Medical Image Processing - TU
Transcript of Medical Image Processing - TU
BU3 Project Proposal
Medical Image Processing
Group Members
1. Ms.Watcharaporn Sitsawangsopon ID: 5422791509
2. Ms. Maetawee Juladash ID: 5422772905
Advisor: Dr. Bunyarit Uyyanonvara (Associate Professor)
School of Information, Computer and Communication Technology,
Sirindhorn International Institute of Technology,
Thammasat University
Semester 1, Academic Year 2014
Date: December 15, 2014
Table of Contents
1 Introduction ....................................................................................................................... 1
2 Background ....................................................................................................................... 4
3 Objectives ......................................................................................................................... 6
4 Outputs and Expected Benefits ......................................................................................... 6
4.1 Outputs ......................................................................................................................... 6
4.2 Benefits ........................................................................................................................ 6
5 Literature Review ............................................................................................................. 7
6 Methodology ..................................................................................................................... 8
6.1 Approach ...................................................................................................................... 8
6.2 Tools and Techniques ................................................................................................ 12
7 Project Schedule ............................................................................................................. 14
8 References ....................................................................................................................... 15
Statement of Contribution
By submitting this document, all students in the group agree that their contribution in the
project so far, including the preparation of this document, is as follows:
1. Ms.Watcharaporn Sitsawangsopon ID:5422791509 50%
2. Ms. Maetawee Juladash ID:5422772905 50%
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 1
Introduction
In the medical profession of a facial skin, many patients that come with several
problems and skin lesion types, compare with the amount of doctors that less than the patients
in many countries around the world. So the technology nowadays can help the medical
profession and doctors at least to analyse the problem on each patient's face. This can
decrease the situation of doctors and patients to be in face to face, help the doctors and
patients that resident in different countries, and also decrease the problem of ratio between
doctors and patients.
The cause of the problems make use of a computer to aid in maintenance to
detect amount of acne instead of using manual method lead to the developing of algorithm to
use to the calculation for positions of acne on the patient's face and also detect and measuring
amount of acne on the patient's face. The developing of algorithm is not accurate only on acne
area, but it can also perform calculations amount of acne and comparative analysis of the
difference between the recorded results in each time. It use the process and the basic steps
from Image processing[1-3] to improve and apply. In this project was select the processes that
involved significant of Face Detection[4], Blob Detection[5], and Color segmentation[6] to
study the methods and procedures as well as the statistics to help determine the position, area,
and color to increase the efficiency and accuracy of detecting. So the developing of algorithm
to apply to the program and medical profession are taken from a variety of knowledge to
generate new works to develop medical technologies.
Technology that can apply to help in medical profession of facial skin in term
of analyse the problem on each patient's face will be about the image processing and face
detection. So mainly thing that necessary is picture of the patient that shown the problem of
facial skin clearly. We observe from several research or previous studies about the Detection
system, Edge detection, Face detection, and we know that the picture of patient's should be
the same view point as possible to decrease the error that can happen when detect the key
points (eyes, nose, and mouth) on patient's face. These technologies can find and detect some
problem on the patient's face, for example, detect all spots, acne, wrinkles, etc. But it was a
thoroughly method and difficult because of it maybe detect the unnecessary details around the
face of patient’s and also include the background of the picture that use to analyse.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 2
The original images and ground truth images experiment methods of blob
detection.
Position mark image Ground truth image
The images data in the real world position mark of the experiment method of
wrinkles detection.
Original image (1) Mark position image (1)
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 3
Original image (2) Mark position image (2)
Original image (3) Mark position image (3)
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 4
Original image (4) Mark position image (4)
Background
In the globalization, improvement of technology, science, economy, society
and the education provide people to have a better standard of living and lifestyles. Relating to
the development of medical treatment facial skin that people become more attention of beauty
and healthy, it affects to the rapid growth in beauty care industry, especially in now. Because
of the people focus on the important of their own face would be good looking, clearly face
and still younger, but the real natural face of people will be change over time. The natural
acne always have a chance was born on the face. No matter how old you are or what gender
you are. The medical of facial skin became important to treatment them.
Since the majority of people are interested and enthusiastic to treat a skin
disease. It is the acne on a facial skin that commonly found in the teenage age. The statistics
from the Institute of Dermatology found that acne is one of the reasons to make a patient
going to meet a doctor increasing. Therefore, the number of patients increased steadily with
the number of doctors. The patients have several kind and different of people such as sex, age,
body, face structure and the location of acne on a face. The doctor used a traditional tool to
notes the result of treatment with written it by hand so the doctor will used the symbol of
mathematic to draw a location of acne into the paper for represent an acne point instead of
drawing into a real face such as draw a circle, rectangle or spot. The traditional way was
found the problem when the doctor keep continue the treatment results in the next time, it
doesn’t work. The spot acne of patient is correct location and difficult to analyse the direction
of quantity acne on the face. For each patient will spend a lot of time to treatment them so it
effect to the management time that is difficult to manage more people in each day. The
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 5
doctors have only one way communicate with the patient by face to face. Instead of using the
traditional way to retention the information of patient without drawing a mark point by hand.
They would have a new innovation of medical facial skin program, to help a doctor and
patient is easier and comfortable.
For the unlimited technology always moved forward, it is the one key factor
that effect in the process of human thought and analysis. The number of internet users and
social media are expanded widely. Everybody can access to a large community that it cause to
the human thinking about how to make themselves look good, because social network links to
worldwide. Therefore, the most users always want to edit the images. To make face look
smooth and clearly without acne on face. So the developer created and released a various
application to respond them. The application was released many version to improve the result
of images to be efficient. They blur the whole images to look smooth and unwanted point will
disappear. This algorithm also made the environment around the face blur.
All the cause of problem, the technology of computer satisfies all the aspect
utility function of the doctor that can detect the quantity of acne. Therefore, we developed
algorithm using some path of successful method to improve our program that the detection
will have more accuracy. The research purposed, concerned to develop a medical facial skin
for detect the acne on the face, and know which method has the efficient of detection. The
important process and method involved to the Face detection, Blob detection and Color
segmentation. Then, ours project used the knowledge of statistic in mathematic subject to
adjust with the algorithm. The statistic method can be calculated the exact area and color that
increase the accuracy of image processing. So that, the several major of knowledge will be
applied to develop the tool of medical facial skin. We created an instance element of image to
tests our method and evaluated the accuracy of each experiment to know the best solution.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 6
Objectives The aim of this project is use of the image processing to develop software for
read and analyse images of medical profession, especially the dermatologist. The doctor will
be able to take the results of program to detect change and treatment acne on the face more
effectively. The program reads the input images to detect only the problem area of acne or the
different color area of surface, and analyse to correct the problem without unwanted area such
as mouth, nose, eyes and hair. After detect it, will be able to analyse the amount of acne and
calculate for represent in the percentage in order to compare the different result with the same
images. The doctor will know how the movement of treatment going to be and can analyse
the next treatment results. This program useful to the doctors, they don’t need to meet their
patients. The programs used only the image of face patient's then detect and display the
specific acne area. So, doctors don’t need to analyse the problems by themselves, they will
know immediately how to cure their patient's problems. Moreover, the technology of image
processing further to developed the mobile application that related to beautify the images such
as photo editor, photo blur and beauty face photo etc. The algorithm adjusted to the one part
of application to be increase efficiency that is changing different tone of the surface and focus
only the specific point.
Outputs and Expected Benefits
4.1 Outputs
The output that we want is ground truth image of patient's face from detected
process. The image will show only the line of wrinkles on the patient's face correctly. So, the
important thing that we concern is how to write a program that can read and detect only the
wrinkles on the patient's face because the images of patients is composition of face, hair,
background, etc. So, we want to make sure that program will not detect something else that
not a wrinkle. We learn from the previous researches that maybe program will detect some
key point on the human face (eyes, nose, and mouth) and hairs that have a line shape similar
to the wrinkles The direction of human face also can effect to the program that it limited for
program to analyse the image. Every images that import to the program, the human face have
to be in the same direction and the best is the human face that look straight to the camera
because it easier to the program to read and detect the wrinkles.
4.2 Benefits
Our project, we write a useful program for the medical profession, especially
in facial skin line. The program will useful with doctors to analyse the methods to cure their
patient's problem by using the ground truth image from program that show the lines of
wrinkles on patient's face. This can decrease the problem that nowadays we lack of doctors
and they not have a chance to discuss or analyse the patient's problems. Doctors can use the
ground truth image from program to follow up the change of the patient's problem in every
period of times, they can know that the problems is better or not by compare the ground truth
images to see the changes of the wrinkles.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 7
The future development of our method can apply with the smart phone on the
application photo feature. The program will detect the wrinkles on human face then it can
develop to blend/blurred the wrinkles that program detected to make the images more
beautiful.
Literature Review
Several ideas and methods that we have study would be the guideline for our
project of acne detection.
Image processing [1] Automatic Facial Skin Defect Detection System : processes of the
proposed approach, which includes face detection, facial feature detection, ROI locating, spot
detection and wrinkle detection.
Image processing [2] Extraction of acne lesion in acne patients from Multispectral Images :
the algorithm for the extraction of acne lesions from MSI. In the preprocessing, background,
hair and normal skin are removed while in the classification step, reddish papule, pustule and
scar are classified.
Image processing [3] Learning-Based Detection of Acne-like Regions Using Time-Lapse
Features : Detecting Acne-like Regions In Skin Images that use algorithm to detect acne
lesions using images acquired under cross-polarized modality.
Face Detection [4] The face detection separates into 4 methods; Skin color segmentation is
the process of rejecting non-skin color from the entire image. It is based on the color of all
races human face, Morphological Processing is to performed the clean up of the image. The
goal is to end up with a mask image that can be applied to the input image to yield skin color
regions without noise and clutter, Connected Region Analysis the output from morphological
processing still contains non-face regions. Most of them are hands, arms, legs, clothing that
match the skin color and some parts on background. In connected region analysis, image
statistics from the training set are used to classify each connected region in the image.
Template Matching is the basic idea of template matching is to compare the image with
another template image that is representative of faces. Finding an appropriate template is a
challenge since ideally the template (or group of templates) should match any given face with
differences of the size and features.
Blob Detection [5] Automatic detection of blobs from image datasets is an important step in
analysis of a large-scale of scientific data. These blobs may represent organization of nuclei in
a cultured colony, homogeneous regions in geophysical data, tumor locations in MRI or CT
data, etc.
Color Segmentation [6] It is based on the color of all races human face. We set the HSV (Hue,
Saturation, Value) color space for segmentation up for only focus on that specific value of H
and S. This information was used to define appropriate thresholds for H and S space that
correspond to faces. The threshold values were embedded into the color segmentation routine.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 8
The skin color segmentation resulting image is converted to HSV color space. All the color
pixels that fall outside the H and S range are rejected as the non-face objects.
Methodology
6.1 Approach
To developing algorithms of detect facial acne, we started to collecting a face
acne images of patient. Those images used to simulate the detection point acne in the
Photoshop program. We imported the image to first layer, and created the second layer to spot
a specific area of acne. Then exported only the paint of spot acne to be a result of the
simulation (Ground truth) that obtained in this process, using it compared with the results of
the testing algorithm in computer processing. Ground truth represents ability of the human
detecting.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 9
To processing the algorithm, import the images into the program and it will automatic
detection the acne.
1. Convert the RGB color images to the Grey scale images
2. Find the maximum value of intensity images with X and Y coordinates on the Grey
scale images.
3. Calculate normalized grey-scale image by divide the value of intensity to 0 or 1 with
X and Y coordinates, to compare with HSV images.
4. Retrieve HSV color images to define the value of H(Hue) = 0 for drop a red color
5. To extract the brightness area (V) from HSV model and define Dark color = 0 and
White color = 1.
6. To subtract by V-Grey scale, the result show the region of maximum lightness
7. Define the value of threshold background is white color otherwise will be a black
color. The images convert to negative binary color
8. To analyse the images for eliminate a tiny spot area.
9. From the result of step 8, divided the area less than 7000.
10. The results from step 7, 8 and 9 will represent the appropriated size of specific
object.
11. Create the square to cover the area of detection
12. Detect the input images and calculate the amount of acne.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 10
The processes of the program consist of two main techniques that include the
Face detection and the Statistic method. The defaults method is to upload the image of face
patients into the system. Then the program will check the acne on the face through two
methods that describe in the previous statement. The both image results are different accuracy
away. Comparing it with the evaluation processed (Evaluation Accuracy) by Qualitative and
Quantitative data.
The structure working process of the program consists of two main techniques.
There are Statistic Methods and Face Detection by way of starting are the same. But the
results in term of accuracy will be difference by can compare the image from the results of the
program (Evaluation Accuracy). As the result image shows that the amount of acne on the
patient's face from the treatment of each time. Using basic manually method to checking each
frame of acne that program detected is correct or not. By the way called Qualitative and
Quantitative and each method are different, as follows.
-Qualitative: Evaluation of Qualitative data is considered on the details of the
images, the result is information that can be seen with the naked eye. Evaluation from the
result images by comparing each image and determine of each frame of acne that program
detected is the actual acne including areas where error of program detection.
-Quantitative: Evaluation of Qualitative data is a consideration about numbers
or data that can be measured as the number of digits. From each image results can be
evaluated by calculating the quantity of acne on the patient's face and determine whether the
average of the frames of acne that program detected is the actual acne including areas where
error of program detection.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 11
From the analysis of Quantitative and Quantitative, each result images will
have different accuracy. The result of analysis will be classified into 4 types (Confusion
Matrix)[7];
- True Positive (TP) is what the program predicted, and says it is true.
- True Negative (TN) is what the program to predict that's not true, and says it is false.
- False Positive (FP) is what the program predicted, but says it is false.
- False Negative (FN) is what the program predicted that's not true, but says it is true.
From the results, each type of classification that is used to calculate the "true
positive rate" called Sensitivity and Specificity[8] which calculate the rate of accuracy of the
results from the program. Sensitivity is the rate of what the program predicted, and says it is
true, also can be calculated as a percentage. And Specificity is the rate of what programs
predict that's not true and says that it is false, also can be calculated as a percentage.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 12
To summary, the structure working process of the program consists of the two
main techniques, Statistic Methods and Face Detection by way of starting are the same that to
upload images of acne on patient's face into the system, but difference in term of accuracy of
the results by can compare from the evaluation precision(Evaluation Accuracy). We will
comparing from the result images from program. The results of the program can tell changes
or the development of acne on the patient's face by a method called Qualitative and
Quantitative. Each method will be able to evaluate the results come out differently. From the
results of each type of classification can be divided into Sensitivity and Specificity those are
the calculation precision of the results of the program. Sensitivity is the rate of what programs
predict that true and said it was true. Specificity is the rate of what programs predict that false
and said that it's not true, both can calculate out as a percentage.
6.2 Tools and Techniques
6.2.1 Tools
Software
- Adobe Photoshop : To simulate the expected result of acne images for compared with the results of experiment of program
- Matlab : To write a program with a basic function of image processing
command
Hardware
- Camera 1 unit
- Computer 1 unit Functional Specification
- Upload image of face acne
- Represent to detected area of acne
- Represent a number measurement of acne
6.2.2 Techniques
The technique using to find the solution of the method is less a mistake, but
most efficiency and more accuracy is following to.
1. Statistic Methods
The detection using image processing technology, in each figure of images
represents the shapes of square surrounding the specific point of acne that is a blob. The blob
determine the scope area of acne problem that is the way to used it developed program have
more accuracy. This method calculated the average area and color of acne problem after that
classify the different of data into the group. Then eliminate the group that is doesn’t close
with the other group. Those areas are the mistake of detection program, and assume it is
unwanted acne area.
2. Face Detection
The technology of face detection always continues development. The
experiences of program have more accuracy. The basic method is detected the region of face
structure such as eyes nose or mouth to locate the right position that you need. Moreover, the
researches show the Connected Component Analysis method to solve the detection mistake.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 13
In each object on the body, define the identical value of each path of body then set the value
of data that we want to deduct it. So program doesn’t detect the identical area and represent
only the region of face. Those methods can reduce the detection mistake and can modify it to
use with only the face structure to less mistake of acne detection.
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 14
Project Schedule Our work schedule is based on the advice of Dr. Bunyarit
Uyyanonvara(advisor) in every week, as well as the duration of the study the research and
also development of program.
Task Description Person Duration Deadline Status
1 Review the previous study and
research. Select which one can be
the guideline for our project
WS &
MJ
1w 18 Sep
14
100% completed
2 Try to learn and study Matlab
program
WS &
MJ
1w 25 Sep
14
3 - Design the progress.
-Select the methods to implement
algorithms of project then try to
work on Matlab
WS &
MJ
2w 9 Oct 14 100% completed
4 Prepare report and paper for the
conferences (NSC & IC-ICTES
2015)
WS &
MJ
1w 16 Oct 14 50% completed
5 -Develop program to split the blob
of acne. (1blob per 1acne only)
-Develop program to count and
show amount of detected acne
WS &
MJ
2w 30 Oct 14 100% completed for
count and show amount
of detected acne
7 Prepare slides for midterm
presentation
WS &
MJ
1w 6 Nov 14 100% completed
8 Submit the report to NSC
conference
WS &
MJ
2w 10 Nov
14
100% completed
9 Learn the progress of the results
images from program
WS &
MJ
1w 17 Nov
14
80% completed
10 Prepare slides for final presentation WS &
MJ
10d 27 Nov
14
50% completed
11 Submit paper to IC-ICTES 2015
conference
WS &
MJ
2w 8 Dec 14 100% completed
12 Submit proposal and slides for final
presentation
WS &
MJ
1w 15 Dec
14
100% completed
Senior Project 2014 Medical Image Processing
School of ICT, SIIT 15
References [1] Chuan-Yu, Shang-Cheng Li, Pau‐Choo Chung, Jui-Yi Kuo, Yung-Chin Tu.
"Atomatic Facial Skin Defect Detection System." Dept. of Computer Science & Information
Engineering, National Yunlin University of Science & Technology, Taiwan. pp.527-‐532,
2010.
[2] Hideaki Fujii, Takashi Yanagisawa, Masanori Mitsui, Yuri Murakami, Masahiro
Yamaguchi, Nagaaki Ohyama, Tokiya Abe, Ikumi Yokoi, Yoshie Matsuoka, and Yasuo
Kubota. "Extraction of acne lesion in acne patients from Multispectral Images". Annual
International IEEE EMBS Conference Vancouver, British Columbia, Canada. pp.4078-4081,
2008.
[3] Siddharth K, Madan and Kristin J, Dana. "Learning-Based Detection of Acne-like Regions
Using Time-Lapse Features". Department of Electrical and Computer Engineering, Rutgers
University NJ, USA. 2011.
[4] Phannapat S, Watcharaporn S, Maetawee J, Guntachai O. "Face Detection" School Of
Information Computer and Communication Technology Sirindhorn International Institute of
Technology, Thammasat University, Thailand. 2013.
[5] Anne Kaspers. "Blob detection". Biomedical Image Sciences,Image Sciences Institute,
UMC Utrecht, 2011.
[6] Chai D, Ngan K.N., "Face segmentation using skin-color map in videophone applications,"
Circuits and Systems for Video Technology, IEEE Transactions on , vol.9, no.4, pp.551,564,
Jun 1999