306
Biomedical Signal and Image Processing
15.9 REGISTRATION WITH MR IMAGES
The registration with MR images can be done for two purposes. Frequently, patients
are imaged with various image modalities for one single diagnostic application. The
first reason for registration roots in the fact while MRI provides some functional
information, there are as significant amount of functional information captured
by other modalities such as PET that need to be superimposed on the anatomical
images. As a typical example, it is known that the anatomical details as well as the
resolution of PET images are very limited, and, often, it is necessary to superimpose
the functional information of PET on MRI or CT.
The second reason to perform registration is the need to register different MR
images captured with different MR machines or even with the same machine at different times. More specifically, sometimes multiple imaging sessions on the same
or different MRI machines are scheduled to track progress and document changes
following a treatment. In f MRI, it is also customary to acquire a series of images in
one session under different conditions (e.g., investigating evoked responses or metabolic parameters). Knowing that each time the exact position of the patient as well as
the exact calibration of the machine might be different, one needs to register these
multiple images with each other. In order to do so, often the polynomial registration
methods, as explained in Part I of the book, are used. These methods often require
tie points to calculate the mapping between the two images.
One popular method in providing the tie points is through the use of external
markers. However, external markers may sometimes interfere with the coils in
the MRI device. An additional factor of concern is the external markers’ position
with respect to the biological medium that is being imaged. Specifically, the MRI
machine has a limited field of view, and the markers may fall outside this view. Other
issues are distortion of the markers due to the location with respect to the sensors
and shadow imaging of the markers, which may hide biological points of reference.
In the cases where the use of external markers may not be a feasible option, computational registration methods are applied that apply the counter of solid objects
such as the skull for registration. In such a procedure, the corresponding locations
on the skull bone in both imaging modalities are identified and registered. Then, the
rest of the pixels in the two images are registered using the same mapping that maps
the contour of the skull in the two images.
When registering MRI with PET, the image sizes play an important role. The typical
MRI machines usually generate 256 × 256 image matrices, and the more advanced MRI
systems even produce 512 × 512 matrices. PET, on the other hand, typically has image
of 128 × 128 pixels. Each pixel in PET slices often represents a region with the areas of
2–8 mm 2 . The PET slice thickness depends mostly on the hardware used. The gray-level
range of PET is also significantly less than that of MRI. These differences call for registration methods to superimpose PET on MRI that address the differences of the two
technologies both in the special and gray-level resolutions. Such registration methods
are often specialized to accurately map certain objects in the brain in the two modalities. One biological object of interest in registration is the gray matter of the brain.
Another image registration technique used for comparing PET and MRI relies
on physiological matches, since both MRI and PET have the capability to extract
Biomedical Signal and Image Processing
15.9 REGISTRATION WITH MR IMAGES
The registration with MR images can be done for two purposes. Frequently, patients
are imaged with various image modalities for one single diagnostic application. The
first reason for registration roots in the fact while MRI provides some functional
information, there are as significant amount of functional information captured
by other modalities such as PET that need to be superimposed on the anatomical
images. As a typical example, it is known that the anatomical details as well as the
resolution of PET images are very limited, and, often, it is necessary to superimpose
the functional information of PET on MRI or CT.
The second reason to perform registration is the need to register different MR
images captured with different MR machines or even with the same machine at different times. More specifically, sometimes multiple imaging sessions on the same
or different MRI machines are scheduled to track progress and document changes
following a treatment. In f MRI, it is also customary to acquire a series of images in
one session under different conditions (e.g., investigating evoked responses or metabolic parameters). Knowing that each time the exact position of the patient as well as
the exact calibration of the machine might be different, one needs to register these
multiple images with each other. In order to do so, often the polynomial registration
methods, as explained in Part I of the book, are used. These methods often require
tie points to calculate the mapping between the two images.
One popular method in providing the tie points is through the use of external
markers. However, external markers may sometimes interfere with the coils in
the MRI device. An additional factor of concern is the external markers’ position
with respect to the biological medium that is being imaged. Specifically, the MRI
machine has a limited field of view, and the markers may fall outside this view. Other
issues are distortion of the markers due to the location with respect to the sensors
and shadow imaging of the markers, which may hide biological points of reference.
In the cases where the use of external markers may not be a feasible option, computational registration methods are applied that apply the counter of solid objects
such as the skull for registration. In such a procedure, the corresponding locations
on the skull bone in both imaging modalities are identified and registered. Then, the
rest of the pixels in the two images are registered using the same mapping that maps
the contour of the skull in the two images.
When registering MRI with PET, the image sizes play an important role. The typical
MRI machines usually generate 256 × 256 image matrices, and the more advanced MRI
systems even produce 512 × 512 matrices. PET, on the other hand, typically has image
of 128 × 128 pixels. Each pixel in PET slices often represents a region with the areas of
2–8 mm 2 . The PET slice thickness depends mostly on the hardware used. The gray-level
range of PET is also significantly less than that of MRI. These differences call for registration methods to superimpose PET on MRI that address the differences of the two
technologies both in the special and gray-level resolutions. Such registration methods
are often specialized to accurately map certain objects in the brain in the two modalities. One biological object of interest in registration is the gray matter of the brain.
Another image registration technique used for comparing PET and MRI relies
on physiological matches, since both MRI and PET have the capability to extract