WO2014170780A1

WO2014170780A1 - Iterative image reconstruction with tissue dependent regularization

Info

Publication number: WO2014170780A1
Application number: PCT/IB2014/060337
Authority: WO
Inventors: Michael Grass; Bernhard Johannes Brendel; Rolf Dieter Bippus; Frank Bergner; Thomas Koehler
Original assignee: Koninklijke Philips N.V.; Philips Deutschland Gmbh
Priority date: 2013-04-19
Filing date: 2014-04-01
Publication date: 2014-10-23

Abstract

A method includes obtaining an update term for a statistical iterative reconstruction. The update term includes a regularization parameter that is a function of a tissue type of a voxel of an initial image of an object. The method further includes iteratively reconstructing a final image by updating the initial image using the update term. A reconstructor (118) includes an image reconstructor (212) that iteratively reconstructs a final image by updating an initial image of an object using an update term that includes a regularization parameter that is a function of a tissue type of a voxel of the initial image being updated.

Description

ITERATIVE IMAGE RECONSTRUCTION WITH TISSUE DEPENDENT REGULARIZATION

The following generally relates to iterative image reconstruction with tissue dependent regularization and finds particular application to computed tomography (CT). The following is also amenable to other imaging modalities.

A CT scanner includes an x-ray tube mounted on a rotatable gantry that rotates around an examination region about a z-axis. A detector array subtends an angular arc opposite the examination region from the x-ray tube. The x-ray tube emits radiation that traverses the examination region. The detector array detects radiation that traverses the examination region and generates projection data indicative thereof. A reconstructor processes the projection data using an iterative or non- iterative reconstruction algorithm which generates an image indicative of the examination region.

Statistical iterative reconstruction approaches generally include a noise and a system model and employ regularization to achieve target noise values in a final image. The regularization has included a global regularization parameter for a data set or a spatially resolved image regularization for the data set, e.g., increasing noise levels in the center of the patient. As a consequence, the regularization applied in a statistical iterative reconstruction depends on the system geometry as well as the physics of the measurement process, while the different properties of the tissues in the image are not represented in the regularization.

An example statistical iterative reconstruction update term with a quadratic penalty in the image domain is shown EQUATION 1 :

EQUATION 1 :

∑a - logy, -∑α₀,μ - fi∑w_kfa - μ,")

μ"⁺ = μ

∑α_ϋγ,∑α_ϋ, + β∑ w,.

where y_t are measured projection values, μ " are image values in an nth iteration, are elements of a system matrix (i.e., an intersection of a blob representing pixel j with a ray belonging to projection value i), Wk are weights used to weight different neighbors of the pixel j, and β is the global regularization parameter for the entire image.

With a global regularization parameter, voxels representing sharp edges (e.g., bone) and low contrast structure (e.g., soft tissue) are similarly smoothed. For example, with a current state of the art approach, a final image noise level is typically used (e.g., decrease image noise by 30%) to determine the global regularization parameter β that provides a uniform decrease in noise across the image. However, it is often desirable to smooth low contrast structure differently than sharp edges. With current state of the art approaches, the data is either reconstructed more than once, with each reconstruction using a β optimized for the particular tissue type (e.g., bone or soft tissue), or there is a compromise and the global regularization parameter β is not optimized for either sharp edges or low contrast structure.

Aspects described herein address the above-referenced problems and others.

The following describes an approach that employs a tissue dependent statistical iterative reconstruction update term in a statistical iterative reconstruction.

In one aspect, a method includes obtaining an update term for a statistical iterative reconstruction. The update term includes a regularization parameter that is a function of a tissue type of a voxel of an initial image of an object. The method further includes iteratively reconstructing a final image by updating the initial image using the update term.

In another aspect, a reconstructor includes an image reconstructor that iteratively reconstructs a final image by updating an initial image of an object using an update term that includes a regularization parameter that is a function of a tissue type of a voxel of the initial image being updated.

In another aspect, a computer readable storage medium is encoded with computer readable instructions, which, when executed by a processer, causes the processor to: reconstruct an initial image of an object, obtain an atlas corresponding to the object, register the initial image and the atlas, determine a probability that a voxel in the initial image corresponds to a particular tissue or a combination of different tissues, generate a statistical iterative reconstruction update term with the probability, and iteratively reconstruct a final image by updating the initial image using the update term.

The invention may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. FIGURE 1 schematically illustrates an example imaging system including a reconstructor that employs a statistical iterative reconstruction algorithm with a tissue dependent regularization parameter.

FIGURE 2 schematically illustrates an example of the reconstructor that employs an atlas registration with an initial image.

FIGURE 3 schematically illustrates an example of the reconstructor that employs an atlas registration with an initial image and at least one atlas registration with an intermediate image.

FIGURE 4 illustrates an example method for iterative reconstruction with a single atlas registration and a tissue dependent regularization term.

FIGURE 5 illustrates an example method for iterative reconstruction with multiple atlas registrations and a tissue dependent regularization term.

FIGURE 1 illustrates an example imaging system 100, such as a computed tomography (CT) scanner. The imaging system 100 includes a stationary gantry 102 and a rotating gantry 104, which is rotatably supported by the stationary gantry 102 and rotates around an examination region 106 about a z-axis 108. A subject support 110, such as a couch, supports an object or subject such as a human or animal subject in the examination region 106.

A radiation source 112, such as an x-ray tube, is rotatably supported by the rotating gantry 104 and rotates therewith, and emits radiation that traverses the examination region 106. A one or two dimensional radiation sensitive detector array 114 subtends an angular arc opposite the radiation source 112 across the examination region 106. The detector array 114 detects radiation traversing the examination region 106 and generates projection data indicative thereof.

A computer serves as an operator console 116. The console 116 includes a human readable output device such as a monitor and an input device such as a keyboard, mouse, etc. Software resident on the console 116 allows the operator to interact with and/or operate the scanner 100 via a graphical user interface (GUI) or otherwise. For example, the console 116 allows the operator to identify a reconstruction algorithm of interest, such as the iterative reconstruction with a tissue dependent regularization term.

A reconstructor 118 processes the projection data and reconstructs volumetric image data indicative thereof. In this example, the reconstructor 118 processes the projection data using one or more reconstruction algorithms from reconstruction algorithm memory 120 such as an iterative reconstruction with regularization algorithm 122. As described in greater detail below, a suitable iterative reconstruction algorithm includes a tissue dependent regularization parameter in the update term.

As such, each voxel in the statistical iterative reconstruction is smoothed based on the tissue it represents (e.g., bone, soft tissue, air, a combination thereof, etc.), and a single statistical iterative reconstruction can be employed (and, in some instance, optimized) for multiple different tissue types of interest. In one instance, this mitigates having to run multiple reconstructions, each with a regularization parameter tuned for a particular tissue type, or a single reconstruction in which there is a compromise between smoothing low contrast structure and retaining sharp edges.

It is to be appreciated that the reconstructor 118 can be implemented via one or more processor (a central processing unit (CPU), a microprocessor, etc.) executing one or more computer executable instructions embedded or encoded on computer readable storage medium, including physical memory and other non-transitory medium. Additionally or alternatively, at least one of the computer executable instructions can be carried by a carrier wave, signal, and other transitory medium. The reconstructor 118 can be part of the imaging system 100 (as shown) or remotely located in a computer based system(s).

FIGURE 2 illustrates an example of the reconstructor 118.

A filtered-backprojection (FBP) reconstructor 202 receives, as an input, the projection data generated by the detector array 114 (FIGURE 1) of the imaging system 100 and/or other imaging system and reconstructs an initial image based on a FBP reconstruction algorithm. In a variation, the initial image is based on another type of reconstruction and/or otherwise created, e.g., through simulation, etc.

A registration component 204 registers an atlas from an atlas bank 206 with the initial image generated by the FBP reconstructor 202. The particular atlas utilized corresponds to the scanned object (or a particular anatomy, where the scanned object is a subject). For example, where the scan is a pelvic scan, the atlas utilized corresponds to the pelvis. In this case, the anatomical atlas represents a probability density function of a spatial location of a voxel given a tissue type. For example, if a human body is located in atlas space, it is likely that bone tissue is located at those positions where the voxel intensity is high. The registration component 204 can employ a rigid/affine registration (e.g., rotation, scaling, translation, etc.) and/or a non-rigid/elastic registration (e.g., warping, rotation, scaling, translation, etc.). A probability determiner 208 determines a probability of tissue type for each voxel based of the initial image based on the registration of the initial image with the anatomical atlas. By way of example, in one non-limiting instance, a voxel representing only soft tissue is given probability of zero, a voxel representing only bone is given probability of one, a voxel representing half bone and half soft tissue is given probability of one half, etc. Other values can alternatively be utilized. This can be achieved by mapping each voxel of the initial image to the anatomical atlas, which, as discussed above, represents a probability density function describing a probability of a spatial location being a voxel of a particular tissue type. Since, with a CT image, the intensity at a location is known and represented in Hounsfield units, a probability of the tissue type can be deduced from the intensity value.

An update term generator 210 generate an iterative reconstruction update term with a regularization parameter that is based on the mapping of each voxel to the anatomical atlas, or a probability density function, as shown in EQUATION 2:

where, again, _y_;- represents the measured projection values, μ " represents the image values in the nth iteration, ay represents the elements of the system matrix (e.g., the intersection of the blob representing pixel j with the ray belonging to projection value i), and Wk represents weights used to weight the different neighbors of the pixel j.

In EQUATION 2, /¾_f represents a voxel dependent regularization value, depending on the most probable tissue in this voxel or the mean over the tissues in this voxel, where the subscript i represents the voxel and the subscript t represents the tissue type. If more than one regularization term is used, this approach can be applied to the regularization parameter for each regularization term individually. The basic value β may be determined by the acquisition parameters (e.g. dose), while only the modifications are determined via the atlas.

EQUATION 2 illustrates a "quadratic" based regularization. This regulation penalizes a quadratic difference between two neighboring voxel, μ" and μ , which is essentially the derivative of this quadratic difference. The regularization in this case can be generalized as a voxel dependent function which depends on the neighbors of this voxel, which penalizes undesired properties like noise, and whose parameters are changed depending on the likeliness of the tissue class. An example parameter is, e.g., the

regularization strength, which is the pre-factor β in EQUATION 2.

However, other regularization terms and/or modifications thereto are also contemplated herein. For example, another suitable regularization is a potential function that penalize certain image properties during the iterative reconstruction. An example is the Huber potential, which is known to penalize fluctuations like noise but preserve the edges in the images. The Huber potential itself has another parameter, which then could be also varied depending on the likeliness of the tissue class. A modification can be to use more than two neighboring pixel for the potential function and/or combining different potential functions to a new one.

An image reconstructor 212 iteratively reconstructs an image using the update term generated by the update term generator 210. Logic 214 determines whether another iteration (or update) is performed. Stopping criteria can include, but is not limited to, a maximum number of iterations, a pre-determined time interval, a difference between consecutive updates images, user provided criteria, and/or other criteria. The output of the image reconstructor 212 is the final reconstructed image.

FIGURE 3 illustrates a variation of FIGURE 2 in which the atlas based registration can also be applied, after at least one image update, to an intermediate image generated by applying the update term. In this example, the logic 214 determines whether the atlas based registration is applied to an intermediate image after an image update.

In FIGURES 2 and 3, the image is generated by the FBP reconstructor 202. In a variation, the initial image could be generated by a different reconstruction approach such as Fourier inversion, iterative, and/or other approach. In another variation, a coarse-to-fine approach can be utilized. With this approach, the image resolution in the reconstruction and the resolution of the atlas is increased for a least one subsequent iteration.

The approach described herein can also be applied to multi-energy or spectral CT imaging. In this instance, either atlas information can be based on the normal absorption images and reused in the energy dependent sub images, or, in case that energy dependent atlas functions are available, can be directly used on the sub images.

FIGURES 4 and 5 illustrate example methods in accordance with the embodiments described herein. It is to be appreciated that the ordering of the acts in the methods described herein is not limiting. As such, other orderings are contemplated herein. In addition, one or more acts may be omitted and/or one or more additional acts may be included.

At 402, projection data is reconstructed, generating an initial image.

At 404, the initial image is registered with an anatomical atlas corresponding to the anatomical structure represented in the initial image.

At 406, a probability that a voxel represents a particular tissue or combination of different tissues is determined based on the registration.

At 408, an update term for a statistical iterative reconstruction is generated, using the probability to provide a tissue dependent regularization parameter.

At 410, the initial image is updated using the update term, producing an intermediate image.

At 412, it is determined whether there will be another update.

If there will be another update, then acts 410-412 are repeated, using the intermediate image in place of the initial image.

If there will not be another update, then a final image generated.

At 502, projection data is reconstructed, generating an initial image.

At 504, the initial image is registered with an anatomical atlas corresponding to the anatomical structure represented in the initial image.

At 506, a probability that a voxel represents a particular tissue or combination of tissues is determined based on the registration.

At 508, an update term for a statistical iterative reconstruction is generated, using the probability to provide a tissue dependent regularization parameter.

At 510, the initial image is updated using the update term, producing an intermediate image.

At 512, it is determined whether there will be another update.

If there will be another update, then at 514, it is determined whether the intermediate image will be registered with the atlas.

If the intermediate image will be registered with the atlas, then acts 504 - 512 are repeated, using the intermediate image in place of the initial image. If the intermediate image will not be registered with the atlas, then acts 510- 512 are repeated, using the intermediate image in place of the initial image.

If there will not be another update, then a final image is generated.

The above may be implemented by way of computer readable instructions, encoded or embedded on computer readable storage medium, which, when executed by a computer processor(s), cause the processor(s) to carry out the described acts. Additionally or alternatively, at least one of the computer readable instructions is carried by a signal, carrier wave or other transitory medium.

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

Claims

CLAIMS:

1. A method, comprising:

obtaining an update term for a statistical iterative reconstruction, wherein the update term includes a regularization parameter that is a function of a tissue type of a voxel of an initial image of an object; and

iteratively reconstructing a final image by updating the initial image using the update term.

2. The method of claim 1, wherein the regularization parameter controls an amount of smoothing applied to each voxel of the initial image.

3. The method of any of claims 1 to 2, further comprising:

obtaining a probability that the voxel corresponds to a particular tissue or a combination of different tissues; and

generating the update term based on the probability.

4. The method of claim 3, further comprising:

obtaining an atlas corresponding to the object;

registering the initial image and the atlas; and

determining the probability based on the registered initial image and atlas.

5. The method of any of claims 3 to 4, further comprising:

generating an intermediate image by updating the initial image using the update term;

registering the intermediate image and the atlas; and

determining a subsequent probability that a voxel in the intermediate image corresponds to the particular tissue or the combination of the tissues based on the registered intermediate image and atlas;

generating a subsequent update term with the subsequent probability, and iteratively reconstructing the final image by updating the intermediate image using the subsequent update term.

6. The method of any of claims 3 to 5, wherein at least one voxel represents a sharp edge of structure and at least another voxel represents soft tissue.

7. The method of any of claims 3 to 6, wherein at least one voxel represents a combination of a sharp edge of structure and soft tissue.

8. The method of any of claims 1 to 7, further comprising:

reconstructing projection data using a filtered-backprojection reconstruction algorithm, generating the initial image.

9. A reconstructor (118), compri sing :

an image reconstructor (212) that iteratively reconstructs a final image by updating an initial image of an object using an update term that includes a regularization parameter that is a function of a tissue type of a voxel of the initial image being updated.

10. The reconstructor of claim 9, wherein the regularization parameter controls an amount of smoothing applied to each voxel of the initial image.

11. The reconstructor of any of claims 9 to 10, further comprising:

an update term generator (210) that generates the update term based on a probability that each voxel in the initial image corresponds to a particular tissue or a combination of different tissues.

12. The reconstructor of claim 11, further comprising:

a probability determiner (208) that determines the probability based on a registration of the initial image and an atlas corresponding to the object.

13. The reconstructor of claim 12, further comprising:

a registration component (204) that registers the initial image and the atlas.

14. The reconstructor of claim 13, wherein the image reconstructor generates an intermediate image by updating the initial image using the update term, the registration component registers the intermediate image and the atlas, the probability determiner determines a subsequent probability that a voxel in the intermediate image corresponds to the particular tissue or the combination of the tissues based on the registered intermediate image and atlas, the update term generator generates a subsequent update term with the subsequent probability, and the image reconstructor generates the final image by updating the intermediate image using the subsequent update term.

15. The reconstructor of any of claims 9 to 14, further comprising:

a filtered-backprojection reconstructor (202) that generates the initial image by reconstructing projection data generated from a CT scan of the object.

16. The reconstructor of any of claims 9 to 15, wherein at least one voxel represents a sharp edge of structure and at least another voxel represents soft tissue.

17. The reconstructor of any of claims 9 to 16, wherein at least one voxel represents a combination of a sharp edge of structure and soft tissue.

18. A computer readable storage medium encoded with computer readable instructions, which, when executed by a processer, causes the processor to:

reconstruct an initial image of an object;

obtain an atlas corresponding to the object;

register the initial image and the atlas;

determine a probability that a voxel in the initial image corresponds to a particular tissue or a combination of different tissues;

generate a statistical iterative reconstruction update term with the probability; and

iteratively reconstruct a final image by updating the initial image using the update term.

19. The computer readable storage medium of claim 18, wherein the processor, when executing the computer readable instructions, further causes the processor to:

generate an intermediate image by updating the initial image using the update term;

register the intermediate image and the atlas; determine a subsequent probability that a voxel in the intermediate image corresponds to the particular type of tissue or combination of tissue based on the registered intermediate image and atlas;

generate a subsequent update term with the subsequent probability; and iteratively reconstruct the final image by updating the intermediate image using the subsequent update term.

20. The computer readable storage medium of any of claims 18 to 19, wherein the update term includes a regularization parameter that is a function of a tissue type of a voxel of the initial image.