medical image analysis

49
Medical Image Analysis Medical Image Analysis Image Segmentation Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Upload: felicia-clements

Post on 30-Dec-2015

49 views

Category:

Documents


2 download

DESCRIPTION

Medical Image Analysis. Image Segmentation. Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003. Edge-Based Image Segmentation. Useful links The Berkeley Segmentation Dataset and Benchmark TurtleSeg. Edge-Based Image Segmentation. Edge-based approach - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Medical Image Analysis

Medical Image AnalysisMedical Image AnalysisImage Segmentation

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Page 2: Medical Image Analysis

Edge-Based Image Edge-Based Image SegmentationSegmentationUseful links

◦The Berkeley Segmentation Dataset and Benchmark

◦TurtleSeg

Page 3: Medical Image Analysis

Edge-Based Image Edge-Based Image SegmentationSegmentation

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Edge-based approach◦Spatial filtering to compute the first-

order or second-order gradient information of the image: Sobel, Laplacian masks

◦Edges need to be linked to form closed regions

◦Uncertainties in the gradient information due to noise and artifacts in the image

Page 4: Medical Image Analysis

Edge Detection Edge Detection OperationsOperationsGradient magnitude and

directional information from the Sobel horizontal and vertical direction masks

101

202

101

xG

121

000

121

yG

yxy GGGGMx

22

Page 5: Medical Image Analysis

Edge Detection Edge Detection OperationsOperationsThe second-order gradient

operator Laplacian can be computed by convolving one pf the following masks

010

141

010

)4(LG

111

181

111

)8(LG

Page 6: Medical Image Analysis

Edge Detection Edge Detection OperationsOperationsA smoothing filter first before

taking a Laplacian of the imageCombined into a single Laplacian

of Gaussian function as

),(),(

),(),(),(2

2

yxfyxg

yxfyxgyxh

Page 7: Medical Image Analysis

Edge Detection Edge Detection OperationsOperations

2

22

2

22

2

22

2

22

2

)(

4

222

2

)(

2

22

)(

2

2

2

)(22

2

),(

yx

yxyx

yx

eyx

ey

ex

eyxg

Page 8: Medical Image Analysis

Edge Detection Edge Detection OperationsOperationsA Laplacian of Gaussian (LOG)

mask of pixels, :

00100

01210

121621

01210

00100

LOGG

LOGG55 2

Page 9: Medical Image Analysis

Boundary TrackingBoundary Tracking

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Edge-linking◦Pixel-by-pixel search to find

connectivity among the edge segments

◦Connectivity can be defined using a similarity criterion among edge pixels

◦Geometrical proximity or topographical properties

Page 10: Medical Image Analysis

Boundary TrackingBoundary TrackingThe neighborhood search method

◦ : edge magnitude◦ : edge orientation◦ : a boundary pixel◦ : a successor boundary pixel◦ , , :pre-determined

thresholds

),( yxe),( yx

jb1jb

1T 2T 3T

Page 11: Medical Image Analysis

Boundary TrackingBoundary Tracking

1)( Tbe j

11)( Tbe j

21)()( Tbebe jj

31 2mod)()( Tbb jj

Page 12: Medical Image Analysis

Boundary TrackingBoundary TrackingA graph-based search method

◦Find paths between the start and end nodes minimizing a cost function that may be established based on the distance and transition probabilities

◦The start and end nodes are determined from scanning the edge pixels based on some heuristic criterion

Page 13: Medical Image Analysis

Start Node

End Node

Figure 7.1. Top: An edge map with magnitude and direction information; Bottom: A graph derived from the edge map with a minimum cost path (darker arrows) between the start and end nodes.

Page 14: Medical Image Analysis

Boundary TrackingBoundary TrackingA* search algorithm

◦1. Select an edge pixel as the start node of the boundary and put all of the successor boundary pixels in a list, OPEN

◦2. If there is no node in the OPEN list, stop; otherwise continue

◦3. For all nodes in the OPEN list, compute the cost function and select the node with the smallest cost . Remove the node from the OPEN list and label it as CLOSED. The cost function may be computed as

)(ztz

z

)(zt

)(zt

Page 15: Medical Image Analysis

Boundary TrackingBoundary TrackingA* search algorithm

◦4. If is the end node, exit with the solution path by backtracking the pointers; otherwise continue

◦5. Expand the node by finding all successors of . If there is no successor, go to Step 2; otherwise continue

)()()( zhzczt

k

jj

k

iii zdzzskc

121 )(),()(

z

zz

Page 16: Medical Image Analysis

Boundary TrackingBoundary TrackingA* search algorithm

◦6. If a successor is not labeled yet in any list, put it in the list OPEN with updated cost as and a pointer to its predecessor

◦7. If a successor is already labeled as CLOSED or OPEN, update its value by

◦ . Put those CLOSED successors whose cost functions were lowered, in the OPEN list and redirect to the pointers from all nodes whose costs were lowered. Go to Step 2

)(),()()( iii zdzzszczc

iz

z

iz

),()(),(min)(' iii zzszczczc

)(' izcz

Page 17: Medical Image Analysis

Hough TransformHough TransformHough transform

◦Similar to the Radon transform◦Detect straight lines and other

parametric curves such as circles, ellipses

◦A line in the image space forms a point in the parameter space

),( cm

cmxy

Page 18: Medical Image Analysis

Hough TransformHough Transform

Figure comes from the Wikipedia, www.wikipedia.org.

Page 19: Medical Image Analysis

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Gradient e

r

p

O

Figure 7.2. A model of the object shape to be detected in the image using Hough transform. The vector r connects the Centroid and a tangent point p. The magnitude and angle of the vector r are stored in the R-table at a location indexed by the gradient of the tangent point p.

Page 20: Medical Image Analysis

Pixel-Based Direct Pixel-Based Direct Classification MethodsClassification Methods

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Example the histogram for bimodal distribution

Find the deepest valley point between the two consecutive major peaks

Tyxf

Tyxfyxg

),( if0

),( if1),(

Page 21: Medical Image Analysis

Figure 7.3. The original MR brain image (top), its gray-level histogram (middle) and the segmented image (bottom) using a gray value threshold T=12 at the first major valley point in the histogram.

Page 22: Medical Image Analysis

Figure 7.3. The original MR brain image (top), its gray-level histogram (middle) and the segmented image (bottom) using a gray value threshold T=12 at the first major valley point in the histogram.

Page 23: Medical Image Analysis

Figure 7.4. Two segmented MR brain images using a gray value threshold T=166 (top) and T=225 (bottom).

Page 24: Medical Image Analysis

Optimal Global Optimal Global ThresholdingThresholdingAssume

◦The histogram of an image to be segmented has two Gaussian distributions belonging to two respective classes such as background and object

◦The histogram)()()( 2211 zpPzpPzp

121 PP

Page 25: Medical Image Analysis

Optimal Global Optimal Global ThresholdingThresholding

The error probabilities of misclassifying a pixel

Tyxf

Tyxfyxg

),( if2 Class

),( if1 Class),(

T

dzzpTE )()( 21

T

dzzpTE )()( 12

Page 26: Medical Image Analysis

Optimal Global Optimal Global ThresholdingThresholding

Assume the Gaussian probability density functions

)()()( 2211 TEPTEPTE

22

22

21

21

2/)(

2

2

2/)(

1

1

2

2)(

z

z

eP

eP

zp

Page 27: Medical Image Analysis

Optimal Global Optimal Global ThresholdingThresholdingThe optimal global threshold T

02 CBTAT

22

21 A

)(2 212

221 B

)/ln(2 211222

21

21

22

22

21 PPC

Page 28: Medical Image Analysis

Pixel Classification Through Pixel Classification Through ClusteringClusteringFeature vector of pixels

◦Gray value, contrast, local texture measure, red, green, or blue components

Clusters in the multi-dimensional feature space◦Group data points with similar feature

vectors together in a single cluster◦Distance measure: Euclidean or

Mahalanobis distancePost-processing

◦Region growing, pixel connectivity

Page 29: Medical Image Analysis

K-Means ClusteringK-Means Clustering◦1. Select the number of clusters

with initial cluster centroids ; ◦2. Partition the input data points into

clusters by assigning each data point to the closest cluster centroid using the selected distance measure

◦3. Compute a cluster assignment matrix representing the partition of the data points with the binary membership value of the th data point to the th cluster such that

kiv ki ,...,2,1

kix

iv

U

ijuU

ji

Page 30: Medical Image Analysis

K-Means ClusteringK-Means Clustering◦4. Re-compute the centroids using the

membership values as

◦5. If cluster centroids or the assignment matrix does not change from the previous iteration, stop; otherwise go to Step 2.

n

jij

n

jjij

i

u

u

1

1

x

v

Page 31: Medical Image Analysis

K-Means ClusteringK-Means ClusteringObjective function

k

i

n

jijwJ

1 1

2),( vxvU

Page 32: Medical Image Analysis

Fuzzy c-Means ClusteringFuzzy c-Means ClusteringThe objective function

c

i

n

jij

mij

c

i

n

jij

mijm

u

duJ

1 1

2

1 1

2),(

vx

vU

Page 33: Medical Image Analysis

Region-Based Region-Based SegmentationSegmentationRegion-growing based segmentation

◦Examine pixels in the neighborhood based on a pre-defined similarity criterion

◦The neighborhood pixels with similar properties are merged to form closed regions

Region splitting◦The entire image or large regions are split

into two or more regions based on a heterogeneity or dissimilarity criterion

Page 34: Medical Image Analysis

Region-GrowingRegion-GrowingTwo criteria

◦A similarity criterion that defines the basis for inclusion of pixels in the growth of the region

◦A stopping criterion that stops the growth of the region

Page 35: Medical Image Analysis

Center Pixel

Pixels satisfying the similarity criterionPixels not satisfying the similarity criterion3x3 neighborhood

5x5 neighborhood

7x7 neighborhood

Segmented region

Figure 7.5. A pixel map of an image (top) with the region-growing process (middle) and the segmented region (bottom).

Page 36: Medical Image Analysis

Figure 7.6. A T-2 weighted MR brain image (top) and the segmented ventricles (bottom) using the region-growing method.

Page 37: Medical Image Analysis

Region-SplittingRegion-SplittingThe following conditions are met:

◦1. Each region, ; is connected

◦2.

◦3. for all , ; ◦4. = TRUE for◦5. = FALSE for ,

where is a logical predicate for the homogeneity criterion on the region

iR ni ,...,2,1

iR

RRn

ii

1

ji RR i j ji )( iRH ni ,...,2,1

)( ji RRH ji )( iRH

Page 38: Medical Image Analysis

R

R4R3R2R1

R2

4

R2

3

R2

2

R2

1

R1

R21 R22

R23

R41

R43

R24

R42

R44

R3

R4

4

R4

3

R4

2

R4

1

Figure 7.7. An image with quad region-splitting process (top) and the corresponding quad-tree structure (bottom).

Page 39: Medical Image Analysis

Recent Advances in Recent Advances in SegmentationSegmentationModel-based estimation methodsRule-based systemsAutomatic segmentation

Page 40: Medical Image Analysis

Estimation-Model Based Estimation-Model Based Adaptive SegmentationAdaptive Segmentation

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

A multi-level adaptive segmentation (MAS) method

Page 41: Medical Image Analysis

Define Classes

Determination of Model Parameters

(From a set ofmanually segmented and labeled images)

Classification of ImagePixels Using Model

Signatures

Identification of TissueClasses and Pathology

Formation ofa New ClassNecessary?

All pixelsclassified?

ParameterRelaxation

Yes

Yes

No

No

Figure 7.8: The overall approach of the MAS method.

Page 42: Medical Image Analysis

Figure 7.9: (a) Proton Density MR and (b) perfusion image of a patient 48 hours after stroke.

Page 43: Medical Image Analysis

Figure 7.10. Results of MAS method with 4x4 pixel probability cell size and 4 pixel wide averaging. (a) pixel classification as obtained on the basis of maximum probability, (b) as obtained with p>0.9.

Page 44: Medical Image Analysis

Image Segmentation Using Image Segmentation Using Neural NetworksNeural Networks

Figures come from the textbook: Medical Image Analysis, by Atam P. Dhawan, IEEE Press, 2003.

Backpropagation neural network for classification

Radial basis function (RBF) network

Segmentation of arterial structure in digital subtraction angiograms

Page 45: Medical Image Analysis

x1

xn

x2

1

Non-Linear Activation Function F

wn+1

w1

w2

wn

n

inii wxwFy

11

Figure 7.11. A basic computational neural element or Perceptron for classification.

Page 46: Medical Image Analysis

Hidden Layer Neurons

Output Layer Neurons

x1 x2 x3xn 1

Figure 7.12. A feedforward Backpropagation neural network with one hidden layer.

Page 47: Medical Image Analysis

RBF Unit 1 RBF Unit 2 RBF Unit n

Input ImageSliding

Image Window

Output

Linear Combiner

RBF Layer

Figure 7.13. An RBF network classifier for image segmentation.

Page 48: Medical Image Analysis

Figure 7.14. RBF Segmentation of Angiogram Data of Pig-cast Phantom image (top left) with using a set of 10 clusters (top right) and 12 clusters (bottom) respectively.

Page 49: Medical Image Analysis

Figure 7.14. RBF Segmentation of Angiogram Data of Pig-cast Phantom image (top left) with using a set of 10 clusters (top right) and 12 clusters (bottom) respectively.