Image analysis – Applications – Biomedical applications
Reexamination Certificate
2000-03-21
2003-05-06
Johns, Andrew W. (Department: 2621)
Image analysis
Applications
Biomedical applications
C382S131000
Reexamination Certificate
active
06560353
ABSTRACT:
BACKGROUND OF THE INVENTION
(1) Field of the Invention
This invention pertains to the field of magnetic resonance imaging and multi-dimensional image reconstruction from partially acquired data. More particularly, this invention pertains to a method of reconstructing a complex MRI image from a portion of asymmetrically acquired data using an iterative multidimensional inverse transform technique. The use of the method reduces data acquisition time by reducing the amount of data needed to reconstruct the image matrix.
(2) Description of the Related Art
Acquiring data in many imaging modalities often takes a significant length of time. Typically, as less
Acquiring data in many imaging modalities often takes a significant length of time. Typically, as less of the data are acquired, the resolution of the image is decreased while the temporal resolution is increased. On the other hand, the collection of more data allows for higher resolution but requires longer acquisition times.
The size of the data matrix used to generate an image is known to affect the resolution. In one dimension, if the data matrix coverage comprises data from −k
max
to k
max
, then the resolution is given by &Dgr;x=1/(2k
max
). Many methods of reducing image acquisition time are based on the theory that it is not necessary to collect all the data for the data matrix since, theoretically, the data is purely real and will therefore express complex conjugation symmetry. By using complex conjugation symmetry, the negative terms in a data matrix s(k) can be generated from the positive terms since s(−k)=s*(k). Therefore, if only half of the data of a given data matrix need to be acquired to produce an image that has the same properties in terms of its reconstructed appearance as if all the data were acquired, a factor of two in time can be saved. On the other hand, if resolution rather than time is the issue, then the same number of data points as previously acquired can be used to create a data matrix twice as large as before, thereby increasing the resolution by a factor of two. In practice, however, the data acquired is never purely real due to artifacts and noise which are invariably present. As such, the images produced by imposing complex conjugate symmetry are only approximations of the images that would have been produced by collecting all of the data for the given data matrix.
In the case of magnetic resonance imaging, reconstruction of a given data matrix, or k-space as it is known in the field, from partially collected data has the potential of improving acquisition speed or resolution as described generally above. Prior art methods of data extrapolation, known as partial or fractional k-space coverage techniques, are well known in the field of magnetic resonance imaging and are set forth in detail in Liang et al.,
Constrained Reconstruction Methods in MR Imaging
, Reviews of Magnetic Resonance in Medicine, Vol. 4, 67-185 (Pergamon Press Ltd. 1992). These methods have included one dimensional partial Fourier methods and constrained reconstruction techniques as described below. While the former can reduce imaging times by at most a factor of two, the savings in time from the latter methods is less predictable and is often less than a factor of two.
Initial attempts at using partial Fourier imaging utilized methods that simply acquired half of the k-space of a given data matrix and then used complex conjugation to recover the missing half. Typically, for a two-dimensioned image acquisition scheme, such methods involve acquiring data s(k
x
, k
y
) for the entire top half of k-space as shown in FIG.
1
. In
FIG. 1
, the k-space shown is two-dimensional data where the two dimensions can represent the read and phase encoding directions or the phase and partition encoding directions. Assuming complex conjugate symmetry (i.e., s(−k
x
,−k
y
)=s*(k
x
, k
y
)), data for the lower half of k-space can then be approximated. This is performed using a one-dimensional approach by first transforming the acquired data along the horizontal direction to create a set of one-dimensional image/data set s(x, k
y
) for each x-position in the image matrix as shown in FIG.
2
. With the data so transformed, the one-dimensional data in k
y
is then complex conjugated for each x-position to produce the missing half of the image matrix. However, as discussed above, such approaches are based on the assumption that the data are purely real and it is well known that complex conjugation cannot produce the missing half of k-space if the object is complex or when other sources of error are present in the data.
For the above reasons, a method was sought to remove any background variations in the raw data caused by the presence of imaginary information in the image itself. Such an attempt was first made by Margosian as disclosed in
Faster MR Imaging—Imaging with half the Data
, SMRM Conference Abstracts, Vol. 2, 1024-25 (1985). This method is well known in the art and is performed by extending the region of acquired data (usually by 8 points for spin echo data) as shown in
FIG. 3
to obtain data for both positive and negative k-space points (i.e., before or after the origin in a given direction for an equal number of points). A low pass estimate of the phase, &phgr;(x, y), of the image is determined by using only the central portion of the acquired k-space data. Additionally, the acquired data is transformed as described in the above half k-space method after zero filling the remainder of the data matrix and applying an asymmetrical Hamming-like filter to the central portion of the data. This transformation produces a first estimate of the image that is complex, not real. An assumption is then made that the phase (&phgr;(x, y)) errors are small enough that the reconstructed image can be phase corrected by multiplying the first complex estimate of the image by e
−i&phgr;(x, y)
. The final image is presumed to be the real part of this second estimate. There are several disadvantages of using this approach. First, if the phase has any high spatial frequency components in it, the method will fail because too few points are used to estimate the phase. Increasing the number of points used would require collecting additional data, thereby reducing the time benefits of the method. Second, the filter used suppresses low spatial frequency information in the image itself, leading to an unnecessary loss in signal-to-noise over and above that inherent in the partial Fourier method itself. Finally, using solely the real part of the image matrix is not representative of the pristine image. Attempted methods of improving the phase estimate have not obviated these difficulties (SR McFall et al.,
Corrections of Spatially Dependant Phase Shifts for Partial Fourier Imaging
, Magnetic Resonance Imaging, Vol. 6, 143-55 (1988)).
A similar approach using an iterative scheme which attempts to force the image to be real was developed by Cuppen et al. as disclosed in U.S. Pat. No. 4,853,635, the disclosure of which is incorporated herein by reference. Extensions to this approach have also been developed by Haacke et al. as disclosed in
A Fast, Iterative, Partial
-
Fourier Technique Capable of Local Phase Recovery
, Journal of Magnetic Resonance, Vol. 92, 126-45 (1991) which is incorporated herein by reference.
In general, the Cuppen approach has the advantage over the Margosian method in its ability to handle phase errors and to do so without an increased reduction of the signal-to-noise ratio above the sqrt(2) expected reduction. In Cuppen's method, a similar one-dimensional approach is used to generate a complex image &rgr;(x). A new complex image, &rgr;
1
(x), is then calculated as &rgr;
1
(x)=&rgr;*(x)e
i2&phgr;x
. Estimated data for the k-space is then calculated by transforming &rgr;
1
(x), thereby generating s
1
(k
x
). The estimated data, s
1
(k
x
), is then used in place of the uncollected data and is merged with the acquired data s(k
x
) to generate s
1new
(k
x
) which can then be transfor
Haacke E. Mark
Xu Yingbiao
Johns Andrew W.
Magnetic Resonance Innovations, Inc.
Nakhjavan Shervin
Thompson & Coburn LLP
LandOfFree
Method of MRI image reconstruction from partially acquired... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of MRI image reconstruction from partially acquired..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of MRI image reconstruction from partially acquired... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3043076