Medical Image Registration

ABSTRACT
Radiological images are increasingly being used in healthcare and medical research. There is,
consequently, widespread interest in accurately relating information in the different images
for diagnosis, treatment and basic science. This article reviews registration techniques used to
solve this problem, and describes the wide variety of applications to which these techniques
are applied. Applications of image registration include combining images of the same subject
from different modalities, aligning temporal sequences of images to compensate for motion
of the subject between scans, image guidance during interventions and aligning images from
multiple subjects in cohort studies. Current registration algorithms can, in many cases,
automatically register images that are related by a rigid body transformation (i.e. where tissue
deformation can be ignored). There has also been substantial progress in non-rigid
registration algorithms that can compensate for tissue deformation, or align images from
different subjects. Nevertheless many registration problems remain unsolved, and this is
likely to continue to be an active field of research in the future. Medical image registration is
a critical process in many areas of healthcare, facilitating the combination and comparison of
images acquired from different modalities, times, or subjects. One of the primary goals of
image registration is to ensure that the anatomical structures in different images align
accurately, which is vital for accurate diagnosis and treatment planning. For instance, in
radiation therapy, accurate alignment of pre-treatment planning images with real-time
imaging can ensure that the targeted areas are precisely irradiated, minimizing damage to
healthy tissues. Similarly, in neurosurgery, registering preoperative and intraoperative images
helps surgeons navigate the brain more precisely, reducing the risk of complications. Rigid
registration methods, where only translation and rotation are applied to align images, are
often sufficient when dealing with structures that do not undergo significant deformation.
These methods are widely used in applications where the patient or subject remains stationary
during image acquisition, such as in CT and MRI scans. However, for more complex
situations where tissue deformation occurs, such as during respiration or organ movement,
non-rigid registration techniques are required. These methods can model and compensate for
changes in shape, making them essential for applications such as cardiac imaging, where the
heart's motion must be accounted for. The development of robust non-rigid registration
algorithms has been an area of active research in recent years. These algorithms aim to map
one image to another by considering local transformations that accommodate anatomical
variations and motion. They are particularly useful in longitudinal studies, where images of
the same subject are taken at different time points, as well as in studies involving multiple
subjects with varying anatomies. Despite significant progress, non-rigid registration still faces
challenges, particularly in handling large deformations, noise, and varying image quality,
making it a topic of ongoing research in the medical imaging field.

INTRODUCTION

Functional imaging using single photon emission computed tomography (SPECT) and
positron emission tomography (PET) is extremely valuable in the diagnosis of various
disorders. Uncertainty in the anatomic definition on SPECT and PET images, however,
sometimes limits their usefulness. To overcome this problem, a combination of magnetic
resonance images (MRI) and X-ray computed tomography (CT) images with functional
SPECT or PET images of the same sections of the body, is used. This provides
complementary anatomic (MRI or CT) and physiological (SPECT or PET) information that is
of great importance to research, diagnosis, and treatment.
Two basic types of medical images are made: functional body images (such as SPECT or
PET scans), which provide physiological information, and structural images (such as CT or
MRI), which provide an anatomic map of the body. Different medical imaging techniques
may provide scans with complementary and occasionally conflicting information. The
combination of images can often lead to additional clinical information not apparent in the
separate images. The goal of image fusion is to impose a structural anatomic framework on
functional images. Often in a functional image, there simply isn't enough anatomic detail to
determine the position of a tumor or other lesion.
Although, the construction of a composite, overlapping medical image - described in the field
as medical image registration has been primarily used in the fusion of functional and
anatomical images, it has also been applied to a series of the same modality images.
Examples of this are registration of SPECT images of the same subject in follow-up studies
or in a comparison of an image with normal uptake properties to an image with suspected
abnormalities. In addition, image registration of SPECT and PET images and the registration
of SPECT and PET images with anatomic atlases, provide an important means to evaluate
comparative uptake properties of SPECT and PET radiopharmaceuticals, and to correlate
uptake properties with anatomy.
Medical image registration has been applied to the diagnosis of breast cancer, colon cancer,
cardiac studies, wrist and other injuries, inflammatory diseases and different neurological
disorders including brain tumors, Alzheimer's disease and schizophrenia. This method has
also been utilized in radiotherapy, mostly for brain tumors, and by cranio-facial surgeons to
prepare for and simulate complex surgical procedures.
One area where image registration plays an important role is in medical imaging in the early
detection of cancers. Radiologists often have difficulty locating and accurately identifying
cancer tissue, even with the aid of structural information such as CT and MRI because of the
low contrast between the cancer and the surrounding tissues in CT and MRI images. Using
SPECT and radioactively labeled monoclonal antibodies it is possible to obtain high contrast
images of the concentration of antibodies in tumors. However, sometimes it is difficult to
determine the precise location of the high concentration of the radioactive isotope in SPECT
or PET images in relation to anatomic structures, such as vital organs and surrounding
healthy tissue. Image registration is a visualization tool that can significantly aid in the early
detection of tumors and other diseases, and aid in improving the accuracy of diagnosis.
Medical image registration is an essential technique in modern diagnostic imaging,
particularly when different imaging modalities are used to capture distinct but
complementary aspects of a patient's condition. By aligning images from modalities such as
MRI, CT, PET, and SPECT, clinicians can obtain a more comprehensive view of a patient's
anatomy and physiology. This is especially important in oncology, where tumors often
exhibit characteristics that are better visualized through functional imaging (like PET or
SPECT) but lack the anatomical detail provided by CT or MRI. By fusing these images,
clinicians can localize tumors more accurately and assess their size, shape, and relation to
nearby structures, thus aiding in treatment planning and monitoring. In neurological
disorders, image registration has proven invaluable in tracking disease progression and
evaluating the effects of treatment. For conditions like Alzheimer's disease, brain tumors, or
epilepsy, imaging modalities like PET and SPECT provide insight into brain activity, while
CT or MRI offers structural information. Combining these modalities can reveal subtle
changes in brain function and structure that might not be apparent when viewed separately.
For example, in Alzheimer's disease, early detection of changes in brain metabolism through
PET scans, when registered with MRI scans, can help in identifying regions of the brain that
are most affected by the disease, potentially guiding more targeted therapeutic interventions.
One of the challenges of medical image registration, particularly when combining functional
and anatomical images, is the need for accurate alignment between the different modalities.
Variations in patient positioning, breathing patterns, or organ motion during image
acquisition can introduce discrepancies between images, making it difficult to achieve precise
registration. To address this, advanced algorithms have been developed to handle these
issues, including rigid and non-rigid registration techniques. Rigid registration involves the
alignment of images by only rotating or translating the data, while non-rigid methods allow
for more flexibility, adjusting for deformations and movements in soft tissues. The choice
between these methods depends on the specific characteristics of the images and the degree
of tissue deformation that needs to be accounted for.
Furthermore, as medical imaging technologies continue to evolve, the integration of artificial
intelligence (AI) and machine learning (ML) techniques has enhanced the accuracy and speed
of image registration. Deep learning algorithms, in particular, have shown promise in
automating the process of registration by learning complex patterns in data and improving the
alignment of images without requiring manual intervention. These technologies have the
potential to reduce the time spent on image processing and enhance the reproducibility and
consistency of results across different healthcare providers and institutions. The future of
image registration is likely to involve further advancements in AI, allowing for real-time
image fusion and analysis, thus transforming clinical workflows and enabling more
personalized, precise healthcare interventions. As medical image registration becomes
increasingly integrated into clinical practice, its applications continue to expand beyond
traditional diagnostic use. One area that is seeing significant growth is in the field of
radiotherapy, where precise tumor localization and treatment planning are critical for
delivering effective and safe radiation doses. In radiation oncology, image registration is used
to align pre-treatment diagnostic images, such as CT and MRI scans, with real-time imaging
during treatment to ensure that radiation is targeted accurately at the tumor while minimizing
damage to surrounding healthy tissues. Advanced registration techniques, particularly those
that can account for tumor deformation and organ motion, are crucial for improving treatment
outcomes and reducing the risk of complications in cancer patients undergoing radiation
therapy.
Another important application of image registration is in surgical planning and navigation,
particularly in complex procedures like neurosurgery, orthopedic surgery, and craniofacial
surgery. In these areas, pre-operative imaging data, such as CT or MRI scans, are registered
with real-time intra-operative imaging, such as ultrasound or endoscopy, to provide surgeons
with a more precise view of the anatomy during the procedure. This fusion of imaging
modalities helps guide surgical instruments accurately, improving the precision of the surgery
and reducing the risk of complications. In some cases, image registration is also used in
robotic surgery, where the robotic system can be programmed to follow a pre-determined
surgical path based on registered images, further enhancing the accuracy and safety of the
procedure. Looking ahead, the future of medical image registration is likely to be influenced
by the continued development of multimodal imaging techniques and the increasing
availability of patient-specific data. For example, the use of 3D imaging technologies and
functional imaging techniques, such as functional MRI (fMRI), may provide even more
detailed information for registration, allowing for better analysis of tissue function and
structure. Additionally, personalized medicine, which tailors treatment based on individual
patient characteristics, could benefit from the integration of image registration with genetic
and molecular data. By combining imaging data with genetic profiles and biomarkers,
clinicians may be able to achieve more accurate diagnoses and design more effective,
individualized treatment plans. As these advancements continue, medical image registration
will play a central role in transforming healthcare, enhancing both diagnostic and therapeutic
capabilities.

Literature survey
A Survey Of Medical Image Registration (1998): The purpose of this paper is to present a
survey of recent (published in 1993 or later) publications concerning medical image
registration techniques. These publications will be classified according to a model based on
nine salient criteria, the main dichotomy of which is extrinsic versus intrinsic methods. The
statistics of the classification show definite trends in the evolving registration techniques,
which will be discussed. At this moment, the bulk of interesting intrinsic methods is based on
either segmented points or surfaces, or on techniques endeavouring to use the full information
content of the images involved.
Within the current clinical setting, medical imaging is a vital component of a large number of
applications. Such applications occur throughout the clinical track of events; not only within
clinical diagnostis settings, but prominently so in the area of planning, consummation, and
evaluation of surgical and radiotherapeutical procedures. The imaging modalities employed
can be divided into two global categories: anatomical and functional. Anatomical modalities,
i.e., depicting primarily morphology, include X-ray, CT (computed tomographya ), MRI
(magnetic resonance imagingb), US (ultrasoundc ), portal images, and (video) sequences
obtained by various catheter “scopes”, e.g., by laparoscopy or laryngoscopy. Some prominent
derivative techniques are so detached from the original modalities that they appear under a
separate name, e.g., MRA (magnetic resonance angiography), DSA (digital subtraction
angiography, derived from Xray), CTA (computed tomography angiography), and Doppler
(derived from US, referring to the Doppler effect measured). Functional modalities, i.e.,
depicting primarily information on the metabolism of the underlying anatomy, include
(planar) scintigraphy, SPECT (single photon emission computed tomographyd), PET
(positron emission tomographye), which together make up the nuclear medicine imaging
modalities, and fMRI (functional MRI). With a little imagination, spatially sparse techniques
like, EEG (electro encephalography), and MEG (magneto encephalography) can also be
named functional imaging techniques. Many more functional modalities can be named, but
these are either little used, or still in the pre-clinical research stage, e.g., pMRI (perfusion
MRI), fCT (functional CT), EIT (electrical impedance tomography), and MRE (magnetic
resonance elastography). The growing need for accurate and reliable medical image
registration methods has led to the development of numerous techniques, each offering its
own strengths and limitations. In the early stages of medical image registration, extrinsic
methods, which rely on external markers or systems for alignment, were the primary
approach. These methods involve the use of physical devices such as fiducial markers or rigid
body tracking systems to provide a reference for spatial alignment. While these methods are
often accurate, they can be cumbersome, time-consuming, and require additional equipment,
making them less ideal for certain clinical settings. On the other hand, intrinsic methods,
which aim to align images based on the content of the images themselves, have gained more
attention in recent years. These techniques, which utilize the image features such as edges,
contours, or intensity patterns, can be more flexible and require fewer external devices,
making them more suitable for real-time clinical applications. One of the major challenges in
medical image registration is dealing with the variability of patient anatomy and the presence
of different tissue types. For example, the liver in two different patients can vary in size,
shape, and internal structure, making it difficult to align images from different modalities. To
address these issues, researchers have developed advanced techniques such as deformable
image registration, which allows for the non-rigid transformation of images to account for
variations in tissue geometry. These techniques use sophisticated algorithms that model tissue
deformation and can achieve a high degree of accuracy in aligning images from different
modalities, even when there are significant differences in the underlying anatomy. However,
deformable image registration is computationally intensive and often requires careful tuning
to avoid errors caused by incorrect assumptions or inaccurate models. Another critical aspect
of medical image registration is the fusion of multi-modal images, where images from
different imaging modalities are combined to provide complementary information. For
example, combining CT or MRI scans with functional imaging data from SPECT or PET can
provide both anatomical and physiological insights, enabling more comprehensive diagnosis
and treatment planning. Registration methods designed for multi-modal fusion must not only
align the images accurately but also handle differences in image characteristics, such as
contrast, resolution, and noise levels. Techniques like mutual information-based registration,
which measures the statistical dependence between images, have been widely used in this
area. This approach is particularly useful in the registration of modalities like MRI and PET,
where differences in contrast and intensity can make direct pixel-to-pixel matching difficult.
The future of medical image registration looks promising, as ongoing advancements in
computational techniques, artificial intelligence, and machine learning continue to enhance
the accuracy, speed, and efficiency of these methods. Deep learning algorithms, in particular,
have shown great potential in automating the registration process by learning complex image
patterns and features from large datasets. These algorithms can improve image alignment by
recognizing subtle anatomical structures that may be difficult for traditional methods to
detect. As the volume of medical imaging data increases, AI-driven registration methods will
likely become indispensable tools in clinical practice, helping clinicians make faster and
more accurate decisions. Additionally, the integration of real-time image registration into
surgical and interventional procedures, such as robotic surgery or image-guided biopsies, will
further enhance the precision and outcomes of medical treatments. The integration of
multimodal data through medical image registration is paving the way for personalized
medicine. As more sophisticated imaging techniques become available, clinicians can obtain
highly detailed information about a patient's anatomy, physiology, and pathology. This
enables more tailored treatment plans that take into account the unique characteristics of each
individual, improving the chances of successful outcomes. For example, in oncology,
combining functional imaging data (such as PET scans) with anatomical data (like MRI) can
help identify tumor regions, assess their metabolic activity, and evaluate how they interact
with surrounding healthy tissue. This comprehensive view allows for more accurate tumor
characterization and precise targeting of treatments such as radiation therapy or
chemotherapy. The development of real-time image registration techniques for dynamic
procedures, such as minimally invasive surgery and radiation therapy, is transforming how
treatments are delivered. Real-time registration ensures that the surgeon or radiologist has
access to the most up-to-date anatomical data during the procedure, even as the patient's
position or the internal anatomy changes. This is particularly critical in applications like brain
surgery, where even slight shifts in position can impact the accuracy of the procedure. Real-
time image registration allows for continuous monitoring and adjustment, enhancing the
safety and effectiveness of the intervention. As medical imaging technology continues to
evolve, the combination of more powerful imaging modalities, advanced computational
techniques, and real-time processing will significantly improve patient care and treatment
outcomes.
Medical Image Registration (2001): Image registration of X-ray images is used to correctly
align the patient before the cancer tumour is treated with sets of high dose X-ray radiation.
Before the first treatment, a simulated X-ray image is taken showing the best possible
localisation of the tumour. From then on, after each of the radiation treatments of the tumour,
a low dose of the X-ray theurapeutic radiation is used to take an image of the tumour and its
surroundings. This so called portal X-ray image is compared to the simulated image to decide
whether the patient should be moved before the next treatment in order to improve the
accuracy of the theurapeutic beam and, hence, prevent the high dose radiation from hitting
other surrounding tissues close to the tumour. Due to differences in quality of the X-ray
plates used for the recording of the simulated and the portal image, the images differ a lot in
contrasts, noise level, and possibly even scale and size. The diverse quality of the images is
the main problem of the image registration task. At present, the comparison of the X-ray
images is carried out by hand. A software for automating the process would decrease the
influence of human intuition on the treatment, decrease the treatment time, and enable less
qualified personnel to carry out the treatment. Much research has already been done within
the field of image registration, many with various results. The preceding pilot study suggests
that a deeper study, that would go beyond the scope of this project, is needed for the task.
This report strives for presenting fields of possible approaches to the problem and tests
performed within these fields. Hopefully, the report will give some guidance on directions
that a future research could take.
Medical imaging plays a pivotal role in modern clinical practice, supporting a wide range of
applications throughout the clinical pathway. It is integral not only in diagnostic settings but
also in the planning, execution, and evaluation of surgical and radiotherapy procedures. The
imaging modalities utilized in healthcare are broadly classified into two categories:
anatomical and functional. Anatomical imaging techniques, such as X-ray, CT (computed
tomography), MRI (magnetic resonance imaging), and ultrasound (US), focus primarily on
visualizing the structure and morphology of organs and tissues. These modalities, including
advanced derivatives like magnetic resonance angiography (MRA) and computed
tomography angiography (CTA), are essential for identifying abnormalities in organ structure
and aiding in surgical and diagnostic decision-making. On the other hand, functional imaging
techniques provide insights into the metabolic processes occurring within the body’s tissues,
offering a deeper understanding of disease progression. Modalities such as SPECT (single-
photon emission computed tomography), PET (positron emission tomography), and fMRI
(functional MRI) are widely used in assessing physiological activities like blood flow, tissue
metabolism, and brain activity. These functional imaging techniques complement anatomical
imaging by revealing underlying disease mechanisms that may not be visible through
structural imaging alone. Additionally, emerging functional modalities like perfusion MRI
(pMRI) and magnetic resonance elastography (MRE) show promise in advancing clinical
diagnostics and treatment planning, although they are still largely in the research phase.
Together, these imaging technologies provide comprehensive diagnostic and therapeutic tools
that enhance patient care and clinical outcomes. The integration of anatomical and functional
imaging techniques has significantly enhanced the capabilities of medical imaging,
particularly in the context of image registration. Image registration, the process of aligning
images from different modalities or time points, is crucial in providing a coherent and unified
view of a patient's condition. In the case of cancer treatment, particularly in radiation therapy,
accurate image registration allows for the precise targeting of tumors while minimizing
damage to surrounding healthy tissues. As mentioned in the study, portal X-ray images taken
during treatment can be compared to simulated images, and any discrepancies can guide
adjustments to the patient’s positioning to ensure optimal radiation delivery. The challenge,
however, lies in the varying quality of the images, including differences in contrast, noise
levels, and scale, which complicates the registration process. Automating this process through
advanced image registration algorithms can help reduce human error, speed up the workflow,
and improve the accuracy of the treatment plan. In addition to cancer treatment, medical
image registration is also essential in various other fields, including neurosurgery,
orthopedics, and cardiac imaging. In neurosurgery, for example, precise image registration is
used to overlay preoperative MRI or CT scans with intraoperative images to guide the
surgeon in locating and removing tumors or other abnormal tissues. Similarly, in orthopedics,
image registration techniques are applied to align preoperative CT scans with intraoperative
X-rays, improving the accuracy of procedures like joint replacement or fracture fixation. In
cardiac imaging, registration of functional and anatomical images allows for better
assessment of myocardial function and tissue viability, aiding in the diagnosis of coronary
artery disease and other cardiac conditions. The ability to accurately correlate functional and
anatomical information provides a clearer and more comprehensive understanding of a
patient's condition, leading to better outcomes. The advancements in medical image
registration are driven by the need for increased accuracy and efficiency in clinical
procedures. Traditional manual methods of image registration are time-consuming and prone
to error, especially when dealing with large datasets or images with varying quality.
Automated image registration algorithms, leveraging machine learning and artificial
intelligence, offer the potential to streamline the process, making it faster and more reliable.
Recent developments in deep learning-based approaches, such as convolutional neural
networks (CNNs), have shown promise in improving the accuracy and robustness of image
registration. These methods can learn to automatically identify corresponding features across
different images, even in the presence of noise, distortions, or variations in image quality.
The integration of such algorithms into clinical practice could greatly enhance the speed and
precision of procedures, ultimately improving patient care. As medical imaging technology
continues to evolve, there is a growing need for standardized image registration protocols that
can be applied across different imaging modalities and clinical settings. While current
techniques are highly effective, challenges remain in dealing with the inherent variability of
patient anatomy and the differences in image quality across various devices. The future of
image registration in healthcare lies in the development of more robust and adaptable
algorithms that can handle these complexities. Research in this field is ongoing, with efforts
focused on improving the performance of image registration systems, enhancing automation,
and making these technologies more accessible to healthcare providers worldwide. With
continued advancements, medical image registration will play an even more crucial role in
enhancing diagnostic accuracy, guiding surgical procedures, and improving overall patient
outcomes. In addition to improving the accuracy and efficiency of medical procedures, the
integration of advanced image registration techniques has the potential to drive innovations in
personalized medicine. By combining multiple imaging modalities, clinicians can gain a
more comprehensive understanding of a patient's unique anatomical and physiological
characteristics. This holistic view enables more tailored treatment plans that take into account
the individual variability in disease presentation and progression. Furthermore, as the
technology advances, the ability to integrate real-time imaging with treatment planning can
facilitate dynamic adjustments during surgeries or radiation therapies, optimizing the
treatment process and minimizing risks. Ultimately, continued research and development in
medical image registration promise to revolutionize the way healthcare providers approach
diagnosis, treatment, and patient management, leading to more precise, efficient, and
personalized care.
Retrospective Registration Of Tomographic Brain Images (1996): In modern clinical
practice, the clinician can make use of a vast array of specialized imaging techniques
supporting diagnosis and treatment. For various reasons, the same anatomy of one patient is
sometimes imaged more than once, either using the same imaging apparatus (monomodal
acquisition ), or different ones (multimodal acquisition). To make simultaneous use of the
acquired images, it is often necessary to bring these images in registration, i.e., to align their
anatomical coordinate systems. The problem of medical image registration as concerns
human brain images is addressed in this thesis. The specific chapters include a survey of
recent literature, CT/MR registration using mathematical image features (edges and ridges),
monomodal SPECT registration, and CT/MR/SPECT/PET registration using image features
extracted by the use of mathematical morphology.
The development of medical imaging apparatus has truly soared in the past decades.
Nowadays, the clinician can employ a vast array of specialized imaging techniques
supporting the diagnostic and treatment track. For various reasons, more than one image is
often acquired of the same anatomy, either monomodally or multimodally. In the monomodal
case, only one imaging apparatus is used. Several monomodal images of the same anatomy
are, e.g., used in growth monitoring, in comparison of rest and stress conditions, in
comparison of ictal and interictal conditions, and in vascular imaging. When there is a need
to surpass the limits set by a single imaging device, multimodal images can be acquired.
Multimodal imagery can be used, e.g., to extend knowledge on anatomy (by combining
anatomical modalities such as CT1 and MRI2), or to link physiological and anatomical
information (by combining an anatomical modality with a functional modality such as
SPECT3 or PET4). It is by no means trivial to use all of the available image information to its
maximum extent. This thesis addresses the highly specialized research area of integration of
different images of the same anatomy. More exactly, we focus on the problem of registration
of the image data involved, i.e., finding the correct geometrical transformation that brings one
image in precise spatial correspondence with another image. Registration is a necessary pre-
processing step in many instances where information from several images is to be combined.
For example, an image of an epileptic brain, constructed by subtracting functional images of
the brain in ictal and interictal condition, is only meaningful if the two original images
involved are properly registered. Another example is the projection of meaningful visual
information into the ocular of an operating microscope during surgery, e.g., indicative of the
surgical target and of vascular structures that are paramount not to be damaged. Needless to
say, the displayed information must be accurately registered with the anatomy visible through
the microscope. The process of medical image registration has become a cornerstone in
modern clinical practice, especially in cases where multiple images of the same anatomy need
to be integrated for a comprehensive understanding of the patient’s condition. This is
particularly critical in the diagnosis and treatment of neurological disorders, where accurate
alignment of brain images from different modalities (such as CT, MRI, SPECT, and PET) is
essential. Each imaging technique provides unique insights: CT and MRI offer detailed
anatomical images, while SPECT and PET provide functional information, such as blood
flow and metabolic activity. Combining these images allows clinicians to make more
informed decisions about treatment plans, monitor disease progression, and guide
interventions with greater precision. However, the challenge in registering these images lies
not only in aligning the coordinate systems but also in addressing differences in image
quality, resolution, and contrast. Different imaging modalities often have varying sensitivities
to the underlying structures, which may result in mismatches during registration. The
development of robust algorithms to handle these discrepancies is crucial for improving the
accuracy and reliability of multimodal image registration. These algorithms often rely on
image features such as edges, ridges, or texture to guide the alignment process, with
mathematical morphology providing an essential tool for extracting such features from the
images. The success of these methods is dependent on the precision with which the geometric
transformation is applied to bring the images into spatial correspondence. The importance of
image registration is particularly evident in the context of brain imaging, where
understanding the spatial relationship between different functional and anatomical regions is
key to treating conditions such as epilepsy, tumors, or neurodegenerative diseases. For
instance, in epilepsy, comparing ictal (seizure) and interictal (non-seizure) brain images can
provide valuable information about the areas of the brain involved in seizures. Proper
registration ensures that these images can be accurately compared, leading to more precise
localization of seizure foci and improved surgical planning. Additionally, for conditions like
brain tumors, the combination of anatomical and functional images enables a better
understanding of tumor boundaries, vascular involvement, and metabolic activity, which is
critical for effective treatment planning. Looking ahead, the continued advancement of image
registration techniques holds great promise for enhancing medical imaging capabilities. As
newer modalities and technologies, such as functional MRI (fMRI) and diffusion tensor
imaging (DTI), emerge, the need for more sophisticated registration algorithms will only
increase. These developments are expected to lead to even more accurate, personalized
treatment options for patients, particularly in the realm of neurosurgery and oncology.
Furthermore, the integration of artificial intelligence and machine learning techniques into
image registration algorithms could improve automation, making it easier and faster for
clinicians to combine and analyze complex imaging data, ultimately enhancing clinical
decision-making and patient outcomes. In the field of radiotherapy, accurate image
registration is essential for ensuring that radiation is delivered to the correct target while
minimizing damage to surrounding healthy tissue. During cancer treatment, patients often
undergo multiple imaging sessions before, during, and after radiation therapy. Registration of
these images, especially when acquired using different modalities (e.g., CT for planning and
PET for assessing tumor metabolism), allows for the precise alignment of the tumor and
surrounding tissues across different time points. This not only improves the effectiveness of
the treatment but also reduces the risk of side effects by preventing radiation from being
delivered to healthy organs. The development of real-time image registration techniques,
potentially aided by machine learning, will further enhance the precision and adaptability of
radiotherapy, improving outcomes for patients undergoing cancer treatment.
Software For Image Registration: Algorithms, Accuracy, Efficacy (2013): Image
registration is finding increased clinical use both in aiding diagnosis and guiding therapy.
There are numerous algorithms for registration, which all involve maximizing a measure of
similarity between a transformed floating image and a fixed reference image. The choice of
the similarity measure depends, to some extent, on the application. Methods based on the use
of the joint intensity histogram have become popular because of their flexibility and
robustness. A distinction is made between rigid-body and non-rigid transformations. The
latter are needed for inter-subject registration or intra-subject registration in cases where the
region of the body of interest is not considered rigid. Non-rigid transformation is normally
achieved using a global model of the deformation but can also be defined by a set of locally
rigid transformations, each constrained to a small block in the image. There is scope for
further research on the incorporation of appropriate constraints, especially for the application
of non-rigid transformations to nuclear medicine studies. Most of the initial practical
concerns regarding image registration have been overcome and there is increasing availability
of commercial software. There are several approaches to the validation of registration
software, with validation of non-rigid algorithms being particularly difficult. Studies have
demonstrated the accuracy on the order of half a pixel for both intra- and inter-modality
registration (typically 2 to 3 mm). Although hardware-based registration has now become
possible by using dual-modality instruments, software-based registration will continue to play
an important role in nuclear medicine.
Image registration is finding increased clinical use both in aiding diagnosis and guiding
therapy. There are numerous algorithms for registration, which all involve maximizing a
measure of similarity between a transformed floating image and a fixed reference image. The
choice of the similarity measure depends, to some extent, on the application. Methods based
on the use of the joint intensity histogram have become popular because of their flexibility
and robustness. A distinction is made between rigid-body and non-rigid transformations. The
latter are needed for inter-subject registration or intra-subject registration in cases where the
region of the body of interest is not considered rigid. Non-rigid transformation is normally
achieved using a global model of the deformation but can also be defined by a set of locally
rigid transformations, each constrained to a small block in the image. There is scope for
further research on the incorporation of appropriate constraints, especially for the application
of non-rigid transformations to nuclear medicine studies. Most of the initial practical
concerns regarding image registration have been overcome and there is increasing availability
of commercial software. There are several approaches to the validation of registration
software, with validation of non-rigid algorithms being particularly difficult. Studies have
demonstrated the accuracy on the order of half a pixel for both intra- and inter-modality
registration (typically 2 to 3 mm). Although hardware-based registration has now become
possible by using dual-modality instruments, software-based registration will continue to play
an important role in nuclear medicine.
Medical Image Registration Using Mutual Information (2003): Analysis of multispectral
or multitemporal images requires proper geometric alignment of the images to compare
corresponding regions in each image volume. Retrospective three-dimensional alignment or
registration of multimodal medical images based on features intrinsic to the image data itself
is complicated by their different photometric properties, by the complexity of the anatomical
objects in the scene and by the large variety of clinical applications in which registration is
involved. While the accuracy of registration approaches based on matching of anatomical
landmarks or object surfaces suffers from segmentation errors, voxel-based approaches
consider all voxels in the image without the need for segmentation. The recent introduction of
the criterion of maximization of mutual information, a basic concept from information theory,
has proven to be a breakthrough in the field. While solutions for intrapatient affine
registration based on this concept are already commercially available, current research in the
field focuses on interpatient nonrigid matching.
With current medical imaging modalities, such as computed tomography (CT), magnetic
resonance imaging (MRI), or positron emission tomography (PET), it is possible to routinely
and minimally invasively acquire three-dimensional (3-D) images of the internal organs of
the human body. These radiological images provide detailed information about anatomy and
function of the imaged organs and play an essential and crucial role in the medical decision
process in diagnosis, therapy planning, and assessment. The classical representation of
medical images, printed on radiological film and visualized using a light box, makes image
interpretation necessarily subjective and qualitative. However, most modern medical image
acquisition systems generate digital images that can be processed by a computer and
transferred over computer networks. Digital imaging allows to extract objective, quantitative
parameters from the images by image analysis. Medical image analysis exploits the
numerical representation of digital images to develop image processing techniques that
facilitate computer-aided interpretation of medical images. The continuing advancement of
image acquisition technology and the resulting improvement of radiological image quality
have led to an increasing clinical need and physician’s demand for quantitative image
interpretation in routine practice, imposing new and more challenging requirements for
medical image analysis. Typical applications include for instance volumetry of organs or
lesions using CT or MRI, morphometry of the brain using MRI, correlation of anatomical
information from MRI with functional information from PET, or the planning of therapeutic
interventions such as surgery or radiotherapy using CT images. However, accurate
quantification of structural and physiological parameters from medical images is often quite
difficult, due to the 3-D nature of the problem and due to limitations intrinsic to the imaging
process itself, such as insufficient contrast, limited resolution, noise, inhomogeneities, or
artifacts. Ambiguity in the interpretation of the images introduces interand intraobserver
variability in the measurements that may be of the same order of magnitude as the parameter
to be quantified itself. This variability affects the reproducibility of the measurements and
undermines the significance of the clinical findings derived from them. There is, therefore,
great need for more automated methods for computer-aided image interpretation to improve
the reliability of the measurements. One of the key challenges in image registration is
determining the most appropriate similarity measure for a given application. Different
imaging modalities and clinical scenarios often require tailored approaches to maximize
accuracy and efficacy. Common similarity measures include mutual information, normalized
cross-correlation, and sum of squared differences. Mutual information has become
particularly popular due to its robustness in handling multimodal images, such as combining
MRI and PET scans. However, each similarity measure has its limitations, and the choice of
measure must align with the specific clinical objective, whether it is for pre-surgical
planning, treatment monitoring, or disease diagnosis. Advancements in non-rigid image
registration techniques have significantly enhanced the ability to account for tissue
deformations, especially in dynamic imaging situations. These transformations are
particularly important in longitudinal studies or when imaging tissues that undergo significant
movement, such as the heart or lungs. For instance, non-rigid registration allows for the
precise alignment of scans acquired at different times, even when the anatomy has shifted due
to patient movement or disease progression. Techniques such as elastography or
biomechanical modeling have shown promise in improving non-rigid registration accuracy by
incorporating tissue properties and physical constraints into the transformation model.
Despite the progress in algorithm development and software availability, image registration
in clinical practice is still not without its challenges. One of the major hurdles is the issue of
computational complexity, especially in the case of non-rigid transformations that involve
processing large datasets. Real-time registration, particularly for dynamic imaging or
interventional procedures, remains a significant challenge. There is ongoing research into
optimizing registration algorithms to speed up processing times without sacrificing accuracy.
The integration of machine learning and artificial intelligence into the registration process is
another area showing promise, where algorithms can learn from large datasets to enhance
both speed and precision. Another important aspect of image registration is the validation
process. Ensuring the accuracy of registration algorithms is crucial for their adoption in
clinical settings. Traditional validation methods rely on comparing registered images with
manual annotations or expert input, but this is not always feasible, especially for complex or
large-scale datasets. Researchers are exploring more advanced validation techniques, such as
phantom-based testing or using synthetic data, to assess algorithm performance in a more
standardized manner. Moreover, a universal validation framework for non-rigid
transformations is still under development, as it is much more difficult to validate these
algorithms compared to rigid-body transformations. As registration technology advances,
ongoing validation efforts will be necessary to ensure the reliability and safety of image
registration systems in clinical practice. In recent years, there has been significant progress in
integrating image registration techniques with other technologies, such as robotic surgery
systems and augmented reality (AR). These integrations offer promising potential for
enhancing real-time image-guided interventions. For instance, during a surgical procedure,
real-time registered images can be overlaid onto the patient's anatomy, providing the surgeon
with precise guidance for targeting tumors or avoiding critical structures. The fusion of
registered images with robotic systems can also enable more accurate and controlled
movements, ultimately improving surgical outcomes. Augmented reality can further enhance
this process by visualizing complex anatomical structures in three dimensions, offering
surgeons a more intuitive understanding of the region of interest. Moreover, with the growing
trend of personalized medicine, image registration plays an essential role in tailoring
treatment plans to individual patients. For example, in radiotherapy, image registration
ensures that radiation beams are precisely delivered to the tumor, minimizing damage to
surrounding healthy tissues. Personalized treatment plans rely heavily on accurate anatomical
and functional mapping of the patient's body, often requiring multi-modal imaging data. The
ability to precisely register and align different imaging modalities enhances the reliability of
treatment planning, improving the chances of successful outcomes. As imaging technology
continues to evolve, the integration of advanced registration methods with personalized
medicine will lead to more effective, patient-specific therapies.
Image registration techniques in nuclear medicine imaging (2006): Nuclear medicine has
traditionally relied on radionuclide images for diagnostic analysis, with qualitative correlation
between images based on anatomical expertise. However, spatial distortions arising from
differences in posture and acquisition techniques have limited the use of complementary
image data for quantitative analysis. Recent advancements in image registration software and
hardware have significantly improved the ability to fuse information from multiple imaging
modalities, enabling more accurate and comprehensive diagnostics. This chapter explores the
challenges and techniques involved in image registration, focusing on algorithmic approaches
and their clinical applications in nuclear medicine. It highlights the importance of both
software and hardware solutions, including dual-modality systems, in overcoming
registration issues and enhancing clinical practice.
Nuclear medicine has a long tradition of incorporating quantitative analysis in its diagnostic
procedures. Until recently, the analysis was based on radionuclide images as the sole input
although the importance of the complementary information available from other modalities or
from earlier scans has long been recognized. Indeed, qualitative correlation between images,
based on anatomical expertise, has always been part of the repertoire of the nuclear medicine
clinician. However, spatial distortion between images, caused largely by differences in
posture and acquisition technique, prevented the extension of these techniques to quantitative
analysis. Recent advances in image registration software and hardware have made it
increasingly possible to utilize that complementary image information in a clinical setting.
Reporting stations now often provide access to data from multiple modalities and permit
various forms of fused display for viewing spatially registered images. Recently introduced
dual-modality systems (described in Chapter 2 of this volume) provide a hardware approach
to spatial registration that is of particular value in those cases where numerical algorithms
fail. However, software approaches remain essential to the solution of many registration
problems in clinical practice. In this chapter, we define the registration problem, introduce
various algorithmic approaches to registering images, and discuss issues of clinical
application with emphasis on nuclear medicine. One of the key challenges in image
registration within nuclear medicine is addressing the inherent spatial distortions that arise
during image acquisition. Variations in patient positioning, as well as differences in the
acquisition techniques used by various imaging modalities, often result in misalignment
between the anatomical and functional images. For example, when combining a CT scan with
a PET scan, the patient’s movement between scans or slight variations in posture can cause
the images to become spatially misaligned, making it difficult to accurately correlate
functional data with anatomical structures. Advanced image registration algorithms seek to
correct these misalignments, enabling precise fusion of functional and structural images,
which is critical for accurate diagnosis and treatment planning. A variety of algorithmic
approaches have been developed to address the challenges of image registration in nuclear
medicine. Rigid-body transformations are useful when images share the same coordinate
system, and the only differences arise due to rotation and translation. However, non-rigid
transformations are required when the subject’s body undergoes deformations or when the
images are from different modalities that do not share the same anatomical framework. Non-
rigid algorithms involve more complex mathematical models that account for tissue
deformations, improving the accuracy of image fusion. For instance, techniques such as
spline-based interpolation or deformation models are used to correct for soft tissue movement
between scans, leading to a more reliable registration process. In clinical applications, the
integration of multi-modality imaging systems is playing an increasingly important role.
Dual-modality systems, such as PET/CT or SPECT/CT scanners, combine functional and
anatomical imaging within a single system, improving the accuracy and speed of diagnosis.
These systems minimize the errors associated with patient movement between separate
imaging sessions. However, even with dual-modality systems, post-processing software is
still necessary to further refine image alignment and ensure that the images are perfectly
matched. Software-driven solutions can help to correct minor misalignments and are often
used to enhance the data fusion process, providing clinicians with a more detailed and
accurate representation of the patient's anatomy and physiology. The clinical importance of
image registration in nuclear medicine extends beyond diagnostics and includes therapeutic
applications, such as treatment planning for radiotherapy. Accurate registration of functional
and anatomical images allows clinicians to precisely target tumors while sparing surrounding
healthy tissue. In radiotherapy, for example, PET and CT images are often fused to provide
both metabolic information from PET and detailed anatomical information from CT, allowing
for better-targeted radiation treatments. As image registration technology continues to evolve,
it is expected that more sophisticated algorithms and multimodal systems will improve the
overall accuracy of clinical decision-making, enhancing patient outcomes across a range of
medical disciplines. Recent advancements in image registration techniques have also brought
about improvements in the detection and monitoring of diseases, particularly in oncology and
neurology. In cancer diagnosis, precise image registration allows clinicians to identify the
exact location of tumors and track their growth over time. By fusing functional images such
as PET or SPECT with anatomical images like CT or MRI, doctors can more accurately
distinguish between healthy and malignant tissues, leading to earlier detection and more
effective treatment plans. Furthermore, image registration can be used to monitor how tumors
respond to therapy, helping to adjust treatment plans in real time based on the tumor’s
behavior, improving the likelihood of successful outcomes. In the field of neurology, image
registration has become a vital tool in studying brain disorders, including Alzheimer's
disease, epilepsy, and brain tumors. By registering functional imaging techniques like fMRI
with structural MRI or CT scans, researchers and clinicians can gain insights into how brain
activity correlates with structural changes in the brain. For example, in Alzheimer's disease,
the ability to register and compare brain scans over time can help identify early signs of
neurodegeneration, allowing for earlier intervention and better management of the disease.
The continued evolution of image registration technologies, paired with artificial intelligence
and machine learning, is expected to further enhance the accuracy and efficiency of brain
imaging, contributing to more personalized and precise treatments for neurological
conditions.
Non-rigid image registration: theory and practice (2004): Image registration is an
important enabling technology in medical image analysis. The current emphasis is on
development and validation of application-specific non-rigid techniques, but there is already
a plethora of techniques and terminology in use. In this paper we discuss the current state of
the art of non-rigid registration to put on-going research in context and to highlight current
and future clinical applications that might benefit from this technology. The philosophy and
motivation underlying non-rigid registration is discussed and a guide to common terminology
is presented. The core components of registration systems are described and outstanding
issues of validity and validation are confronted.
Image registration is a key enabling technology in medical image analysis that has benefited
from 20 years of development [1]. It is a process for determining the correspondence of
features between images collected at different times or using different imaging modalities.
The correspondences can be used to change the appearance – by rotating, translating,
stretching etc. – of one image so it more closely resembles another so the pair can be directly
compared, combined or analysed (Figure 1). The most intuitive use of registration is to
correct for different patient positions between scans. Image registration is not an end in itself
but adds value to images, e.g. by allowing structural (CT, MR, ultrasound) and functional
(PET, SPECT, functional MRI (fMRI)) images to be viewed and analysed in the same
coordinate system, and facilitates new uses of images, e.g. to monitor and quantify disease
progression over time in the individual [2] or to build statistical models of structural variation
in a population [3]. In some application areas image registration is now a core tool; for
example (i) reliable analysis of fMRIs of the brain requires image registration to correct for
small amounts of subject motion during imaging [4]; (ii) the widely used technique of voxel
based morphometry makes use of image registration to bring brain images from tens or
hundreds of subjects into a common coordinate system for analysis (so-called ‘‘spatial
normalization’’) [5]; (iii) the analysis of perfusion images of the heart would not be possible
without image registration to compensate for patient respiration [6]; and (iv) some of the
latest MR image acquisition techniques incorporate image registration to correct for motion
[7]. Non-rigid image registration is particularly crucial in the medical field due to the
complex nature of biological structures, where anatomical deformations such as organ
movement, tissue stretching, and variations in shape need to be accounted for. Unlike rigid
registration, which assumes that images can be aligned using simple translations and
rotations, non-rigid registration allows for more sophisticated transformations that can adapt
to the local deformations that occur due to patient movement, disease progression, or
anatomical variations. This makes non-rigid registration ideal for tasks like mapping
functional images onto structural images, where precise alignment of tissues is critical for
accurate diagnosis and treatment planning. Several non-rigid image registration techniques
have emerged over the years, ranging from intensity-based methods, which rely on pixel or
voxel intensity values, to feature-based methods, which focus on detecting specific points,
edges, or surfaces in the images. Intensity-based methods are particularly useful for
modalities like CT or MRI, where the images are rich in structural information, while feature-
based methods are often employed in situations where specific anatomical landmarks need to
be aligned. Both approaches have their strengths and weaknesses, with intensity-based
methods generally offering better performance in cases where detailed anatomical
information is available, while feature-based methods may be more robust in cases where
images are noisy or contain significant artifacts. In clinical practice, non-rigid registration has
found applications in a variety of fields, including neurology, oncology, and cardiovascular
medicine. In neurology, for instance, non-rigid registration is used to map fMRI scans to
structural MRI images of the brain, allowing researchers to examine how brain activity
corresponds to anatomical regions. In oncology, non-rigid registration is used to align pre-
and post-treatment images of tumors, enabling oncologists to assess tumor response to
therapy and adjust treatment plans accordingly. Similarly, in cardiovascular medicine, non-
rigid registration is applied to cardiac imaging to account for heart motion during the cardiac
cycle, improving the accuracy of measurements such as myocardial perfusion and ventricular
volume. Despite the advances in non-rigid image registration, several challenges remain. One
of the primary issues is the computational complexity of these algorithms, particularly when
dealing with large 3D datasets or real-time applications. Achieving high accuracy in non-
rigid registration often requires sophisticated optimization techniques, which can be
computationally expensive and time-consuming. Additionally, there is still the challenge of
incorporating prior knowledge, such as anatomical constraints or tissue properties, into the
registration process. While some methods attempt to address these issues by incorporating
physical models of deformation, this remains an active area of research. Validation of non-
rigid image registration techniques is another significant challenge. Unlike rigid registration,
where errors can be easily quantified by comparing the transformation parameters, non-rigid
registration involves more complex transformations that are harder to validate. A variety of
strategies have been proposed to address this issue, including the use of synthetic datasets
with known ground truth, cross-validation with multiple image modalities, and clinical
validation studies. Despite these efforts, there is still a need for standardized benchmarks and
validation protocols to ensure the reliability and robustness of non-rigid registration
algorithms across different clinical settings. As the field continues to evolve, improving the
validation and performance of non-rigid registration will be crucial for its broader adoption in
clinical practice. One promising area of development in non-rigid image registration is the
integration of machine learning and deep learning techniques. Recent advancements in these
fields have led to the creation of more sophisticated algorithms that can learn complex
deformations directly from data, reducing the reliance on hand-crafted models. Convolutional
neural networks (CNNs), for instance, have been applied to learn image features that are
particularly useful for registration, allowing for more accurate and robust transformations.
These deep learning-based methods can automatically adapt to varying image characteristics,
which could greatly improve registration accuracy in challenging clinical scenarios, such as
those involving motion artifacts or low-quality images. Additionally, the increasing
availability of multimodal imaging data presents both opportunities and challenges for non-
rigid registration. In many clinical applications, data is obtained from different imaging
modalities, such as CT, MRI, PET, and ultrasound, each offering unique information about
the patient's anatomy and physiology. Combining these different data types requires careful
alignment to ensure that the images correspond to the same anatomical structures. Non-rigid
registration plays a key role in this process by compensating for the spatial distortions that
arise due to differences in resolution, modality-specific artifacts, or patient movement
between scans. As multimodal imaging becomes more common in clinical practice, the
demand for robust non-rigid registration techniques that can handle these complex datasets
will continue to grow.

Conclusion
Image registration by maximization of mutual information considers all voxels in the images
to be registered to estimate the statistical dependence between corresponding voxel
intensities, which is assumed to be maximal when the images are correctly aligned. The MMI
criterion is histogram based rather than intensity based and does not impose limiting
assumptions on the specific nature of the relationship between corresponding voxel
intensities, making it applicable to a large variety of multimodality image combinations of
various body regions. Since its introduction in 1995 by Collignon et al. [17] and by Viola and
Wells [122] in the medical imaging research community, MMI has attracted large interest
within this field and has sparked a fury of publications on multimodal medical image
registration and its clinical applications. Its mathematical elegance and algorithmic simplicity
made it easy for other groups to quickly adopt and implement the new approach. The success
of MMI for multimodal image registration can be explained by the fact that it got rid of the
need for image segmentation or preprocessing as required with previous registration
algorithms and that it allows for completely automated registration without need for user
interaction, making the method very well suited for application in clinical practice. In the
mean time, MMI has become a standard in the field and is the method of choice for
multimodal image registration in a wide range of applications. In recent years, the continued
success and adoption of mutual information-based image registration (MMI) have spurred
further improvements in computational techniques. Researchers have focused on optimizing
the algorithm to handle more complex image datasets, such as those involving higher-
dimensional or time-varying data. For example, advancements in hierarchical and multi-
resolution methods have enabled the MMI approach to handle images with varying
resolutions more effectively, ensuring better alignment and reducing computation time.
Furthermore, the integration of machine learning models into the MMI framework is
enhancing its robustness, particularly when dealing with noisy, low-quality images or images
with significant motion artifacts. Despite its widespread success, there remain some
challenges in the application of MMI, especially when dealing with non-rigid deformations.
For instance, while MMI is highly effective for rigid and affine registration, its performance
can degrade when applied to images that require non-rigid transformations, such as those
involving anatomical changes over time or due to diseases. To address this, hybrid methods
combining MMI with other non-rigid transformation algorithms, such as B-splines or demons
algorithms, have been proposed to improve registration accuracy in such cases. These
combined approaches aim to maintain the benefits of MMI while addressing its limitations in
capturing complex deformations. The future of image registration, particularly multimodal
image registration using MMI, will likely benefit from further advances in computational
power and algorithmic sophistication. The development of real-time registration methods,
where images are registered during acquisition, could revolutionize clinical workflows by
providing immediate feedback to clinicians. Additionally, the continued integration of
artificial intelligence and deep learning techniques holds great potential for refining image
registration algorithms, making them more adaptable to diverse clinical scenarios and capable
of handling more intricate image data. As a result, MMI and its variants are expected to play
an increasingly central role in both diagnostic and therapeutic applications in medicine.
References
[1] J. Ashburner and K. Friston, “Nonlinear spatial normalization using basis functions,”
Human Brain Mapping, vol. 7, no. 4, pp. 254–266, 1999.
[2] J. Ashburner and K. J. Friston, “Multimodal image coregistration and partitioning—A
unified framework,” NeuroImage, vol. 6, no. 3, pp. 209–217, 1997.
[3] R. K. Bajcsy and S. Kovacic, “Multiresolution elastic matching,” Comput. Vis., Graph.
Image Process., vol. 46, pp. 1–21, 1989.
[4] B. H. Brinkmann, T. J. O’Brien, S. Aharon, M. K. O’Connor, B. P. Mullan, D. P. Hanson,
and R. A. Robb, “Quantitative and clinical analysis of SPECT image registration for epilepsy
studies,” J. Nucl. Med., vol. 40, no. 5, pp. 1098–1105, 1999.
[5] M. Bro-Nielsen, “Rigid registration of CT, MR and cryosection images using a GLCM
framework,” in Proc. 1st Joint Conf. Computer Vision, Virtual Reality and Robotics in
Medicine and Medical Robotics and Computer-Assisted Surgery (CVRMED-MRCAS’97),
vol. 1205, Lecture Notes in Computer Science, J. Troccaz, E. Grimson, and R. Mösges, Eds.,
Grenoble, France, Mar. 1997, pp. 171–180.
[6] M. Bro-Nielsen and C. Gramkow, “Fast fluid registration of medical images,” in Proc.
Visualization in Biomedical Computing (VBC’96), vol. 1131, Lecture Notes in Computer
Science, R. Kikinis and K. H. Hoehne, Eds., Hamburg, Germany, Sept. 1996, pp. 267–276.
[7] L. G. Brown, “A survey of image registration techniques,” ACM Comput. Surv., vol. 24,
no. 4, pp. 325–376, Dec. 1992.
[8] T. Butz and J.-P. Thiran, “Affine registration with feature space mutual information,” in
Proc. Medical Image Computing and Computer-Assisted Intervention (MICCAI’00), vol.
1935, Lecture Notes in Computer Science, S. L. Delp, A. M. DiGioia, and B. Jaramaz, Eds.,
Berlin, Germany, Oct. 2000, pp. 549–556.
[9] T. M. Buzug, J. Weese, C. Fassnacht, and C. Lorenz, “Image registration: Convex
weighting functions for histogram-based similarity measures,” in Proc. 1st Joint Conference
Computer Vision, Virtual Reality and Robotics in Medicine and Medical Robotics and
Computer-Assisted Surgery (CVRMED-MRCAS’97), vol. 1205, Lecture Notes in Computer
Science, J. Troccaz, E. Grimson, and R. Mösges, Eds., Grenoble, France, Mar. 1997, pp.
203–212.
[10] M.Cˆ apek and I. Krekule, “Alignment of adjacent picture frames captured by a CLSM,”
IEEE Trans. Inform. Technol. Biomed., vol. 3, pp. 119–124, June 1999.
[11] A. D. Castellano-Smith, T. Hartkens, J. A. Schnabel, D. R. Hose, H. Liu, W. A. Hall, C.
L. Truwit, D. J. Hawkes, and D. L. G. Hill, “Constructing patient specific models for
correcting intraoperative brain deformation,” in Proc. Medical Image Computing and
ComputerAssisted Intervention (MICCAI ’01), vol. 2208, Lecture Notes in Computer
Science, W. J. Niessen and M. A. Viergever, Eds., Utrecht, The Netherlands, Oct. 2001, pp.
1091–1098.
[12] H. Chang and J. M. Fitzpatrick, “A technique for accurate magnetic resonance imaging
in the presence of field inhomogeneities,” IEEE Trans. Med. Imag., vol. 11, pp. 319–329,
Sept. 1992.
[13] J. Y. Chiang and B. J. Sullivan, “Coincident bit counting – A new criterion for image
registration,” IEEE Trans. Med. Imag., vol. 12, pp. 30–38, Mar. 1993.
[14] G. E. Christensen, R. D. Rabbitt, and M. I. Miller, “Deformable templates using large
deformation kinetics,” IEEE Trans. Image Processing, vol. 5, pp. 1435–1447, Oct. 1996.
[15] G. E. Christensen, R. D. Rabbitt, M. I. Miller, S. C. Joshi, U. Grenander, T. A. Coogan,
and D. C. Van Essen, “Topological properties of smooth anatomic maps,” in Proc.
Information Processing in Medical Imaging (IPMI’95), vol. 3, Computational Imaging and
Vision, Y. Bizais, C. Barillot, and R. Di Paola, Eds., Ile de Berder, France, June 1995, pp.
101–112.
[16] M. J. Clarkson, D. Rueckert, A. P. King, P. J. Edwards, D. L. G. Hill, and D. J. Hawkes,
“Registration of video images to tomographic images by optimising mutual information using
texture mapping,” in Proc. Medical Image Computing and Computer-Assisted Intervention
(MICCAI’99), vol. 1679, Lecture Notes in Computer Science, C. Taylor and A. Colchester,
Eds., Cambridge, U.K., Sept. 1999, pp. 579–588.
[17] A. Collignon, F. Maes, D. Delaere, D. Vandermeulen, P. Suetens, and G. Marchal,
“Automated multimodality medical image registration using information theory,” in Proc.
14th Int. Conf. Information Processing in Medical Imaging (IPMI’95), vol. 3, Computational
Imaging and Vision, Y. Bizais, C. Barillot, and R. Di Paola, Eds., Ile de Berder, France, June
1995, pp. 263–274.