333
Ultrasound Imaging
need to be eliminated to increase the contrast and resolution obtained from true
discrete discontinuities. Because of this noise, it is almost impossible to detect any
activation or image feature without resorting to statistical analysis. This analysis
requires a model of noise usually assuming a Gaussian distribution.
The noise often comes from discontinuities at angles with the sonic beam, from
gradient differences, and from thermal and other motion artifacts. All these suggest that in the analysis of ultrasound images in the frequency domain, the highfrequency contents could be often attributed to noise, and, therefore, the filters can
be designed to filter these high-frequency noises out.
Wavelet-based filters are essentially based on an approach that aims at obtaining
an optimal trade-off between good signal averaging over homogeneous regions and
minimal resolution degradation of image details. Similar to the use of wavelets in
other imaging techniques, the wavelet transform has a compression effect since, as
mentioned before, it has the tendency to bundle the signal of interest into a significantly fewer number of identifiers with respectively large coefficients. On the other
hand, noise can be reduced evenly in the wavelet domain. This in turn results in an
enhanced signal-to-noise ratio for those coefficients where the signal is concentrated
and hence improves the detection rate. This allows the use of a conservative decision
strategy to keep the false-detection rate to a minimum.
A crucial point in the selection of wavelet transforms is the appropriate choice of
the wavelet base form, the order of the transform and the iteration depth. Different
basis functions offer different compromises. The iteration depth controls the accuracy
and the sensitivity. The order, depending on the basis functions, has a great influence
on the number of detected coefficients and consequently on the sensitivity and specificity. According to the literature, the Daubechies wavelet analysis seems to be the
best fit for most ultrasound image processing applications since the reflection is a more
or less binary event, which matches the Daubechie mother wavelet. However, other
wavelet analysis may fit the gradient response of inhomogeneous reflection conditions.
The wavelet coefficients, area, volume or size of the object, intensity, eccentricity,
elongation, and Fourier coefficients are among the most popular features used in the
analysis of ultrasound images.
16.12 IMAGE REGISTRATION
Image registration in ultrasound is used to combine images from an array of transducers to present a single image slice. In most cases, the direction of the beam is used
as a reference with respect to the adjacent ray. In other cases, specific landmarks
need to be identified to correlate the separately acquired images to form a continuous image. In general, the feedback mechanism of the ultrasound device gives the
location and direction of each ray of ultrasound, and the Cartesian representation of
the rays serves as the map to reconstruct the image with.
If an ultrasound pulse strikes a tissue interface at an oblique angle, the direction of
travel will be changed if the speed of sound is different on either side of the interface.
For most soft-tissue interfaces, the effect is small. However, if the propagation path
includes fluid (such as in pelvic scans by the transabdominal route), the effect can
be significant. The effect of refraction is to diverge the path of the ultrasound beam.
Ultrasound Imaging
need to be eliminated to increase the contrast and resolution obtained from true
discrete discontinuities. Because of this noise, it is almost impossible to detect any
activation or image feature without resorting to statistical analysis. This analysis
requires a model of noise usually assuming a Gaussian distribution.
The noise often comes from discontinuities at angles with the sonic beam, from
gradient differences, and from thermal and other motion artifacts. All these suggest that in the analysis of ultrasound images in the frequency domain, the highfrequency contents could be often attributed to noise, and, therefore, the filters can
be designed to filter these high-frequency noises out.
Wavelet-based filters are essentially based on an approach that aims at obtaining
an optimal trade-off between good signal averaging over homogeneous regions and
minimal resolution degradation of image details. Similar to the use of wavelets in
other imaging techniques, the wavelet transform has a compression effect since, as
mentioned before, it has the tendency to bundle the signal of interest into a significantly fewer number of identifiers with respectively large coefficients. On the other
hand, noise can be reduced evenly in the wavelet domain. This in turn results in an
enhanced signal-to-noise ratio for those coefficients where the signal is concentrated
and hence improves the detection rate. This allows the use of a conservative decision
strategy to keep the false-detection rate to a minimum.
A crucial point in the selection of wavelet transforms is the appropriate choice of
the wavelet base form, the order of the transform and the iteration depth. Different
basis functions offer different compromises. The iteration depth controls the accuracy
and the sensitivity. The order, depending on the basis functions, has a great influence
on the number of detected coefficients and consequently on the sensitivity and specificity. According to the literature, the Daubechies wavelet analysis seems to be the
best fit for most ultrasound image processing applications since the reflection is a more
or less binary event, which matches the Daubechie mother wavelet. However, other
wavelet analysis may fit the gradient response of inhomogeneous reflection conditions.
The wavelet coefficients, area, volume or size of the object, intensity, eccentricity,
elongation, and Fourier coefficients are among the most popular features used in the
analysis of ultrasound images.
16.12 IMAGE REGISTRATION
Image registration in ultrasound is used to combine images from an array of transducers to present a single image slice. In most cases, the direction of the beam is used
as a reference with respect to the adjacent ray. In other cases, specific landmarks
need to be identified to correlate the separately acquired images to form a continuous image. In general, the feedback mechanism of the ultrasound device gives the
location and direction of each ray of ultrasound, and the Cartesian representation of
the rays serves as the map to reconstruct the image with.
If an ultrasound pulse strikes a tissue interface at an oblique angle, the direction of
travel will be changed if the speed of sound is different on either side of the interface.
For most soft-tissue interfaces, the effect is small. However, if the propagation path
includes fluid (such as in pelvic scans by the transabdominal route), the effect can
be significant. The effect of refraction is to diverge the path of the ultrasound beam.