background noise without disrupting key components of the image. Following
image enhancement, measurement extraction is used to gather useful information
from an enhanced image. Image defects which could be caused by the digitization
process or by faults in the imaging set-up (for example, bad lighting) can be
corrected using Image Enhancement techniques.

Measurement extraction

Once the image is in good condition, the Measurement Extraction


operations can be used to obtain useful information from the image. It is also
known as feature extraction. Feature extraction involves simplifying the amount of
resources required to describe a large set of data accurately. When performing
analysis of complex data one of the major problems stems from the number of
variables involved. Analysis with a large number of variables generally requires a
large amount of memory and computation power or a classification algorithm
which overfits the training sample and generalizes poorly to new samples.

3.5 APPLICATIONS OF IMAGE PROCESSING

Application areas

 Photography and printing


 Satellite image processing
 Machine Vision
 Medical image processing
 Face detection, feature detection, face identification
 Microscope image processing

26
 Car barrier detection

The future

Today, we are in the middle of a revolution sparked by rapid progress in


imaging and computer technology. But contrary to common belief, computers are

not as fast as humans in computations related to the analysis or processing of


images.

Biological or human vision is still by far the most powerful and versatile
perceptual mechanism. Scientist Mead notes that “the visual system of a single
human being does more image processing than the entire world’s supply of
supercomputers.” However, some tasks such as image compression, enhancement
and data extraction from paper via technologies such as OMR, OCR and ICR, etc,
can now be performed on desktop computers available today. It should be noted
that the present computer methods, by contrast, could provide answers only to
precisely stated questions that are not ill-defined.

But a ray of hope surely comes from the distributed and parallel
computing paradigms that are expected to boost real-time response for many image
processing solutions. Image processing technology is waiting to address many
unanswered questions. There is every reason to believe that once this technology
achieves a level of competence that even modestly approaches that of human
vision, and at a competitive cost, the imaging applications will virtually explode.

27
CHAPTER 4

PROJECT DESCRIPTION

4.1 OVERVIEW OF THE PROJECT

The histogram equalization is an approach to enhance a given image.


The approach is to design a transformation T(.) such that the gray values in the
output is uniformly distributed in [0, 1].Histogram equalization yields an image
whose pixels are (in theory) uniformly distributed among all gray levels.
Sometimes, this may not be desirable. Instead, we may want a transformation that
yields an output image with a pre-specified histogram. This technique is called
histogram specification.

The Sobel operator is used in image processing, particularly within edge


detection algorithms. Technically, it is a discrete differentiation operator,
computing an approximation of the gradient of the image intensity function. At
each point in the image, the result of the Sobel operator is either the corresponding
gradient vector or the norm of this vector. The Sobel operator is based on
convolving the image with a small, separable, and integer valued filter in
horizontal and vertical direction and is therefore relatively inexpensive in terms of
computations. On the other hand, the gradient approximation which it produces is
relatively crude, in particular for high frequency variations in the image.
28
It use many situations in medical image processing for Brain MRI new
lesion detection that is surgical purpose, diagnosis, tumor detection.

4.2 MODULES DESCRIPTION

 Get training image.

 Preprocessing of the training image.

i) Histogram equalization.

 Voxel-wise Bayesian classification.

 Get test imsge.


 Preprocessing of the test image.
i) Median filtering.
ii) Edge detection using SOBEL.

 Training image/Test image matching

i) Test image rotation for image matching


ii) Less level representation using close up searching method.

iii) Corner detection using convolution method.

 Lesion level random forest classification.

Get the training image

29
In this first module any one of the training image get from one
particular data base. In that image is gray level which have few levels compare to
RGB image.

Preprocessing of the train image

Histogram equalization

Histogram provides a global description of the appearance of the image.


If we consider the gray values in the image as realizations of a random variable R,
with some probability density, histogram provides an approximation to this
probability density. The shape of a histogram provides useful information for
contrast enhancement. The histogram equalization is an approach to enhance a
given image. The approach is to design a transformation T(.) such that the gray
values in the output is uniformly distributed in [0, 1].Let us assume for the moment
that the input image to be enhanced has continuous gray values, with r = 0
representing black and r = 1 representing white. We need to design a gray value
transformation s = T(r), based on the histogram of the input image, which will
enhance the image. In terms of histograms, the output image will have all gray
values in “equal proportion” .This technique is called histogram equalization. This
technique is called histogram equalization.

Voxel-wise bayesian classification

30
The Bayesian Classification represents a supervised learning method as
well as a statistical method for classification. Assumes an underlying probabilistic
model and it allows us to capture uncertainty about the model in a principled way
by determining probabilities of the outcomes. It can solve diagnostic and predictive
problems Bayesian Classification provides a useful perspective for understanding
and evaluating many learning algorithms.

It calculates explicit probabilities for hypothesis and it is robust to noise


in input data. Bayes classifiers are among the most successful known algorithms
for learning to classify text documents.
Bayes classifier employs single words and word pairs as features. It
allocates user utterances into nice, nasty and neutral classes, labelled +1, -1 and 0
respectively.This numerical output drives a simple first-order dynamical system,
whose state represents the simulated emotional state of the experiment's
personification, Ditto the donkey. Bayesian reasoning is applied to decision
making and inferential statistics that deals with probability inference. It is used the
knowledge of prior events to predict future events.
The Bayesian classifier that uses the Naïve Bayes assumption and
computes the MAP hypothesis is called Naïve Bayes classifier One of the most
practical learning methods
Bayesian spam filtering has become a popular mechanism to
distinguish illegitimate spam email from legitimate email (sometimes called "ham"
or "bacn").[4] Many modern mail clients implement Bayesian spam filtering. Users
can also install separate email filtering programs.

31
Figure 4.1: Schematic diagram of Bayesian classifier

Applications

 Medical Diagnosis.
 Text classification.

Get test image

In this module get any one of the test image from one particular data
base.

Preprocessing or the test image

Median filtering

The median filter is a nonlinear digital filtering technique, often used to


remove noise. Median filtering is very widely used in digital image processing
because it preserves edges while removing noise. The median filter is a sliding-
window spatial filter. It replaces the value of the center pixel with the median of
the intensity values in the neighborhood of that pixel.

32
Median filtering is a nonlinear operation often used in image processing
to reduce "salt and pepper" noise. A median filter is more effective than
convolution when the goal is to simultaneously reduce noise and preserve edges.
Median filters are particularly effective in the presence of impulse noise, also
called ‘salt – and – pepper’ noise because of its appearance as white and black dots
superimposed on an image. For every pixel, a 3x3 neighborhood with the pixel as
center is considered. In median filtering, the value of the pixel is replaced by the
median of the pixel values in the 3x3 neighborhood.

 Computes a weighted average of pixels in the window


 Less blurring, less noise cleaning for the same size
 The family of binomial filters can be defined recursively
 The coefficients can be found from (1+x)n

Color median filter

There is no natural ordering in 3D (RGB) color space. Separate


filtering on R,G and B components does not guarantee that the median selects a
true sample from the input window. Vector median filter, defined as the sample
minimizing the sum of absolute deviations from all the samples. Computing the
vector median is very time consuming, although several fast algorithms exist.

Edge detection using sobel

The Sobel operator performs a 2-D spatial gradient measurement on an


image. Typically it is used to find the approximate absolute gradient magnitude at

33
each point in an input grayscale image. The Sobel edge detector uses a pair of 3x3
convolution masks, one estimating the gradient in the x-direction (columns) and
the other estimating the gradient in the y-direction (rows). A convolution mask is
usually much smaller than the actual image. As a result, the mask is slid over the
image, manipulating a square of pixels at a time or the norm of this vector. The
Sobel operator is based on convolving the image with a small, separable, and
integer valued filter n horizontal and vertical direction and is therefore relatively
inexpensive in terms of computations.

On the other hand, the gradient approximation which it produces is


relatively crude, in particular for high frequency variations in the image.

Figure 4.2: Edge detection

The Sobel operator represents a rather inaccurate approximation of the


image gradient, but is still of sufficient quality to be of practical use in many
applications. More precisely, it uses intensity values only in a 3×3 region around
each image point to approximate the corresponding image gradient, and it uses

34
only integer values for the coefficients which weight the image intensities to
produce the gradient approximation.

Train image/test image matching

Test image rotation for image matching

In this module the test image going to rotate by using geometrical rotated
signals for match the position of both train and test image.

Less level representation using close up searching method

In this module test and train image is compare to detect differences b/w
that images.

Corner detection using convolution method

The convolution of two functions is an important concept in a number of


areas of pure and applied mathematics such as Fourier Analysis, Differential
Equations, Approximation Theory, and Image Processing.Nevertheless
convolutions often seem unintuitive and difficult to grasp for beginners. This
project explores the origins of the convolution concept as well as some computer
graphic and physical interpretations of convolution which illustrate various ways
the technique of smoothing can be used to solve some real world problems

35
• New Pixel Value Computed from Neighboring Pixel Values
• Convolution of an N x N Matrix (Kernel) with the Image

Different filters can be formed by applying convolution with the


former equation, we get linear filters, which each is a summation of weighted pixel
intensities and then is divided by a constant value, or weight.

Lesion level random forest classification

Introduction

This section gives a brief overview of random forests and some


comments about the features of the method.We assume that the user knows about

the construction of single classification trees. Random Forests grows many


classification trees. To classify a new object from an input vector, put the input
vector down each of the trees in the forest. Each tree is grown as follows:

1. If the number of cases in the training set is N, sample N cases at random -


but with replacement, from the original data. This sample will be the training
set for growing the tree.
2. If there are M input variables, a number m<<M is specified such that at each
node, m variables are selected at random out of the M and the best split on
these m is used to split the node. The value of m is held constant during the
forest growing.
3. Each tree is grown to the largest extent possible. There is no pruning.

36
In the original paper on random forests, it was shown that the forest error rate
depends on two things:

 The correlation between any two trees in the forest. Increasing the
correlation increases the forest error rate.
 The strength of each individual tree in the forest. A tree with a low error rate
is a strong classifier. Increasing the strength of the individual trees decreases
the forest error rate.

Reducing m reduces both the correlation and the strength. Increasing it increases
both. Somewhere in between is an "optimal" range of m - usually quite wide.
Using the oob error rate (see below) a value of m in the range can quickly be
found.

The convolution of two functions is an important concept in a number of


areas of pure and applied mathematics such as Fourier Analysis, Differential
Equations, Approximation Theory, and Image Processing.Nevertheless
convolutions often seem unintuitive and difficult to grasp for beginners. This
project explores the origins of the convolution concept as well as some computer
graphic and physical interpretations of convolution which illustrate various ways
the technique of smoothing can be used to solve some real world problems

Features of Random Forests

 It is unexcelled in accuracy among current algorithms.


 It runs efficiently on large data bases.
 It can handle thousands of input variables without variable deletion.

37
 It gives estimates of what variables are important in the classification.
 It generates an internal unbiased estimate of the generalization error as the
forest building progresses.
 It has an effective method for estimating missing data and maintains
accuracy when a large proportion of the data are missing.
 It has methods for balancing error in class population unbalanced data sets.
 Generated forests can be saved for future use on other data.
 Prototypes are computed that give information about the relation between
the variables and the classification.
 It computes proximities between pairs of cases that can be used in clustering,
locating outliers, or (by scaling) give interesting views of the data.
 The capabilities of the above can be extended to unlabeled data, leading to
unsupervised clustering, data views and outlier detection.

Remarks

Random forests does not overfit. You can run as many trees as you want.
It is fast. Running on a data set with 50,000 cases and 100 variables, it produced
100 trees in 11 minutes on a 800Mhz machine. For large data sets the major
memory requirement is the storage of the data itself, and three integer arrays with
the same dimensions as the data. If proximities are calculated, storage requirements
grow as the number of cases times the number of trees.

How random forests work

To understand and use the various options, further information about how
they are computed is useful. Most of the options depend on two data objects

38
generated by random forests. When the training set for the current tree is drawn by
sampling with replacement, about one-third of the cases are left out of the sample.
This data is used to get a running unbiased estimate of the classification error as
trees are added to the forest. It is also used to get estimates of variable importance.

After each tree is built, all of the data are run down the tree,
and proximities are computed for each pair of cases. If two cases occupy the same
terminal node, their proximity is increased by one. At the end of the run, the
proximities are normalized by dividing by the number of trees. Proximities are
used in replacing missing data, locating outliers, and producing illuminating low-
dimensional views of the data.

The out-of-bag (oob) error estimate

In random forests, there is no need for cross-validation or a separate


test set to get an unbiased estimate of the test set error. It is estimated internally,
during the run, as follows: Each tree is constructed using a different bootstrap
sample from the original data. About one-third of the cases are left out of the
bootstrap sample and not used in the construction of the kth tree. Put each case left
out in the construction of the kth tree down the kth tree to get a classification. In
this way, a test set classification is obtained for each case in about one-third of the
trees. At the end of the run, take j to be the class that got most of the votes every
time case n was oob. The proportion of times that j is not equal to the true class of

39
n averaged over all cases is the oob error estimate. This has proven to be unbiased
in many tests.

Variable importance

In every tree grown in the forest, put down the oob cases and count the
number of votes cast for the correct class. Now randomly permute the values of
variable m in the oob cases and put these cases down the tree. Subtract the number
of votes for the correct class in the variable-m-permuted oob data from the number
of votes for the correct class in the untouched oob data. The average of this number
over all trees in the forest is the raw importance score for variable m.If the values
of this score from tree to tree are independent, then the standard error can be
computed by a standard computation. The correlations of these scores between
trees have been computed for a number of data sets and proved to be quite low.

therefore we compute standard errors in the classical way, divide the raw score by
its standard error to get a z-score, ands assign a significance level to the z-score
assuming normality. If the number of variables is very large, forests can be run
once with all the variables, then run again using only the most important variables
from the first run.For each case, consider all the trees for which it is oob. Subtract
the percentage of votes for the correct class in the variable-m-permuted oob data
from the percentage of votes for the correct class in the untouched oob data. This is
the local importance score for variable m for this case, and is used in the graphics
program.

Gini importance

40
Every time a split of a node is made on variable m the gini impurity
criterion for the two descendent nodes is less than the parent node. Adding up the
gini decreases for each individual variable over all trees in the forest gives a fast
variable importance that is often very consistent with the permutation importance
measure.

Interactions

The operating definition of interaction used is that variables m and k


interact if a split on one variable, say m, in a tree makes a split on k either
systematically less possible or more possible. The implementation used is based on
the gini values g(m) for each tree in the forest. These are ranked for each tree and
for each two variables, the absolute difference of their ranks are averaged over all
trees.

This number is also computed under the hypothesis that the two variables
are independent of each other and the latter subtracted from the former. A large
positive number implies that a split on one variable inhibits a split on the other and
conversely. This is an experimental procedure whose conclusions need to be
regarded with caution. It has been tested on only a few data sets.

Proximities

These are one of the most useful tools in random forests. The
proximities originally formed a NxN matrix. After a tree is grown, put all of the
data, both training and oob, down the tree. If cases k and n are in the same terminal

41
node increase their proximity by one. At the end, normalize the proximities by
dividing by the number of trees.

Users noted that with large data sets, they could not fit an NxN matrix
into fast memory. A modification reduced the required memory size to NxT where
T is the number of trees in the forest. To speed up the computation-intensive
scaling and iterative missing value replacement, the user is given the option of
retaining only the nrnn largest proximities to each case. When a test set is present,
the proximities of each case in the test set with each case in the training set can
also be computed. The amount of additional computing is moderate.

Missing value replacement for the training set


Random forests has two ways of replacing missing values. The first way is
fast. If the mth variable is not categorical, the method computes the median of all
values of this variable in class j, then it uses this value to replace all missing values

of the mth variable in class j. If the mth variable is categorical, the replacement is
the most frequent non-missing value in class j. These replacement values are called
fills.
The second way of replacing missing values is computationally more
expensive but has given better performance than the first, even with large amounts
of missing data. It replaces missing values only in the training set. It begins by
doing a rough and inaccurate filling in of the missing values. Then it does a forest
run and computes proximities. If x(m,n) is a missing continuous value, estimate its
fill as an average over the non-missing values of the mth variables weighted by the
proximities between the nth case and the non-missing value case. If it is a missing
42
categorical variable, replace it by the most frequent non-missing value where
frequency is weighted by proximity. Now iterate-construct a forest again using
these newly filled in values, find new fills and iterate again. Our experience is that
4-6 iterations are enough. When there is a test set, there are two different methods
of replacement depending on whether labels exist for the test set. If they do, then
the fills derived from the training set are used as replacements. If labels no not
exist, then each case in the test set is replicated nclass times (nclass= number of
classes).
The first replicate of a case is assumed to be class 1 and the class one fills
used to replace missing values. The 2nd replicate is assumed class 2 and the class 2
fills used on it.This augmented test set is run down the tree. In each set of
replicates, the one receiving the most votes determines the class of the original
case

4.3 FLOW DIAGRAM

Train image or
data set

Preprocessing of
the train image

Voxel-wise Bayesian
classifier for brain
tissues

43
Lesion level
Preprocess
Test Random forest
image Test image
classifier

New lesion
image

Figure 4.3: Flow diagram

4.4 INPUT DESIGN

Input design is the process of converting user-oriented input to a


computer-based format. Input design is a part of the overall system design, which
requires very careful attention. Input design must be in such a way that it must
have the Input Images, Gray level Image and etc. It must be simple. The input
deign must ensure user-friendly screen, with simplicity, providing ease of viewing
& entering the data. The main objective of designing input focuses on
 Input Images
 Gray Level Images

 Color Transformation

44
Ensure that the input is acceptable to and understood by the staff. The goal of
designing input data is to make entry easy, logical and free from errors as possible.
The entering data entry operators need to know the allocated space for each field,
field sequence and which must match with that in the source document. The
processor analyzes the input required. It is then accepted or rejected.

Input types can be categorized as


 Figure

 Gray Level Images


4.5 OUTPUT DESIGN

The normal procedure in developing a system is to design the output in


detail first and then move back to the input. The output will be in the form of views

and reports. The output from the system is required to communicate the result of
processing to the users. They are also used as the permanent copy for later
verifications.

45
CHAPTER 5

PROJECT IMPLEMENTATION

To take gray level brain MRI image because the gray scale image
having few levels (0-255) compare to RGB image. In that gray level image going
to apply histogram equalization. Histogram equalization I which is used to
distribute or equalize the pixels to provide the contrast image. Then, three types of
brain tissues are there that is white matter, gray matter, cerebro spial fluid(CSF),the
Bayesian classifier is used to classify the brain tissue. Then, Lesion level Random
classifier is applied to detect the difference between reference and follow image
for detecting an inflammatory disease that is called injuries, unwanted cells.

46
We apply two stage of classifier so, that it gives high performance and
accuracy. Here, Bayesian classifier is used to classify the Brain tissues and
Random forest classifier is used to detect the new lesions or an inflammations.

5 RESULTS AND DISCUSSIONS

5.1GET TRAINING IMAGE

47
Figure 5.1: Training image

In this module brain MRI training image is a gray level image because
the gray level image have few levels (0-255) compare to RGB image.

5.2 PRE PROCESSING OF TRAINING IMAGE

HISTOGRAM EQUALIZATION

48
Figure 5.2: Visualize the histogram equalization

To visualize the histogram equalization is the distribution or equalization


of the pixels to provide brightness and contrast image.

AFTER THE HISTOGRAM EQUALIZATION

49
Figure 5.3: After the histogram equalization

The histogram equalization is done in this module. After the equalization


the brightness level of all pixels become flat compare to the input image.

5.3 VOXEL-WISE BAYESIAN CLASSIFIER FOR TRAINING IMAGE

50
Figure 5.4: Voxel-wise Bayesian classification

In this module the Brain tissues are classified by using Bayesian


classifier. There are three types that is i)white matter, Ii)gray matter, iii)CSF.

54 5GET TEST IMAGE

51
Figure 5.5: Test image

In this module going to take anyone of the test image from the data set.

5.5PREPROCESSING OF TEST IMAGE

MEWDIAN FILTERING

52
Figure 5.6: After the median filtering the test image

In this module the median filter is applied to the test image for replacing
the pixels and reduce the noise. It used to provide quality of image.

5.6 EDGE DETECTION OF TEST IMAGE USING SOBEL

53
Figure 5.7: Edge detecting of test image

In this module the edge detection is done by using SOBEL. It is used to


cover the particular area of the test image.

5.7 TRAINING IMAGE/TEST IMAGE MATCHING

TRAINING IMAGE

54
Figure 5.8: Training image

5.8 ROTATED TEST IMAGE FOR IMAGE MATCHING

55
Figure 5.9: Rotated test image

In this module the test image rotation is done by using geometrical


rotated signal. It is used to determine the differences between both train and test
image accurately.

5.9 LESS LEVEL PERRESENTATION FOR DETECT DIFFERENCE

56
Figure 5.10: Less level representation

5.10 CORNER DETECTION USING CONVOLUTION METHOD

57
Figure 5.11: Corner detection

5.11 RANDOM FOREST CLASSIFIER

58
Figure 6.12: Detect an inflammations in Brain image

5.12OUTPUT IMAGE

INFLAMMATION DETECTION IN TEST IMAGE

59
Figure 5.13: Inflammation detection

In this module the new lesions or an inflammation disease detected by


using compare both train and test image.

CHAPTER 6

CONCLUSION AND FUTURE ENHANCEMENT


60
6.1 CONCLUSION

I presented a framework for automated detection of new MS lesions


using a two-stage classifier that first performed a joint Bayesian classification of
tissue classes at each voxel of reference and follow-up scans using intensities and
intensity differences, and then performed a lesion-level classification using a
random forest classifier. The new lesion classifier allows for trade-off of sensitivity
and specificity through the use of a user-specified confidence threshold (or target
sensitivity). Sample points of operation showed that our classifier is able to detect
new lesions as small as three voxels with a sensitivity and false detection rate and
a sensitivity with false detection rate as compared to a reference standard.
Comparisons to manual identification of new lesions using only sequential FLAIR
scans showed better performance than any individual expert rater and comparable
performance to consensus segmentation combining manual identification of new
lesion from nine independent raters.
6.2 FUTURE ENHANCEMENT
 Both classification of tissues and detection of new lesions in brain MRI by
using Principal Component Analysis Method.

 It gives fast detection and high accuracy.


 It can use VHDL code.

REFERENCES

[1]Cohn Elliott,Dougles,Arnold,D.Lous Colliius,Member,IEEE and Tal


Arbal”Temporally Consistene Prabablistic Detection of New Multiple

61
Sclerosis Lesions in Brain MRI”Medical imaging,vol 32,no.8,PP,1345-
1357,2013.

[2] Y. Duan, P. Hildenbrand, M. Sampat, D. Tate, I. Csapo, B. Moraal, R. Bakshi,


F. Barkhof, D. Meier, and C. Guttmann, “Segmentation of subtraction images for
the measurement of lesion change in multiple sclerosis,” Am. J. Neuroradiol., vol.
29, no. 2, pp. 340–346, 2008.

[3] B.Moraal et al., “SubtractionMR images in a multiple sclerosis multicenter


clinical trial setting1,” Radiology, vol. 250, no. 2, pp. 506–514, 2009.
[4] A. Evans, J. Frank, J. Antel, and D. Miller, “The role of MRI in clinical trials of
multiple sclerosis: Comparison of image processing techniques,” Ann. Neurol.,
vol. 41, no. 1, pp. 125–132, 1997.
[5] P.Molyneux, P. Tofts,A. Fletcher, B.Gunn, P. Robinson,H.Gallagher,
I.Moseley, G. Barker, and D. Miller, “Precision and reliability formeasurement of
change inMRI lesion volume inmultiple sclerosis: A comparison of two computer
assisted techniques,” J. Neurol., Neurosurg.
Psychiatry, vol. 65, no. 1, pp. 42–47, 1998.
[6] C. Guttmann et al., “Quantitative follow-up of patients with multiple sclerosis
using MRI: Reproducibility,” J. Magn. Reson. Imag., vol. 9, no. 4, pp. 509–518,
1999.

[7] P. Molyneux, D. Miller, M. Filippi, T. Yousry, E. Radü, H. Ader, and F.


Barkhof, “Visual analysis of serial t2-weighted MRI in multiple sclerosis: Intra-

62
and interobserver reproducibility,” Neuroradiology, vol. 41, no. 12, pp. 882–888,
1999.
[8] E. Ashton, C. Takahashi, M. Berg, A. Goodman, S. Totterman, and S. Ekholm,
“Accuracy and reproducibility of manual and semiautomated quantification of MS
lesions by MRI,” J. Magn. Reson. Imag., vol. 17, no. 3, pp. 300–308, 2003.
[9] B.Moraal et al., “Long-interval t2-weighted subtractionmagnetic resonance
imaging: A powerful new outcome measure in multiple sclerosis trials,” Ann.
Neurol., vol. 67, no. 5, pp. 667–675, 2010.
[10] M. P. Sormani, B. Stubinski, P. Cornelisse, S. Rocak, D. Li, and N. D.
Stefano, “Magnetic resonance active lesions as individual-level surrogate for
relapses in multiple sclerosis,” Multiple Sclerosis J., vol. 17, no. 5, pp. 541–549,
2011.
[11] K. Van Leemput, F. Maes, D. Vandermeulen, A. Colchester, and P. Suetens,
“Automated segmentation of multiple sclerosis lesions by model outlier detection,”
IEEE Trans. Med. Imag., vol. 20, no. 8, pp. 677–688, Aug. 2001.
[12] X. Wei, S. K. Warfield, K. H. Zou, Y. Wu, X. Li, A. Guimond, J. P. Mugler,
R. R. Benson, L.Wolfson, H. L.Weiner, and C. R. Guttmann, “Quantitative
analysis of MRI signal abnormalities of brain white matter with high
reproducibility and accuracy,” J. Magn. Reson. Imag., vol. 15, no. 2, pp. 203–209,
2002.

63
[13] M. Bosc, F. Heitz, J.-P. Armspach, I. Namer, D. Gounot, and L. Rumbach,
“Automatic change detection in multimodal serial MRI: Application to multiple
sclerosis lesion evolution,” NeuroImage, vol. 20, no. 2, pp. 643–656, 2003.
[14] A. Zijdenbos, R. Forghani, and A. Evans, “Automatic “pipeline” analysis of
3-d MRI data for clinical trials: Application to multiple sclerosis,” IEEE Trans.
Med. Imag., vol. 21, no. 10, pp. 1280–1291, Oct.2002.

64
65
66