Imaging, mainly due to its impact on medicine and biology, has been
selected as one of the greatest achievements of the twentieth century
by the National Academy of Engineering. In the last several decades,
medical imaging systems have advanced in quantum leaps. There
have been substantial improvements in characteristics such as sensitivity,
resolution, and acquisition speed. Multislice, 64-Slice, and very
soon, 256-Slice computer tomography (CT) scanners, for instance,
allow the visualization of the entire coronary tree, even atherosclerotic
plaques within the coronaries with extremely high accuracy and detail.
Similar advances have occurred in other medical imaging modalities
such as magnetic resonance imaging (MRI) and positron emission
tomography (PET).
Substantial effort has been put into the integration of different modalities.
These systems are also called hybrid systems. The integration of CT
and PET scanners has enabled physicians to localize biochemical activity
(functional) with a high degree of certainty in the human body and
will significantly impact molecular imaging, which can be defined as in
vivo imaging of biochemical or molecular activity in the organ. There
is also significant development in small animal imaging modalities. In
vivo and in vitro molecular imaging has already been contributing to
the advancement of the study of the genome and efficacy of new drugs.
With the help of imaging, now biologists can get a snapshot of almost the
entire range of genomic activity (expression or disexpression of genes)
within a diseased tissue in a matter of days. It will not be long before
physicians can visualize, in vivo, the biochemical processes triggered by
a disease. All of this may soon result in a paradigm shift in healthcare.
It may open up the possibility of designing drugs as per a patient’s individual
genetic profile.
Advanced techniques of image processing and analysis find widespread
use in biology and medicine. In medical and biological fields,
image data are ubiquitously used in clinical as well as scientific studies
to infer details regarding the process under investigation whether
it be a disease process or a biological or biochemical process. Today,
perhaps, health care institutions alone produce the largest amount of
image data, which are used in diagnosis and treatment of patients.
Information provided by medical images has become an indispensable
part of today’s patient care. As the number of images produced
increases, utilization and handling of image data are becoming an
increasingly formidable task for engineers, scientists, and medical
physicists.
There are two main issues that concern the field of image processing
and analysis applied to medical applications. These are the following:
- Improving the quality of the acquired image data
- Extraction of information (i.e., feature) from medical image data
in a robust, efficient, and accurate manner
Image enhancement techniques such as noise filtering, contrast and edge
enhancement; and image restoration techniques that focus on removing
degradations in images, all fall within the former category, whereas image
analysis methods deal primarily with the latter issue.
The sheer size of images in medical applications has been increasing
rapidly with the advent of imaging technologies; hence, transfer and storage
issues are also challenging tasks. The main goal of developing efficient
image data compression techniques is to address these two issues.
Unlike the images produced in industrial applications, the images generated
in medical and biological applications are complex and vary substantially
from application to application. In addition, as one can imagine,
the field of image processing and analysis has to tackle a diverse and
complex set of problems. Because this is such a vast subject, we focus on
certain topics that we consider important in the fields of medicine and
biology.
Some concepts in image processing and analysis are theory-intensive
and may be difficult for beginners to grasp. Explaining complex topics
in image processing through examples and MATLAB algorithms is the
principal aim of this book. While working on this book, we tried to strike
a balance between theory and practice. We wanted to keep it neither shallow
nor complex, so that readers from diverse fields would comprehend
without difficulty. Image processing techniques in general are ad-hoc
in the sense that they are optimized and tailored to solve a particular
problem in hand, although they are based on solid mathematical theories.
That is, they are not applicable to a wide range of applications or situations.
This lack of generalizability often forces scientists and researchers
to resort to the method of trial-and-error. The algorithms provided in this
book will help the scientists and researchers to quickly identify the most
effective method of solution for a particular problem at hand.
This book will help readers understand advanced concepts through
algorithms applied to real-world problems in medicine and biology.
The
examples and exercises included in every chapter will make the book
suitable for use as a textbook for students at the senior undergraduate or
graduate level who are studying image processing and analysis for the
first time; or as a reference book for researchers, scientists, and biologists
in the related fields.
In addition to fundamental topics in image processing and analysis,
the book covers new areas such as nonlinear diffusion filtering (NDF) or
partial differential equation (PDE)-based image filtering, and relatively
advanced topics such as segmentation methods based on Markov random
field (MRF) modeling. Statistical and stochastic modeling in image processing
are emphasized in this book.
In the past, computation times and memory demand for 3-D algorithms
were unrealistic, but with the advent of computer (or CPU) technology,
processing time and memory needs in 3D are no longer a prohibitive factor.
Therefore, we have described applications to 3-D volume images. We
tried to extend the techniques (algorithms) in this book to 3D whenever we
could.
Finally, the reader with a moderate level of calculus, linear algebra, and
probability and statistics background will find this book reasonably easy
to comprehend.
The content of this book can be summarized as follows.
Chapter 1 discusses major imaging modalities in diagnostic radiology.
They include CT, MRI, gamma cameras, and single photon emission
tomography (SPECT) systems and PET.
In Chapter 2, we discuss fundamental image processing techniques. We
have presented basic but useful as well as advanced image processing and
analysis techniques with MATLAB codes or functions. Most of these techniques
are not available in the Image Processing Toolbox, and are hence
unique. Some of these techniques have also be used in the subsequent
chapters of the book.
Chapter 3 covers the theory of probability and statistics on which some
image processing and analysis methods are built. This chapter will help
the readers build a background that may them help follow the other chapters
such as chapters 6 and 7.
Chapter 4 introduces the 2-D fast Fourier transform with unique
examples. We also briefly discuss the tomographic image reconstruction
method filtered back projection as it is one of the medical applications of
the Fourier transform.
Chapter 5 deals with nonlinear diffusion filtering as well as some of
the PDE-based image denoising techniques. This relatively recent class of
filters has found many applications in medical imaging because of their
superior performance in removing noise and preserving edge sharpness.
Chapter 6 discusses most of the intensity-based image segmentation
methods. It discusses the thresholding techniques based on betweenclass
variance, the Kullback function, entropy, and mixture modeling. We
also discuss K-means and fuzzy C-means clustering techniques and their
application to image segmentation.
Chapter 7 discusses the image segmentation method based on MRF in
detail. By MRF modeling, we model the spatial dependency of the intensities
in a local neighborhood. The conditional density of the intensities
and the MRF local dependency model is combined under the Bayesian
framework in which the MRF model is viewed as a prior. This formulation
leads to the maximum a posteriori (MAP) estimate of the true image (i.e.,
the image not deteriorated by the noise and the imaging system). We have
discussed both the deterministic and probabilistic methods of finding the
MAP estimate.
Chapter 8 describes fundamental image analysis methods applicable to
a wide range of problems in image analysis. These methods include, for
example, regions’ properties, boundary analysis, curvature analysis, and
line and circle detection using the Hough transform.
Chapter 9 discusses deformable models and their application to image
segmentation. The theory of both parametric and geometric deformable
models has been covered.
Chapter 10 includes three applications of image processing and analysis.
Unique approaches to image processing and analysis problems have
been described. Through these applications, we wish that the reader will
also gain experience in tackling a problem at hand.
We would like to thank Dr. Ersin Bayram for contributing the Deformable
Model chapter and the MRI section of Chapter 1. We are thankful to
Mr. Kostas Chantziantoniou for contributing to the Computer Tomography
section in Chapter 1.
We would also like to thank Edward D. Carroll and John Schneider for
proofreading some chapters and for pointing out places where the text
was very clear.