WO1999030187A1 - Mri with removal of field inhomogeneity artifacts - Google Patents

Abstract

Artifacts in magnetic resonance imaging are removed by a third Fourier transform step that discriminates image data affected by susceptibility induced biasing differences applicable to particular voxels in a sample slice. The third Fourier transform effectively collects data on the affected voxels as if they were displaced along the slice selection Z or B0 axis, thereby collecting the data applicable to the slice regardless of biasing inhomogeneities. The spatial record representing the image of the slice is integrated together with one or more additional frames incrementally spaced along the Z axis. In this manner, image data displaced by inhomogeneities of biasing due to susceptibility variations of tissues, organs, openings or sinuses and the sample periphery, is associated together, providing a substantially artifact free image of the slice.

Description

MRI WITH REMOVAL OF FIELD INHOMOGENEITY ARTIFACTS

Background of the Invention 1. Field of the Invention

The invention relates to the field of gradient echo nuclear magnetic resonance imaging, namely collecting image data on spaced slices of a sample, using T₂ ^* weighted imaging. An improved method and apparatus is provided for processing the gradient echo data collected for individual imaging slices, so as to recover for each slice certain image data that is effectively masked by intravoxel phase dispersion due to susceptibility effects or other causes of localized main magnetic field inhomogeneity. Phase dispersion can be caused, for example, by localized differences in the magnetic permeability/susceptibility of discrete portions of the sample, for example human tissues, which causes image artifacts exemplified by loss of image contrast data at spatial positions where local permeabilities differ markedly, such as adjacent to the peripheries of organs, cranial sinuses and similar locations. A two dimensional image of an X-Y plane or slice is obtained initially for each slice from gradient echo data, by two conventional Fourier transform steps. According to the invention, a third Fourier transform is undertaken on the two dimensional image to provide an image profile along an axis perpendicular to the local field gradient. This selects for image data otherwise forming artifacts produced by low-order local field gradients, from the high- order microscopic field gradients responsible for T₂ ^* contrast. The third Fourier transform data is integrated into the two dimensional image data and thereby re- associated with the slice to complete the image and substantially remove the artifacts. 2. Prior Art

Nuclear magnetic resonance (NMR) gradient echo imaging essentially involves applying a longitudinal static magnetic field to at least a portion of the subject along a Z axis, often identified as the B₀ axis or, in connection with the acquisition of a plurality of slice images, in the slice selection direction. The field causes the nuclear spins of atoms and molecules in the subject to assume an alignment to the biasing field at rest, effectively magnetizing the subject along the Z axis. The subject is then irradiated with a radio frequency (RF) excitation pulse. The excitation pulse tips or realigns the nuclear spins of the atoms and molecules by a predetermined angle (the "flip" angle) away from the Z axis. After cessation of the excitation pulse, the spins precess about the Z axis at a predetermined frequency (the Larmor frequency) as they molecules return to the rest alignment determined by the biasing field. The precession of the spins produces a signal which is captured, digitized, discriminated for time, frequency and phase, and used to generate an X-Y planar image of a slice through the subject. The slice corresponds to an image slice at a predetermined orientation, typically (but not always) perpendicular to the B₀ field, and perpendicular to the local field gradient.

The transverse component of the signals emitted by the precessing atoms and molecules, namely in the X-Y plane normal to the Z axis, is detected as the output signal. The timing, amplitude and/ or phase of the output signal is processed by Fourier transforms to resolve the echo response of discrete spatial volumes (voxels) residing in the particular X-Y plane being imaged (the slice). The process can be repeated sequentially for successive slices, spaced along the Z axis, while advancing the subject relative to the apparatus (or vice versa). Each slice comprises an X-Y field of picture elements (pixels), typically displayed as an image having a luminance that varies as a function of the response of the tissues located at the corresponding X-Y position of each volume increment of voxel or predetermined slice width, at a particular timing, frequency and phase corresponding to the X-Y position of the voxel.

A set of slices for spaced positions along the Z axis forms a visual representation of the internal tissues through the volume of the subject. The NMR imaging technique is useful for visualizing tissue structures generally, and can be used in a variety of ways for diagnostic and measurement purposes in which detected relative differences in the magnetic permeability of spatially discrete areas of tissue are indicative of tissue characteristics. Tissue location and density can be distinguished. Solid tissues versus sinuses, tissues that are more or less dense, and tissues that inherently are more or less magnetically permeable provide contrast that makes tissues and organs distinctly visible in the image.

The contrast to have an adjacent to issues can be enhanced by perfusion agents, namely liquid substances infused into the blood stream, have a magnetic permeability distinct from that of adjacent issues. Similarly, the magnetic permeability of blood can be distinguished from that of tissue, to detect ischemic tissue due, for example, to stroke. In so-called functional MRI (fMRI) imaging of brain tissue and the like, image data for a given slice or set of slices is collected over time while applying a stimulus. Brain functioning in response to the stimulus causes variations in magnetic permeability of tissues (e.g., from alteration of blood flow patterns). Thus, MRI is useful not only for the static spatial mapping of tissues for diagnostic purposes, but also for more dynamic studies.

With the introduction of fMRI, techniques that optimize T₂ ^* contrast have become increasingly important in both clinical imaging of the brain and in neurological research. Functional MRI (fMRI) has emerged as a unique neuroimaging method that allows noninvasive monitoring of brain function with spatial and temporal resolution not possible with previous imaging modalities. The mechanism for signal changes observed in fMRI experiments can be based on blood oxygen level dependent contrast, which is a form of T₂ ^* contrast. In brain cortex, task activation or cognitive processes introduce regional variation in blood flow, and a change in the oxy/deoxyhemoglobin ratio. Because deoxyhemoglobin contains unpaired electrons, it increases the blood bulk magnetic susceptibility. Differences in magnetic susceptibility between the blood and the adjacent brain parenchyma result in intravoxel magnetic field gradients, which decrease the pixel signal intensity on heavily T₂ ^* weighted images. Neuronal activation produces a sharp, transient increase in the oxy/deoxyhemoglobin ratio; lowering the magnetic susceptibility of blood, and thereby decreasing the intravoxel magnetic field gradients. The increased uniformity of the magnetic field results in a few percent increase in amplitude of the heavily T₂ ^* weighted MRI signal. The sensitivity of detection of brain activities with fMRI is therefore directly proportional to the degree of T₂ ^* contrast produced by the imaging technique. In order to reliably detect neuronal activation, specialized MRI methods are needed to optimize T₂ ^* contrast. In addition to fMRI, examples of neuroimaging methods dependent on T₂ ^* contrast include: i. Assessment of relative brain iron concentration: Brain iron dysregulation and resultant deposition have been implicated in several neuropathologies such as Alzheimer's disease (AD), Parkinson's disease (PD), multiple sclerosis, Pick's disease, Huntington's disease, Hallervorden-Spatz disease and aceruloplasminemia. Qualitative and quantitative assessment of brain iron distribution with T₂ ^* contrast provides a useful tool for studying these neurodegenerative diseases. ii. Characterization of intracranial hemorrhage and calcification, iii. Qualitative assessment of regional cerebral blood flow (rCBV) during the dynamic injection of susceptibility based contrast agents. The rCBV measurements have been clinically used for detentions and evaluations of cerebral ischemic lesions in the hyperacute stage of a stroke ( < 6 hours), vascular stenosis or occlusions and brain tumor. Reliable and consistent T₂ ^* contrast is crucial for qualitative and quantitative evaluations of these diseases with T₂ ^* images and T₂ ^* measurements.

The importance of T₂ ^* contrast in brain research is shown by the increase of the number of high field human MRI systems (3 - 8 Tesla) among leading research groups. A motivation of investments in high field systems is to enhance T₂ ^* contrast for brain research. Magnetic resonance imaging techniques rely on biasing the tissue or other sample with a magnetic field (the Z axis B₀ field). However, differences in the magnetic permeability of localized portions of the sample affect the magnetic flux density in local areas. That is, the B₀ biasing field conforms locally to the character of the sample. Where there are localized permeability variations, the biasing field likewise varies. Higher permeability tissues tend to confine the flux lines. Less-permeable tissues allow flux lines to disperse. Assuming that a homogeneous biasing field is applied along that the Z axis, the magnetic bias incident on particular tissue areas is affected by the nature of adjacent tissues, specifically by the susceptibility /permeability of adjacent tissues. Whereas susceptibility /permeability characteristics vary throughout the sample, background field gradients are introduced that render the biasing of the overall image inhomogeneous. The imaging apparatus is attuned to decode and select for particular frequencies and phase angles as indicative of the gradient echo of tissues at preselected positions in the slice that is biased by the B₀ field and excited by the illuminating RF pulse. The technique for decoding the echo signal into a two dimensional contrast map or image is based on an assumption or approximation, namely that the biasing field is homogeneous. The localized permeability differences cause biasing irregularities that result in phase dispersion. The phase dispersion is such that the echo response of discrete volumes or voxels in the slice is not decoded or selected for association with other voxels spatially positioned in the slice. In this manner, localized permeability differences, which represent the very attribute of the sample that causes contrast for providing a useful image, also causes phase dispersion that detracts from the extent to which image data is recovered.

Phase dispersion from biasing field inhomogeneities affects the degree of contrast that can be obtained in the image. The phase dispersion can also be considered to deteriorate the signal to noise ratio over the slice. However, phase dispersion due to tissue permeability variations affecting particular voxels is different than random noise modulating the biasing field and the resulting echo responses. It is a steady state effect wherein the biasing field applied to a voxel is modified relative to an otherwise-homogeneous biasing field by the character of adjacent tissues, which character is of course not variable. If the modification of the biasing was the same for all the voxels, it would be possible to shift the echo response to account for the modification. However, the bias modification is different for different voxels because the adjacent tissues vary in placement and permeability.

Phase dispersion effects from local field inhomogeneity vary with the character of the tissue but are particularly evident when imaging tissues that are immediately adjacent to tissue voids. For example when imaging a cranial slice adjacent to a sinus, the amplitude of the local B₀ field may vary substantially from the average over the slice, thereby likewise shifting the echo response in k-space for the voxels adjacent to the sinus. By decoding the echo data assuming an average bias and discriminating the data closely as required to limit the collected contrast data to a spatially thin slice, the contrast information for the voxels adjacent to the sinus can be effectively shifted out of the image collected for the slice. These voxels simply disappear from the image of the slice. Alternatively, the contrast data can be decoded with a lower level of discrimination, thereby giving up resolution that would be available if the B₀ field was linear and homogeneous over the slice, and assuming repetitive slices, over the succession of images for the slices. What is needed, and what is provided according to the invention, is a method and apparatus for recovering or normalizing the echo response data for voxels subject to local variations in the level of the B₀ field. Advantageously, such echo response data is recovered notwithstanding the fact that the level of the local B₀ field applied to a given voxel may be higher or lower than the average B₀ field for voxels in the image slice field.C

Summary of the Invention It is an object of the invention to cancel the effects of macroscopic magnetic field inhomogeneity in an imaged sample, to obtain an optimum image and imaging technique with respect to image accuracy, image contrast, signal to noise ratio and image collection time.

It is a further object to provide a technique to recover image signal on voxels in a discrete slice, which signal appears for particular voxels to have been shifted in k-space from other voxels in the slice which are subject to a different level of bias.

It is another object to recover such image data while maximizing the contrast and accuracy of the resulting composite of voxels and displaced voxels in a finished image of the slice.

These and other objects are accomplished by removing artifacts in magnetic resonance imaging by using a third Fourier transform step that discriminates image data affected by susceptibility induced biasing differences applicable to particular voxels in a sample slice. The third Fourier transform effectively collects data on the affected voxels as if they were displaced along the slice selection Z or B₀ axis, thereby collecting the data applicable to the slice regardless of biasing inhomogeneities. The spatial record representing the image of the slice is integrated together with one or more additional frames incrementally spaced along the Z axis. In this manner, image data displaced by inhomogeneities of biasing due to susceptibility variations of tissues, organs, openings or sinuses and the sample periphery, is associated together, providing a substantially artifact free image of the slice.

MRI imaging is hampered by artifacts produced by B₀ field inhomogeneities in the excited slices. According to an inventive aspect, a gradient-echo slice excitation profile imaging (GESEPI) technique recovers the signal lost due to intravoxel phase dispersion in T₂ ^*- weighted images.

Artifacts produced by B₀ field inhomogeneity in the slice selection direction

(along the axis of the B₀ biasing field) occur because local magnetic susceptibility variations produce variations in the level of the B₀ field applied to particular voxels. The extent of these variations is large at relatively high field strengths due to stronger magnetic susceptibility effects. Intravoxel phase dispersion constitutes a time domain echo shift in the slice selection direction. However, the shift for a particular voxel may be positive or negative compared to a nominal echo response and varies from voxel to voxel in the slice. According to the invention, an incremental gradient offset is imposed on the slice refocusing gradient to sample k-space over the full range of spatial frequencies of the excitation profile. More particularly, in addition to collecting gradient echo data in the conventional manner employing two Fourier transforms for discriminating the response of the voxels at particular locations in an X-Y field defining the slice, a third Fourier transform is performed along the spatial slice selection (e.g., B₀) axis on the two dimensional image set obtained from the first two Fourier transforms. This generates an image set in which the artifacts resulting from low-order B₀ inhomogeneity in the sample are separated and removed from the high-order microscopic field gradients responsible for T₂ ^* contrast.

One reason that the invention removes field inhomogeneity artifacts while maintaining useful T₂ ^* contrast stems from the observation that the static magnetic field inhomogeneities that produce artifacts and those that yield contrast are different in scale relative to the image voxel. In the static dephasing regime, field inhomogeneities can be divided into three categories according to their scale relative to the image voxel. Macroscopic inhomogeneities are larger than or on the order of the scale of an image voxel. These produce undesirable image artifacts and are removed according to the invention. Mesoscopic inhomogeneities are smaller than the image voxel but much larger than an atomic and/or molecular scale. These variations produce useful tissue-specific T₂ ^* contrast. The invention retains the representation of mesoscopic variations, removed with a 180° radio-frequency (RF) pulse. Microscopic inhomogeneities are many orders of magnitude smaller than voxel size and are on the order of the atomic and molecular scale. Their fluctuations contribute to T₂ and T₂ ^* contrast. Microscopic variations are not removed with a 180° RF pulse or by the invention.

By taking advantage of differences in the interactions between the spin system and magnetic field inhomogeneities at various scales, the inventive method removes macroscopic field inhomogeneity artifacts while maintaining T₂ ^* contrast. The compensation gradient, G_c, effectively acts as a phase-encoding gradient in the slice selection direction (the Z or B₀ axis direction). The third Fourier transform data set consists essentially of N images of thin slices stacked along the slice selection direction relative to the nominal center of the slice on the Z axis. The thickness of these N slices depends on the maximum strength and duration of the compensation gradient G_c which in turn are determined by local field gradients. It has been found effective in removing the artifacts in the inferior human cortex at 3.0 T using a G_c that produces a 1 mm thickness. Eight to sixteen G_c increments are employed. The effective sensed volume in slice selection direction covered by these N images is thus larger than the nominal excited region.

Brief Description of the Drawings There are shown in the drawings certain exemplary embodiments of the invention as presently preferred. It should be understood that the invention is not limited to the embodiments disclosed as examples, and is capable of variation within the scope of the appended claims. In the drawings,

FIGURE 1 is a schematic illustration of a magnetic resonance imaging method and apparatus according to the invention.

FIGURE 2 is a time graph showing a pulse sequence according to the invention in which a compensation gradient offset is superimposed on a slice refocusing gradient and certain variables are indicated.

FIGURE 3 is a diagram illustrating the physical positions of a slice and a sample according to an experimental application of the invention.

FIGURE 4 illustrates three spaced images collected by analysis of the slice of FIGURE 3. FIGURE 5 is a set of amplitude graphs versus phase (k space) and position along the Z axis, corresponding to the images in FIGURE 4.

FIGURE 6 is a comparison of sixteen frame images collected according to a conventional processing technique involving two time/frequency-phase and frequency- phase/spatial transforms (6a) versus sixteen frames collected according to the invention. FIGURE 7 is a comparison of composite images showing a human brain gradient echo response (7a) having substantial susceptibility artifacts, and a human brain response according to the invention (7b) in which the artifacts are removed.

FIGURE 8 is a comparison of composite images of an immature rat brain at high biasing field strength (9.4 T) in which the invention (7a, 7c, 7d) is compared to a T₂ ^* weighted gradient echo image (7b) at equal TR/TE.

FIGURE 9 is a diagram illustrating an embodiment in which multi- variable slab thickness method is employed. Detailed Description of the Preferred Embodiments The invention is illustrated with respect to high field imaging of brain tissue but is applicable to various field strengths and for imaging purposes generally. With high field brain imaging susceptibility variations are marked and brain imaging, particularly adjacent sinuses, shows the effectiveness of the technique in recovering data otherwise lost to the artifacts. In the following examples, 3.0 T imaging is applied to human cranial imaging and 9.4 T for immature rats, demonstrating significant improvement in quality of the T₂ ^*-weighted contrast images for discerning the character and location of tissues with improved overall precision and contrast compared to known techniques. Intravoxel variation in the phase of echo responses in magnetic resonance imaging occurs due to local static magnetic field gradients in the sample. The variations in echo response provide a useful source of contrast using gradient-echo techniques, but also produce troubling image artifacts that detract from imaging effectiveness. Changes in tissue magnetic susceptibility with certain pathological processes provide a source of image contrast that is diagnostically useful. Clinical applications of gradient-echo T₂ ^*- weighted images include, for example, assessment of the relative brain iron concentration, characterization of intracranial hemorrhage and calcification, evaluation of the trabecular pattern of bone marrow, qualitative assessment of perfusion during the dynamic injection of susceptibility based contrast agents, localization of task-specific brain activation in functional magnetic resonance imaging (fMRI) using based on the endogenous deoxyhemoglobin, etc.

High magnetic field strengths enhance the contrast derived from tissue magnetic susceptibility. However, the diagnostic utility of gradient-echo images is limited by artifacts which are pronounced at high fields and are caused by macroscopic magnetic field gradients. These magnetic field gradients may be due to spatial differences in tissue magnetic susceptibility, imperfect static magnetic field (B₀) adjustments and similar factors. With heavily T₂ ^*- weighted images, severe signal loss occurs at interfaces between tissues of different character and near air-tissue interfaces. For example, tissues of the inferior frontal and lateral temporal lobes of the brain may be obscured due to these artifacts. Field inhomogeneities likewise occur near implanted paramagnetic surgical devices and produce marked image distortion. With small-bore micro-imagers, field inhomogeneities span a significant region of interest and are difficult to remove, thereby limiting the application of gradient-echo techniques. According to an inventive aspect, gradient-echo techniques, especially at higher field strengths, are provided that optimize image contrast while minimizing image artifacts.

The invention provides a means to differentiate differences in echo response due to static magnetic field gradients (i.e. , artifacts) and those that yield contrast distinguishing voxels within a selected slice. In general, the field gradients that produce artifacts are macroscopic in scale within the tissue (on the order of the image voxel). Their variation across the image voxel is either linear or has a low-order spatial dependence. The field gradients that generate useful magnetic susceptibility based contrast, on the other hand, are microscopic in scale and vary either randomly or with various high-order spatial dependencies applicable to the voxel. In addition, these microscopic gradients may vary temporally with changes in tissue magnetic susceptibility. The invention improves image quality by removing the low-order field gradients that result in artifacts from the high-order microscopic intravoxel gradients that yield desirable contrast.

Field inhomogeneity artifacts have been the subject of various studies. The signal loss due to linear static field gradients of a known specific magnitude can be partially recovered by adjusting the slice refocusing gradient as suggested by Frahm et al. in "Direct FLASH MR Imaging of Magnetic Field Inhomogeneities by Gradient Compensation, " Magn. Reson. Med. 6, 474-480 (1988). Cho et al. in "Reduction of Susceptibility Artifact in Gradient-Echo Imaging, " Magn. Reson. Med. 23, 193-200 (1992) demonstrated that at a cost of a lower signal to noise ratio (SNR), signals from the homogeneous region and regions with specific static field gradients could be acquired simultaneously using a tailored excitation pulse. Haacke et al. have shown that intravoxel dephasing artifacts are partially reduced in three dimensional gradient-echo images. "Reduction of T₂ ^* Dephasing in Gradient Field-Echo Imaging, " Radiology 170, 467-469 (1989) and "Theory and Application of Static Field Inhomogeneity Effects in Gradient-Echo Imaging, " JMRI 7, 266-279 ( 12997) . Ro and Cho proposed an innovative technique to analyze the susceptibility effect, which digitized the intravoxel phase dispersion for spectral decomposition of local magnetic field gradients. "Susceptibility Magnetic Resonance Imaging Using Spectral Decomposition," Magn. Reson. Med. 33, 521-528 (1996).

The present invention improves on the foregoing methods. Images are acquired with an incremental slice refocusing gradient offset and integrated (along the BO direction) with a Fourier transform. This method, referred to as Gradient-Echo Slice Excitation Profile Imaging (GESEPI), effectively images the slice excitation profile. The technique separates and removes the intravoxel phase dispersion artifacts in the slice dimension caused by local field gradients, while maintaining T₂ ^* contrast and improving image signal to noise ratio (SNR).

The technique is schematically illustrated in FIGURE 1. A theoretical analysis of the method and experimental results demonstrating its utility for high field brain imaging of humans at 3.0 T and immature rats at 9.4 T follows, examining the intravoxel phase dispersion due to the local field gradient along the slice selection direction Z in a gradient-echo image.

The local field gradients directed along the slice selection axis, G_z generate a phase dispersion within the excited slice along the Z direction. At a time TE, the magnetization distribution along axis Z is:

M_TE(z)=M(z)e ^{'i Gi Ez}

where γ is the gyromagnetic ratio, and M(z) is the initial magnetization density which includes all other factors that affect the voxel signal. In a two-dimensional (2D) gradient-echo imaging technique, the image pixel intensity, I^2D _VOχ_> remains in the time-domain (k-space) with respect to the slice selection direction. To examine the influence of the phase dispersion factor on the time-domain signal, a Fourier transform (FT) is applied:

I^_x(k )--FiM(z)e -ⁱ^^TEz}

where F is the Fourier transform operator. For the case of a linear intravoxel field gradient (first order approximation), G_z ^l is a constant across the voxel. The shift theorem can be applied to the foregoing equation to yield:

I_v ² _oxD(k_z)=M(k_z-K_zfi)=M(a)

where k.₀ = γG^TE and α = k_^- _^.₀. The signal intensity reaches a maximum (echo peak) when the argument, a, is zero. Thus, the phase factor produced by G_z ^l causes an echo shift to k^.

The above Fourier operation can be experimentally carried out with a pulse sequence as shown in FIGURE 2, in which a compensation gradient offset, G_c, is superimposed on the slice refocusing gradient in a conventional gradient-echo sequence. A series of N images are acquired with sequential increments, ΔG_C, of G_c within a range of ±GT - Application of a two dimensional Fourier transform to each of the N acquired two dimensional k-space data sets generates N images (2D-FT data set), in which the intensity of each pixel varies with k_j (1^ = γG_ct_c) as described above. In conventional two dimensional imaging methods, the image signal represents a single point, i.e., l = 0. For voxels in the magnetically uniform regions where G_z ^! = 0, the signal intensity is maximum at the center of k-space. For voxels where G_z ^! ≠ O, the signal maximum shifts from the center to k^, and thus the gradient-echo image acquired conventionally with k_^ centered in k-space results in a loss of signal compared to the maximum. The signal loss is recovered according to the invention by effectively moving the acquisition point to k~ = k, ₀. The final artifact-reduced image is generated by summation of the 2D-FT magnitude image set. However, combining these images by simple addition decreases the signal to noise ratio in the resultant image. This problem is solved according to the invention by integrating the 2D-FT images with a third inverse complex FT of the 2D-FT data set with respect to k_z, to generate a 3D-FT data set with voxel signal intensity, I_vox ^3D(z), given by:

where F¹ is an inverse Fourier transform operator. This equation shows that the signal intensity for the 3D-FT data set in image-space is modulated by the excitation profile in the slice selection direction. I_Voχ^3D thus represents the image of the slice (or slab) excitation profile M_TE(z) in the first equation. The k-space signals that are shifted by different G_z' and refocused in separate images with corresponding G_c are now all associated in the same image field within the excited slice in image-space.

In this case, the intravoxel field gradients only cause phase differences, which have no effect on the final magnitude image. The intravoxel phase dispersion artifacts are removed by the gradient-echo slice excitation profile (GESEPI) imaging method as described. The signal intensity of the images within the excited slice in the 3D-FT image set is proportional to the echo magnitude, regardless of how the echo shifts in k-space. In general, the intravoxel gradients are non-linear. In this case, the phase factor can be expanded using a Fourier series such that:

-i ^β TEz

and the time domain signal may be expressed as:

*_z)=∑„ c„ (*.-tf_Z)„)

From this expression, the time domain signal in k-space can be regarded as a superposition of multiple echoes, each with magnitude (c_n) and shift position in (k^). A Fourier transformation of the foregoing equation once again generates the image set expressed above where:

Therefore, this method also, in principle, removes the intravoxel phase dispersion due to non-linear terms of the local field gradients.

In the GESEPI method, G_c effectively acts as a phase encoding gradient in the slice, selection direction. Thus the excitation profile is imaged. Consequently, each image slice thickness (Δz) in the 3D-FT data set is different from the slice thickness for the 2D-FT data set, and depends on the maximum strength and duration of the compensation gradient offset:

The images within the excited slice profile of the 3D-FT data set can be summed to produce a resultant image with the same slice thickness as that of the 2D-FT image set.

The 3D-FT image set acquired by GESEPI is similar to 3D gradient-echo images, but there are some essential differences. A first difference is that since the purpose of G_c is to compensate for the intravoxel phase dispersion due to G_z\ the maximum compensation gradient offset G7³ is determined by the maximum value of G_z ^! in the image slice and by TE. That is, G^L_. > G^TE (compare FIGURE 2), in order to include the shifted echo peak from all the voxels within the acquisition window. The maximum phase encoding gradient for the excitation-selected slab 3D gradient echo image is determined by a user defined slice thickness. The phase encoding gradients for the 3D-GE method are not optimized to compensate for the intravoxel phase dispersion.

As another difference, the acquisition bandwidth for the phase encoding produced by G_c is N/γG^ (where N is the number of G_c increment steps). This is much larger than the bandwidth of the slice selective excitation RF pulse. GESEPI over-samples the image-space in the second phase encoding direction by G_c. Over-sampling in GESEPI plays an essential role in removing the image aliasing artifacts due to the high-order field gradients.

Conventional gradient-echo performance and performance of the GESEPI technique as described above were compared using three test situations: a phantom model, human brain images, and immature rat brain images. With the GESEPI technique, pixel signal intensity of the 2D-FT data set varies with k_z as described. Fourier transformation of the 2D-FT data set generates a set of images in which pixel intensity is modulated by the excitation profile, and is referred to as the 3D-FT data set. The images within the excited profile were summed to yield an image with the same slice thickness as the images in the 2D-FT data set.

The magnitude of the increment, ΔG_C, and the range of the compensating slice refocusing gradient offset, ±G_c ^max are reported as a percent of the slice refocusing gradient. As explained in the previous section, G ™* is determined by the criterion G^t. > G_Z'TE for all the voxels in the slice. Since G_z ^: is a function of slice orientation and position of the image slice, G ™* was carefully determined experimentally to ensure that the largest echo shift is included within the sampling window. Subsequently, G_c ¹™*^ defines the slice thickness of the 3D-FT data set. The number of G_c increment steps, N, which defines the acquisition bandwidth in the slice direction once G,.¹™* is determined, was also determined experimentally.

The phantom diagram illustrated in FIGURE 3 shows the arrangement of two air-filled spheres positioned inside a gelatin filled cylindrical container with its longitudinal axis perpendicular to B₀. An axial image slice was taken such that the two spheres are located at an equal distance on opposite sides of the image plane. With this arrangement, the intravoxel field gradients G_z' at points P and Q are equivalent in magnitude, but are in the opposite direction. The voxel at point R is in a region of uniform magnetic field.

The phantom and human brain images (compare the incremental frames in FIGURE 6 parts a and b, and composites in FIGURE 7 parts a and b) were acquired using a MEDSPEC S300 3.0 T research whole body imager (Bruker Instruments, Inc., Karlsruhe, Germany). The selective excitation pulse was a 2 ms five lobe sine pulse. The phantom images were obtained with a flip angle of 25°, TR/TE = 80/25 ms, field of view (FON) = 25 cm, image matrix = 128 x 128, slice thickness (ST) = 5 mm, and acquisition bandwidth = 50 kHz. The slice refocusing gradient offset was systematically varied in steps of 64 phase encodings to produce a set of 64 frames or images with G ™* ranging by ±70% . Accordingly, the data representing the 5mm slice thickness was discriminated to provide 64 data or image frames representing voxels that appear in the data (due to biasing inhomogeneities) to be incrementally spaced along the Z axis. A conventional 3D-gradient-echo image set of 0.3 mm slice thickness in the axial direction was acquired with the same parameters as in the gradient echo image.

The axial human brain image set consisted of 16 images with G ™* also ranging by +70% . The image plane was 2 mm above the spheroid sinus with TRITE = 100/25 me, FON = 25 cm, matrix = 256 x 256, and ST =5 mm. This set provided from the 5 mm slice thickness a set of 16 images of voxels apparently incrementally spaced along the Z axis.

Coronal brain images of immature rat pups (7-16 days old) were acquired with a small-bore micro- imaging probe at 9.4 T (Bruker Instruments, Inc., AM 400 WB) (FIGURE 8). The inhomogeneity of the incident field was initially adjusted to within 0.1 ppm across the slice profile. Images were obtained with a 2 ms single lobe sine excitation pulse with a 35° flip angle, TR = 80 ms, FON = 20 mm, ST = 1 mm, matrix = 128 x 128 or 256 x 256, and acquisition bandwidth = 25 kHz. The compensating slice refocusing gradient consisted of 16 or 32 steps, with a range of G_c = 125% .

FIGURE 3 shows a schematic layout and FIGURE 4 shows the conventional 2D gradient-echo (4b), GESEPI (4c) and conventional 3D gradient-echo (3d) images from the same slice of the phantom arrangement shown in FIGURE 3, wherein the slice is taken entirely through a cross section of gelatin between and spaced from two air filled spheres as shown in FIGURE 3. The dark areas at points P and Q in FIGURE 4b are the typical magnetic susceptibility artifacts that appear adjacent to air-filled spheres, similar to artifacts in images of tissue adjacent to sinuses in cranial imaging. As demonstrated in FIGURE 4c, these artifacts are removed in the GESEPI image. As shown in the image in FIGURE 4d, it is possible by adding 16 conventional 3D gradient-echo images from 0.3 mm slice frames to improve the problem of artifacts; however, significant signal loss artifacts still appear.

FIGURE 5 is a plot of the voxel intensity at points P, Q, and R from the phantom image is plotted following both the initial two dimensional Fourier transform, as a function of k_^, and the third Fourier transform, as a function of the position z along the slice selection direction. With respect to voxel R in the homogeneous region, the intensity as a function of k, has a sine shaped profile corresponding to the wave-form of the excitation pulse. Signal intensity is maximum where 1^ = 0 rad cm^"1 (32nd image). The signal intensities from voxels P and Q are much reduced in this central image, since the signal maxima are shifted in opposite directions by the opposing local field gradients, G_z', to the 47^th and 19^th images, respectively. This clearly demonstrates that the signal loss due to intravoxel dephasing can be treated as an echo shift in k_^, and can be recovered by shifting the acquisition point to 1^ ₀. Following the initial 2D-FT, this signal spreads amongst all images of the data set according to the strength and direction of G_z'. Each image in the 2D-FT data set contains only a portion of the imaged object. After performing the third Fourier transform with respect to k_^ the voxel intensity of magnitude images is modulated by the excitation profile. Unlike the signal intensity plots for P, Q, and R in the 2D-FT data set, the voxel intensities in the homogeneous region R' and inhomogeneous regions, P' and Q' are all at maximum in the center of the slice profile.

Five images are shown in the excited slice position. Summation of these five images generates an image, free of artifacts, with the same slice thickness as the conventional gradient image shown in FIGURE 4b. The signal to noise ratio for the GESEPI image in FIGURE 4c is 122. The signal to noise ratio for the conventional gradient-echo image is 90. In general, the SNR for GESEPI is the same or better than the conventional gradient-echo image, depending on the extent of over-sampling.

The intrinsic difference in signal characteristics between k-space and image-space is that for an ensemble of spins in a voxel with various phases, the signal intensity at any point in k-space is the vector sum while for the magnitude image-space, the signal is proportional to the scalar sum (echo peak). Phase dispersion in the spin ensemble reduces the vector sum and has no affect on the scalar sum. As long as the time-domain acquisition window includes all of the shifted echo peaks from the voxels of the excited slice, the signal attenuation due to intravoxel phase dispersion can be recovered. The role of the third Fourier transform is to effectively bring the pixel signals that are spread into different images in the 2D-FT data set (i.e. displaced along the Z axis) to the images within the excited slice profile. Thus, the GESEPI images are substantially free of signal loss artifacts due to intravoxel phase dispersion in the slice direction.

FIGURE 6 shows an example of a human brain axial image of the entire 16 2D-FT data set and the corresponding 3D-FT data set. Only images within the excited slice (images 4 to 9) show strong signal intensity. As discussed previously, the image-space bandwidth defined by N/7G_c ^maxt_c is wider than the excitation pulse bandwidth such that regions outside the excited slice are also imaged. This appears to be wasteful or counterproductive, since presumably there is no useful signal coming from these regions to image. There are two reasons why over-sampling is necessary in order to remove the phase dispersion artifacts. First, the sampling in 1^ must be fine enough to capture and define the echo peak that is shifted due to the heterogeneous distribution of G_z' in the excited slice. The intensity of the echo peak determines the pixel intensity in the 3D-FT images. Accurate sampling of the echo peak improves the final image quality. Secondly, the local field gradients (especially the non-linear components) in the voxels produce frequency shifts which may exceed the slice bandwidth. Thus, it can be seen explicitly from FIGURE 5b that the images outside the excited slice exhibit an increasing proportion of signal intensity from the inhomogeneous areas. Without over-sampling, these signals would alias into the central images, creating aliasing artifacts.

The image in FIGURE 7a is a sum of sixteen magnitude images of 0.3 mm thick slice acquired with a standard 3D gradient method. The slice is adjacent to the human sphenoid sinus, and serious signal loss and aliasing artifacts can be seen in such image even with such a relatively thin slice. Comparing with the artifact free GESEPI image in FIGURE 7b, in addition to the magnetic susceptibility artifacts that are still present conventional 3D-GE images, aliasing artifacts degrade image quality significantly. The extent of over-sampling needed depends on the distribution of local field gradients within a specific imaging slice. For the slice of the human brain at 3.0 T shown in FIGURE 6, sixteen G_c steps are adequate to remove the susceptibility artifacts substantially completely. For the phantom image in FIGURE 4, 64 G_c increment steps were used to obtain an the depicted precise representation of the echo shape, although only 32 increment steps were found to be necessary substantially to remove the signal loss artifact in the areas around the voxels P and Q. This excess of over-sampling produces an increase in SNR in the GESEPI image in FIGURE 4c due to multiple sampling on the echo peak. Thus, excess oversampling is preferable to signal averaging for generating T₂ ^*-weighted images in high field micro-imaging, where the voxel volume is small and the magnetic susceptibility effect is strong.

FIGURE 7 shows a conventional gradient-echo axial human brain image and the corresponding 5 mm thick GESEPI image which was obtained by summing the four center images of the 3D-FT data set. The signal losses in the gradient-echo image which produce large, dark artifacts in both the frontal and temporal areas and at the brain-skull interface are completely recovered in the GESEPI image. Image blurring due to the field gradients are also removed. The overall signal intensity is more uniform than the gradient-echo image, reflecting the signal recovered from losses attributed to inherent field inhomogeneities. The overall SNR of the GESEPI image is better than the corresponding gradient-echo image because the signal intensity in the latter is attenuated by the intravoxel field gradients, while the GESEPI image is amplified due to the summation of the component slice images.

The GESEPI method removes only the intravoxel phase dispersion due to the macroscopic local field gradient, G_z'. The contrast in GESEPI images is still dominated by T₂ ^* weighting. FIGURE 8 shows a rat brain GESEPI image along with the corresponding conventional gradient echo T_r and T₂- weighted spin-echo images from the same slice. The contrast in the GESEPI image is distinctively different from that of the spin-echo images. Unlike the T₂- weighted image, the cerebral spinal fluid appears darker in the GESEPI image due to the field gradients on the brain surface, which is rich in blood vessels and close to the air-tissue interface. The improved T₂ ^*-weighted GESEPI image reveals a detailed cortical vascular structure that cannot be seen in the T_r and T₂- weighted spin-echo images. These vascular structures are indistinct or obscured in the conventional T₂ ^*-weighted gradient-echo image of the immature rat brain. The main factor that contributes to the contrast in the GESEPI brain image is presumably the susceptibility effect created by the deoxygenated capillary blood in the voxel. The susceptibility effects that cause this T₂ ^* contrast are intrinsically different from those that cause the artifacts. The former are due to gradients that vary randomly in space at the microscopic scale. The latter are due to the macroscopic gradients that vary smoothly in space within the voxel. Only the effects that cause artifacts can be altered by the application of external macroscopic gradients. Thus, the GESEPI method removes the unwanted intravoxel phase dispersion artifacts caused by the macroscopic gradients, while still retaining the desirable T₂ ^* weighting for the magnetic susceptibility contrast.

Conventional gradient-echo images at short TE times may appear blurry with uneven signal intensity over the image. These artifacts become progressively worse as TE increases. The GESEPI images, however, provide superior image quality. At TE = 22.4 ms, the GESEPI image displays significant anatomical details that are not visible in the conventional gradient-echo image. The gradient-echo image at TE = 33 ms is completely distorted by the artifacts, while details from the GESEPI image at TE = 50 ms are still discernible with enhanced T₂ ^* weighting. Thus, the GESEPI technique extends the TE observation window for magnetic susceptibility contrast at very high field micro-imaging.

According to a further embodiment, and with reference to FIGURE 9, effective removal of field inhomogeneity artifacts with maximum volume coverage and minimum acquisition time is further accomplished according to the invention. The success of GESEPI allows acquisition of artifact-free heavily T₂ ^*-weighted images. To make the GESEPI method readily applicable to clinical examinations and dynamic studies (e.g. fMRI), the whole-brain volume coverage and high temporal resolution can be achieved with multiple variable slab thickness (MNST) method, illustrated in FIGURE 9. A human brain is imaged with a set of slabs with variable thickness determined by location in the brain. The 3D imaging sampling bandwidth (SBW) in the slab selection direction is wider than the bandwidth of the excited slab (EBW). A continuous 3D image set covering the total imaging region is obtained after discarding the slices in the over-sampled regions. The thicker slabs will be used for the regions not influenced by magnetic susceptibility artifacts. Referring to FIGURE 9, this acquisition method excites a set of slabs of variable thickness in any orientation to cover the entire brain during one TR period. These excited slabs are subsequently partitioned into multiple slices by the G_c gradient. Multiple slab excitation in a signal TR time allows multiple fold reduction of GESEPI/3D imaging time. Taking advantage of the similarities in data acquisition between GESEPI and

3D-GE methods, the excitation slab thickness was selectively chosen according to the slab positions in the brain. Each slab is phase-encoded with the same number of compensation gradient steps (N) but with different strength (G_c). For the inhomogenous regions (inferior brain regions in Slab 3 in FIGURE 9) stronger G_c is required to remove the artifacts, which leads to thinner slice and slab thickness (sum of thickness of the thin slices) for a given N. Thus, the slab thickness in these regions is determined by local field gradient as in GESEPI. Likewise, weaker G_c leads thicker excited slab in the areas with weaker local field gradients (Slab 2 in FIGURE 9). For the slabs in the superior region with minimal local field gradients, the strength of G_c is determined by the desirable slice thickness as in the conventional 3D imaging methods to cover a larger volume (Slab 1 in FIGURE 9). Thus, the excited slabs have increasing thickness in the inferior-superior direction as depicted. The slab thickness is varied according to the strength of the local field gradients. Variable slab thickness allows removing field inhomogeneity artifacts with GESEPI and optimal spatial coverage with conventional 3D-GE method.

Uniform slice thickness for final whole-brain image set is achieved by group-wise adding the magnitude images from adjacent slices in the inferior regions to match the slice thickness in the superior region.

This message has a number of advantages. For example, it removes magnetic susceptibility artifacts in T₂ ^*- weighted GE and EPI images; and achieves multiple fold reduction of 3D volume imaging time. Clinical applications include true whole-brain scan (including the brain areas that are obscured by the artifact) with T₂p* weighting, as with the following scenarios. In a first scenario (MVST-GE), it is assumed desirable to acquire an GE axial image set with 24 images of 6 mm thick, TR = 80 ms, TE= 25 ms, and matrix = 128 x 128. Using MVST, the image set consists of 4 slabs with thickness 6, 6, 36, 48 and 48 mm inferior-superior. Each slab is partitioned into 8 slices, resulting a total of 32 (4 x 8) images. The total image time is 5.4 min. For final image that covers 144 mm in axial direction can be formed with 24 images with 6 mm slice thickness. In another scenario (MVST-EPI), it is conservatively assumed that a 64 x 64 image of single slice is to be acquired in 100 ms (normally 50 ms or less) with EPI. The total image time to acquire the above image set is 3.2 sec. A further possibility is spin-echo 3D imaging and angiography.

The invention generally and the foregoing embodiment in particular have clinical utilization for fast or dynamic examinations (fMRI, perfusion, diffusion) and for imaging of whole brain rapidly. The significant pathologies that occur in the inferior frontal and temporal brain regions were previously inaccessible for diagnostic examinations due magnetic susceptibility artifacts. Cardiac imaging and angiography also frequently suffer from magnetic susceptibility artifacts and can be corrected with this method. The method can be implemented in three steps. To avoid the influence of the inherent artifacts in EPI, the MVST excitation strategy is first implemented with conventional gradient-echo sequence (MVST-GE). With MVST-GE, the optimal parameters for MVST technique are determined by finding the proper tradeoff between volume coverage, temporal resolution and effective artifact removal. The parameters that need to be optimized are the number of iterations for the excited slab (G_c, N), and the thickness of each slab in the brain at specific TE.

The optimal G_c, N and slab thickness required for effective artifact removal at each specific location in the inferior brain regions can be first determined using single slab GESEPI method. The inferior frontal brain surface can be used as the reference for slab locations. Since the artifact also depends on TE, this process can be repeated, for example, four times for each slab at TE = 5, 15, 25, 35ms. Optimal parameter settings for other TE values will be obtained by interpolation or extrapolation.

Since the average superior-inferior dimension of a human brain is about 15-16 cm, the number of slabs will be determined subsequently for a given FOV in the inferior-superior direction. It is believed that the entire brain can be imaged with eight slabs: two 5 mm thick slabs right above the inferior frontal cortex and three 50 mm slabs covering the remaining areas. Experience has shown that the artifacts in a 5 mm slab in the inferior frontal and temporal brain areas can be removed effectively with eight G_c steps on our 3.0 Tesla system. In this process, regional SNR and CNR (contrast to noise ratio) are be used to judge the image quality under different parameter settings. The SNR from the same tissue type (e.g., gray or white matters) from homogeneous regions and inhomogeneous regions in the same slices are compared to determine how effective the artifact is reduced. The regional CNR between gray and white matters are also calculated to determine if the T₂ ^* contrast is altered under different parameter settings. Using MVST-EPI with eight G_c steps in each slab, the acquisition time of the entire brain with 64 x 64 in-plane pixel resolution is estimated to be possible in less than 5 sec.

According to the foregoing method and image capture apparatus, signal loss due to local field gradients is ascribed to an echo shift in the slice selection direction in k-space. This signal, normally not acquired with 2D-GE and under-sampled with conventional 3D-GE techniques, is fully recovered by the GESEPI method. The 3.0 T in vivo results shown demonstrate that the GESEPI method significantly improves the quality and consistency of gradient-echo images, thus strengthening the use of T₂ ^*-weighted contrast as a reliable diagnostic modality. The GESEPI experimental results at 9.4 T show that this method is especially valuable for providing susceptibility contrast images at long TE times, with small bore micro-imagers, and at the highest field strengths.

The invention having been disclosed in connection with the foregoing variations and examples, additional variations will now be apparent to persons skilled in the art. The invention is not intended to be limited to the variations specifically mentioned, and accordingly reference should be made to the appended claims rather than the foregoing discussion of preferred examples, to assess the scope of the invention in which exclusive rights are claimed.

Claims

We claim:

1. A method for magnetic resonance imaging of a sample having magnetic susceptibility differences among discrete volumes in the sample, comprising the steps of: magnetically biasing a sample along a Z axis direction, thereby aligning magnetic spins in the sample toward a rest position relative to the Z axis, said biasing being inhomogeneously variable due to said magnetic susceptibility differences; illuminating the sample with an electromagnetic pulse transverse to the Z axis, thereby tipping the magnetic spins in the sample, the spins precessing following said electromagnetic pulse; sampling and digitizing an electromagnetic response of the spins while precessing to provide data representing amplitude and timing characteristics of the response; generating from said data a record of the response according to at least one of time delay, frequency and phase, and a spatial record representing a response of spins located at voxels at discrete spatial positions in an X-Y slice normal to the Z axis, said spatial record corresponding to an image of the voxels in the slice; further generating a further spatial record representing a response of the spins appearing to be located at least at one predetermined space along the Z axis from the X-Y slice; and, integrating together the spatial record and the further spatial record to provide data representing an image of the slice, whereby image data displaced by inhomogeneities of said biasing is associated together with undisplaced image data representing the slice.

2. The method of claim 1 , wherein said generating according to at least one of time delay, frequency and phase comprises a first Fourier transform, said generating of the spatial record comprises a second Fourier transform, and said further generating comprises a third Fourier transform.

3. The method of claim 1 , wherein said further generating step comprises discriminating the data for responses of the spins appearing to be located at a plurality of predetermined spaces along the Z axis from the X-Y slice.

4. The method of claim 1 , wherein said illuminating of the sample with the RF pulse is accomplished over a slice thickness greater than a selected thickness of a resulting slice in the X-Y plane.

5. The method of claim 3 , wherein said illuminating of the sample with the RF pulse is accomplished over a slice thickness greater than a selected thickness of a resulting slice in the X-Y plane, and said plurality of predetermined spaces along the Z axis comprise positions along the Z axis in said slice thickness greater than the selected thickness.

6. The method of claim 3, wherein said plurality of predetermined spaces includes at least four positions along the Z axis.

7. The method of claim 6, wherein said plurality of predetermined spaces comprise at least sixty four positions along the Z axis.

8. The method of claim 3, wherein the sample comprises human tissue.

9. The method of claim 8, further comprising applying a stimulus to the sample, repeating said biasing, illuminating, sampling, generating, further generating and integrating steps, and comparing results before and after the stimulus for identifying functional variations in the sample.

10. The method of claim 1, further comprising varying a slab thickness of an excited slice according to a local field gradient.

11. The method of claim 1 , further comprising varying an orientation of the excited slice according to a local field gradient.