![](bg4.png)
Specificity measures the accuracy among
positive instances, and is calculated by dividing the
true negatives by the number of all other organ
slices. Sensitivity measures the accuracy among
negative instances, and is calculated by dividing the
number of true positives by the total number of that
specific organ slices. Precision measures show how
consistent the results can be reproduced. Accuracy
reflects the overall correctness of the classifier, and
is calculated by adding the true positives and
negatives together and dividing by the entire number
of organ slices.
4 WAVELET - RIDGELET
COMPARISON
Tables 2-5 in the Appendix show a comparison of
accuracy, precision, specificity, and sensitivity
results, for each tissue of interest for the three
wavelet-based texture features and the ridgelet-based
texture features respectively. Within all the
wavelets, the Haar wavelet outperformed all others
for most organs and performance measures. The
only exception is the backbone, for which the
Daubechies and Coiflet wavelets produce slightly
better results. The performance for the Haar-based
descriptors in all other organs was significantly
higher, thus indicating that these descriptors yield
the highest discriminating power among the
wavelet-based features.
The results also show that the ridgelet-based
texture features outperform all wavelet-based
descriptors. Accuracy rates for Wavelet-based
texture descriptors range between 85 - 93%, while
ridgelet-based accuracy rates are in the 91 - 97%
range. Precision rates for the wavelets are between
55 - 91%, compared to 73 - 93% for ridgelets.
Specificity rates for the wavelets are in the 82-97%
range, while specificity for the ridgelet descriptors is
in the 92-98% range. Furthermore, sensitivity rates
for the wavelets are in the 35-87% range, whereas
ridgelets are between 72-94%. The lower bound of
the sensitivity range for wavelets is due to the poor
performance of those descriptors (especially Coiflets
and Daubechies) for Heart and Spleen. The texture
of the images for these two organs is quite similar
and the classifier often mistakes the two organs for
one another. Further investigation is needed to
determine the underlying cause for the poor
performance of the Heart and Spleen.
Overall, the ridgelet-based descriptors have
significantly higher performance measures, with
accuracy rates approximately four percent higher
than any other feature set for all individual organs.
This was expected due to the fact that the ridgelet
transform is able to capture multi-directional
features, as opposed to the wavelet transform which
focuses mainly on horizontal, vertical, and diagonal
features, which are not dominant in medical CT scan
images. One of the limitations of using ridgelet-
based descriptors is the fact that ridgelets are most
effective in detecting linear radial structures, which
are not the main component of medical images. A
recent extension of ridgelets is the curvelet
transform. Curvelets have been proven to be
particularly effective at detecting image activity
along curves instead of radial directions (Starck
Donoho & Candes 1999). We are currently
investigating the use of curvelet-based texture
descriptors and we expect this to further improve the
ability of our classifier to successfully classify each
tissue sample.
REFERENCES
Xu, D., Lee, J., Raicu, D.S., Furst, J.D., & Channin, D.
(2005). Texture Classification of Normal Tissues in
Computed Tomography. The 2005 Annual Meeting of
the Society for Computer Applications in Radiology.
Channin, D., Raicu, D.S., Furst, J.D., Xu, D.H., Lilly, L.,
& Limpsangsri, C. (2004). Classification of Tissues in
Computed Tomography using Decision Trees. Poster
and Demo. The 90
th
Scientific Assembly and Annual
Meeting of Radiology Society of North America.
Semler, L., Dettori, L., & Furst, J. (2005). Wavelet-Based
Texture Classification of Tissues in Computed
Tomography. Proceedings of the 18th IEEE
International Symposium on Computer-Based Medical
Systems, 265-270.
Do, M.N., & Vetterli, M. (2003) The Finite Ridgelet
Transform for Image Representation. IEEE
Transactions on Image Processing, 12, 16 - 28.
LeBorgne, H.L., & O’Connor, N. (2005). Natural Scene
Classification and Retrieval Using Ridgelet-based
Image Signatures. Advanced Concepts for Intelligent
Vision Systems, 20-23.
Starck, J.L., Donoho, D.L., & Candes, E.J. (1999).
Astronomical Image Representation by the Curvelet
Transform. Astronomy &Astrophysics, 398, 785-800.
Semler, L., Dettori, L., & Kerr, B. (2006). Ridgelet-Based
Texture Classification in Computed Tomography,
submitted to IEEE Computer Society Conference on
Computer Vision and Pattern Recognition.
Dara, B & Watsuji, N. (2003). Using Wavelets for Texture
Classification. IJCI Proceedings of International
Conference on Signal Processing, ISN 1304-2386.
VISAPP 2006 - IMAGE ANALYSIS
288