6. Chapter 1:
Introduction
Problem Statement:
“Mammographic images usually noisy and
have poor contrast with wide range of
anatomical patterns leads to greater
number of false positive cases among
radiologist.” Image enhancement might
help to increase the detection
performance.
7. Chapter 1:
Introduction
Research Goal:
To
perform the quantitative image
analysis using Receiver Operating
Characteristic (ROC) analysis of
enhanced and original images.
8. Chapter 1:
Introduction
Research Objectives:
To
develop
image
enhancement
techniques and determine whether the
techniques improves the image quality.
To evaluate subjectively mammographic
phantom image by subjective evaluation
rating scales.
To compare the quality of images with and
without enhancement techniques by using
ROC analysis.
9. Chapter 1:
Introduction
Research Scope:
Mammographic phantom containing
micronodules, nodules and fibrils was developed.
The mammographic phantom images were
obtained using Digital Mammography System.
For preprocessing techniques, the mammographic
images were denoised using low pass Gaussian
filter.
The structures of images were enhanced using
morphological techniques and 2D wavelet
transform.
Observers evaluated the mammographic
phantom image subjectively.
11. Chapter 2:
Literature Review
Previous Studies in Wavelet Applications
Song et al. (2006) claimed that wavelet transform and
morphological techniques gave less false positives (FPs)
[2].
Amutha et al. (2012)proved that Biorthogonal filter with
two level of decomposition combined with
morphological techniques improved the image quality
[1] [3].
References:
[1] Amutha, S., Ramesh Babu, D.R., Ravi Shankar, M. and Harish Kumar, N. (2011).
Mammographic Image Enhancement using Modified Mathematical Morphology and BiOrthogonal Wavelet. IEEE Transaction On Medical Imaging: 548 - 552.
[2] Bozek J., Mustra, M., Delac, K. and Grgic, M. (2009). A Survey of Image
Procesing Algorithms in Digital Mammography in Grgic, M., Delac, K. and
Ghanbari, M., (Eds), Recent Advances in Multimedia Signal Processing and
Communication, Berlin Heidelberg, Springer, 631- 657.
12. Chapter 2:
Literature Review
Previous Studies in Morphological Techniques
Kimori (2011) proved that the shape parameter of a
structuring element which set to the shape of the
structures improved the image contrast [4].
Kumar et al. (2012) proved that the mathematical
morphology enhanced the image contrast and
wavelet for denoising improved the image quality [3].
References:
[3] Harish, K.N., Amutha, S. and Ramesh, B.D.R. (2012). Enhancement of
Mammographic Images using Morphology and Wavelet Transform. International
Journal of Computer Technology & Applications. 3: 192 – 198.
[4] Kimori, Y. (2011). Mathematical Morphology-based Approach to the
Enhancement of Morphological Features in Medical Images. Journal of Clinical
Bioinformatics. 1: 1 – 10.
13.
Flow charts of research activities
Development of anthropomorphic
mammographic phantom.
Image acquisition
Image Dataset
Image Preprocessing
Image Enhancement Techniques
Measuring image quality
Image Scoring
Receiver operating characteristic (ROC) analysis
14. Chapter 3:
Research Methodology
Flow charts of Research Activities
Develop a mammographic phantom
that contains micronodules, nodules and
fibrils accuracy detection.
Image Acquisition
Obtain mammographic phantom
images using Hologic Selenia Full Field
Digital Mammography System at
Hospital Sultan Ismail, Johor Bahru.
Image Dataset
Convert mammographic phantom
images in DICOM formats into TIFF
format.
15. Chapter 3:
Research Methodology
Image
Enhancement
Image without
enhancement
Image Preprocessing using
Low Pass Gaussian filter to
denoise images
Morphological
techniques using disk
structuring elements
2D wavelet transform
using Biorthogonal 2.8 filter
16. Chapter 3:
Research Methodology
Measuring image quality by using Peak
Signal to Ratio (PSNR) and MSE (mean
squared error).
Image Scoring
Observers interpreted the structures in
each mammographic images
subjectively.
ROC analysis
Operating points were
calculated using Excel
2010, CORROC2 and ROCFIT.
19. Chapter 3:
Research Methodology
Image Acquisition
Hologic Seleria Full Field
Digital Mammography
with focal spot 0.3 mm
at Diagnostic Imaging
Department, Hospital
Sultan Ismail, Johor
Bahru
24. Image Enhancement:
Morphological techniques
The images were dilated using a „disk‟ shaped structuring
elements to improve the image contrast.
MATLAB command:
%morphological techniques enhancement
se = strel('disk', 4);
b = imdilate(I2, se);
figure (3), imshow(b, []);
I3=imclose(b, se);
figure (4), imshow(I3, []);
25. Image Enhancement:
2D Wavelet Transform
The mammographic images were enhanced using 2
level decomposition of Biorthogonal 2.8 filter.
28. Image Scoring
..Msc
Thesis2APPENDIX A.docx
Each observers interpreted the embedded structures
in mammographic phantom images subjectively
based on 5 confidence levels and subjective
evaluation on contrast visibility, sharpness and
overall image quality.
29. Receiver Operating Characteristics (ROC) Analysis
The interpretation score were calculated using
Microsoft Excel as attached in ..latest.xls.
CORROC2 and ROCFIT software was used to
process the clustered data from the ROC scoring
and operating calculation dataset as attached in
..OMORN1.RESnotepad.txt and
..MORM1.RESpresent.txt.
30. Chapter 4:
Results And Discussion
Chapter 4: Results and Discussion
Visual performance of original and enhanced
images.
Receiver Operating Characteristics (ROC) analysis
The comparison of ROC Az values between
observers
Subjective Evaluation Rating Scales
31. Visual Performance of Original and Enhanced
Images
Original image with
29 kVp, mAs = 124.6
under 5.8 cm
compression using Rh
filter.
Morphological enhanced
image.
Wavelet transform enhanced image using
Biorthogonal 2.8 wavelet filter, two level
decomposition
32. Figures below shows the graphs of MSE parameter versus
PSNR values obtained for original
images, morphological enhanced images and wavelet
transform enhanced images.
33. Receiver Operating Characteristics (ROC)
Curves
ROC curves were plotted using CORROC2
and ROCFIT. The significance of differences
of areas could be determined by the p
values.
38. Detection Of Micronodules
Comparison of original and morphological
enhanced images.
observer 1 (detection of micronodules)
1
observer 2 (detection of micronodules)
1
original images (Az = .9973)
morphological enhanced images (Az = 0.9888)
0.6
0.4
0.2
original images (Az = 0.9742)
0.8
TPF (sensitivity)
0.8
0.6
0.4
0.2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
0.1
0.2
0.3
FPF (1 - specificity)
0.4
0.5
0.6
FPF (1 - specificity)
observer 3 (detection of micronodules)
1
TPF (sensitivity)
0
morphological enhanced images (Az = 0.9993)
0.8
0.6
0.4
0.2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
FPF (1 - specificity)
0.7
0.8
0.9
1
0.7
0.8
0.9
1
39. Detection of micronodules
Comparison of original and wavelet transform
enhanced images.
observer 3 (detection of micronodules)
observer 1 (detection of micronodules)
1
1
original images (Az = 0.9985)
wavelet transform enhanced images (Az = 0.9936)
0.8
TPF (sensitivity)
0.8
0.6
0.4
0.2
0.6
0.4
0.2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
wavelet transform enhanced images (Az = 0.9933)
0.7
0.8
0.9
1
0
0
0.1
0.2
0.3
FPF (1 - specificity)
0.4
0.5
observer 4 (detection of micronodules)
TPF (sensitivity)
1
wavelet transform enhanced images (Az = 0.9918)
0.8
0.6
0.4
0.2
0
0
0.6
FPF (1 - specificity)
0.1
0.2
0.3
0.4
0.5
0.6
FPF (1 - specificity)
0.7
0.8
0.9
1
0.7
0.8
0.9
1
40. The comparison of ROC Az values between observers based on
trapezium method.
The comparison of Az in detection of nodules from original,
morphological enhanced and wavelet transform enhanced
images by all observers.
41. Detection of Fibrils
1
0.98
ROC Az Values
0.96
0.94
Original Images
0.92
Morphological
Enhancement Images
0.9
Wavelet Transform
Enhancement Images
0.88
0.86
Observer 1 Observer 2 Observer 3 Observer 4
Observers
The comparison of Az in detection of fibrils from original,
morphological enhanced and wavelet transform enhanced
images by all observers.
42. Detection of Micronodules
1.02
1
ROC Az Values
0.98
0.96
Original Images
0.94
Morphological Enhancement Images
0.92
Wavelet Transform Enhancement
Images
0.9
0.88
0.86
Observer 1
Observer 2
Observer 3
Observer 4
Observers
The comparison of Az in detection of micronodules from
original, morphological enhanced and wavelet transform enhanced
images by all observers.
43. Subjective Evaluation Rating Scales (Radiologist)
Image Dataset
Contrast Visibility/
σ value
Sharpness/
σ value
Overall image quality/
σ value
a)Original images
4.7
1.3
4.9
1.7
5.1
1.5
b) Morphological
Enhancement Images
3.2
0.8
3.6
0.9
3.6
0.9
c) Wavelet transform
Enhancement images
4.5
1.8
5.1
1.9
5.0
1.8
Mean rating scale values based on contrast visibility, sharpness
and overall image quality and standard deviation from radiologist.
44. Subjective Evaluation Rating Scales (OridinaryObserver)
Mean rating scale values based on contrast visibility, sharpness and
overall image quality and standard deviation from ordinary observer.
45. Discussions
Based on ROC curves Az values, morphological
enhancement improved the detection for nodules and
micronodules .Morphological enhancement improved
the detection of nodules based on trapezium method.
In morpholgical enhancement by using „disk‟ stucturing
elements, the dark details from the structures are
reduced. Dilation add pixels to the structures and
closing reduced the background pixels.
Wavelet transform enhancement could improve the
detection for fibrils and micronodules based on the Az
values from ROC curves and trapezium method.
46. The detection of fibrils and micronodules are
improved by reducing noisy pixels of images by
applying decomposition using Discrete Wavelet
Transform.
Both observers rated that original images are better
than enhanced images for contrast visibility, sharpness
and overall image quality based on the subjective
evaluation rating scales.
AEC function and higher effective energy beam (Rh
target and Rh filter) improved the quality of image by
minimizing quantum noise.
47. Chapter 5:
Conclusions And Future Work
Conclusions
The enhancement methods could not increase the
detection performance based on ROC analysis and
subjective evaluation rating scales.
PSNR values for all images did not affect the quality
of image and detection performance.
The best technical factors using AEC function
(mAs, kVp, filtration and target material) improved
the quality of original images .
48. Chapter 5:
Conclusions And Future Work
Future Work
The algorithms for the enhancement methods
have to improve to increase
effectiveness, sensitivity and the quality of images
with better visualization.
The algorithms should enhance contrast, spatial
resolution details, edge response and remove
noise without changing the morphology of the
structures.