304
Biomedical Signal and Image Processing
Even though the estimated slice is not truly measured during the data acquisition
process, if the estimation is a reliable one, this process doubles the resolution in the
z-direction. A common estimation method use for MR image interpolation is the
simple averaging of the pixels in the two slices around the estimated slice.
Another commonly used group of MR image processing methods are the algorithms used to register MR images with the images captured by other modalities
such as positron emission tomography (PET). The registration process between MRI
and other image modalities is described in more detail later in this chapter.
Processing of f MR images are often not based on the image regions. Rather,
when the voxels responding to a particular stimulus are identified, the time signals of
these voxels collected throughout the experiment are treated as 1-D signals and are
analyzed using methods such as DFT and discrete wavelet transform (DWT). The
feature extraction methods applied for the analysis of these time signals are exactly
the same as those applied for processing of the evoked potential EEG recordings.
15.7.1 SOURCES OF NOISE AND FILTERING METHODS IN MRI
Even though the overall resolution and signal-to-noise ratio of the MR images are
very high, there are several distinct sources of noise that can be identified in MRI
technology. These sources are thermal noise, subject motion, physiological activity,
low frequency drift, spontaneous neural and vascular fluctuations, shear and strain
noise, and artifacts arising from rapid imaging methods.
Thermal noise results from the fact that temperature is a form of kinetic energy
of atoms and molecules. The nuclear spin orientation of certain individual molecules
and atoms may, as a result, be affected by the heat produced by the MRI machine
and the fact that the patient is enveloped by the MRI machine, retaining some heat
produced by the patient’s body.
The scanning process relies on the assumption that detected spins stem from one
particular location that does not change in time. Any motion by the patient will
disturb the location registration algorithm and will produce a motion artifact. The
main noise and artifacts that need compensation are often motion artifacts. When
MRI is taken, patients are supposed to stay still, but, in reality, it is often the case
that during the long duration of the data acquisition process, patients do move. This
is more significant in children or claustrophobic patients who are not comfortable
inside the MRI machine. This makes motion artifacts the main sources of noise in
MRI. Motion artifact is often removed using deblurring filters described in Part I of
the book as well as more specialized filters designed for this purpose.
The main share of the physiological noise originates from breathing, additionally
heartbeat, and peristaltic motions. A rather more fundamental source of noise deals
with the way the actual readings are collected. As a general rule, when scanning
thinner slices of the tissue, higher levels of noise are registered by the detectors. As
a result, if thin slices of images are needed, often the dose of the enhancer agents
needs to be adjusted. In many applications, image improvements must be obtained
by using an experimentally optimized dose that balances contrast and saturation.
Nerve tone changes and nervous twitches can result from adaptation and irritation and may not be avoided. This type of noise will need to be recognized as
Biomedical Signal and Image Processing
Even though the estimated slice is not truly measured during the data acquisition
process, if the estimation is a reliable one, this process doubles the resolution in the
z-direction. A common estimation method use for MR image interpolation is the
simple averaging of the pixels in the two slices around the estimated slice.
Another commonly used group of MR image processing methods are the algorithms used to register MR images with the images captured by other modalities
such as positron emission tomography (PET). The registration process between MRI
and other image modalities is described in more detail later in this chapter.
Processing of f MR images are often not based on the image regions. Rather,
when the voxels responding to a particular stimulus are identified, the time signals of
these voxels collected throughout the experiment are treated as 1-D signals and are
analyzed using methods such as DFT and discrete wavelet transform (DWT). The
feature extraction methods applied for the analysis of these time signals are exactly
the same as those applied for processing of the evoked potential EEG recordings.
15.7.1 SOURCES OF NOISE AND FILTERING METHODS IN MRI
Even though the overall resolution and signal-to-noise ratio of the MR images are
very high, there are several distinct sources of noise that can be identified in MRI
technology. These sources are thermal noise, subject motion, physiological activity,
low frequency drift, spontaneous neural and vascular fluctuations, shear and strain
noise, and artifacts arising from rapid imaging methods.
Thermal noise results from the fact that temperature is a form of kinetic energy
of atoms and molecules. The nuclear spin orientation of certain individual molecules
and atoms may, as a result, be affected by the heat produced by the MRI machine
and the fact that the patient is enveloped by the MRI machine, retaining some heat
produced by the patient’s body.
The scanning process relies on the assumption that detected spins stem from one
particular location that does not change in time. Any motion by the patient will
disturb the location registration algorithm and will produce a motion artifact. The
main noise and artifacts that need compensation are often motion artifacts. When
MRI is taken, patients are supposed to stay still, but, in reality, it is often the case
that during the long duration of the data acquisition process, patients do move. This
is more significant in children or claustrophobic patients who are not comfortable
inside the MRI machine. This makes motion artifacts the main sources of noise in
MRI. Motion artifact is often removed using deblurring filters described in Part I of
the book as well as more specialized filters designed for this purpose.
The main share of the physiological noise originates from breathing, additionally
heartbeat, and peristaltic motions. A rather more fundamental source of noise deals
with the way the actual readings are collected. As a general rule, when scanning
thinner slices of the tissue, higher levels of noise are registered by the detectors. As
a result, if thin slices of images are needed, often the dose of the enhancer agents
needs to be adjusted. In many applications, image improvements must be obtained
by using an experimentally optimized dose that balances contrast and saturation.
Nerve tone changes and nervous twitches can result from adaptation and irritation and may not be avoided. This type of noise will need to be recognized as