LABORATORY MODULE

UEMB 3244
MEDICAL IMAGING

LAB 3:
Introduction to Medical Image Processing and Analysis II – Voxel-
based Registration

NAME : ____________________________________________

STUDENT ID : ____________________________________________

DEPARTMENT OF MECHATRONIC AND BIOMEDICAL ENGINEERING

LEE KONG CHIAN FACULTY OF ENGINEERING AND SCIENCE
 Universiti Tunku Abdul Rahman

Faculty Lee Kong Chian Faculty of Engineering and

Science
Department: Department of Mechatronics and BioMedical
Engineering
Unit Code and Name UEMB3244 Medical Imaging
Experiment No.: 3
Title of Experiment: Introduction to Medical Image Processing and
Analysis II – Voxel-based Registration
Laboratory Room No. and Name: Any computer lab with Matlab image processing
toolbox
Experiment Duration (hour): 3 hours
Number of Students per Group 1 student

Objectives (Optional)

Introduction

Equipment and Materials

Quantity estimation
Item Description *Item category (e.g. per set/group of student)
MATLAB with image processing toolbox S 1 per student

*Item category
SP Sample or specimen
C Consumable
CH Chemical
W Labware, glassware, tool, and
components
E Equipment
S Software

Caution Statement (Optional)

Procedure

Questions & Report Writing

2
 UNIVERSITI TUNKU ABDUL RAHMAN (UTAR)

LABORATORY

INTRODUCTION TO IMAGE PROCESSING AND ANALYSIS

OBJECTIVES

• To understand the basic concept of image processing and analysis.

• To apply the concept of image registration to align images.
• To explain the principles of different image registration techniques.

SOFTWARE AND EQUIPMENTS

• Matlab with image processing toolbox.

• A set of sampled images and codes can be downloaded from the links at the end of this
document.

INTRODUCTION

This lab is designed to introduce image processing and analysis, specified focus will be on
different image registration techniques.

In the lab, the students will work on the main concepts of voxel-based image registration.
Throughout this practical, the students need to align two images, which are referred to as the
target image (which remains fixed throughout the process) and the source image.

In order to maintain a reasonable computational complexity in Matlab, and to facilitate

visualizing the results, all the experiments will be carried out in two dimensional space. However,
all the algorithms used here can be extended to n-dimensional space.

INSTRUCTIONS

i. Open Matlab and create a directory (folder) named MI_Lab.

ii. Download the images and Matlab codes from the links at the end of this document.
iii. In order to use the downloaded images and codes in the MI_Lab folder, the students have
to first add the path of MI_Lab to the Matlab working or search directory.
Go to “Set Path” -> “Add with Subfolders” -> select the location of the MI_Lab
folder and click the “Save” button.
iv. If the students are unsure how a command works, try ‘help’ or ‘doc’ followed by the name
of the command.

3
VOXEL-BASED REGISTRATION

Voxel similarity based registration algorithms generally consist of the calculation of the
geometrical transformation which, when applied to the source image, maximizes a similarity
function. This similarity function estimates how similar the images are once the transformation
has been applied. There are several possible similarity functions; two of them, namely sum of
squared differences (SSD) and normalized mutual information (NMI) will be studied in this
practical. In order to focus on similarity values, the transformation will be restricted to a simple
translation and an optimizer will not be used in this part of the practical.

i. Use imread () to read in the images, brain_t1.tif and brain_t1_trans.tif, and display them
as the above (using equal axes and gray colormap). Find the difference of images by
subtracting one from the other to appreciate that the two images are identical except for
a translation.

ii. Use the provided functions in the ‘registration’ folder, translate_image () and
mean_square_difference () to implement a “brute-force” approach to register the
images, by calculating the mean square difference measure for a matrix of possible x and
y translation values, and then calculating the position of the minimum of the matrix.
NOTE: the input image of translate_image function should be in double format.

Write a function which does this: save the similarity values in an array called sim_values
and show them using the surf () function.

❖ Include your algorithm and codes below (only the important part):

iii. Is there a clear minimum in the similarity values? The presence of a clear, global minimum
will make it easier for a minimization algorithm to work. Does the minimum correspond
to the correct translation? Check this by inspecting the difference between the two
aligned images.

❖ Insert below the difference images before and after registration, and the surface (surf):

4
 iv. Open and display now once more the image brain_t2.tif. Apply the same voxel-similarity
registration program you just wrote to try to align brain_t2 and brain_t1_trans.

❖ Does it work? Why? Or if not, why not?

v. Try now an alternative similarity function: mutual_information(). The function returns the
negative of the actual mutual information value so that the minimum of the returned
values can be used to register the T1/T2 image pair. Explain and include your registration
result below.

vi. As a final task, substitute the brute-force approach (which is obviously very inefficient)
with a proper optimization algorithm or any other minimization routine such as
fminserach in Matlab. Apply this to any of the question above and include the registration
result below. Briefly discuss the advantages of this new method when compared to the
brute-force approach.
(Note: please make sure you understand how to do this because you need to use

5
 fminserach to fit the data in the assignment.)

================================== THE END ===================================

Reference: Image Registration Practical by Dr Schnabel.

Download the images and codes from the following links:

images – images
registration – codes

EXTRA

In order to make this lab to become more practical and interesting, please download the
Parkinson’s Disease data from this link: Extra PD Data ; first scan in 2014 and follow-up scan in
2015.

Use any of the registration technique you learnt from this lab to register these two 3D T1 data.

6
In the lab report, prepare it according to the following format:

-------------------------------------------------------------------------------------------------------------------------------

Font: Times New Roman

Font size: 12
Line spacing: 1.5

INTRODUCTION & METHODS (1 – 2 pages, 20%)

Write a short introduction of voxel-based registration and describe briefly what have been done
in this lab.

RESULTS & DISCUSSION (5 – 7 pages, 70%)

Show the results and explain them and answer the questions in this section. Do not include the
Matlab codes here, unless stated otherwise by the question.

CONCLUSION (1/2 to 1 page, 5%)

Summarize the work done and provide a conclusion for the findings.

OTHERS (references, grammar & format, 5%)

-------------------------------------------------------------------------------------------------------------------------------

The report should not be more than 8 PAGES, all inclusive (including cover page, references and
appendix).

Please upload a pdf copy of your report with the naming of the file as
[FullSurname][FirstLetterofGivenName], e.g. if my name is Tee Yee Kai, the pdf filename is
TeeYK.pdf, to the link provided on WBLE. Due before 5 pm on 28/04/2023.

Plagiarism is strictly prohibited and late submission in the absence of acceptable extenuating
circumstances will be subject to a penalty.

7
Marking Rubrics

Aspects (Weightage) 1 2 3 4 5
Poor Below average Average Good Excellent
Introduction & The background information Correct background Some introductory Introduction and methods are Introduction and methods are
Methods and methods are incorrect. information and methods but information and methods are almost complete; missing complete and well-written;
too little details. provided, but still missing some minor points. provides all necessary
some major points. background principles for the
experiment.
Results Most of the important data/ Some data/ figures/ graphs Most data/ figures/graphs are All data/ figures/ graphs are All data/ figures/graphs are
figures/ graphs are missing. are not provided. provided but with minor provided but with minor neatly presented and with
errors. errors or could still be proper labeling, captions, and
improved. descriptions.
Discussion Incorrect or improbable Least/weak interpretation of Basic interpretation and Almost all of the results have All important trends and data
interpretation about results. the results. discussion of the results. been correctly interpreted and comparisons have been
discussed; only minor interpreted correctly and
improvements are needed. discussed; good
understanding of results is
conveyed.
Conclusion Missing most of the Conclusions regarding some Conclusions regarding most All important conclusions All important conclusions
important points in the of the major points are of the major points are have been drawn, could be have been clearly made.
conclusion. drawn. drawn. better stated.
References Numerous citation and/or Some of the citations and Most of the citations and All citations and bibliography All citations and bibliography
bibliography errors. bibliography are correct, with bibliography are correct, with are shown according to the are shown exactly according
a couple of major errors. a couple of minor errors. guidelines with very minor to the guidelines, with no
inaccuracies. errors.
Grammar and style Numerous grammatical or Many obvious grammatical Some obvious grammatical or Very few grammatical or No grammatical or spelling
spelling errors are present or spelling errors are present spelling errors are present spelling errors can be found errors.
which cause most part(s) of which negatively affect which affect which does not affect the
the report to be understandability of the understandability of the report's quality.
incomprehensible. report. report.
Appearance and Sections out of order, sloppy Some of the sections are out Most sections are in order, All sections in order, All sections in order, well-
formatting formatting. of order, rough formatting. formatting are rough but formatting generally well but formatted, very readable.
readable. could still be improved.