Texture Analysis Methods for Medical Image Characterisation
95
Feature Method Kruskall-Wallis
Mann-Whitney
Mann-Whitney
Mann-Whitney
vCJD/spor CJD t1 vCJD/spor CJD t2 spor CJD t1/ spor CJD t2
Value Power
Value
Power
Value
Power
Value
Power
f
1
GTSDM
0.052
-
0.556
-
0.071
-
0.057
-
f
2
GTSDM
0.005
+
0.111
-
0.036
+
0.057
-
f
3
GTSDM
0.004
+
0.016
+
0.036
+
1.000
-
f
4
GTSDM
0.050
+
0.111
-
0.143
-
0.114
-
f
5
GTSDM
0.000
+
0.016
+
0.036
+
0.057
-
f
6
GTSDM
0.004
+
0.016
+
0.036
+
0.857
-
f
7
GTSDM
0.002
+
0.016
+
0.036
+
0.400
-
f
8
GTSDM
0.007
+
0.016
+
0.571
-
0.571
-
f
9
GTSDM
0.061
-
0.016
+
0.786
-
0.400
-
f
10
GLRLM
0.020
+
0.111
-
0.143
-
0.057
-
a
b
f
GLRLM
0.065
-
0.111
-
0.250
-
0.114
-
11
f
12
GLRLM
0.000
+
0.016
+
0.036
+
0.057
-
f
13
GLRLM
0.050
+
0.111
-
0.143
-
0.114
-
f
14
GLRLM
0.050
+
0.111
-
0.143
-
0.114
-
f
15
GLRLM
0.000
+
0.016
+
0.036
+
0.114
-
f
16
GLRLM
0.001
+
0.032
+
0.036
+
0.057
-
f
Fourier
17
(Ring 1)
0.002
+
0.016
+
0.036
+
0.400
-
f
Fourier
18
(Ring 2)
0.002
+
0.016
+
0.036
+
0.400
-
f
Fourier
19
(Ring 3)
0.002
+
0.016
+
0.036
+
0.400
-
f
Fourier
20
(Ring 4)
0.002
+
0.016
+
0.036
+
0.400
-
f
Fractal
21
0.004
+
0.016
+
0.036
+
0.857
-
Table 1. Power of the features used to discriminate between: vCJD and sporadic CJD type 1;
vCJD and sporadic CJD type 2; sporadic CJD type 1 and sporadic CJD type 2. The exact
c
d
significance value is a measure of the discriminatory power of each feature. (+ indicates
good discriminatory power, - indicates poor discriminatory power).
Fig. 12. Microscopic images showing the distribution of prion protein in the molecular layer
of the cerebellum of cases of vCJD (a), sporadic CJD type 1 (b) and sporadic CJD type 2 (c).
The choice of ring filters used was based on extensive trials on similar data sets. The radii of
Figure d shows the distribution of prion protein in a larger region formed from nine
the final set of filters chosen is summarised in Table 2. The main objective of this work was
overlapping microscopic images.
to use texture analysis to investigate and study the distribution of prion protein found in
vCJD and sporadic CJD. All of the Fourier ring features used showed significant
discriminatory power in discriminating between vCJD and sporadic CJD types 1 and 2.
However, they were unable to discriminate between sporadic CJD types 1 and 2.
96
Biomedical Imaging
9. References
Filter Inner Radius Outer Radius
Bastian, F.O. (1991). Creutzfeldt-Jakob disease and other transmissible human spongiform
1
r
r
2
5
1
2
encephalopathies. St Louis: Mosby Year Book.
Bajscy, R. (1973). Computer description of textured surfaces, Proceedings of the Joint
2
r
r
2
10
1
2
Conference on Artificial Intelligence, August, pp. 572-579.
3
r
r
Brodatz, P. (1966). Textures: a photographic album for artists and designers. Dover Publications,
2
10
1
5
ISBN 0-486-40699-7, New York.
4
r
r
2
20
1
10
Bruce, V.; Georgeson, M. & Green, P. (2003). Visual Perception: Psychology and Ecology.
Table 2. Inner and outer radii of the ring filters used to discriminate between: vCJD and
Psychology Press Publications, ISBN 1841692387, UK.
sporadic CJD type 1; vCJD and sporadic CJD type 2; sporadic CJD type 1 and sporadic CJD
Conners, R. & Harlow, C. (1980). Theoretical comparison of texture algorithms. IEEE
type 2.
Transactions on Pattern Analysis and Machine Intelligence, Vol. 2, No. 3, pp. 204-222.
Devijver, P.A. & Kittler, J. (1982). Pattern recognition: a statistical approach. Prentice Hall, 0-13-
7. Summary
654236-0, London.
Distani, R.; Nappi, M. & Riccio, D. (2006). A range/domain approximation error-based
Texture analysis methods are useful for discriminating and studying both distinct and
approach for fractal image compression. IEEE Transactions on Image Processing, Vol.
subtle textures in multi-modality medical images. Practical implementation requires careful
15, No. 1, pp. 89-97.
consideration of the power of the individual features to discriminate between textures. This
Duda, R.; Hart, P. & Stork, D. (2001). Pattern Classification, John Wiley & Sons, ISBN 0-471-
is essential to reduce the influence that heavily correlated features, and features with little
05669-3.
discriminatory power, have on the overall classification. Statistical texture analysis
Esgiar, A.; Naguib, R.; Sharif, B.; Bennet, M. & Murray, A. (2002). Fractal analysis in the
techniques are constantly being refined by researchers and the range of applications is
detection of colonic cancer images. IEEE Transactions on Information Technology in
increasing. Fractal approaches, which offer the convenience of characterising a textured
Biomedicine, Vol. 6, No. 1, pp 54-58.
region by a single measure, appear more application-specific than statistical approaches and
Feder, J. (1988). Fractals. Plenum Press, ISBN 0-306-42851-2, New York.
require more research. Algorithmic advances have been made on the use of full 3D texture
Galloway, M.M. (1975). Texture analysis using grey-level run lengths. Computer Graphics and
analysis approaches and the publications in this area demonstrate that this is a promising
Image Processing, Vol. 4, pp. 172-179.
area of research. This is particularly important given that biomedical image data with near
Gonzalez, R.C. & Woods, R.E. (2001). Digital Image Processing, Prentice Hall, ISBN 0-201-
isotropic resolution is becoming more common in clinical environments. However, it has
18075-8.
been shown that there is minimal loss of discriminatory power when 2D techniques are
Haralick, R.M.; Shanmugam K. & Dinstein, I. Texture features for image classification. IEEE
applied in the coronal and sagittal planes.
Transactions on Pattern Analysis and Machine Intelligence, Vol. SMC-3, No. 6, Nov
1973, 610-621.
Hartigan, J. (1975). Clustering Algorithms. John Wiley & Sons, ISBN 0-471-35645-X.
8. Acknowledgements
ICRU, (1999). International Commission on Radiation Units and Measurements. ICRU
The support of Dr. A.T. Redpath, Dr. D.B. McLaren and Professor S. McLaughlin is
Report 62: Prescribing, Recording and Reporting Photon Beam Therapy
gratefully acknowledged for helpful discussion and preparation of the results presented in
(Supplement to ICRU 50), Oxford University Press.
Case Study 1. I am also grateful to Professor J.W. Ironside and Dawn Everington for their
Ironside, J.W. (1998). Neuropathological findings in new variant Creutzfeldt-Jakob disease
assistance in obtaining the results presented in Case Study 2. The support of the laboratory
and experimental transmission of BSE. FEMS Immunology & Medical Microbiology,
staff at the National CJD Surveillance Unit, UK is acknowledged for preparation of the
Vol. 21, No. 2, pp. 91-95.
tissue samples used in Case Study 2. The support of the staff in the NHS Department of
Ironside, J.W.; Head, M.W.; Bell, J.E.; McCardle, L. & Will, R.G. (2000). Laboratory diagnosis
Oncology Physics and the School of Engineering at the University of Edinburgh is gratefully
of variant Creutzfeldt-Jakob disease, Histopathology, Vol. 37, No. 1, pp. 1-9.
acknowledged. The James Clerk Maxwell Foundation is gratefully acknowledged for the
Jain, A.K. & Chandrasekaran. (1982). Dimensionality and sample size considerations. In:
provision of financial assistance to continue this work.
Pattern Recognition in Practice, Krishnaiah, P.R. & Kanal, L.N., Vol. 2, pp. 835-888.
North Holland.
Jain, A.K. & Zongker, D. (1997). Feature selection: evaluation, application and small sample
performance. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 2,
No. 19, pp. 153-158.
Julesz, B. (1975). Experiments in the visual perception of texture. Scientific American, Vol.
232, pp 34-43.
Texture Analysis Methods for Medical Image Characterisation
97
9. References
Filter Inner Radius Outer Radius
Bastian, F.O. (1991). Creutzfeldt-Jakob disease and other transmissible human spongiform
1
r
r
2
5
1
2
encephalopathies. St Louis: Mosby Year Book.
Bajscy, R. (1973). Computer description of textured surfaces, Proceedings of the Joint
2
r
r
2
10
1
2
Conference on Artificial Intelligence, August, pp. 572-579.
3
r
r
Brodatz, P. (1966). Textures: a photographic album for artists and designers. Dover Publications,
2
10
1
5
ISBN 0-486-40699-7, New York.
4
r
r
2
20
1
10
Bruce, V.; Georgeson, M. & Green, P. (2003). Visual Perception: Psychology and Ecology.
Table 2. Inner and outer radii of the ring filters used to discriminate between: vCJD and
Psychology Press Publications, ISBN 1841692387, UK.
sporadic CJD type 1; vCJD and sporadic CJD type 2; sporadic CJD type 1 and sporadic CJD
Conners, R. & Harlow, C. (1980). Theoretical comparison of texture algorithms. IEEE
type 2.
Transactions on Pattern Analysis and Machine Intelligence, Vol. 2, No. 3, pp. 204-222.
Devijver, P.A. & Kittler, J. (1982). Pattern recognition: a statistical approach. Prentice Hall, 0-13-
7. Summary
654236-0, London.
Distani, R.; Nappi, M. & Riccio, D. (2006). A range/domain approximation error-based
Texture analysis methods are useful for discriminating and studying both distinct and
approach for fractal image compression. IEEE Transactions on Image Processing, Vol.
subtle textures in multi-modality medical images. Practical implementation requires careful
15, No. 1, pp. 89-97.
consideration of the power of the individual features to discriminate between textures. This
Duda, R.; Hart, P. & Stork, D. (2001). Pattern Classification, John Wiley & Sons, ISBN 0-471-
is essential to reduce the influence that heavily correlated features, and features with little
05669-3.
discriminatory power, have on the overall classification. Statistical texture analysis
Esgiar, A.; Naguib, R.; Sharif, B.; Bennet, M. & Murray, A. (2002). Fractal analysis in the
techniques are constantly being refined by researchers and the range of applications is
detection of colonic cancer images. IEEE Transactions on Information Technology in
increasing. Fractal approaches, which offer the convenience of characterising a textured
Biomedicine, Vol. 6, No. 1, pp 54-58.
region by a single measure, appear more application-specific than statistical approaches and
Feder, J. (1988). Fractals. Plenum Press, ISBN 0-306-42851-2, New York.
require more research. Algorithmic advances have been made on the use of full 3D texture
Galloway, M.M. (1975). Texture analysis using grey-level run lengths. Computer Graphics and
analysis approaches and the publications in this area demonstrate that this is a promising
Image Processing, Vol. 4, pp. 172-179.
area of research. This is particularly important given that biomedical image data with near
Gonzalez, R.C. & Woods, R.E. (2001). Digital Image Processing, Prentice Hall, ISBN 0-201-
isotropic resolution is becoming more common in clinical environments. However, it has
18075-8.
been shown that there is minimal loss of discriminatory power when 2D techniques are
Haralick, R.M.; Shanmugam K. & Dinstein, I. Texture features for image classification. IEEE
applied in the coronal and sagittal planes.
Transactions on Pattern Analysis and Machine Intelligence, Vol. SMC-3, No. 6, Nov
1973, 610-621.
Hartigan, J. (1975). Clustering Algorithms. John Wiley & Sons, ISBN 0-471-35645-X.
8. Acknowledgements
ICRU, (1999). International Commission on Radiation Units and Measurements. ICRU
The support of Dr. A.T. Redpath, Dr. D.B. McLaren and Professor S. McLaughlin is
Report 62: Prescribing, Recording and Reporting Photon Beam Therapy
gratefully acknowledged for helpful discussion and preparation of the results presented in
(Supplement to ICRU 50), Oxford University Press.
Case Study 1. I am also grateful to Professor J.W. Ironside and Dawn Everington for their
Ironside, J.W. (1998). Neuropathological findings in new variant Creutzfeldt-Jakob disease
assistance in obtaining the results presented in Case Study 2. The support of the laboratory
and experimental transmission of BSE. FEMS Immunology & Medical Microbiology,
staff at the National CJD Surveillance Unit, UK is acknowledged for preparation of the
Vol. 21, No. 2, pp. 91-95.
tissue samples used in Case Study 2. The support of the staff in the NHS Department of
Ironside, J.W.; Head, M.W.; Bell, J.E.; McCardle, L. & Will, R.G. (2000). Laboratory diagnosis
Oncology Physics and the School of Engineering at the University of Edinburgh is gratefully
of variant Creutzfeldt-Jakob disease, Histopathology, Vol. 37, No. 1, pp. 1-9.
acknowledged. The James Clerk Maxwell Foundation is gratefully acknowledged for the
Jain, A.K. & Chandrasekaran. (1982). Dimensionality and sample size considerations. In:
provision of financial assistance to continue this work.
Pattern Recognition in Practice, Krishnaiah, P.R. & Kanal, L.N., Vol. 2, pp. 835-888.
North Holland.
Jain, A.K. & Zongker, D. (1997). Feature selection: evaluation, application and small sample
performance. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 2,
No. 19, pp. 153-158.
Julesz, B. (1975). Experiments in the visual perception of texture. Scientific American, Vol.
232, pp 34-43.
98
Biomedical Imaging
Kachouie, N. & Fieguth, P. (2007). A medical texture local binary pattern for TRUS prostate
Press, W.; Flannery, B.; Teulkolsky, S. & Vetterling, W. Numerical recipes in C: the art of
segmentation, Proceedings of the 29th International Conference of the IEEE Engineering
scientific computing. Cambridge University Press, ISBN 0-521-35465-X, UK.
and Biology Society, Lyon, France, August 23rd – 26th.
Russ, J.C. (1994). Fractal surfaces. Plenum Press, ISBN 0-306-44702-9, New York.
Karahaliou, A.; Boniatis, I; Skiadopoulos, S. & Sakellaropoulos, F. (2008). Breast cancer
Sammon, J. (1969). A nonlinear mapping for data structure analysis. IEEE Transactions on
diagnosis: analyzing texture of tissue surrounding microcalcifications. IEEE
Computing, Vol, C-18, pp. 401-409.
T