computer-aided diagnosis of breast cancer: towards the ...ranga/enel697/cadbcex.pdf ·...
TRANSCRIPT
![Page 1: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/1.jpg)
Computer-Aided Diagnosis of Breast Cancer: Towards the
Detection of Early and Subtle Signs
Rangaraj M. RangayyanJ. E. Leo DesautelsFábio José Ayres
Department of Electrical and Computer Engineering, University of CalgaryAlberta Cancer Board: Screen Test
Calgary, Alberta, CANADA
![Page 2: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/2.jpg)
2
Breast cancer statistics
Lifetime probability of developing breast cancer is one in 8.8 (Canada)
Lifetime probability of death due to breast cancer is one in 27 (Canada)
Prevalence: 1% of all women living with the disease
Screening mammography has been shown to reduce mortality rates by 30% to 70%
![Page 3: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/3.jpg)
3
X-ray imaging of the breast
![Page 4: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/4.jpg)
4
Mammography
Signs of Breast Cancer:
Masses
Calcifications
Bilateral asymmetry
Architectural distortion (subtle, often missed)
![Page 5: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/5.jpg)
5
Two standard views per breast:Cranio-caudal and Mediolateral oblique
![Page 6: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/6.jpg)
6
Masses
Breast cancer causes a desmoplastic reaction in breast tissue
A mass is observed as a bright, hyper-dense object
Mammogram with a mass
![Page 7: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/7.jpg)
7
Calcification
Deposits of calciumin breast tissue
Mammogram with calcification
![Page 8: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/8.jpg)
8
Bilateral AsymmetryDifferences in the overall appearance of one breast with reference to the other
![Page 9: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/9.jpg)
9
Architectural Distortion
Third most common mammographic sign of nonpalpable breast cancer
The normal architecture of the breast is distorted
No definite mass visible
Spiculations radiating from a point
Focal retraction or distortion at the edge of the parenchyma
Mammogram with architectural distortion
![Page 10: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/10.jpg)
10
Objectives of computer-aidedprocessing of mammograms
Enhancement of image qualityDetection of subtle signs of cancerQuantitative analysis of featuresObjective aids to diagnostic decisionAccurate and consistent analysisEarlier detection of breast cancer!
![Page 11: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/11.jpg)
11
Some important problems
Detection of:• Breast boundary (skin – air boundary) • Pectoral muscle (in MLO views) • Fibro-glandular disc• Calcifications• Masses and tumors• Curvilinear structures• Bilateral asymmetry (asymmetric densities)• Architectural distortion
![Page 12: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/12.jpg)
12
Computer-aided diagnosis (CAD)
Increased number of cancers detected1
by 19.5%
Increased early-stage malignancies detected1
from 73% to 78%
Recall rate increased1 from 6.5% to 7.7%
50% of the cases of architectural distortion missed2
1 (Freer and Ulissey, 2001) 2 (Baker et al., 2003)
![Page 13: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/13.jpg)
13
Simultaneous contrast
![Page 14: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/14.jpg)
14
Simultaneous contrast
![Page 15: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/15.jpg)
15
Just-noticeable difference
![Page 16: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/16.jpg)
16
Contrast enhancement
Original mammogram Enhanced imagewith calcifications using adaptive-neighborhood
contrast enhancement
![Page 17: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/17.jpg)
17
Examples of benign and malignant calcifications
![Page 18: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/18.jpg)
18
Detection of calcifications by region growing
![Page 19: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/19.jpg)
19
Detection of calcifications by error of prediction
(a) Part of original mammogram
(b) Seeds detected using prediction error
(c) Calcifications detected by region growing
![Page 20: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/20.jpg)
20
Detection of masses by density slicing and texture flow-field analysis
Most benign masses have smooth shapes with convex lobules.
![Page 21: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/21.jpg)
Detection and analysis of tumors
The green parts of the boundary represent concave segments, indicating malignancy.
![Page 22: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/22.jpg)
22
Detection and analysis of tumors
Tumor + FP detectedOrientation field Coherence
![Page 23: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/23.jpg)
23
Detection of a subtle tumor
![Page 24: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/24.jpg)
24
Radiological characterization of masses (BI-RADS)
![Page 25: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/25.jpg)
25
Analysis of masses: feature extraction
Shape analysis:Fractional concavity
Mass region
Normals to contour for computation ofedge sharpness (acutance)
Ribbon for computation of texture features
![Page 26: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/26.jpg)
26
Objective representation of breast masses
(a) b145lc95Fcc = 0.00A = 0.07F8 = 8.11
(b) b164ro94 Fcc = 0.42A = 0.08F8 = 8.05
(c) m51rc97Fcc = 0.64A = 0.09F8 = 8.15
(d) m55lo97Fcc = 0.83A = 0.01F8 = 8.29
benign benign malignant malignantcircumscribed macrolobulated microlobulated spiculated
![Page 27: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/27.jpg)
27
Rank-ordering using shape: Fcc
![Page 28: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/28.jpg)
28
Rank-ordering using acutance
![Page 29: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/29.jpg)
29
Classification of masses
Logistic regression Mahalanobis(pooled)
Linear discriminant analysis KNN = 7 Recall
Features Sens Spec Avg Sens Spec Avg Sens Spec Avg Az Sens Spec Avg Avg
Fcc 90 97.3 94.7 90 97.3 94.7 100 97.3 98.2 0.99 90 97.3 94.7 90.4
A 50 94.6 78.9 75 67.6 70.0 75.0 73.0 73.7 0.73 45 91.7 73.7 63.6
F8 30 86.5 66.7 65 56.8 59.6 75.0 54.0 61.4 0.68 25 67.6 52.6 53.5
Fcc, A 90 97.3 94.7 90 97.3 94.7 100 97.3 98.2 0.98 90 100 96.5 84.6
Fcc, F8 90 97.3 94.7 90 97.3 94.7 100 97.3 98.2 0.99 90 97.3 94.7 85.6
A, F8 55 86.5 75.4 60 70.3 66.7 75.0 73.0 73.7 0.76 55 89.2 73.7 61.6
Fcc, A, F8 90 97.3 94.7 95 97.3 96.5 100 97.3 98.2 0.99 90 97.3 94.7 83.4
14 texture * * * 70 50.0 64.9 65.0 64.9 64.9 0.67 # # # #
![Page 30: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/30.jpg)
30
Content-based retrieval and analysis: benign mass
Query
![Page 31: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/31.jpg)
31
Content-based retrieval and analysis: malignant tumor
Query
![Page 32: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/32.jpg)
Detection of the pectoral muscle edge and the breast boundary using Gabor filters and active contour models
32
![Page 33: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/33.jpg)
Analysis of bilateral asymmetry using Gabor filters
33
The directional distribution of fibroglandulartissue differs between the left and right breasts
![Page 34: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/34.jpg)
34
Architectural distortion
spiculated focal retraction incipient mass
![Page 35: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/35.jpg)
Normal vs. architectural distortion
35
![Page 36: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/36.jpg)
Normal vs. architectural distortion
36
![Page 37: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/37.jpg)
37
Detection of architectural distortion
1. Extract the orientation field
2. Filter and downsample the orientation field
3. Analyze orientation field using phase portraits
4. Post-process the phase portrait maps
5. Detect sites of architectural distortion
![Page 38: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/38.jpg)
38
Gabor filter
( ) ( )fxyxyxyxyx
πσσσπσ
2cos21exp
21,g 2
2
2
2
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛+−=
Gabor parametersDesign parameters
⎥⎦
⎤⎢⎣
⎡′′
⎥⎦
⎤⎢⎣
⎡ −=⎥
⎦
⎤⎢⎣
⎡=
==
yx
yx
l
f
xy
x
θcosθsinθsinθcos
;
2ln22;1
σσ
τστ• line thickness τ
• elongation l• orientation θ
![Page 39: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/39.jpg)
Design of Gabor filters
l = l0τ = τ0θ = θ0
l > l0τ = τ0θ = θ0
l = l0τ > τ0θ = θ0
l = l0τ = τ0θ > θ0 39
![Page 40: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/40.jpg)
40
Extracting the orientation field
Compute the texture orientation (angle) for each pixel
Gabor filtering(line detection)
![Page 41: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/41.jpg)
41
Extracting the orientation field
( )yx,I( )yx,g2
( )yxK ,g
Original image
( )yx,g1
M
Gaborfilter bank (K = 180)
( )
( )K
kyx
yxk kk
ππθ max
max
2,
|},{|maxarg
+−=
= I
Orientation field
( )yx,1I
( )yx,2I
( )yxK ,I
Image resolution: 200 µm/pixelFiltering
![Page 42: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/42.jpg)
Filtering and downsamplingthe orientation field
( )yx,θ
Orientation field
( )[ ]yx,2sin θ
( )[ ]yx,2cos θ
Gaussianfiltering
Gaussianfiltering
( )( )⎥⎦
⎤⎢⎣
⎡yxcyxs
,,arctan
21
( )yxs ,
( )yxc ,
( )yxf ,θ
Filteredorientation field
4↓Downsample ( )yxd ,θ
Downsampledorientation field
Image resolution: 200 µm/pixel
Filtering
Downsampling
Image resolution: 800 µm/pixel
![Page 43: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/43.jpg)
Orientation field:architectural distortion
Original image Gabor magnitude Filtered orientation field
![Page 44: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/44.jpg)
Orientation field: normal case
Original image Gabor magnitude Filtered orientation field
![Page 45: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/45.jpg)
45
Phase portraits
( ) bA , +⎟⎟⎠
⎞⎜⎜⎝
⎛=⎟⎟
⎠
⎞⎜⎜⎝
⎛=
yx
vv
yxy
xvr
node saddle spiral
![Page 46: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/46.jpg)
Phase portrait type
Eigenvaluesof matrix A Streamlines Orientation
field
Node Real, same sign
Real, opposite sign
Complex conjugate
Saddle
Spiral
46
![Page 47: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/47.jpg)
47
Model error
( ) ( )∑∑∆=x y
yx,, 22 bAε
( ) ( ) ( )[ ]bA,,,sin, yxyxyx φθ −=∆
( )bA,, yxφModel-generated field
( )yx,θOrientation field
Local error measure
Sum of the squared error measure
![Page 48: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/48.jpg)
48
Texture analysis usingphase portraits (step 1 of 3)
1. Fit phase portrait model to the moving analysis window
⎥⎦
⎤⎢⎣
⎡−−
=
⎥⎦
⎤⎢⎣
⎡−
=
9.78.4
7.12.03.01.1
b
A
![Page 49: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/49.jpg)
49
Texture analysis usingphase portraits (step 2 of 3)
2. Find phase portrait type and location of fixed point
⎥⎦
⎤⎢⎣
⎡−−
=
⎥⎦
⎤⎢⎣
⎡−
=
9.78.4
7.12.03.01.1
b
A Type: node
Fixed point: x=3, y=5
![Page 50: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/50.jpg)
50
Texture analysis usingphase portraits (step 3 of 3)
3. Cast a vote in the corresponding phase portrait map
Saddle SpiralOrientationfield
Node
![Page 51: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/51.jpg)
51
Post-processing and detection
1. Filter the node map with a Gaussian mask2. Detect peaks in the node map larger than
the other peaks within a radius of 6.4 mm (8 pixels)
3. The peaks indicate the locations of architectural distortion
![Page 52: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/52.jpg)
Phase portrait maps: architectural distortion case
node[0, 1.1]
saddle[0, 0.3x10-3]
spiral[0, 0]
![Page 53: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/53.jpg)
Phase portrait maps: normal case
node[0, 0.98]
saddle[0, 0.2x10-4]
spiral[0, 0]
![Page 54: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/54.jpg)
54
Initial results of detection (2004)
Test dataset: 19 mammograms with architectural distortion(MIAS database)
Sensitivity: 84%
18 false positives per image
![Page 55: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/55.jpg)
0 5 10 15 20 25 300
10
20
30
40
50
60
70
80
90
100
FP/image
Sens
itivi
ty (%
)FROC analysis
55
![Page 56: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/56.jpg)
Reduction of false positives
56
![Page 57: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/57.jpg)
57
Rejection of confounding structures
Confounding structures includeEdges of vesselsIntersections of vesselsEdge of the pectoral muscleEdge of the fibro-glandular disk
![Page 58: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/58.jpg)
58
Detection of curvilinear structures (CLS)
Nonmaximal suppressionIf a pixel in the magnitude image is greater than its neighbors along the direction perpendicular to the local orientation field angle, the pixel is a core CLS pixel
CLSGabor magnitude outputCore CLS pixelNeighboring pixels along normal
![Page 59: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/59.jpg)
59
Nonmaximal suppression
Gabor magnitude output
ROI with a vessel Output of nonmaximalsuppression (NMS)
![Page 60: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/60.jpg)
60
Rejection of confounding structures
Main feature of confounding structures:
Angle from the orientation field and direction perpendicular to the gradient vector differ by less than 30 degrees
(Adaptation of a method by Karssemeijerand te Brake: IEEE TMI 1996)
![Page 61: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/61.jpg)
61
Rejection of confounding CLS
Core CLS pixels detected(Output of NMS)
CLS pixels rejected from further analysis
![Page 62: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/62.jpg)
Rejection of confounding CLS
Core CLS pixels detected (Output of NMS)
CLS pixels rejected from further analysis 62
![Page 63: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/63.jpg)
63
Improved phase portrait analysis
Local error measure weighted by smoothed and downsampled map of CLS pixels
Simulated annealing (SA) applied to obtain initial estimate of phase portrait parameters at every position of analysis window
Global optimization of weighted sum of squared error measure over 6-D space of A and b
Parameters further refined by nonlinear least squares
![Page 64: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/64.jpg)
Improved detection of sites of architectural distortion
Node map without CLS analysis
Node map with CLS analysis
![Page 65: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/65.jpg)
65
Result of detection ofarchitectural distortion
![Page 66: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/66.jpg)
FROC analysis (2005)
0 5 10 15 200
20
40
60
80
100FROC curve
False positives per image
sens
itivi
ty (%
)
With CLS analysis
Without CLS analysis
66
![Page 67: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/67.jpg)
Effect of conditioning number of matrix A on the orientation field
67
![Page 68: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/68.jpg)
Improved results (2006)
• 19 cases of architectural distortion• 41 normal control mammograms (MIAS)
• Symmetric matrix A: node and saddle only• Conditioning number of A > 3 : reject result
• Sensitivity: 84% at 4.5 false positives / image
• Sensitivity: 95% at 9.9 false positives / image68
![Page 69: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/69.jpg)
FROC analysis with symmetric A (2006)
0 5 10 15 200
20
40
60
80
100
FP/image
sens
itivi
ty (%
)
69
![Page 70: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/70.jpg)
70
Conclusion and future work
Phase portraits can be used to detect architectural distortion
Need to reduce false positives further
Evaluate method with a large database
Test method with screening mammograms taken prior to mass formation:
earlier detection of breast cancer
![Page 71: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/71.jpg)
71
Applications of computer-aided diagnosis
Screening program or diagnostic clinic:Consultation by radiologistsDecision support
Second opinionComparison with cases of known diagnosis
Training:Teaching, continuing medical education
Teleradiology, telemedicine:When local expertise is not available
![Page 72: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/72.jpg)
72
Local computer
Remote host computer
Select ROI
Extract features[0.09, 0.02, 0.04]
Network interface Network interface
Mobile software agent Mobile software agent
Secure communication
link
Retrieval results
Query data
Results of retrieval
Viewed onmonitor #2
Query mammogram
Viewed on monitor #1
Indexed atlas
Mammography database
Comparativeanalysis
K-nearest casesand notes for comparative analysis by radiologist
Use of the University of Calgary indexed atlas with mobile agents
![Page 73: Computer-Aided Diagnosis of Breast Cancer: Towards the ...ranga/enel697/CADbcex.pdf · Architectural distortion (subtle, often missed) 5 Two standard views per breast: Cranio-caudal](https://reader035.vdocuments.site/reader035/viewer/2022071103/5fdceafe07eed56387529c79/html5/thumbnails/73.jpg)
73
AcknowledgmentNatural Sciences and Engineering Research Council of CanadaAlberta Heritage Foundation for Medical ResearchCanadian Breast Cancer FoundationScreen Test: Alberta Program for the Early Detection of Breast CancerMy research students, coworkers, and collaborators