Abstract
A method for improving image homogeneity of image data acquired from balanced Steady-State Free Precision (bSSFP) sequences in magnetic resonance imaging. Multiple bSSFP sequences are performed with different radio frequency phase increments to create multiple bSSFP image volumes with different phase offsets . Each image has voxels whose intensity M is a function of a nuclear resonance signal (or magnetization) measured by the MR imaging apparatus. Per-voxel fitting of a mathematical signal model onto the measured magnetization of the field of view in function of the phase offsets . Then the spin density M.sub.0, the relaxation time ratio and the local phase offset are determined from the fit for each voxel. A homogeneous image of the object is generated by calculating the signal intensity in each voxel, using the spin density M.sub.0 and the relaxation time ratio , wherein is chosen such that =0.
Claims
1. A method for improving intensity homogeneity of image data acquired using a balanced steady-state free precision (bSSFP) sequence in magnetic resonance (MR) imaging, the method comprising: performing multiple bSSFP sequences with different radio frequency (RF) phase increments by using a MR imaging apparatus, to generate multiple bSSFP image volumes with different phase offset of an object to be examined, each image volume containing voxels having intensities M that are a function of a nuclear resonance signal, and wherein a position and a size of the voxel is defined by a field of view (FOV) and a desired resolution; a per-voxel fitting of a mathematical signal model onto a measured magnetization M of the FOV as a function of the phase offset ; determining from the fit a spin density M.sub.0, a relaxation time ratio and a local phase offset for each voxel; generating a homogeneous image of the object by calculating a new signal intensity M () for each voxel, using the previously obtained spin density M.sub.0 and the relaxation time ratio , and choosing so that =0 , in order to obtain the homogeneous image of the object; and using the following mathematical signal model for fitting the measured magnetization M of the FOV as a function of the phase offset : wherein: M.sub.0 is the spin density of a voxel in the FOV; is the relaxation time ratio T1/T2; is the applied flip-angle of the bSSFP MR sequence; and is the local phase offset caused by field inhomogeneity for the voxel.
2. The method according to claim 1, which comprises undersampling an acquisition of the multiple bSSFP images with different RF phase increments in order to reduce an acquisition time.
3. The method according to claim 2, which comprises reconstructing the undersampled data by using one or a combination of advanced reconstruction methods selected from the group consisting of: a model-based reconstruction method; a compressed sensing method; and a parallel imaging method.
4. The method according to claim 1, wherein each step is automatically performed.
5. The method according to claim 1, which comprises determining at least one of the quantitative parameter maps selected from the group consisting of: a quantitative parameter map representing the spin density M.sub.0 for all voxels; a quantitative parameter map representing the relaxation time ratio for all voxels; and a quantitative parameter map representing the local phase offset for all voxels.
Description
BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING
(1) FIG. 1 is an illustration of a signal void at the nasal cavity in a bSSFP image of the human brain caused by local field inhomogeneities (arrow).
(2) FIG. 2 is an illustration of a CISS image reconstructed using two phase-increments combined calculating the sum of squares of the two images as well as the remaining artefacts in the final CISS image contrast (arrows).
(3) FIG. 3 is an illustration of the method according to the invention.
(4) FIG. 4 is an example of a trueCISS image contrast obtained according to the present invention.
(5) FIG. 5 is an illustration of a transformation of eight bSSFP images with different phase increments into a sparse representation by applying a wavelet transform in the spatial domain and a Fourier transform along the phase increment dimension, providing a good fundament for a compressed sensing reconstruction.
(6) FIG. 6 is a flowchart of a trueCISS image reconstruction using compressed sensing according to the invention.
(7) FIG. 7 is a flowchart of the trueCISS image reconstruction using model-based iterative optimization according to the invention.
DETAILED DESCRIPTION OF THE INVENTION
(8) FIG. 1 is an illustration of a transverse slice of the brain acquired with a bSSFP sequence according to prior art bSSFP techniques. A typical signal void occurred at the nasal cavity due to off-resonances (see arrow).
(9) FIG. 2 shows a CISS image C reconstructed using two phase increments (=0 and =180) combined by calculating the sum of squares of the two images intensities I1, I2 for each voxel of the images A and B. While the CISS image C shows fewer signal voids than the pure bSSFP image, residual banding artefacts are visible (arrows) and signal intensities differ from a genuine bSSFP signal.
(10) FIG. 3 exemplarily shows at the top eight acquisitions of a brain slice image with different phase increments (also referred to as phase-cycled bSSFP acquisition), at the bottom left, signal intensities of one voxel along the phase increments (dots) and the fitted mathematical signal model (dotted line) according to Eq. 1, and at the bottom right, the three resulting parameter maps M.sub.0, and obtained after fitting the mathematical signal model onto the acquired bSSFP image data. The resulting parameter maps are then used to synthesize an on-resonant bSSFP image by applying the mathematical signal model (Eq. 1) with =0, resulting in an image with no signal voids and genuine bSSFP signal contrast. A resulting image contrast is shown in FIG. 4, where it can be seen that no residual phase bands are present. The contrast depends purely on the genuine bSSFP signal. Furthermore and advantageously, the parameter maps may be directly used for the quantification of tissue properties such as the transverse relaxation T2 or proton density.
(11) FIGS. 5 to 7 illustrate the present invention in the particular case of using advanced image reconstruction methods for subsampling the acquisition of the multiple bSSFP images with different RF phase increments in order to achieve a shorter acquisition time. TrueCISS images based on a undersampled acquisition can be reconstructed using for example Compressed Sensing (see FIG. 5 for an example of rendering the undersampled images sparse as well as FIG. 6 showing an applicable algorithm for a CS reconstruction) or model-based reconstruction (see FIG. 7).
(12) FIG. 6 presents a preferred embodiment of the method according to the invention that starts with an accelerated acquisition 1 of data for a FOV defined for an object to be examined, using a standard bSSFP sequence which, however, acquires only a subset of the usually required data by skipping data samples, that results in an undersampled dataset 2. Then, according to the preferred embodiment illustrated by FIG. 6, the method according to the invention comprises a compressed sensing reconstruction 31. Compressed sensing relies on a sparse representation of the image of the object in order to recover the missing data samples, which were skipped during acquisition, using an iterative optimization. In order to transform the acquired image data into a sparse representation, the present invention proposes to use preferentially a wavelet transformation in the spatial domain and a Fourier transformation along the phase increment dimension. The wavelet transformation is particularly convenient since its transformed domain is sparse for typical medical images. Applying a spatial wavelet transformation together with a Fourier transformation along the phase increment dimension (see FIG. 5 for an exampleblack pixels mean zero or near-zero values, i.e. the more black, the sparser) provides the advantage that the result is a very sparse representation of the acquired image data due to the harmonic nature of the bSSFP signal-model. According to the invention, the method comprises using a conventional iterative compressed sensing optimization technique to yield or create multiple bSSFP images 4 without aliasing artifacts using the previously proposed sparse representation of the image data (see, e.g., Lustig et al., supra, for more details on compressed sensing). Subsequently, the method comprises a voxel-wise model-fitting 5, i.e. fitting the mathematical signal model of Eq. 1 for each sub-volume onto the corresponding voxel intensities of the multiple bSSFP images which are a function of the RF phase offset, determining the three resulting parameter maps 6. Finally, the method according to FIG. 6 comprises the generation 7 of an on-resonant bSSFP image in order to obtain the trueCISS image 8.
(13) FIG. 7 presents another preferential embodiment of the method according to the invention that starts also (as for FIG. 6) with an accelerated acquisition 1 of data for a FOV, defined for an object to be examined, resulting in an undersampled dataset 2. According to said other embodiment, the method according to the invention proposes to use a model-based iterative optimization 32 to reconstruct the trueCISS image 8. In this case, the mathematical signal model given by Eq. 1 is used as prior knowledge to recover missing samples of the undersampled data, using an iterative optimization, intrinsically estimating the required parameter-maps 6 (see Sumpf et al. previously cited for more details). The parameter maps 6 are then used to synthesize 7 the trueCISS image 8 as previously described.
(14) To summarize, the present invention proposes the following: to use the mathematical signal model given by Eq. 1 for fitting a series of bSSFP images with different phase offsets in order to obtain quantitative parameter maps, wherein the fitting provides the advantage of automatically separating image information that depends on the environment (i.e. local field offset ) from tissue parameters of interest (i.e. M.sub.0,). Advantageously, the parameter maps may be directly used for a quantification of tissue properties such as relaxometry or proton density; and to synthesize a phase band free image, i.e. said trueCISS image, with intensities according to the genuine bSSFP signal by applying the mathematical signal model Eq. 1 onto the parameter-maps and assuming that no local field inhomogeneity is present (=0), wherein the synthesized image is advantageously free of signal voids. Additionally, the image intensities represent the genuine bSSFP signal in contrary to the conventional CISS method; and optionally an undersampling of the multiple bSSFP image acquisition in order to accelerate the measurement, notably by using image reconstruction techniques such as compressed sensing, parallel imaging or model-based iterative optimization.
