Methods of spatial normalization of positron emission tomography images

Abstract

An adaptive template image for registering a PET or a SPECT image includes a template image model including variability of values for each voxel in a template image according to one or more control parameters.

Claims

1. An adaptive template image system to perform spatial normalization and registration of an acquired image using a single image registration procedure, the system comprising: memory configured to store and provide an adaptive template image including a template image model including an adjustable variability of values for one or more voxels in the adaptive template image, the variability of values, wherein the adaptive template image is in a reference space and is used for registering the acquired image in the reference space during the single image registration procedure; and at least one processor configured to at least: initiate a first registration of the acquired image into the reference space; spatially normalize the acquired image to the adaptive template image in the first registration by: (i) adjusting the template image model using a first control parameter, the adjusting to include adjusting among the variability of values for one or more voxels in the adaptive template image based on the first control parameter, (ii) rendering the adaptive template image from the adjusted template image model, (iii) performing a spatial normalization operation by adjusting the acquired image using spatial transformation parameters, (iv) evaluating the adjusted adaptive template image and the adjusted acquired imaging to determine a convergence of the adjusted adaptive template image to the adjusted acquired image, wherein convergence is determined by comparing voxels of the adjusted adaptive template image to corresponding voxels of the adjusted acquired image, and wherein the image registration between the adjusted adaptive template image and the adjusted acquired image is not performed until the adjusted adaptive template image converges to fit the adjusted acquired image, and (v) when the adjusted adaptive template image does not converge to fit the adjusted acquired image, adaptively altering both the adjusted adaptive template image and the adjusted acquired image by selecting an updated control parameter to render the altered adjusted adaptive template image and updated spatial transformation parameters to spatially normalize the altered adjusted acquired image for re-evaluation at (iv) within the single image registration procedure and without performing an image registration until an altered adjusted adaptive template image converges to fit the altered adjusted acquired image; and when the altered adjusted adaptive template image converges to fit the altered adjusted acquired image, register the altered adjusted acquired image in the reference space to complete the first registration, wherein said spatially normalizing further comprises performing linear and non-linear transformations on the acquired image to find a match between the template image model and the acquired image, and wherein said linear transformation comprising at least one of translation, rotation, and scaling, and wherein the values of the voxels are defined by an equation:
I.sup.i.sub.template=I.sup.i.sub.0+I.sup.i.sub.slopex wherein I.sup.i.sub.template is the value of the voxel i in the adaptive template image, I.sup.i.sub.0 represents a value corresponding to a normal level of uptake of an imaging agent within that voxel, I.sup.i.sub.slope corresponds to a maximum additional amount of uptake of the imaging agent corresponding to a high level uptake of the imaging agent in that voxel with respect to a control parameter x.

2. The adaptive template image system of claim 1, wherein the acquired image includes at least one of a positron emission tomography (PET) image or a single photon emission tomography (SPECT) image.

3. The adaptive template image system of claim 1, wherein the variability of values for each voxel ranges from a value corresponding to a normal level of uptake of an imaging agent to an abnormal level of uptake of the imaging agent.

4. The adaptive template image system of claim 3, wherein the level of uptake is measured in gray matter of a brain.

5. The adaptive template image system of claim 3, wherein the imaging agent includes a dopamine transport agent.

6. The adaptive template image system of claim 5, wherein the imaging agent includes DaTSCAN® and the level of uptake is determined for a striatum.

7. The adaptive template image system of claim 1, wherein the variability of values for each voxel ranges from a value corresponding to a normal level of amyloid in grey matter to a raised amyloid level in grey matter.

8. The adaptive template image system of claim 1, wherein at least one of the first control parameter or the updated control parameter is selected as part of a spatial normalization optimization process.

9. The adaptive template image system of claim 1, wherein I.sub.0 and I.sub.slope are determined using regression analysis of multiple images spanning different levels of disease state.

10. A method of performing spatial normalization and registration of an acquired image using a single image registration procedure to an adaptive template image including a template image model including an adjustable variability of values for one or more voxels in the adaptive template image, wherein the adaptive template image is in a reference space and is used for registering the acquired image in the reference space during the single image registration procedure, the method comprising: initiating a first registration of the acquired image into the reference space; spatially normalizing the acquired image to the adaptive template image in the first registration by: (i) adjusting the template image model using a first control parameter, the adjusting to include adjusting among the variability of values for one or more voxels in the adaptive template image based on the first control parameter, (ii) rendering the adaptive template image from the adjusted template image model, (iii) performing a spatial normalization operation by adjusting the acquired image using spatial transformation parameters, (iv) evaluating the adjusted adaptive template image and the adjusted acquired imaging to determine a convergence of the adjusted adaptive template image to the adjusted acquired image, wherein convergence is determined by comparing voxels of the adjusted adaptive template image to corresponding voxels of the adjusted acquired image, and wherein the image registration between the adjusted adaptive template image and the adjusted acquired image is not performed until the adjusted adaptive template image converges to fit the adjusted acquired image, and (v) when the adjusted adaptive template image does not converge to fit the adjusted acquired image, adaptively altering-both the adjusted adaptive template image and the adjusted acquired image by selecting an updated control parameter to render the altered adjusted adaptive template image and updated spatial transformation parameters to spatially normalize the altered adjusted acquired image for re-evaluation at (iv) within the single image registration procedure and without performing an image registration until an altered adjusted adaptive template image converges to fit the altered adjusted acquired image; and when the altered adjusted adaptive template image converges to fit the altered adjusted acquired image, registering the altered adjusted acquired image in the reference space to complete the first registration, wherein said spatially normalizing further comprises performing linear and non-linear transformations on the acquired image to find a match between the template image model and the acquired image, and wherein said linear transformation comprising at least one of translation, rotation, and scaling, and wherein the values of the voxels are defined by an equation:
I.sup.i.sub.template=I.sup.i.sub.0+I.sup.i.sub.slopex wherein I.sup.i.sub.template is the value of the voxel i in the adaptive template image, I.sup.i.sub.0 represents a value corresponding to a normal level of uptake of an imaging agent within that voxel, I.sup.i.sub.slope corresponds to a maximum additional amount of uptake of the imaging agent corresponding to a high level uptake of the imaging agent in that voxel with respect to a control parameter x.

11. The method of claim 10, wherein the acquired image includes at least one of a positron emission tomography (PET) image or a single photon emission tomography (SPECT) image.

12. The method of claim 10, wherein the variability of values for each voxel ranges from a value corresponding to a normal level of uptake of an imaging agent to an abnormal level of uptake of the imaging agent.

13. The method of claim 12, wherein the imaging agent includes a dopamine transport agent.

14. The method of claim 10, wherein the variability of values for each voxel ranges from a value corresponding to a normal level of amyloid in grey matter to a raised amyloid level in grey matter.

15. The method of claim 10, wherein at least one of the first control parameter or the updated control parameter is selected as part of a spatial normalization optimization process.

16. The method of claim 10, wherein I.sub.0 and I.sub.slope are determined using regression analysis of multiple images spanning different levels of disease state.

17. A non-transitory computer readable storage medium comprising computer readable program code including instructions for performing spatial normalization and registration of an acquired image using a single image registration procedure to an adaptive template image including a template image model including an adjustable variability of values for one or more voxels in the adaptive template image, the variability of values adjustable using a control parameter, wherein the adaptive template image is in a reference space and is used for registering the acquired image in the reference space during the single image registration procedure, wherein execution of the computer readable program code causes at least one processor to at least: initiate a first registration of the acquired image into the reference space; spatially normalize the acquired image to the adaptive template image in the first registration by: (i) adjusting the template image model using a first control parameter, the adjusting to include adjusting among the variability of values for one or more voxels in the adaptive template image based on the first control parameter, (ii) rendering the adaptive template image from the adjusted template image model, (iii) performing a spatial normalization operation by adjusting the acquired image using spatial transformation parameters, (iv) evaluating the adjusted adaptive template image and the adjusted acquired imaging to determine a convergence of the adjusted adaptive template image to the adjusted acquired image, wherein convergence is determined by comparing voxels of the adjusted adaptive template image to corresponding voxels of the adjusted acquired image, and wherein the image registration between the adjusted adaptive template image and the adjusted acquired image is not performed until the adjusted adaptive template image converges to fit the adjusted acquired image, and (v) when the adjusted adaptive template image does not converge to fit the adjusted acquired image, adaptively altering-both the adjusted adaptive ktemplate image and the adjusted acquired image by selecting an updated control parameter to render the altered adjusted adaptive template image and updated spatial transformation parameters to spatially normalize the altered adjusted acquired image for re-evaluation at (iv) within the single image registration procedure and without performing an image registration until an altered adjusted adaptive template image converges to fit the altered adjusted acquired image; and when the altered adjusted adaptive template image converges to fit the altered adjusted acquired image, register the altered adjusted acquired image in the reference space to complete the first registration, wherein said spatially normalizing further comprises performing linear and non-linear transformations on the acquired image to find a match between the template image model and the acquired image, and wherein said linear transformation comprising at least one of translation, rotation, and scaling, and wherein the values of the voxels are defined by an equation:
I.sup.i.sub.template=I.sup.i.sub.0+I.sup.i.sub.slopex wherein I.sup.i.sub.template is the value of the voxel i in the adaptive template image, I.sup.i.sub.0 represents a value corresponding to a normal level of uptake of an imaging agent within that voxel, I.sup.i.sub.slope corresponds to a maximum additional amount of uptake of the imaging agent corresponding to a high level uptake of the imaging agent in that voxel with respect to a control parameter x.

18. The computer readable storage medium of claim 17, wherein the acquired image includes at least one of a positron emission tomography (PET) image or a single photon emission tomography (SPECT) image.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 provides an illustration of typical Flutemetamol uptake patterns in an amyloid positive (Aβ+) and amyloid negative (Aβ−) scan.

(2) FIG. 2 depicts an outline of the complete image registration procedure including the novel extension.

(3) FIG. 3 depicts an illustration of the calculated mean SUVR values of the composite region COM.sub.SUVR for all subjects in two different groups of Aβ− and Aβ+ scans.

(4) FIG. 4 depicts the resulting slope and intercept images according to one embodiment of the present invention.

(5) FIG. 5 shows template images of the present invention.

(6) FIG. 6 depicts a system for performing the instant invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

(7) The present invention provides a spatial normalization method when using imaging agents. While a specific embodiment for scans with the [.sup.18F]-Flutemetamol amyloid imaging agent, which only depends on the Flutemetamol scan itself, are described in detail, the present invention is also applicable to other imaging agents used for other imaging applications. For example, the present invention may also be employed for DaTSCAN imaging where low uptake in the brain is indicative of disease.

(8) The uptake pattern in Flutemetamol scans can differ much between normal subjects (Aβ−) and subjects with high amyloid in gray matter (Aβ+). Therefore, the spatial normalization method has been extended with an adaptive template mechanism, which during the image registration process alters the template image along with the spatial transformation parameters to make it more similar to the scan being registered. By using this extension, the method can accurately spatially normalize scans ranging over the whole amyloid scale from Aβ− to very high Aβ+.

(9) The present invention solves this by creating a model over how the typical uptake pattern in the whole image varies along the scale going from Aβ− to Aβ+. In the model presented hereinbelow, a one parameter linear regression model for each voxel in the image is used to express the dependence, but generally a more advanced model expressing uptake pattern dependence on multiple parameters, is also contemplated by the present invention. The model is built once for all in a separate pre-processing step and is then fixed as an internal part of the method. Although the method described below has been developed for Flutemetamol it should be noted that other PET amyloid tracers exhibit exactly the same characteristic. Moreover, the method described is general and there is nothing limiting the method of the present invention to PET amyloid scans only. As long as a model over the typical variation modes of the images being registered can be created, the adaptive template method has potential to be used for other imaging agents as well as for both PET and SPECT applications. While the examples provided hereinbelow are directed PET applications, the present invention also applies to SPECT applications.

(10) As noted hereinabove, the present invention may alter one or more of the parameters for the image, the image template, or both. For example, parameters for spatial transformation may include translation along each axis (providing three possible parameters), scaling along each axis (providing another three possible parameters), and rotation about each axis (providing an additional three parameters), for a total of up to nine parameters which may be altered by the present invention. Additionally, parameters which account for shape differences between the image and the adaptive template provide additional parameters which may be altered by the present invention. An illustration of typical Flutemetamol uptake patterns in an Aβ+ and Aβ− scan is shown in FIG. 1. It can be noted that the Aβ+ have much more activity in cortical gray matter and even though white matter activity is about the same in the two cases the relative activity pattern between white and gray matter is to some extent reversed. That is from having highest activity in white matter in Aβ− to the opposite with highest, or at least as high, activity in gray matter in Aβ+.

(11) Spatial normalization is the process of transforming a scan from patient space to a standard space thus allowing for comparison of data from different subjects. Typically, this is performed through an iterative process where the patient scan is compared with a template image and where it is geometrically transformed to be as similar to the template as possible. However, by observing the patterns above it can be seen that there is no obvious way of selecting a similarity metric for a spatial normalization method which would be able to perform accurate and robust registrations of both of these types of images, by comparing them to a common template image.

(12) Therefore, the present invention provides an extension to be used together with a standard image registration method, where the number of possible template images is infinite. Basically, no definite selection of one specific template image occurs. Instead the intensities in the template are continuously altered during the registration, to make the template converge along with the spatial transformation parameters, to optimally fit the image being registered. An outline of the complete image registration procedure including the novel extension is shown in FIG. 2.

(13) First, the step of providing a template model 100 is performed. Step 100 is a pre-processing step performed once during a design phase to prepare the adaptive template of the present invention. Separately, a moving image (ie, the patient image) 110, such as a PET or SPECT scan, is taken of a subject and the step of choosing starting parameters 120 for the moving image is performed. Including within step 120 is the step of choosing a template control start parameter. In short, the following steps are then iterated until convergence is reached. An adaptive template is then built 130 using the control parameter for the adaptive template and the model from step 100. A transforming step 140 is performed in which the moving image is adjusted using the start parameters. Then, a comparison step 150 is performed which compares the similarity between the transformed moving image and the adaptive template. In this step, the similarity metric is evaluated from the voxel values in the template image and the transformed moving image. If the moving image and the adaptive template are found to be sufficiently or maximally converged, the registration of the moving image is completed, step 160. Should the moving image and the adaptive template not be sufficiently or maximally converged, the step of choosing new parameters 170, for both the moving image and the template control parameter, is performed. Then transforming step 140 is repeated, this time using the parameters chosen during step 170. Additionally, step 130 is repeated in which the existing adaptive template is also modified by the new template control parameter chosen during step 170. The comparison step 150 is again performed using the newly transformed moving image and adaptive temple. If the method has not converged, another set of adaptive template and transformation parameters is selected by the optimization method and a new evaluation of the similarity metric is made. The present invention contemplates that the optimization method may be any numerical optimization method.

(14) Convergence may be sufficiently achieved when the value of a function for describing the similarity between the images has reached a level deemed suitable for proceeding. Convergence is desirably achieved when the value of a function for describing the similarity between the images has reached its maximum, ie, showing maximum similarity between the template image in reference space and the transformed image. The similarity metric could be based, for example, on correlation or on mutual information, although the present invention contemplates that any suitable similarity metric may be applied.

(15) The pre-processing step 100 is performed once and the extra parameter or parameters controlling the template model are added to the registration parameters and altered by the optimization method in a similar way as the parameters controlling the geometry change. The following will now in detail describe the creation of a one-parameter linear template model, which has been found to be suitable for Flutemetamol PET scans.

(16) The present invention is contemplated to provide a non-transitory computer-readable storage medium with an executable program for performing the method of the present invention. This computer-readable storage medium includes computer-readable program code includes instructions for performing a method of the present invention, such that execution of the computer-readable program code causes a processor to perform the steps of the method of the present invention. Additionally, the present invention is contemplated to use an adaptive template model of the present invention for performing the spatial normalization method of the present invention. The starting parameters may be fixed for all procedures and the updated parameters may be selected automatically by the process.

(17) Adaptive Template Image Model

(18) In the following example, Montreal Neurological Institute (MNI) reference space sampled to a uniform 2×2×2 mm voxel size is used as standard space for the template image. A set of Flutemetamol input images is provided to create a template of the present invention, which images have already been spatially normalized to this space using an MR-based spatial normalization method.

(19) Normally when creating a template image for spatial normalization one takes these spatially-normalized input images and calculates a mean image from these. It follows that this mean image will also be located in the reference space and can be used for spatial normalization of new unseen images by using an image registration method to find the transformation making the new images match the template mean image.

(20) However, as described hereinabove, the problem with this approach is that the mean image will only be representative for one typical image pattern for a specific level of amyloid, i.e. either the pattern of a Aβ−, Aβ+, or something in between. To be able to create a model of the image pattern variability the present invention instead calculates a regression model for each voxel in the template image, where the dependent variable which is the voxel intensities in the input images is regressed on some variable which ideally for each input image expresses a “true” value of where on a scale from Aβ− to very severe Aβ+ it is located.

(21) In practice, it is impossible to extract such true values but, being principally interested in finding a robust estimated value for each scan on a one-dimensional scale of how much Aβ the gray matter contains, this approach provides such a robust estimate. Therefore, the present invention calculates for all input images a mean standard uptake ratio value (SUVR) inside a large composite region (COM) covering regions known to be most affected by Aβ in Alzheimer's disease patients. These regions were the prefrontal cortex, parietal cortex, precuneus, lateral temporal cortex, and the anterior and posterior cingulate cortices.

(22) An illustration of the calculated SUVR values of the composite region COM.sub.SUVR for all subjects in two different groups of Aβ− and Aβ+ scans is shown in FIG. 3.

(23) It is thus seen that the present invention uses a good sampling over the whole spectrum from Aβ− subjects with very low Flutemetamol uptake inside COM to Aβ+ subjects with a very high Flutemetamol uptake.

(24) In the next step, the linear regression model was applied for each voxel i in the template image matrix to express the dependence of that particular voxel, y.sub.i, on the COM.sub.SUVR values according to Equation 1:
y.sub.i={circumflex over (α)}.sub.i+β.sub.iCOM.sub.SUVR+ε.sub.i  Equation 1
Linear regression model
Applying this model over all input images regressed on the corresponding COM.sub.SUVR values provides an intercept image and a slope image according to Equation 2:
I.sub.0=[{circumflex over (α)}.sub.1{circumflex over (α)}.sub.2 . . . {circumflex over (α)}.sub.n]
I.sub.Slope=[{circumflex over (β)}.sub.1{circumflex over (β)}.sub.2 . . . {circumflex over (β)}.sub.n]  Equation 2
Intercept image, I.sub.0, and slope image, I.sub.slope, created from linear regression, where n is the number of voxels in the images.
The intercept image will correspond to a pattern of a fully normal subject located in the lower part of the COM.sub.SUVR scale. The slope image, on the other hand, will have the highest values for the parts of the image where the largest changes occur when going from low to high on the COM.sub.SUVR scale. These resulting images are shown in FIG. 4.

(25) From these images the present invention creates synthetic simulated images, along the whole COM.sub.SUVR scale over the range from Aβ− to Aβ+. This is done by adding multiplicatively scaled portions of I.sub.Slope to I.sub.0 according to Equation 3:
I.sub.template=I.sub.0+I.sub.slope x  Equation 3
Template image equation
A value of the scale factor x of 0.0 will correspond to I.sub.0 itself, which represents a pattern of an Aβ− subject, and a value of 1.0 will correspond to an Aβ+ subject in the high end of the COM.sub.SUVR scale. From this formula the present invention can create synthetic images for any x along the linear path defined by I.sub.slope. FIG. 5 shows such synthetic images covering values of x from 0.0 to 1.0 in steps of 0.2.

(26) Such synthetic images are used as the adaptive template in a spatial normalization method. The slope image, I.sub.Slope, and intercept image, I.sub.0, is the adaptive template model is controlled by one single parameter x. This parameter x is then adjusted by the image registration algorithm in a similar way as any of the other parameters, and the template image is updated accordingly any time the x parameter is changed.

(27) Referring now to FIG. 6, the method of the present invention is contemplated to be performed by system 200. System 200 includes a scanner 210 for performing PET or SPECT scans, a computer 220 for receiving scan images, e.g., the patient image (aka, the moving image), and a database 230 for providing the adaptive template. Computer 220 typically includes a display 222, an input device 224 such as a keyboard 224a and a mouse 224b, and a processor 226. Processor 226 typically includes software for performing the method of the instant invention using the adaptive template from database 230 and the moving image from scanner 210. Additionally, processor 226 is contemplated to include non-transitory computer readable storage medium with an executable program for using an adaptive template of the present invention and for registering a PET image to an adaptive template image according to a method of the instant invention using the moving image from scanner 210. The non-transitory computer readable storage medium includes computer-readable program code including instructions for using an adaptive template of the present invention and for registering a PET image to an adaptive template image. Connections between scanner 210, computer 220, and database 230 are contemplated to be by any means known to the art, such as hardwire, wireless, or any combination thereof. Additionally, the present invention contemplates that processor 226 and database 230 are connected such that processor 226 may return the upgraded adaptive template to database 230 upon completion of the method of the present invention.

(28) While the particular embodiment of the present invention has been shown and described, it will be obvious to those skilled in the art that changes and modifications may be made without departing from the teachings of the invention. The matter set forth in the foregoing description and accompanying drawings is offered by way of illustration only and not as a limitation. The actual scope of the invention is intended to be defined in the following claims when viewed in their proper perspective based on the prior art.