Institute of Chemical Technology, Prague
Department of Computing and Control Engineering
BIOMEDICAL SIGNAL AND IMAGE
PROCESSING
Ph.D. Thesis
Victor Musoko
Chemical and Process Engineering
Technical Cybernetics
Prague July 2005
ii
Declaration
The research work described in this dissertation was done by the author from September
2002 till July 2005 at the Department of Computing and Control Engineering, Institute
of Chemical Technology, Prague. The author do hereby declare that the contents of the
thesis except where indicated are entirely original and are a result of the research work
he has carried out.
Victor Musoko
Acknowledgements
I would like to take this opportunity to thank all those who have contributed to this thesis.
First and foremost my big thanks go to my supervisor Prof. AleˇsProch´azka, who gave me
the opportunity to pursue the research. I appreciated his guidance, encouragement and the
helpful discussions. The work presented here would certainly not have been accomplished
without his influence and support. From my supervisor, I have learned a great deal about
problem solving and presentation of the results of my research work.
Last but not least, I would also like to express my deep thanks to Ing. AleˇsPavelka,
for his tireless help at the time it was needed most and Assoc. Prof. Jarom´ır Kukal, for
introducing me to the world of programming in C/C
++
and for his important discussions.
Thanks in general to all my colleagues in the Department of Computing and Control
Engineering.
iv
Abstract
The main goal of the thesis is to show the de-noising algorithms based upon the discrete
wavelet transform (DWT) that can be applied successfully to enhance noisy multidimen-
sional magnetic resonance (MR) data sets i.e two-dimensional (2-D) image slices and
three-dimensional (3-D) image volumes. Noise removal or de-noising is an important task
in image processing used to recover a signal that has been corrupted by noise. Random
noise that is present in MR images is generated by electronic components in the instrumen-
tation. The thesis present both 2-D image decomposition, thresholding and reconstruction
and the 3-D de-noising of MR image volumes using the DWT as a new approach which
can be used in the processing of biomedical images. A novel use of the complex wavelet
transform for the study and application in the de-noising of MR images is presented in the
main part of the thesis. Segmentation of image textures using a watershed transform and
a wavelet based feature extraction for classification of image textures by a competitive
neural network will be shown.
Further topic of interest to be presented in the thesis is the visualization of 2-D MR
slices and 3-D image volumes using some MATLAB functions. This makes it possible
to visualize images without the need of special glasses especially for 3-D image volumes
or using special expensive software programs which will need some bit of expertise. The
application of the proposed algorithms is mainly in the area of magnetic resonance imaging
(MRI) as an imaging technique used primarily in medical field to produce high quality
images of the soft tissues of the human body. An insight to the visualization of MRI data
sets i.e. 2-D image slices or 3-D image volumes is of paramount importance to the medical
doctors.
The thesis presents the theory of the fundamental mathematical tools (discrete Fourier
transform (DFT) and DWT) that are used for the analysis and processing of biomedical
images. DWT plays an increasingly important role in the de-noising of MR images. 3-D
digital image processing, and in particular 3-D DWT, is a rapidly developing research area
with applications in many scientific fields such as biomedicine, seismology, remote sensing,
material science, etc. The 3-D DWT algorithms are implemented as an extension of the
existing 2-D algorithms. The performance of the de-noising algorithms are quantitatively
assessed using different criteria namely the mean square error (MSE), peak signal-to-noise
ratio (PSNR) and the visual appearance. The results are discussed in accordance to the
type of noise and wavelets implemented. The properties of wavelets make them special
in that they have a good time and frequency localization which make them ideal for the
processing of non-stationary signals like the biomedical signals (EEG, ECG,..) and images
(MR). The traditional Fourier transform only provides the spectral information of a signal
and thus it is not suitable for the analysis of non-stationary signals.
A novel complex wavelet transform (CWT) which was introduced by Dr. Nick Kings-
bury of Cambridge University is analyzed and implemented in the main part of the thesis.
The description of the dual tree implementation of CWT is followed by its analysis and
discussions devoted to its advantages over the classical wavelet transform. This enhanced
transform is then applied to MRI data analysis. Experimental results show that complex
wavelet de-noising algorithm can powerfully enhance the PSNR in noisy MRI data sets.
The further part of the thesis devoted to the description of basic principles of a wa-
tershed transform for segmentation of MR images. After its verification for simulated
textures it is used for segmentation of a human knee MR image. The anatomical regions
of the knee which includes the muscle, bone and tissue can be easily distinguished by this
algorithm. Segmentation is an important field in medical applications and can be used for
disease diagnosis e.g. detecting brain tumor cells.
Texture analysis of artificial textures based on image wavelet decomposition is consid-
ered as a pre-processing method for the classification of the textures. The wavelet features
are obtained by using the mean or the standard deviations of the wavelet coefficients. For
the classification of the textures these feature vectors form the inputs to a competitive
neural network. The work also presents own algorithms for class boundaries evaluation.
Texture analysis is used in a variety of applications, including remote sensing, satellite
imaging, medical image processing, etc.
Finally, I conclude and give suggestions for future research work. The thesis also gives
a review of the de-noising and visualization of biomedical images on the web using the
Matlab Web Server (MWS).
Keywords
Time-Frequency and Time-Scale Signal Analysis, Discrete Wavelet Transform, Complex
Wavelet Transform, Image De-noising, Biomedical Image Processing, Watershed Algo-
rithm, Segmentation, Feature Extraction, Image Visualization