The present invention relates to method of reducing Magnetic Resonance Imaging (MRI) acquisition times, and more particularly to a method of using a partial k-space data set with a full k-space data set of different phase to selectively suppress a chemical species when generating one or more MRI images.

A number of suppression techniques have been implemented to differentiate chemical species in MRI imaging. Suppression techniques, known as fat suppression, are often used to in MRI to differentiate fat from water in an imaged object. The majority of fat suppression techniques require specialized Radio Frequency RF excitation schema to selectively excite water protons or saturate fat protons. Binomial and CHESS are known excitation schema which are extremely effective at 1.5T. However, these techniques have encountered limitations clinically. For example, at higher field strengths imposed Specific Absorption Rate (SAR) constraints limit the range of spectrally selective excitations for fat suppression. At lower field strengths, T1 and T2 relaxation causes problems, and the increased duration of the RF excitations needed for the low field strengths can cause undesirable extension to the overall imaging time.

Known multi-point Dixon fat suppression methods take advantage of the relative difference in precession frequency of fat and water to create water and fat images from at least two full acquisitions. Higher order Dixon methods can be used to correct for field inhomogeneities and/or susceptibility artifacts in the fat suppressed images. Dixon fat suppression is obtained without specialized RF excitation pulses, thereby eliminating the SAR constraints of binomial and CHESS excitations. However, the multi-point Dixon methods

significantly extend the overall imaging time by requiring multiple full acquisitions. For example, the 2-point Dixon method for fat-suppression requires two full acquisitions, one full set of"Water-Fat"k-space data and one full set of"Water+Fat"k-space data to generate the fat-suppressed image.

It is desirable to provide a short acquisition time for rapid imaging applications such as cardiac imaging, especially on ultra-high field MRI systems where SAR limits the use of spectrally selective RF excitation pulses.

It is therefore, desirable to reduce the acquisition times used to generate Dixon fat-suppressed images.

SUMMARY OF THE INVENTION According to the present invention, a new and improved method of generating an MRI image by selectively suppressing one of two chemically shifted species is provided.

In accordance with a first aspect of the invention, the method includes acquiring a full data set and a partial data set with the echo times (TE's) set such that the magnetization of the second chemical species is 180° out of phase from its orientation in the first data set. The full and partial data sets are combined to generate an MRI image that is perceptually equivalent to a multi-point Dixon image.

In accordance with a second aspect of the invention, the invention includes acquiring the minimum amount of data in the partial data set to generate an image which is perceptually equivalent to a multi-point Dixon image.

Other features, benefits and advantages of this invention will become apparent to those skilled in the art from the following detailed description of the preferred embodiments, when read in light of the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS The invention may take form in certain components and structures, preferred embodiments of which will be illustrated in the accompanying drawings wherein: Fig. 1 illustrates the steps of the invention;

Fig. 2 is a diagram illustrating the combination of the full and partial data sets in accordance with the invention; Fig. 3 illustrates MRI images generated in accordance with the invention compared to a multi-point Dixon image; Fig. 4 is a regression plot to compare the error from a Perceptual Difference Model with human observer error scores; and Fig. 5 is a plot of the number of keyhole lines of k-space data needed to generate an MRI image that is perceptually equivalent to a multi-point Dixon image.

DETAILED DESCRIPTION OF THE INVENTION It is to be understood that the specific devices and processes illustrated in the attached drawings, and described in the following specification are simply exemplary embodiments of the inventive concepts defined in the appended claims.

Referring to Fig. 1, the invention includes a method of generating an MRI image by selectively suppressing one of two chemically shifted species as shown generally at 10. For the purposes of example, the suppression technique described herein is a fat suppression technique and the first and second chemical species are fat and water. However, this example should not be considered as limiting. The invention is applicable to any other two chemical species exhibiting a relative difference in precession frequencies during MRI acquisitions.

The method includes, at 12, acquiring a full first data set of the first and second chemically shifted species at a first TE. The full first data set can be acquired in any suitable known manner for generating known Dixon images.

For the examples described herein, A FLASH (Fast Low Angle SHot) sequence was implemented on a 1.5T Siemens Sonata scanner (Siemens Medical Solutions, Erlangen Germany) to obtain all of the k-space data sets with TR=20ms, although any suitable known sequence can be used. The data sets each include a matrix having a number of lines, each line having a number of data points. For the example described herein, the full data set

included 192 lines each having 256 data points, although any suitable amount and configuration of data can be used.

The method also includes, at 14, acquiring a partial second data set, known as a keyhole, of the first and second chemically shifted species at a second TE, wherein the TE is set such that the second chemical species is 180° out of phase from its orientation to the first data set. The first and second TE can be any suitable lengths of time appropriate to creating known Dixon images for the first and second chemical species. For the water and fat example used herein, the first TE was set at 6.6 ms and the second TE was set at 8.8ms. The partial data set should include lines which sample the low spatial frequencies of k-space.

The method also includes, at 16, combining the first and second data sets to obtain an MRI image of the first chemical species and suppress the image of the second chemical species. The MRI image can be obtained in any suitable known manner upon combining the first and second data sets.

The steps of acquiring the first full data set and the partial second data set can include acquiring the data sets in two dimensions or three dimensions.

Multi-slice or 3 dimensional volume acquisitions can be obtained using this method.

The invention was substantiated and the minimum number of lines in the partial data set was determined by generating a series of 96 keyhole Dixon images of a water (saline) and fat (baby oil) phantom. The 96 keyhole Dixon images were created by combining partial k-space"Water+Fat"data sets, referred to as keyhole k-space"Water+Fat"data sets, to the full k-space "Water-Fat"data set.

Referring to Fig. 2, one example of this combination of data sets is shown generally at 18. In this example, the full"Water-Fat"data set 20 includes all 192 k-space lines of the"Water-Fat"data set acquired as described above. The keyhole"Water+Fat"data set 22 includes a partial k- space data set. The full"Water-Fat"k-space data set 20 and the partial k- space data set 22 are combined to create a keyhole Dixon data set shown at 24 for generating the MRI keyhole Dixon image.

The 96 different images were created, by successively increasing the amount of k-space data for the"Water+Fat"data set which was combined with the full"Water-Fat"data set. Successively larger keyhole k-space "Water+Fat"data sets were obtained by using more lines of k-space, 2 lines, 4 lines, 6 lines, etc. , up to all 192 lines. Centrically symmetric portions were created by using lines ky-i and ky1, then lines ky 2 though ky2, then lines ky 3 though ky3, etc. , up to 192 lines using lines ky-ge through ky96. (In this<BR> example there is no line kyo. ) Of the 96 different images generated, 95 were keyhole Dixon images obtained by combining a keyhole k-space data set with the full k-space data set. The image created by combining all 192 k-space lines of the"Water-Fat" data set with all 192 k-space lines of the"Water+Fat"data set was used as the reference 2-point Dixon image, which is known in the art. Although the acquisition and generation of 2-point Dixon images have been discussed, these should be considered examples and should not be limiting. The invention can also be used to reduce the acquisition times of any suitable multi-point Dixon images by acquiring a partial k-space data set in a similar manner.

A subset of keyhole Dixon phantom images and the reference 2-point Dixon image generated as described above are shown in Fig. 3. A 4-line keyhole Dixon image using a keyhole width of 4 lines of k-space data, shown at 30, includes a water phantom 32 and a fat phantom 34. A 96-line keyhole Dixon image shown at 36, includes a water phantom 38 and a fat phantom 40.

A 2-point Dixon image, shown at 42, includes a water phantom 44 and a fat phantom 46. Small keyhole widths (2-10 k-space lines) produced excessive blurring of the fat phantom 34, however, the visual quality of the phantom keyhole images 30,36 improved with increasing keyhole width and the blurring and edge enhancement of the fat phantom diminished.

The invention also includes the determination of the minimum amount of k-space data, such as the minimum number of k-space lines, needed in the partial second data set which when combined with the full first data set generates an image that is perceptually equivalent to a 2-point Dixon image.

A perceptual difference model (PDM) was used to analyze the differences

between the keyhole Dixon images and a 2-point Dixon image for determining the minimum amount of k-space data needed. The output of the model is a two-dimensional spatial map representing the likelihood that a human observer will perceive a difference between the two images at each pixel location. All values in the region of visual interest in the PDM error map are combined to give a scalar PDM error score. PDM errors were then analyzed to determine the minimum number of keyhole lines that provided an image that was perceptually equivalent to the reference 2-point Dixon image in terms of fat suppression, SNR, and artifacts.

The PDM models the functional anatomy of the human visual system including nonlinear gray level responses, the human contrast sensitivity function, visual channels, and various contrast mechanisms. The PDM provides an objective means to quantify image quality that more accurately reflects human perception than either contrast-to-noise ratio or mean- squared-error measurements. The model contains human visual system processing similar to known Image Difference Models and is designed to mimic the functional anatomy of the visual pathway. Grayscale nonlinearity of the retina, contrast sensitivity function, spatial frequency channels found in the visual cortex, and a measure of the contrast and visual detection are among the components of the human visual system that are modeled by the PDM.

The PDM has been shown to reliably represent computationally what a human observer detects as changes in an image. The accuracy of the PDM was verified by human observers using the water and fat phantom keyhole Dixon images which confirmed the correlation between PDM scoring and human observer ratings of image difference.

A second human observer experiment established the PDM error threshold below which humans can perceive no difference in two images.

Two observers were asked to classify keyhole Dixon images as being visually equivalent or visually different from a reference full Dixon image. Observers were presented with a two-panel display with the full Dixon magnitude image on the left and a randomly selected keyhole Dixon magnitude image on the right.

All of the 96 keyhole Dixon phantom images were presented and evaluated. Data were processed by calculating the percent of observer responses for which the images were classified as being the same as a function of the keyhole size. Using the sigmoidal function of the form shown in Equation 1 below, the data were fit to the model using nonlinear least squares regression.

P = 1/ (1 + exp [-A* (x-B) ] (1) In this equation, P is the probability of the keyhole Dixon image being classified as the same as the full 2-point Dixon image, A and B are model parameters, and x is the keyhole width.

The PDM error scores displayed a good linear correlation with the human observer visual ratings (R = 0.9083) of the phantom keyhole Dixon images as shown in Fig. 4. The nonlinear least squares regression to fit the sigmoidal model resulted in a threshold keyhole width of 96/192 k-space lines.

Therefore, the minimum amount of k-space data, also referred to as the minimum keyhole width, needed in the partial second data set to generates an MRI image that is perceptually equivalent to a 2-point Dixon image was found to be 96 lines of the full 192 lines, resulting in a 25% reduction in acquisition time over known 2-point Dixon imaging. This threshold keyhole Dixon image corresponded to a PDM score of 1.5 and was closest to having a 50% probability of being classified as the same as the full 2-point Dixon reference image. A PDM score of 1.5 was therefore set as the visual threshold for perceptual equivalence for clinical imaging applications.

It should be appreciated that any suitable known PDM can be used in determining the minimum amount of k-space data needed to generate a keyhole Dixon image which is perceptually equivalent to a multi-point Dixon image. Further, any other suitable known method of determining the minimum amount of k-space data needed to generate a keyhole Dixon image which is perceptually equivalent to a multi-point Dixon image can be used.

For example, Signal to Noise Ratios or Contrast to Noise Ratios can be used.

The user/operator can also use visual inspection of the image as it is generated. Using visual inspection, the user can simply look at the images as they are reconstructed as new lines of k-space are acquired. In this manner,

the user can terminate the reconstruction upon achieving any desired level of suppression.

A region-of-interest (ROI) analysis was also performed to generate individual PDM error scores for the water and fat phantoms. These were <BR> <BR> compared to the global PDM error. All three PDM error scores (i. e. , fat, water and global) were plotted as a function of keyhole width as shown in Fig. 5 to determine the threshold keyhole width needed to obtain perceptual equivalence with the full 2-point Dixon reference image. Results also provide information about the relative contributions of the water and fat phantoms to the global error and the perceptual equivalence threshold.

Referring to Fig. 5, plots of PDM error as a function of keyhole width show that phantom image error decreases with increasing keyhole width. The error from the global image shows a sudden drop in image error followed by a steady decline in PDM score (solid line). In the ROI analysis, the PDM errors associated with the water phantom (Fig. 5, small dashed line) remained below the equivalence threshold (PDM error = 1.5) at all keyhole widths. At the same time, the errors associated with the fat phantom ROI mirrored the global error results (large dashed line). The PDM confirmed visual observations that the keyhole artifacts are largely associated with the fat phantom.

In addition to the phantom keyhole Dixon images, clinical keyhole Dixon images from three different anatomical locations (knee, orbit, and abdomen) were analyzed with the PDM. The FLASH sequence was modified (TR=100ms, Number of averages = 4) to obtain clinically relevant 2-point Dixon images of a volunteer's knee, orbits, and abdominal region. These images were analyzed with the PDM to determine the amount of time savings possible using the keyhole Dixon fat suppression technique for clinical imaging.

Again, 96 keyhole Dixon images were compared to the full Dixon image, and PDM analysis was performed. Error scores were calculated in a manually selected region of interest encompassing the relevant clinical structures for each application. Error scores were plotted versus keyhole size. The visual threshold, as determined from the human observers, was applied to the data showing the possible time-savings with the technique

describe above. The PDM score of 1.5 was set as the visual threshold for perceptual equivalence as used above.

For a knee, the threshold keyhole width was found to be 88 lines of the full 192 lines, resulting in a 27% reduction in acquisition time over a 2-point Dixon image. For an orbit, the threshold keyhole width was found to be 44 lines of the full 192 lines, resulting in a 38% reduction in acquisition time over a 2-point Dixon image. For abdominal images, the threshold keyhole width was found to be 50 lines of the full 192 lines, resulting in a 37% reduction in acquisition time over a 2-point Dixon image.

Therefore, the step of acquiring the partial second data set at 14 described above can include acquiring less than approximately 50% of the lines acquired in the full data set. It can include acquiring less than approximately 40% of the lines acquired while acquiring the full data set. It can include acquiring less than approximately 30% of the lines acquired while acquiring the full data set. Further, the total acquisition time can be reduced by approximately 25% to 40% over a two point Dixon technique in two dimensions.

The threshold keyhole widths for the clinical images were smaller than that that required for the phantom images, perhaps due to the presence of very high frequency fat information hidden within the surrounding anatomy.

Alternatively, the reduction in threshold may be related to the sensitivity of the PDM to small edge effects either included in or excluded from the particular ROI.

Nevertheless, the PDM results show that the threshold keyhole width among the three clinical applications is fairly consistent, resulting in a 25-40% reduction in acquisition time. With the exception of the knee images, PDM scores approach the visual threshold slowly, suggesting that there is only a small change in image error on either side of the limit. Therefore, the PDM visual threshold is not necessarily a hard limit, but allows some flexibility in acquiring keyhole Dixon images. A less complicated image could be acquired with less imaging time by using a keyhole smaller than the threshold, and a more sensitive application may be performed using a more conservative, slower acquisition. A nominal keyhole width of 33% of the full acquisition

providing a 33% reduction in acquisition time may provide a useful compromise between speed and image quality for most clinical applications.

The keyhole acquisition strategy and reconstruction technique can affect the quality of the keyhole images. Therefore, applying other known keyhole reconstruction techniques to the keyhole Dixon method, besides concentrically symmetric techniques, in the step of acquiring the partial second data set at 14 may reduce the level of these artifacts and allow the threshold keyhole width, and thus acquisition time, to be reduced even further.

Examples of other suitable keyhole reconstruction techniques, which should not be considered as limiting, can include CURE, generalized series reconstruction, hybrid techniques, and non-cartesian techniques including but not limited to as spiral and radial techniques, although any other suitable known keyhole reconstruction techniques can be used.

The keyhole Dixon method provides the advantage of reducing the acquisition time of the 2-point Dixon method by 25-40% with no perceptual difference in image quality. The reduction in acquisition time improves the temporal resolution obtainable with the Dixon fat suppression methods making them more useful for rapid imaging applications, such as cardiac imaging and interventional MRI.

The implementation of the keyhole Dixon method offers opportunities for improved fat suppression for both low and high field scanner systems. In addition, the keyhole Dixon method can be implemented in much shorter acquisition times than inversion recovery sequences where-100-300 ms inversion times and long TR's (> 1000 ms) can be expected.

The invention has been described with reference to preferred embodiments. Obviously, modifications and alterations will occur to others upon reading and understanding the preceding specification. It is intended that the invention be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.