KR101356881B1 - Method and apparatus for restructuring image for parallel processing of positron emission tomography with high resolution - Google Patents

Abstract

The present invention relates to image reconstruction, and more particularly, to a method and apparatus for reconstructing an image for parallel processing in high resolution positron emission tomography.
According to an embodiment, the method of reconstructing an image may include a line-of-response (LOR) detected from detectors in response to radiation irradiated to a measurement object to form a cosine curve in the form of a cone beam. converting to a sinogram format, performing back projection on the basis of the converted response lines, and reconstructing an image using the result of performing the back projection.

Description

METHOD AND APPARATUS FOR RESTRUCTURING IMAGE FOR PARALLEL PROCESSING OF POSITRON EMISSION TOMOGRAPHY WITH HIGH RESOLUTION}

The present invention relates to a method and apparatus for reconstructing an image for parallel processing in high resolution positron emission tomography.

The image reconstruction algorithm of Positron Emission Tomography (PET) mainly uses the Ordered Subset Expectation Maximization (“OSEM”) algorithm. Here, OSEM is a method of reconstructing an image by dividing projection data into several subsets, and then performing projection and reverse projection on a subset basis. Therefore, in the OSEM, the convergence speed of the algorithm increases as the number of subsets increases. However, since OSEM requires a lot of computation, it takes a very long time to perform calculations on a personal computer.

In general, in positron emission tomography (PET), data (eg, a line-of-response (LOR)) may be obtained from a ring type detector as shown in FIG. 3. In this case, in order to reduce the amount of OSEM calculation, as shown in FIG. 4, the reaction line LOR may be arc-corrected and converted into a cosine curve (sinogram) having a uniform interval. However, additional interpolation is involved in this process, which may reduce the resolution of the reconstructed image.

In order to solve this problem, a method of reconstructing an image by calculating LOR accurately without performing arc-correction as shown in FIG. 4 has been proposed, which is called 'LOR-based OSEM'.

A parallel processing method for improving the processing speed of the LOR-based OSEM may be used, and all of these parallel processing methods use the line-driven back projection method as shown in FIG. 5. However, the line-driven backprojection method is not suitable for parallel processing.

According to an embodiment of the present invention, when the LOR-based OSEM image reconstruction is performed, the format of the cosine curve (sinogram) may be converted to be suitable for parallel processing to enable computation optimized for parallel processing.

According to an embodiment, a method of reconstructing an image for parallel processing may include response lines (LORs) detected from detectors in response to radiation emitted to a measurement object in the form of cone beams. Converting to a cosine curve format, performing a back projection based on the converted response lines, and reconstructing an image using the result of performing the back projection.

The converting may include converting the reaction lines into a cosine curve format in the form of a cone beam having a vertex of a crystal belonging to a first detector of a first region.

The performing of the back projection may perform a voxel-driven back projection based on the converted response lines.

The reconstructing the image may be a step of reconstructing the image using a LOR-based OSEM parallel algorithm based on a result of performing the reverse projection.

Reconstructing the image may be reconstructed using a gather operator.

According to an embodiment, an apparatus for reconstructing an image for parallel processing may include response lines (LORs) detected from detectors in response to radiation irradiated to a measurement object in a sinogram format in the form of a cone beam. Conversion unit for converting to; An execution unit performing back projection based on the converted response lines; And a reconstruction unit for reconstructing the image using the result of performing the reverse projection.

The converter may convert the reaction lines into a cosine curve format in the form of a cone beam having a vertex of a crystal belonging to a first detector of a first region.

The execution unit may perform voxel-driven back projection based on the converted reaction lines.

The reconstruction unit may reconstruct the image using a LOR-based OSEM parallel algorithm based on a result of performing the reverse projection.

The reconstruction unit may reconstruct the image using the gather operator.

One embodiment of the present invention is to parallelize by converting the format of the cosine curve (sinogram) in the forward projection and back projection during the OSEM (Ordered Subset Expectation Maximization) based image reconstruction It can enable operations optimized for processing.

One embodiment of the present invention can perform a voxel-driven back projection method suitable for parallel processing by aligning the response lines (LOR) in the form of a cone (cone).

In addition, an embodiment of the present invention can prevent the performance degradation of parallel processing by performing the entire OSEM process with the gather operator.

1 is a view for explaining an imaging device for positron emission tomography (PET).
FIG. 2 is a diagram for explaining a method of sampling cosine sinograms in a positron emission tomography (PET) system.
FIG. 3 shows a sinogram obtained from a ring shaped detector in positron emission tomography (PET).
4 shows arc-correction of the response lines (LORs) to a uniformly spaced sinogram.
FIG. 5 is a diagram illustrating a line-driven back projection method of an image reconstruction method based on an ordered subset expectation (OSEM).
FIG. 6 is a diagram illustrating a voxel-driven back projection method of an OSEM-based image reconstruction method.
FIG. 7 illustrates reconstruction of LOR-based OSEMs based on cone-beam in a method of reconstructing images for parallel processing in high resolution positron emission tomography according to an embodiment.
8 is a flowchart illustrating a method of reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment.
FIG. 9 is a diagram for describing a method of reconstructing an image based on a cone-beam in a method of reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment.
10 is a block diagram of an apparatus for reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. Also, like reference numerals in the drawings denote like elements.

1 is a view for explaining an imaging device for positron emission tomography (PET).

Referring to FIG. 1, an imaging device (hereinafter, referred to as an imaging device) 100 for positron emission tomography (PET) administers the positron 120 inside the measurement object 110. When the positron 120 meets the electrons inside the measurement object 110, the positron 120 emits two gamma photons at an angle of about 180 degrees and disappears. In this case, by measuring two gamma photons reaching the detector 140 at about the same time, projection information on the distribution of the positron 120 inside the measurement object 110 along the corresponding line 130 may be obtained.

The projection information may be classified according to the angle of the corresponding line 130 of the positron 120 where annihilation occurs and the distance from the origin. Using this projection information, a 'sinogram' may be generated.

FIG. 2 is a diagram for explaining a method of sampling cosine sinograms in a positron emission tomography (PET) system.

For sampling of the cosine curves, a positron emission tomography (PET) system can obtain several sets of sampling data by regularly circularly moving the entire detector or a subject (eg, a patient).

Referring to FIG. 2, the positron emission tomography (PET) system may acquire the cosine curves 211 at each position by moving the entire detector or the measurement object using an imaging device. These cosine curves 211 can be used to obtain a set of cosine curves 210.

In this case, a super resolution algorithm may be applied to the obtained cosine curves to reconstruct a super-resolution positron emission tomography (PET) image.

Here, since the set of cosine curves 210 is data measured directly in a positron emission tomography (PET) system, the relationship between high-resolution (HR) and low-resolution (LR) is more clear.

A positron emission tomography (PET) system can obtain a high resolution cosine curve 221 through the imaging device 220.

More specifically, in FIG. 1, the cosine curve set 210 may be obtained by moving the entire detector 140 of the imaging device of the positron emission tomography (PET) system. In addition, the same effect as moving the entire detector 140 by moving the measurement object 110 can be seen.

At this time, the sample of each cosine curve is kept blur by at least one of the Positron range of the radiation, non-colinearity of the radiation and the size of the detector. Therefore, even if the positron emission tomography (PET) image is simply reconstructed with the cosine curves with an increased number of samples, the overall resolution does not change significantly.

When moving the entire detector or the object to be photographed, additional samples may be added between existing samples. Therefore, samples of different cosine curve sets are positioned between samples of one cosine curve, thereby obtaining a high resolution cosine curve as shown in FIG. 2.

FIG. 3 shows a sinogram obtained from a ring shaped detector in positron emission tomography (PET), and FIG. 4 shows a uniformly spaced sinogram by arc-correcting the response lines (LOR). To convert to.

When using accurate response lines (Exact LORs) as shown in Figure 3, it can be seen that the reaction lines (LOR) are not aligned with each other.

In this case, as described above, in order to reduce the amount of calculation of the OSEM, the reaction line LOR may be arc-corrected and converted into a sinogram having a uniform interval as shown in FIG. 4. However, additional interpolation is involved in this process, which may reduce the resolution of the reconstructed image.

In order to solve this problem, an embodiment uses a 'LOR-based OSEM' method of reconstructing an image by accurately calculating an LOR without performing arc-correction as shown in FIG. 4.

FIG. 5 is a diagram illustrating a line-driven back projection method of an image reconstruction method based on an ordered subset expectation (OSEM).

The line-driven back projection method may also be referred to as a ray-driven back projection method.

Line-driven back projection is a method of tracking the light in an image, especially by directly discretizing the line integrals so that each line integration is close to the ideal line. It is estimated by the weighted sum of the voxels.

The line-driven back projection method connects the line from the focal spot of the image to the center of the detector cell of interest. At this time, as the number of tracked line-paths per projection beam increases, the accuracy of the projection also improves.

FIG. 6 illustrates a voxel-driven back projection method used in a method of reconstructing an image for parallel processing in high resolution positron emission tomography according to an exemplary embodiment.

As mentioned above, the line-driven back projection method is an operation that is not suitable for parallel processing. Therefore, in one embodiment, the image reconstruction performance is optimized by using a voxel-driven back projection method as shown in FIG. 6.

To this end, in one embodiment, the reaction lines LOR are aligned in the form of cones (or quadrangular pyramids) as shown in FIG. 7 below, and using this, a voxel-driven back projection suitable for parallel processing is used. Can be performed.

FIG. 7 illustrates reconstruction of LOR-based OSEMs based on cone-beam in a method of reconstructing images for parallel processing in high resolution positron emission tomography according to an embodiment. In FIG. 7, the left view shows the reaction lines (Exact LORs), and the right view shows the reconstruction of the response lines of the left view on a cone-beam basis.

In one embodiment, the format of the sinogram in OSEM is converted from a general parallel-beam to a cone-beam to process the forward projection method and the back projection method. All of the methods can be adapted for parallel processing.

When reconstructing an image in an iterative manner, the largest portion of the computation is reconstructing the final image data on the image domain using forward projection and projection data to generate projection data from the image or scan object. It is a reverse projection process. Moreover, when reconstructing the 3D image, the amount of computation and the computation time are further increased due to the huge increase in the number of response lines (LOR).

Thus, parallel processing techniques can be used to reduce this computation time. Parallel processing is a method of dividing work of the same kind in a plurality of arithmetic cores and simultaneously calculating them. A central processing unit (CPU) or a graphics processing unit (GPU) with multiple cores may be used for parallel processing. In particular, GPUs are typically equipped with dozens or hundreds of cores to perform massively parallel processing, and can handle mathematical operations as well as graphic operations.

In the case of using accurate LORs as shown in FIG. 3, since the LORs are not aligned with each other, it is difficult to apply the voxel-driven reverse projection method of FIG. 6 in performing parallel processing. In order to solve this problem, in one embodiment, the voxel-driven reverse projection method can be easily applied in parallel processing by aligning the response lines LOR in a cone-beam shape as shown in FIG. 7.

8 is a flowchart illustrating a method of reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment.

Referring to FIG. 8, an apparatus for reconstructing an image for parallel processing according to an exemplary embodiment (hereinafter, referred to as a “reconstruction apparatus”) may include a response line detected from detectors in response to radiation irradiated to a measurement object (Line-Of). Response LORs are converted into a sinogram format in the form of a cone beam (810).

In this case, the reaction lines may be converted into a cosine curve format in the form of a cone beam having a vertex of a crystal belonging to the first detector of the first region. The response lines converted to the cone-shaped cosine curve format are spread out in the form of cones in a second region where the second detector is located at a vertex of a crystal belonging to the first detector of the first region. In this case, the second detector (or second region) may be in a position facing the first detector (or first region) on a circle.

The reconstruction apparatus performs back projection based on the converted response lines (830). In this case, the reconstruction apparatus may perform voxel-driven back projection suitable for parallel processing based on the converted reaction lines. The reconstruction device may be implemented by a graphics processing unit (GPU).

The reconstruction apparatus reconstructs the image by using the result of performing the reverse projection (850). The reconstruction apparatus may reconstruct an image using a LOR-based OSEM parallel algorithm based on the result of the reverse projection. In one embodiment, LOR-based OSEM may be reconstructed based on a cone beam to perform calculations optimized for parallel processing.

In addition, the reconstruction apparatus may reduce performance degradation of parallel processing by reconstructing an image using a gather operator.

FIG. 9 is a diagram for describing a method of reconstructing an image based on a cone-beam in a method of reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment.

The most time-consuming part of the OSEM algorithm process is the process of forward projection and back projection. Both of these processes may be performed by a line-driven method and a voxel-driven method, which may be performed using different operators as shown in Table 1 below.

[Table 1] shows the case where the Scatter operation and the Gather operation are performed in OSEM, respectively.

If you run the Scatter operator in parallel, data loss occurs. The atomic operator can be used to prevent this loss, but due to its high cost, the parallel processing performance is greatly reduced. In particular, in the graphics processing unit (GPU), the gather operation may be performed more efficiently than the scatter operation.

Therefore, in order to maximize parallel processing performance, the embodiment uses a gather operator.

When paralleling OSEM, it is best to perform forward projection with the line driven method and reverse projection with the voxel-driven method. However, in the general parallel beam type sinogram format, it is not easy to perform reverse projection by voxel-driven method. In order to perform the voxel-driven method, a combination of each voxel and neighboring response lines (LOR) must be obtained, because the process of calculating the combination is complicated in a general cosine curve format.

Therefore, in one embodiment, to solve this problem, the image is reconstructed based on the cone-beam as shown in FIG. 9. First, as shown in FIG. 9, the entire response line LOR is formed between a cone 1 between a detector 1 and a second header 2 that are arbitrary detectors belonging to the first header detector 1. Or square pyramid).

In this case, a combination of response lines (LOR) can be obtained with a small amount of calculation through a cone having crystal 1 as a vertex, and voxel driven reverse projection can be performed by a gather operator.

Even if you change the cosine curve format to cone, you can continue to use line-driven for forward projection, so the entire OSEM process can be performed with the Gather operator suitable for the GPU.

10 is a block diagram of an apparatus for reconstructing an image for parallel processing in high resolution positron emission tomography according to an embodiment. Referring to FIG. 10, an apparatus for reconstructing an image for parallel processing according to an exemplary embodiment (hereinafter, referred to as a “reconstruction apparatus”) 1000 may include a converter 1010, an execution unit 1030, and a reconstruction unit 1050. Include.

The conversion unit 1010 converts the response lines (Line-Of-Response; LOR) detected from the detectors into a sinogram format in the form of a cone beam in response to the radiation irradiated to the measurement object. do.

More specifically, the converter 1010 may convert the reaction lines into a cosine curve format in the form of a cone beam having a vertex of a crystal belonging to the first detector of the first region. At this time, the response lines converted to the cone-shaped cosine curve format appear in the second region where the second detector is located. In this case, the second detector may be in a position facing the first detector on a circle.

The execution unit 1030 performs back projection based on the converted response lines.

In this case, the execution unit 1030 may perform voxel-driven back projection based on the converted response lines.

The reconstruction unit 1050 reconstructs the image using the result of performing the reverse projection. The reconstruction unit 1050 may reconstruct an image using an LOR-based ordered-subset expectation maximization (OSEM) parallel algorithm based on the result of the reverse projection.

The reconstruction unit 1050 may reconstruct an image by using a gather operator.

The above-described methods may be embodied in the form of program instructions that can be executed by various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. This is possible.

Therefore, the scope of the present invention should not be construed as being limited to the embodiments described, but should be determined by equivalents to the appended claims, as well as the appended claims.

1000: reconstruction unit
1010: converter
1030: execution
1050: reconstruction unit

Claims (10)

Converting the line-of-response (LOR) detected from the detectors into a sinogram format in the form of a cone beam in response to the radiation irradiated to the measurement object;
Performing back projection based on the converted response lines; And
Reconstructing the image using the result of performing the reverse projection
Reconstructing the image for parallel processing comprising a. The method of claim 1,
The converting step
Converting the response lines into a cone beam shaped cosine curve format having a vertex of a crystal belonging to a first detector of a first region. The method of claim 1,
Performing the back projection is
Performing a voxel-driven back projection on the basis of the transformed response lines. The method of claim 1,
Reconstructing the image
And reconstructing the image for parallel processing using a LOR-based ordered-subset expectation maximization (OSEM) parallel algorithm based on a result of performing the reverse projection. The method of claim 1,
Reconstructing the image
Reconstructing the image for parallel processing, the step of reconstructing the image using a gather operator. A conversion unit for converting the line-of-response (LOR) detected from the detectors into a sinogram format in the form of a cone beam in response to the radiation irradiated to the measurement object;
An execution unit performing back projection based on the converted response lines; And
Reconstruction unit for reconstructing the image using the result of performing the reverse projection
Apparatus for reconstructing the image for parallel processing comprising a. The method according to claim 6,
The conversion unit
And converting the reaction lines into a cosine curve format in the form of a cone beam having a vertex of a crystal belonging to a first detector of a first region. The method according to claim 6,
The performing unit
And a voxel-driven back projection based on the converted response lines. The method according to claim 6,
The reconstruction unit
And reconstructing the image using a LOR-based ordered-subset expectation maximization (OSEM) parallel algorithm based on a result of performing the reverse projection. The method according to claim 6,
The reconstruction unit
An apparatus for reconstructing an image for parallel processing, wherein the image is reconstructed using a gather operator.

Images