SEMINAR SERIES ON ADVANCED MEDICAL IMAGE PROCESSING INTRODUCTION

1
SEMINAR SERIES ONADVANCED MEDICAL IMAGE
PROCESSING INTRODUCTION

• MINGYUE DING, LIXU GU
• Robarts Research Institute
• London, Ontario, Canada
• June 28, 2002

2

OBJECTIVES

• Provide background knowledge on medical image
  processing
• Explain some important methods in medical image
  segmentation
• Introduce how to use these algorithms on your
  computer (VTK, ITK, MATLAB, IDL )

3
IMAGE PROCESSING BOOKS

• Digital image processing
• Rafael C. Gonzalez, Richard E. Woods.
• 1977,1987,1992, 2002
• Taylor Library TA1632 G66

4
WHAT IS AN IMAGE?

• A digital image is defined a 2-D function,
  f(x,y), where x and y denote spatial coordinates
  and f is called intensity or gray level at the
  point of (x,y)
• Digitization
• Must choose pixel size (maximum possible
  resolution)
• Must choose number of gray levels, dynamic range
• Loss of information
• Data
• Dimensionality (2 or 3, sometimes more)
• Large (usually 10MB to over 1GB)
• Every voxel maps to a point in physical space

5
TYPES OF IMAGE

• Binary image, gray level image, color image,
  false color image
• Sensor optical, infrared, ultraviolet, x-ray,
  Radar, MRI (RF), ultrasound, microwave
• Dimension 2D, 3D, 4D,
• Sensing platform satellite (geography),
  microscope, X-ray film, MRI machine

6
TYPES OF IMAGE
Binary image
Color image
Gray level image
7
INFRARED IMAGE
Gray level image
False color image
8
ULTRAVIOLET IMAGE
Sun UV-image
9
SYNTHETIC APERTURE RADAR (SAR) IMAGE
SAR satellite image
SAR aerial image
10
MICROSCOPIC IMAGE
Pollen image
Myocardium image
11
MICROWAVE IMAGE
Optical
Microwave
RADAR
12
LASER RANGE FINDER IMAGE
Perspective view of 3D laser image
13
MEDICAL IMAGING
MRI IMAGE
14
MEDICAL IMAGING
ULTRASOUND IMAGE
X-RAY IMAGE
15
Positron Emission Tomography(PET)
LUNG
BRAIN
16
3D ULTRASOUND IMAGE
17
IMAGE SEQUENCE (VIDEO)
18
IMAGE DATA STRUCTURE

• An image is represented by an array.
• 2D image f(i,j)
• 3D image f(i,j,k)
• 4D image f(i,j,k,t)
• Here i,j,k,t and f(i,j,k,t) are integers

19
CAMERA PROJECTION GEOMETRY

• Field of view (FOV)
• Origin, orientation, distance
• Diameter of lens, focal distance
• Camera coordinate(xc,yc,zc)
• Object coordinate (xo,yo,zo)
• World coordinate (xw,yw,zw)
• Pixel index coordinate (i,j)
• Transform matrix

20
CAMERA PROJECTION GEOMETRY
yw
zw
yo
zo
xw
ow
xo
oo
yc
Zc
oc
xc
FOV
21
PLANE COORDINATE
(0,0)
y
j
x
i
(0,0)
Pixel (voxel) indices,(i,j)
2-D coordinate, (x,y)
22
IMAGE FORMAT

• 2D image
• raw image, .bmp, .tif, .jpg, .gif,
• 3D image
• rawextra info, .vox, .mnc, .dicom,

23
IMAGE READER/WRITER

• We have many image readers/writers built in our
  software platforms reference to user manual for
  detail
• vtkTIFFReader/Writer
• vtkBMPReader/Writer
• vtkPMGReader/Writer
• vtkMINCReader/Writer
• vtkDICOMReader/Writer

24
SHAPE INVARIANCE UNDER SIMILARITY TRANSFORM

• Similarity Transform shape is unchanged
• Shape invariance of similarity transform
  fundamental feature of pattern recognition
• Nonlinear transform e.g. perspective view or
  warp deformation

25
SIMILARITY TRANSFORMATION

• Suppose the coordinate before and after the
  transform are (x,y) and (x,y)
• Translation
• Rotation

26
MATH TRANSFORMATION

• Scale change
• Order of transform
• Translation followed by scale change
• Scale change followed by translation

27
MATH TRANSFORMATION

• Scale change followed by rotation
• Rotation followed by scale change

28
INVARIANCE OF SIMILARITY TRANSFORM
Original
Rotation (300)
Translation
Scale change (60)
29
TRANSFORM IN VTK/ITK

• VTK
• Linear vtkTransform, vtkLandmarkTransform,
• vtkImageShiftScale,vtkIm
  ageResample
• Nonlinear vtkPerspectiveTransform,
• vtkGridTransform,
• vtkSphericalTransform
• vtkThinPlateSplineTransform
• ITK
• Linear itkAffineTransform, itkRigid2D/3DTransfor
  m
• Nonlinear itkElasticBodySplineKernelTransform
• itkThinPlateSplineKernelTransform

30
WHAT IS IMAGE PROCESSING?

• Image processing is a data processing procedure
  of image using computer in order to perform a
  special task
• Signal processing methods are useful
• Task math modeling solving method software
  programming

31
CATEGORIES OF IMAGE PROCESSING

• Low level from image to image, often called
  image filtering
• Intermediate level from image to symbolic
  representation, also called image segmentation
• High level from symbolic to functional
  description, also called image understanding or
  pattern recognition

32
TYPICAL IMAGE PROCESSING SYSTEM FOR MEDICAL
IMAGING
Image Acquisition
Output
-Low Level- Pre-Processing (filtering) Enhancemen
t (sharpen, smooth, ) Interpolate, reduce noise,
crop
-High Level- Modeling Volume Rendering,
Deformable Models, Mathematical Models
-High Level- Recognition Region feature analysis
(position, orientation, size), Object Matching
-Intermediate Level- Segmentation Region (or
Contour) Extraction, Labeling, Grouping
33
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 1

• Segmentation (Extraction)
• Thresholding
• Differential Top-hat
• Noise Reduction
• Region Reconstruction

EXAMPLE
34
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 2
Volume View 1.vertebra, 2.rib, 3.pelvis, 4.left
kidney, 5.spleen, 6.stomach,7.right kidney,
8.liver, 9.artery, 10.vein, 11.heart
EXAMPLE
35
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 3

• Bone segmentation and reference point
• Extract bones
• Separate spine from ribs
• Find the 12th rib
• Find the vertebra connected with the 12th rib
• Set the center of gravity of the vertebra as the
  reference point (origin of our specified
  coordinate system)

EXAMPLE
36
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 4
Pattern Spectrum -- size analysis
EXAMPLE
37
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 5
Noise Reduction
EXAMPLE
38
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 6
Subdivision into regions
EXAMPLE
39
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 7
Threshold segmentation
EXAMPLE
40
EXTRACTION OF ORGANS FROM 3D ABDOMINAL CT IMAGE
- 8

• Segmentation using Recursive Erosion (RE) and
  Geodesic Influence(GI)
• RE region shrinking to generate all the
  candidate seeds
• GI region reconstruction to recover separated
  organs

EXAMPLE
41
ORGAN RECOGNITION

• Organ recognition
• Feature analysis (size, position)
• Match the data base or dictionary
• Label the objects with an unique symbol (a number
  or an anatomic name)

EXAMPLE
42
ROAD MAP

• FUNDAMENTAL CONCEPTS --- (Lixu Gu)
• Thresholding 2)  Connected components Labeling
• Region properties 4)  Neighbourhood operators
• Noise reduction
• FUNDAMENTAL MATH (I) --- (Mingyue Ding)
• 1)   Edge detection 2)   Curvature and
  gradient  
• 3) Hough transform 4) B-spline fitting
• DEFORMABLE MODELS --- (Hanif.M.Ladak )
• MEDICAL IMAGE PROCESSING TOOLS IN MATLAB
• --- (Hanif.M.Ladak )

43
ROAD MAP
5. FUNDAMENTAL MATH (II) --- (Mingyue Ding)
1) Least square estimation (LSE) 2) 
 Principal component analysis (PCA) 2)
Singular value decomposition (SVD) 4) Bayesian
estimation 6. MATHEMATICAL MORPHOLOGY (I)
--- (Lixu Gu) 1)   Binary morphology i.
Dilation, Erosion, Opening, Closing ii. 
Shape feature extraction 2)  Connectivity 3)  Pa
ttern spectrum and Skeleton 7. MATHEMATICAL
MORPHOLOGY (II) --- (Lixu Gu) 1)     Gray
scale morphology 2)    Top-hat and Differential
Top-hat 3)     Ultimate Erosion (UE) 4)
Geodesic influence (GI)
44
ROAD MAP

• LEVEL-SET SEGMENTATION --- (Mingyue Ding)
• What is the problem with snake? 2) What is a
  level set?
• Zero-level set representation of front 4)
  Hamilton-Jacobi Equation
• Swallow-tail free solution 6) Determination of
  front speed
• Extension of speed 8) Stopping criteria
• Discrete solution of Hamilton-Jacobi Equation
• Level-set based segmentation algorithm
• REGION GROWING AND WATERSHED --- (Lixu Gu)
• Euclidean and Geodesic distance
• Conditional dilation (binary and gray level)
• SKIZ and Influence zone 4) Watershed
• MEDICAL IMAGE PROCESSING TOOLS IN IDL --- (Hua
  Qian)

45
Not Enough?
Go Library!
46
Discussion