Transcript Graduate Seminar Talk at Old Dominion University

Old Dominion University (ODU)
Norfolk, Virginia
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Presenter:
Dr. Abu Asaduzzaman
Assistant Professor of Computer Architecture and Director of CAPPLab
Department of Electrical Engineering and Computer Science (EECS)
Wichita State University (WSU), Wichita, Kansas
April 1, 2016
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
►
Outline
■ Introduction
 Breast cancer and some related information (Q/A)
■ ‘Digital’ Mammogram Image
 Mammogram Image to Digital Matrices
■ Fast Effective Analysis




QUESTIONS? Any time, please!
Existing Methods
Problem Description, Proposed Method, Key Contributions
Preliminary Results
What is next?
■ Computer Arch and Parallel Prog Lab (CAPPLab)
 Laboratory, Research
■ Discussion
Dr. Zaman
2
Thank you, all!
■ ECE Department, ODU
 Dr. Dimitrie C. Popescu
 Khan M. Iftekharuddin
Dr. Zaman
3
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction
■ Dr. (Ph.D.) vs Doctor (M.D.)
■ Doctor of Philosophy Vs Doctor of Medicine
■ Computer Engineers/Scientists Vs Cancer Physicians
Vs
■ Dr. David Patterson, University of California at Berkeley (UCB) [1]
 Algorithms, Machines, and People Laboratory (AMPLab)  3 D., 10 F., 2 V.R., 7
P.D., 44 GR, and 2 UG
■ Brown University and Washington University in St. Louis [2, 3]
■ Vanderbilt-Ingram Cancer Center at Vanderbilt University, Nashville [4]
Dr. Zaman
4
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction (Q/A)
Cancer – Types – Rank
■ Cancer is the name given to a collection of related
diseases. In all types of cancer, some of the body’s
cells begin to divide without stopping and spread into
surrounding tissues [1]. It’s all bad!
■ This uncontrollable cell division turns into tumors or
lump. Two types tumors are common: Benign tumor
and Malignant tumor [2].
■ Cancer is suspected to become the leading cause of
death (2.3 million new cancer cases) in the United
States by 2030 [4].
Dr. Zaman
5
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction (Q/A)
Benign Vs Malignant
Dr. Zaman
6
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction (Q/A)
Breast Cancer – Mobility
■ Breast cancer is a malignant tumor that starts in the
cells of the breast, commonly in women.
■ According to mobility, breast cancers are two types:
Non-Invasive and Invasive [3].
Normal milk-duct
Dr. Zaman
Non-Invasive
7
Invasive
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction (Q/A)
Breast Cancer – Severity
■ Breast cancer is most common in American women.
■ According to recent reports, about 231,840 new cases
of invasive breast cancer are diagnosed in women and
about 40,290 women died from breast cancer in 2015.
■ About 12% (one out of eight) U.S. women are
suspected to grow breast cancer in their life-times.
■ Many of them are suspected to die due to the breast
cancer.
Dr. Zaman
8
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Introduction (Q/A)
Breast Cancer – Common Practice
■ The white dots present in the image are
tumors. These tumors contain Calcium
which makes the tumor brighter than the
surrounding.
■ Commonly used image types include
magnetic resonance imaging (MRI) and
mammogram (soft x-ray) [3].
MRI Image
Dr. Zaman
9
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
►
Outline
■ Introduction
 Breast cancer and some related information (Q/A)
■ ‘Digital’ Mammogram Image
 Mammogram Image to Digital Matrices
■ Fast Effective Analysis




QUESTIONS? Any time, please!
Existing Methods
Problem Description, Proposed Method, Key Contributions
Preliminary Results
What is next?
■ Computer Arch and Parallel Prog Lab (CAPPLab)
 Laboratory, Research
■ Discussion
Dr. Zaman
10
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Mammogram Images
■ MRI (a medical imaging technique for radiology)
uses strong magnetic fields, radio waves, and
field gradients to form images of the body.
■ Mammography (aka, mastography) is the process
of using low-energy X-rays (around 30 Peak
kilovoltage (kVp))
■ Digital mammography is a specialized form of
mammography that uses digital receptors and
computers instead of x-ray film to help examine
breast tissue for breast cancer.
■ ‘Digital’ mammogram image is a digital matrix
generated for each region of interest (ROI) using
the equivalent pixel values.
Dr. Zaman
11
MRI Image
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Digital Mammogram Image
■ Mammogram imaging technique
uses low-power x-ray (~ 30 kVp).
■ A mammogram image can be 8-bit,
12-bit, or 16-bit depth.
■ A smallest addressable point of
mammogram is called pixel. Each
pixel can have a value depends
upon bit depth.
■ For example, an 8-bit pixel value
ranges between 0 and 255; where 0
represents pure black and 255 pure Digital Mammogram Image
white. Similarly, an 16-bit pixel value
ranges between 0 and 65535.
Dr. Zaman
12
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
‘Digital’ Mammogram Image
‘Digital’ Mammogram Image
MRI Image
Dr. Zaman
Digital Mammogram Image
13
Digital Matrix of Pixel Values
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Computer Aided Diagnosis (CAD)
■ Current Images: Four pictures taken
from different angle.
■ Previous Images: Patients old
pictures.
■ Anamnesis Data: Includes patients
family records and personal health
parameters
■ Preprocessing : Image enhancement
and basic parenchymal pattern
determined.
Dr. Zaman
14
CAD System
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Computer Aided Diagnosis (CAD)
■ Detection: Several algorithm applied
here for including neural network
,texture analysis, heuristic methods,
wavelet transform based method etc.
to identify the tumors.
■ Validation: If suspicious location
detected then again an algorithm is
used to justify or reject the detection.
■ Classification: This consists of
findings and probability like measure
reliability of the diagnosis [5].
Dr. Zaman
15
CAD System
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Problems with CAD
■ Computer Aided Diagnosis system
has 72.55% accuracy. Due to this
reason , radiologists need to review
after each screening process.
■ Breast cells are consists of fatty
tissue, CAD system is unable to
separate fatty tissue and
microcalcifications.
■ Mammogram or breast tomosynthesis
has high rate of false positive and
false negative. As a result many
people undergoing delayed treatment.
Dr. Zaman
16
CAD System
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Fourier transform and log
transform [8]
■ Pros: This method is very easy to
program. Digital image can be used in
this technique. It takes very less time
to process the image.
■ Cons: The output image is very
unclear, this method o detects the
tumors or the mass regardless it is
malignant or benign. Size of the tumor
is unknown.
Algorithm
Dr. Zaman
17
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Top hat Image processing
■ Pros: Malignant and benign
differentiation is considered.
■ Cons: Step 3 fill the gaps between
suspected regions may give a wrong
result. Sometimes it may change the
actual tumor size. The output they
have produced has major drawbacks.
Dr. Zaman
18
Algorithm
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Existing Methods
Matlab & LabVIEW
■ In this paper, authors try to highlight the suspicious region and
extracted from the main image. Then they have done a few
feature extraction and record the malignant and benign tumors
properties or features [9].
■ Drawbacks: Breast cancer is not easily predictable. They have
tested their algorithm on 20 patients which is very less to say
something strongly about any feature. They have not validate
their algorithm that is sensitivity to tumor or false positive
calculation.
Dr. Zaman
19
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Problem Description
■ Existing methods to analyze breast cancer are not 100%
effective! Three points: accuracy, faster results, and cost
Major Contributions
■ By processing mammogram images, classify fatty tissues of
(benign and malignant) tumors.
■ By processing mammogram images, select region of interest (ROI)
and generate matrix of pixel values for each ROI.
■ Using ROIs and matrices, extract geometrical (example: area of
tumor) and textural (example: mean pixel value) features.
■ For faster processing of mammogram images, discover/apply
multithreaded GPU-assisted parallel programming techniques.
Dr. Zaman
20
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Proposed Method
■ Work-flow







Breast mammography images are taken.
Each image is ‘pre-processed’ to make it ready ROI selection.
ROIs are selected.
Matrix for each ROI is generated using the pixel values.
Feature values (tumor area, mean pixel value, etc.) are extracted.
Feature values are analyzed for accurate assessment.
GPU-assisted high performance computing are applied for faster and
cheaper analysis/solution.
■ Software/Tools
 Matlab
 CUDA/GPU Computing
Dr. Zaman
21
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Proposed Method
Pre-processing Image
1
2
1
2
3
4
4
3
Images in each processing stage
Dr. Zaman
22
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Proposed Method
Selecting ROI
1
4
4
5
5
6
Images in each processing stage
Dr. Zaman
23
6
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Proposed Method
Feature Extraction
1
■ Geometrical Features
6
 Area, perimeter, radius, and shape of the tumor
■ Textural Features
 Statistical functions: mean, global mean, standard devaition,
entropy, and skewness of pixel values
Dr. Zaman
24
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
►
Outline
■ Introduction
 Breast cancer and some related information (Q/A)
■ ‘Digital’ Mammogram Image
 Mammogram Image to Digital Matrices
■ Fast Effective Analysis




QUESTIONS? Any time, please!
Existing Methods
Problem Description, Proposed Method, Key Contributions
Preliminary Results
What is next?
■ Computer Arch and Parallel Prog Lab (CAPPLab)
 Laboratory, Research
■ Discussion
Dr. Zaman
25
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Area of tumor (a geometrical feature)
A benign tumor shows area
value in the malignant tumor
region. (Not correct!)
A malignant tumor shows area
value in the benign tumor
region. (Not correct!)
The area value due to a benign tumor is between 0 and 12191.
The area value due to a malignant tumor is between 907 and 95114.
Therefore, we consider area value as a lower decision making factor.
Dr. Zaman
26
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Mean of pixel values (a textural feature)
The mean pixel value due to a benign tumor is between 95 and 155.
The mean pixel value due to a malignant tumor is between 175 and 195.
Therefore, we consider mean pixel value as a higher decision making factor.
Dr. Zaman
27
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Radius of tumor (a geometrical feature)
200
150
100
50
0
Radius of malignant tumors
197
69
00
30
33 4031
0000 0
0
68
Radius
Radius
Radius of benign tumors
48 49
0 0 0
200
150
100
50
0
174
114
67 54
47 37
17
Image file name
Image file name
The radius value due to a benign tumor is between 0 and 69.
The radius value due to a malignant tumor is between 17 and 174.
Therefore, we consider radius value as a lower decision making factor.
Dr. Zaman
70
28
29
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Perimeter of tumor (a geometrical feature)
Perimeter of malignant tumors
Perimeter
1400 1237
1200
1000
800
433
427
600
301 307
400
207 251194
188
200
0 0
0 0 0 0
0
0
0
0
0
0
mdb001.pgm
mdb002.pgm
mdb003.pgm
mdb004.pgm
mdb005.pgm
mdb006.pgm
mdb007.pgm
mdb008.pgm
mdb009.pgm
mdb010.pgm
mdb011.pgm
mdb012.pgm
mdb013.pgm
mdb014.pgm
mdb015.pgm
mdb016.pgm
mdb017.pgm
mdb018.pgm
mdb019.pgm
mdb020.pgm
Perimeter
Perimeter of benign tumors
Image file name
1400
1200
1000
800
600
400
200
0
1093
716
420
339
439
295
232
106
Image File name
The perimeter value due to a benign tumor is between 0 and 1237.
The perimeter value due to a malignant tumor is between 106 and 1093.
Therefore, we consider perimeter value as a lower decision making factor.
Dr. Zaman
29
182
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Global mean of pixel values (a textural feature)
1.15486
1.1856
Global Mean
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
Global mean, malignant tumors
0.784563
mdb001.pgm
mdb002.pgm
mdb003.pgm
mdb004.pgm
mdb005.pgm
mdb006.pgm
mdb007.pgm
mdb008.pgm
mdb009.pgm
mdb010.pgm
mdb011.pgm
mdb012.pgm
mdb013.pgm
mdb014.pgm
mdb015.pgm
mdb016.pgm
mdb017.pgm
mdb018.pgm
mdb019.pgm
mdb020.pgm
Global Mean
Global mean, benign tumors
Image File name
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
4.2365
3.213
2.6523
2.365 2.165 2.365 2.265 2.3654
Image file Name
The global mean value due to a benign tumor is between 1.0 and 1.1856.
The global mean value due to a malignant tumor is between 2.3 and 4.2.
Therefore, we consider global mean value as a higher decision making factor.
Dr. Zaman
30
2.365
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Preliminary Results
Standard Deviation (S.D.) of pixel values (a textural feature)
10.2354
Standard Deviation
35
30
25
20
15
10
5
0
S. D. of malignant tumors
7.2
1.2356
mdb001.pgm
mdb002.pgm
mdb003.pgm
mdb004.pgm
mdb005.pgm
mdb006.pgm
mdb007.pgm
mdb008.pgm
mdb009.pgm
mdb010.pgm
mdb011.pgm
mdb012.pgm
mdb013.pgm
mdb014.pgm
mdb015.pgm
mdb016.pgm
mdb017.pgm
mdb018.pgm
mdb019.pgm
mdb020.pgm
Standard Deviation
S. D. of benign tumors
Image File Name
35
30
25
20
15
10
5
0
20.32
32.215
30.236
30.12530.21530.215
25.123623.236
22.325
Image file name
The standard deviation value due to a benign tumor is less than 10.
The standard deviation value due to a malignant tumor is between 20 and 32.
Therefore, we consider standard deviation value as a higher decision making factor.
Dr. Zaman
31
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
What do we have?
Feature
Benign tumor
Malignant tumor
0 to 121922
907 to 95114
Radius
0 to 197
29 to 174
Perimeter
0 to 1237
106 to 1093
Mean
100 to 160
184 to 225
Global Mean
1.0 to 1.23
2.3 to 4.2
< 10
20 to 32
9.2 to -32
-85 to -198
Area
Standard Deviation
Entropy
Skewness
What do we do with it?
Dr. Zaman
32
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
Problem Description
■ Existing methods to analyze breast cancer are not 100%
effective! Three points: accuracy, faster results, and cost
Major Contributions
■ By processing mammogram images, classify fatty tissues of
(benign and malignant) tumors.
■ By processing mammogram images, select region of interest (ROI)
and generate matrix of pixel values for each ROI.
■ Using ROIs and matrices, extract geometrical (example: area of
tumor) and textural (example: mean pixel value) features.
■ For faster processing of mammogram images, discover/apply
multithreaded GPU-assisted parallel programming techniques.
Dr. Zaman
33
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
What do we do with it?
■ By processing many mammogram images, we plan to classify
fatty tissues of (benign and malignant) tumors.
Dr. Zaman
34
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
What is next?
■ High Performance Computing in Healthcare Technology
 Customize image processing / pattern recognition for cancer research
 Discover / apply GPU-assisted multithreaded parallel programming
 Apply data regrouping and task/thread regrouping based optimization
Dr. Zaman
35
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
►
Outline
■ Introduction
 Breast cancer and some related information (Q/A)
■ ‘Digital’ Mammogram Image
 Mammogram Image to Digital Matrices
■ Fast Effective Analysis




QUESTIONS? Any time, please!
Existing Methods
Problem Description, Proposed Method, Key Contributions
Preliminary Results
What is next?
■ Computer Arch and Parallel Prog Lab (CAPPLab)
 Laboratory, Research
■ Discussion
Dr. Zaman
36
Computer Architecture and Parallel Programming
Laboratory (CAPPLab)
CAPPLab
■ Computer Architecture & Parallel Programming
Laboratory (CAPPLab)




Physical location: 245 Jabara Hall, Wichita State University
URL: http://www.cs.wichita.edu/~capplab/
E-mail: [email protected]; [email protected]
Tel: +1-316-WSU-3927
■ Key Objectives
 Lead research in advanced-level computer architecture, highperformance computing, embedded systems, and related fields.
 Teach advanced-level computer systems & architecture, parallel
programming, and related courses.
Dr. Zaman
37
Computer Architecture and Parallel Programming
Laboratory (CAPPLab)
“People First”
■ Students




PhD Students: Kishore K. Chidella, Ahmed E. Aziz
MS Theses: Parthib Mitra, Shanta Mazumder, Jainish R. Jain
MS Projects: Venkatesh Mabbu, Avinash Chintam
UG Students: Suveen R. Emmanuel
■ Collaborators





Dr. H. Neeman, Director of OSCER, University of Oklahoma, OK
Dr. M. Islam, Hematology/Oncology Specialist, UPMC, PA
Dr. Larry Bergman, NASA Jet Propulsion Laboratory (JPL), CA
Mr. J. Metrow, Director of HiPeCC, Wichita State Univ. (WSU), KS
Dr. K. Cluff, Ast. Prof. of Biomedical Engineering, WSU, KS
Dr. Zaman
38
Computer Architecture and Parallel Programming
Laboratory (CAPPLab)
Resources
■ Hardware
 3 CUDA Servers –
 CPU: Xeon E5506, 2x 4-core, 2.13 GHz, 8GB DDR3;
 GPU: GTX Titan X (24x 128 cores, 12GB GDDR5), Telsa K40 (15x
192 cores, 12GB GDDR5) and C2075 (14x 32 cores, 6GB GDDR5)
 Supercomputer (Opteron 6134, 32 cores per node, 2.3 GHz, 64
GB DDR3, Kepler card) via remote access to WSU (HiPeCC)
 2 CUDA enabled Laptops
 More …
■ Software
 CUDA, OpenMP, and Open MPI (C/C++ support)
 MATLAB, VisualSim, CodeWarrior, more (as may needed)
Dr. Zaman
39
Computer Architecture and Parallel Programming
Laboratory (CAPPLab)
Scholarly Activities
■ NVIDIA “GPU Research Center” for 2015-2017
 Grants from NSF, NetApp, CybertronPC, Wiktronics, M2SYS
 Research in Computer Architecture and Parallel Programming
■ Publications
 Journal: 21 published; 1 under review, 3 under preparation
 Conference: 57 published, 4 accepted, 2 under review, 6 under pre
 Book Chapter: 2 published; 1 under preparation
■ Outreach
 USD 259 Wichita Public Schools
 Wichita Area Technical and Community Colleges
 Open to collaborate
Dr. Zaman
40
Computer Architecture and Parallel Programming
Laboratory (CAPPLab)
Research Grants/Proposals
■ Grants/Awards





WSU: URCA, Flossie West
NSF: KS NSF EPSCoR First Award
NetApp: NFS Connector for Spark Systems
NVIDIA GPU Research Center at Wichita State Award
Xilinx University/Teaching (Hardware/Financial) Award
■ Proposals




WSU: URCA, Flossie West
NSF: XPS, IUSE
U.S. Govt.: DoD Breast Cancer Research Program
Industry: CybertronPC, American Association for Cancer Research
Dr. Zaman
41
Old Dominion University (ODU), Norfolk, Virginia, 2016
“Fast Effective Analysis of ‘Digital’ Mammogram
Images for Breast Cancer Treatment”
QUESTIONS? Feedback?
Contact: Abu Asaduzzaman
E-mail: [email protected]
Phone: +1-316-978-5261
http://www.cs.wichita.edu/~capplab/
Thank You!