WO2010116124A1

WO2010116124A1 - Diffusion-weighted nuclear magnetic resonance imaging

Info

Publication number: WO2010116124A1
Application number: PCT/GB2010/000647
Authority: WO
Inventors: Matthew David Blackledge; David John Collins; Martin Osmund Leach
Original assignee: The Institute Of Cancer Research, Royal Cancer Hospital; Royal Marsden Nhs Foundation Trust
Priority date: 2009-04-08
Filing date: 2010-03-31
Publication date: 2010-10-14
Also published as: GB0906149D0

Abstract

A method is provided for producing a diffusion-weighted magnetic resonance image of a region of interest. The method includes the steps of: (a) acquiring at least two images of the region of interest at respective and different b-values; (b) determining a map of the apparent diffusion coefficient in the region of interest from the acquired images, and (c) calculating, from the map determined at step (b), an image of the region of interest at a b-value.

Description

DIFFUSION-WEIGHTED NUCLEAR MAGNETIC RESONANCE IMAGING

The present invention relates to nuclear magnetic resonance (NMR) imaging methods and systems, and particularly to NMR diffusion weighted imaging.

NMR imaging allows anatomical and physiological features of living human and animal bodies to be observed.

Typical NMR imaging of a region of interest involves performing a sequence of NMR measurement cycles. The received NMR signals are then processed to reconstruct the magnetic resonance image.

During each measurement cycle, signal localisation is obtained by controlling the strength of magnetic fields which have the same direction as the main polarizing field, but which impose gradients along the x, y and z directions.

Recently, there has been interest in the magnetic resonance technique known as diffusion weighted imaging (DWI) for the detection of cancer where the image signal is dependent upon the diffusivity of the tissue. In DWI, pulse sequences are employed which contain magnetic field gradients known as diffusion gradients that sensitize the magnetic resonance signal to spin motion. An example pulse sequence contains temporally separated first and second diffusion gradient lobes of equal size.

In a DWI pulse sequence, the detected magnetic resonance signal intensity decreases in relation to the speed of water diffusion in a given volume of tissue. The first moment of the diffusion gradient, known as the "b-value", summarises the diffusion sensitising sequence parameters. The b-value may be adjusted by varying the area of the two lobes of the diffusion magnetic field gradient, or by varying the time interval between them. The magnetic resonance signal intensity, S (b) , at the center of the echo using a spin-echo diffusion-weighted pulse sequence can be related to the b-value by the Stejskal-Tanner equation:

S(b) = S(O)e^~b-^Do

where b is the b-value, S(O) is the MR signal magnitude with no diffusion weighting (i.e. b = 0) , and Do is the diffusion coefficient (in mrtiVs) of the tissue from which the signal derives. Or, more generally:

S(b) = S(bo)e^~Ab-^Do

where S (bo) is the MR signal magnitude at a reference b-value, bo (which is not necessarily b = 0), and Δb = b - bo. The equation assumes mono-exponential decay.

DWI is especially useful in cancer detection, as tumour tissue has been shown to have low diffusivity relative to other tissues and so appears highly intense relative to background in DWI images. Often it is desirable to obtain images at high b-values. However, while noise levels remain generally constant at different b-values, signal levels attenuate as b- values increase. Thus at high b-values, the low signal-to- noise ratio (SNR) can cause significant deterioration in image quality.

It would be desirable to be able to produce high b-value DWI images with improved SNRs .

Accordingly, in a first aspect, the present invention provides a method for producing a diffusion-weighted magnetic resonance image of a region of interest, the method including the steps of:

(a) acquiring (e.g. with a magnetic resonance imaging apparatus) at least two images of the region of interest at respective and different b-values;

(b) determining a map of the apparent diffusion coefficient (ADC) in the region of interest from the acquired images; and

(c) calculating, from the map determined at step (b) , an image of the region of interest at a different b-value . Typically, at step (c) the image is calculated at a first b- value which is higher than the b-values used at step (a) .

Referring to the Stejskal-Tanner equation, for a given acquired image, S(b) and b are known. Thus by acquiring two images, it is possible to estimate Do (i.e. as the ADC), which provides the map at step (b) . From the map, a value for S (b) corresponding to the different b-value of step (c) can then be obtained from the equation (bearing in mind that S(O) or S (bo) is also obtained from one of the acquired images) .

In step (c), the calculated image typically has a b-value which is significantly higher than the b-values used to acquire the images at step (a) . However, advantageously, in this case, the calculated image does not suffer from the same drop in SNR associated with images acquired at the same b- value using magnetic resonance imaging apparatuses. This is because while, in a calculated image, the signal attenuates with increasing b-value, the noise attenuates with increasing b-value as well, leading to a higher SNR.

Typically, each of the acquired images is acquired by- averaging a plurality of images at the respective b-value. Alternatively or additionally, in step (a) more than two images can be acquired, each at a respective and different fa- value, and in step (b) the ADC map can be determined from the more than two images, e.g. by regression analysis or nonlinear fitting. Both techniques, and particularly the image averaging, can improve the accuracy of the ADC map, and thereby improve the quality of the calculated image. In general, the two acquired images can have any two different b-values. However, preferably, one of the images is acquired with a zero b-value (i.e. a zero diffusion gradient) . A zero b-value image (which is typically a T2 weighted image) generally provides more detailed anatomical information than non-zero b-value images due to signal attenuation at higher b- values . Acquiring a zero b-value also helps to reduce noise in the ADC determination (assuming mono-exponential signal attenuation) .

The method may further include a preliminary step of determining the desirable difference between the b-values for the acquisition of the acquired images on the basis of a provided or estimated value of the true diffusion coefficient in a portion of the region of interest. In particular, if the true diffusion coefficient of a tumour in the region of interest is known or can be estimated, the difference may be determined in order to reduce error in the ADC map determination and thereby improve the visibility of the tumour in the calculated image. Preferably, the images in step (a) are acquired with the number of signal averages (NSA) on a 1:3 ratio and the difference between the b-values (i.e. Δb) is about 1.25/DQ, where D₀ is the true diffusion coefficient of the tissue being imaged. However, if a different NSA is used, the preferred difference changes, i.e. Δb is preferably about 1.11/Do for an NSA on a 1:1 ratio and is preferably about 1.34/Do for an NSA on a 1:5 ratio (the preferred Δb values assume that the noise is Gaussian distributed and independent of b-value, and the effects of T2 decay can be ignored - see Jones, D. K., Horsfield, M. A., Simmons, A.: Optimal Strategies for Measuring Diffusion in Anisotropic Systems by- Magnetic Resonance Imaging, Magn. Reson. Med. 42:515-525 (1999) ) . A second aspect of the invention provides a method of displaying 3D diffusion-weighted magnetic resonance image data, the method including the steps of: performing the method of the first aspect (optionally including any one or any combination of the optional features of the method of the first aspect) for each of a plurality of parallel slices through a volume of interest to obtain for each slice a corresponding calculated image, and displaying the calculated images as a projected volumetric image, e.g. as a maximum intensity projection of the volume of interest .

A further aspect of the invention provides a computer system operatively configured to perform the method of the first or second aspect. The system may have one or more optional features corresponding to any one or combination of the optional features of the method of the first or second aspect.

For example, a computer system for producing a diffusion- weighted magnetic resonance image of a region of interest may include : memory for storing at least two images of the region of interest acquired at respective and different b-values; and a processor or processors for determining a map of the apparent diffusion coefficient in the region of interest from the acquired images; and calculating, from the map thus- determined, an image of the region of interest at a different b-value .

The computer system may include a display for displaying the calculated image and/or (if the system is operatively configured to perform the method of the second aspect) the projected volumetric image. A further aspect of the invention provides a magnetic resonance imaging apparatus having a computer system of the previous aspect.

Yet further aspects of the invention provide a computer program for performing the method of the first or second aspect, and a computer program product carrying a computer program for performing the method of the first or second aspect.

Embodiments of the invention will now be described by way of example with reference to the accompanying drawings in which:

Figure 1 is a diagram showing an overview of a DWI calculation;

Figure 2 shows a plot of variance of image noise against fa- value for acquired and calculated DWI images;

Figure 3 shows at left hand side examples of conventionally acquired b = 1400 s/mm² images and at right hand side examples of calculated b = 1400 s/mm² images for two contiguous slices of an iliac bone containing lymphoma deposits;

Figure 4 shows (a) an acquired b = 1000 s/mm² conventional image of a prostate with a confirmed tumour, (b) a corresponding b = 2000 s/mm² calculated DWI image, and (c) the same image as (b) but with respective perimeters drawn around the diseased node and the peripheral zone of the prostate; and

Figure 5 shows (a) a conventionally acquired b = 1000 s/mm² MIP of a patent with multiple skeletal metastases (arrowed) , (b) a calculated b = 1700 s/mm² MIP of the same patient, and (c) a segmented cDWI volume based on sample statistics of a single lesion (arrowed) .

Studies were performed which compared the SNR and image contrast of calculated and acquired images. The studies showed that for b-values within a range typically applied in cancer detection, calculated DWI (cDWI) is superior to conventional DWI. Furthermore, the studies showed that when cDWI images were extrapolated to b-values not readily applicable using clinical scanners, lesions that had not previously been observed became visible.

A method of producing cDWI is demonstrated schematically in Figure 1. Firstly, images are acquired at a number of b- values (2 or more) for the region of interest, including preferably an image at b = 0 s/mm². These are then used to calculate an ADC map for the region. The map is used along with the image obtained with b = 0 s/mm² to produce calculated images at any b-value based on a model (e.g. the Stejskal- Tanner equation) for signal S (b) .

If a b = 0 s/mm² image is not used in the ADC calculation, then another b-value will suffice. However, -Δb.Do then becomes the exponent in the Stejskal-Tanner equation rather than -b.D₀, Δb being the difference between the acquired and sought after fa- value. Further, S(Q) in the equation becomes the value for the MR signal magnitude at that other b-value.

Preferably, the noise in the ADC maps is reduced such that there is minimal propagation of noise through to the cDWI images. For example if the noise is Gaussian distributed, and the effects of T2 decay are ignored, then two images can be acquired at b = 0 s/mm² and b_opt = 1.25/Do, where D₀ is the true diffusion coeffcient of the tissue being imaged. However, optimum imaging parameters may not be achievable throughout the entire slice as diffusion coeffcients are generally inhomogeneous within tumours and vary between different tissues. Nonetheless, optimisation can be performed for a tissue of interest (e.g. a cancer) . A number of experiments were performed on a 1.5T Siemens scanner (Avanto) to compare the noise between cDWI and conventionally acquired images over a range of b-values. A restriction was applied such that the acquisition time of both techniques was the same, limiting the number of signal averages (NSA) that could be acquired for cDWI, which requires an extra b = 0 s/mm² measurement for the estimation of ADC values. Images of a uniform, cylindrical CuSO₄ phantom were acquired for b-values in the range 0,100,...,1500 s/mm², once using NSA = 1:3 for b = 0 s/mm² and each of the other b-values (from which cDWI data could be calculated) and then again using NSA = 4 for all b-values (for conventional DWI data) . This experiment was then repeated using the same imaging parameters such that noise statistics (the standard deviation of pixel values within a region of interest (ROI) ) could be derived from the difference images. Noise statistics were calculated for each b-value using NSA = 4 images and then data from all b-values were used to calculate a mean ADC for the phantom within the same ROI. From this an optimum b-value for calculating the ADC based on two point measurements was derived (b_opt) and this was used along with the b = 0 s/mm² image to calculate cDWI images using the NSA = 3 images for the same range of b-values. The variance of pixel values for the cDWI images was then plotted against b-value on the same axis as for conventional DWI to observe the b-value dependence of the noise for both methods. All other scanning parameters were kept constant: field of view (FOV) = 300 x 300 mm², slice thickness (ST) = 5.0 mm, repetition time (TR) = 1100 ms, echo time (TE) = 228 ms and bandwidth in the read direction (BW) = 1812 Hz/px.

Using the phantom data acquired with NSA = 4, a mean ADC of 2.05 x 10^~3 mm²/s was calculated for the imaged region, suggesting that the optimum single b-value for measuring ADC was 1.25/2.05 x 10^"3 = 609.8 mm²/s. The acquired image that was closest to this for the NSA = 3 data was 600 s/iriiα² which was used along with the b = 0 s/mm² acquired image to calculate the ADC maps used for cDWI . Figure 2 shows a plot of the variance of image noise against b-value. For the acquired image data (solid line) noise was roughly constant across all b-values with a mean variance of 6.08. However, the image noise of the calculated images (points marked "x") varied across the range of b-values, being larger than that for the acquired images when b < 700 s/mm² and smaller when b > 700 s/mm². This reduction in image noise for the calculated images allows high SNRs to be achieved.

The observed change in variance of image noise for cDWI is consistent with theory. In particular it is possible to derive the following expression for the variance in noise of cDWI images (σ² _c) :

where σ² ₀ is the variance in noise of a conventionally acquired b = 0 s/mm² image and x = (calculated b-value/b_opt) • The derivation assumes that:

1. The diffusion coefficient in the tissue of interest (tumour) is homogeneous (D₀) .

2. The acquisition protocol is optimised for ADC determination, i.e. 2 b-value images are acquired, one at b = 0 s/mm² and another at b_opt = 1.25/Do, with NSA for each image being on a 1:3 ratio.

3. The image noise follows a Gaussian distribution such that the propagation of errors formula may be applied:

4. Image noise is independent of b-value.

The curve of the above expression for σ² _c is plotted on Figure 2 as a dotted line, and can be seen to match well with the actual variances of the calculated images.

Next, the SNR of calculated and acquired b = 1400 s/mm² images from a patient with lymphoma deposits in the iliac bone were compared. The patient was scanned axially using b-values of 50, 100, 250 and 750 s/mm² applied in 3 orthogonal directions for ADG calculations and then again with a single b-value of 1400 s/mm². Imaging was performed utilizing the double-spin echo technique to minimise eddy current induced distortions, and with the following acquisition parameters: slice thickness of 5 mm, a TR = 3000ms, TE = 80ms and 4 averages. ADC values were determined from the lower b-value range using semi-log plots and these values were incorporated into the mono- exponential model of the Stejskal-Tanner equation to calculate b = 1400 s/mm² images.

The average SNR of the calculated and acquired images was determined for 3 foci over 3 contiguous slices by dividing the mean signal within regions of interest (ROIs) around the lesions by the standard deviation of background noise within an ROI drawn just adjacent to the lesions in a region where no obvious signal was observed. One standard deviation in the pixel values within the lesion ROIs was taken to be the uncertainty in measurement.

Figure 3 shows at left hand side examples of the conventionally acquired b = 1400 s/mm² images and at right hand side examples of the b = 1400 s/mm² cDWI images for two contiguous slices. The three lower right arrows in the top left hand image indicate positions of the lymphoma deposits and the upper left arrow in the top left hand image indicates the position of the region where no obvious signal was observed. There is an apparent improvement in the image contrast between the lymphoma deposits and the background tissue for the calculated images. This is born out by the average SNRs, which were 27.6+7.3 for the calculated images and 19.7+7.9 for the conventionally acquired images.

Further experiments were performed with the 1.5T Siemens scanner (Avanto) to compare the image contrast between cDWI and conventionally acquired images .

Thus, to assess the viability of extrapolating calculated images to a high b-value of 2000 s/mna², the contrast between the mean signal of a histologically confirmed primary prostate tumour and the mean signal of the peripheral zone of the prostate was calculated for a b = 2000 s/mm² cDWI image and compared to a conventionally acquired b = 1000 s/mm² image (b = 1000 s/mm² being at the high end of conventionally applicable b-values) . The definition of contrast used in this context was the Michelson contrast defined as (S_tUmour - S_prostate) /Stumour- The scanning parameters for both imaging methods were TE = 75 ms, TR = 3700 ms, ST = 6.0 mm and FOV = 320 x 270 mm².

Figure 4 shows (a) the acquired b = 1000 s/mm² conventional image and (b) the b = 2000 s/mm² cDWI image. In the acquired image, it is very difficult to distinguish the diseased tissue (arrowed) from healthy tissue. In contrast, in the calculated image, the diseased node stands out much more clearly from the background tissue due to the higher contrast between these tissues. Figure 4 (c) shows again the calculated b = 2000 s/mm² cDWI image, but with respective perimeters drawn around the tumour and the peripheral zone of the prostate. The Michelson contrast between the mean signal within the lesion and the mean signal within the peripheral zone was -0.10 and 0.54 for the acquired and calculated images respectively. Evidently, the lesion stands out from the image background much more clearly using the high b-value available to cDWI • Next, cDWI images were incorporated into a 3D maximum intensity projection (MIP) tool for segmentation of lesions throughout the body. This tool can be used with the following procedure:

1. Firstly the radiologist or clinician loads in all ADC maps and b = 0 s/mm² images from which a MIP volume is calculated.

2. The b-value of the computed MIP is varied until the image contrast is adequate to see any lesions. On a 2.6 GHz dual core processor and for a typical whole body data set, the new image can take approximately 1 second to compute for each change in b-value .

3. To improve visualisation of tumours, a voxel within a lesion of choice can be selected and statistics from within a 3x3x3 grid (the size may be altered if necessary) centred at that location are computed for both the ADC' and b = 0 s/mm² data. The entire patient volume can then be segmented according to a threshold which is set e.g. to within two standard deviations of the mean calculated for the grid.

The MIP tool was tested on a patient who had multiple metastases throughout the axial skeleton and spine, following this procedure. Figure 5 shows an example result for the patient, demonstrating the differences between conventionally acquired (a) and calculated (b) high b-value volumes and also the effect (c) of thresholding the data based on the ADC and b = 0 s/mm² statistics of a lesion of interest. Figure 5 shows that, relative to the conventionally acquired images, the calculated higher b-value image has a reduced signal from the background tissue and hence provides improved visual delineation of the metastases. Further, the figure shows that thresholding can additionally increase the contrast between metastases and background tissue. For example, in Figure 5(b) much of the soft tissue anatomy has been removed such as the spleen, kidneys and spinal cord, making tumour visualisation easier.

In conclusion, noise is reduced at high b-values in cDWI, thus improving the SNR. Further, the freedom to freely generate images at any b-value can improve the contrast observed between diseased and healthy tissues. Artefacts such as eddy- current distortions, ghosting and susceptibility induced geometric distortions typically plague conventionally acquired high b-value images. The effects of these artefacts are usually amplified at higher b-values (see LeBihan, D. , Poupon, C. Amadon, A. Lethimonnier, F.: Artifacts and Pitfalls in Diffusion MRI, J. Magn. Reson. Imaging 24:478-488 (2006)) . Thus provided that the ADC calculations are performed using images acquired at lower b-values and preferably with minimal image distortions, the use of cDWI should improve image integrity.

While the invention has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. For example, any diffusion model, such as bi-exponential decay (which allows intra-voxel incoherent motion to be modelled) , may be incorporated into cDWI (e.g. by appropriate modification of the Stejskal-Tanner equation) if the source diffusion weighted images provide sufficient data to support its use. In particular, a bi-exponential decay model would require acquired images at four or more different b-values in order to determine the ADC map. Further, cDWI is not limited to cancer detection or to the visualisation of the specific regions imaged in the examples. Further, cDWI is not limited to any specific acquisition protocol or geometry, and can be applied to single slices, multiple slices, non-contiguous slices as well as full 3D acquisitions with any combination of sampling. Accordingly, the exemplary embodiments of the invention set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the invention.

All references mentioned herein are incorporated by reference.

Claims

1. A method for producing a diffusion-weighted magnetic resonance image of a region of interest, the method including the steps of:

(a) acquiring at least two images of the region of interest at respective and different b-values;

(b) determining a map of the apparent diffusion coefficient in the region of interest from the acquired images; and

(c) calculating, from the map determined at step (b) , an image of the region of interest at a b-value.

2. A method according to claim 1, wherein at step (c) the image is calculated at a first b-value which is higher than the b-values used at step (a) .

3. A method according to claim 1 or 2, wherein at step (a) one of the acquired images is acquired at a zero b-value.

4. A method according to any one of the previous claims, further including a preliminary step of determining the difference between the b-values for the acquisition of the acquired images on the basis of a provided or estimated value of the diffusion coefficient in a portion of the region of interest .

5. A method of displaying 3D diffusion-weighted magnetic resonance image data, the method including the steps of: performing the method of any one of the previous claims for each of a plurality of regions of interest which are respective parallel slices through a volume of interest, and thereby obtaining for each slice a corresponding calculated image; and displaying the calculated images as a projected volumetric image .

6. A computer system operatively configured to perform the method of any one of the previous claims.

7. A magnetic resonance imaging apparatus having the computer system of claim 6.

8. A computer program for performing the method of any one of the claims 1 to 5.

9. A computer program product carrying the program of claim