JP4910979B2 - Nuclear medicine diagnostic apparatus and image processing method for nuclear medicine data - Google Patents

Description

The present invention relates to a nuclear medicine diagnostic apparatus for obtaining nuclear medicine data of a subject based on radiation generated from the subject to which a radiopharmaceutical is administered, and an image processing method of the nuclear medicine data .

  As the above-described nuclear medicine diagnosis apparatus, that is, an ECT (Emission Computed Tomography) apparatus, a PET (Positron Emission Tomography) apparatus will be described as an example. The PET apparatus detects a plurality of gamma rays generated by annihilation of positrons, that is, positrons, and reconstructs a tomographic image of a subject only when the gamma rays are simultaneously detected by a plurality of detectors. It is configured.

  Particle noise is mixed in the image of the projection data before reconstruction. Usually, the particle noise signal included in the particle noise is considered to be included in the high frequency component. A technique called “coring” is shown in which an original image is divided into a low-frequency component and a high-frequency component, and the value of the high-frequency component is reduced to “0” within a range of ± threshold (for example, a patent) Reference 1).

In addition, there are techniques that have been developed for application to multi-resolution wavelet transform and Laplacian pyramid representation (see, for example, Patent Documents 2 and 3 and Non-Patent Documents 1 to 3). Non-Patent Documents 1 to 3 described above correspond to a so-called “wavelet degeneration” technique in which high-pass (high-frequency band) components of wavelet transform coefficients are correlated by threshold processing or nonlinear threshold processing. In the above-mentioned Patent Documents 2 and 3, the core by non-linear threshold processing that gently treats the vicinity of the threshold for the high-pass component expressed in the Laplacian pyramid (the Laplacian component that is decomposed into Gaussian and Laplacian). Ring technology is disclosed.
US Pat. No. 4,523,230 US Pat. No. 5,467,404 US Pat. No. 5,805,721 JB Weaver, X. Yansun, DM Healy, Jr. and LD Cromwell, “Filtering Noise from Images with Wavelet Transforms”, Magnetic Resonance in Medicine, vol. 21, no. 2, pp.288-295, 1991. RA DeVor and BJ Lucier, “Fast wavelet techniques for near-optimal image processing”, IEEE Military Communications Conf. Rec. San Diego, Oct. 11-14, 1992, vol. 3, pp. 1129-1135. DL Donoho, “De-noising by soft-thresholding”, IEEE. Trans Inform. Theory, Vol. 41, pp. 613-627, 1995.

  However, when the conventional method for removing noise from high-frequency subbands subjected to multi-resolution conversion is applied to a digital image after actual reconstruction, there is a problem that protruding noise or streak-like noise remains. Such noise is particle noise that depends on the detector and depends on the Poisson distribution.

The present invention has been made in view of such circumstances, and provides a nuclear medicine diagnostic apparatus and an image processing method for nuclear medicine data capable of removing detector-dependent noise and Poisson distribution particle noise. The purpose is to do.

In order to achieve such an object, the present invention has the following configuration.
That is, the invention according to claim 1 is a nuclear medicine diagnostic apparatus for obtaining nuclear medicine data of a subject based on radiation generated from the subject to which a radiopharmaceutical is administered, wherein the radiation is detected to detect the radiation Detection means comprising each detector for detecting and collecting nuclear medicine data, and projection direction that is each projection position for each nuclear medicine data in the slice direction that is each detector position in the body axis direction Image processing means for performing image processing using three-dimensional sinogram data mapped in the channel direction which is a detector direction, and the image processing means includes (a) three two-dimensional sinogram data. In the two-dimensional space, the sinogram data of each two-dimensional space is decomposed into two or more two-dimensional spaces, and a composite having a sequentially lower resolution is compared with each decomposition. Multi-resolution sinogram dividing means for generating the low-frequency sinogram data of each of the two or more two-dimensional spaces, and a plurality of high-frequency sinogram data having successively lower resolution than each decomposition, and (b ) Noise removal processing means for performing noise removal processing on each of the two or more two-dimensional spaces for each of the low-frequency sinogram data and the high-frequency sinogram data; and (c) the noise-removed low frequency Sinogram data synthesis means for synthesizing the noise-removed sinogram data for each of the two or more two-dimensional spaces based on the sinogram data and the noise-removed high-frequency sinogram data, and (d) the synthesized two-phase data. Reconstruction processing means for performing reconstruction processing on sinogram data in a dimensional space It is characterized in.
The invention according to claim 2 is an image processing method for nuclear medical data of a subject based on radiation generated from a subject to which a radiopharmaceutical has been administered. When performing image processing using three-dimensional sinogram data mapped in the slice direction that is each detector position in the projection direction that is each projection position and the channel direction that is the detector direction, (A) the cubic Among the three two-dimensional spaces of the original sinogram data, the two-dimensional space sinogram data is decomposed for each two or more two-dimensional spaces, and sequentially has a lower resolution than each decomposition. A plurality of low-frequency sinogram data and a plurality of high-frequency sinogram data having successively lower resolution than each decomposition are divided into the two or more two-dimensional spaces. (B) noise removal processing is performed for each of the two or more two-dimensional spaces for each of the low-frequency sinogram data and the high-frequency sinogram data, and (C) the noise-removed low Based on the frequency sinogram data and the noise-removed high-frequency sinogram data, the noise-removed sinogram data is synthesized for each of the two or more two-dimensional spaces, and (D) a sinogram of the synthesized two-dimensional space. It is characterized in that a reconstruction process is performed on the data.

[Operation / Effect] In conventional multi-resolution conversion such as wavelet transform, two-dimensional space projection data is decomposed, multi-resolution conversion is performed, noise removal processing is performed, and then filtered back projection (FBP) is performed. : Reconstruction processing is performed using Filtered Back Projection), so noise is affected by convolution (convolution integration) filter characteristics, making it difficult to remove noise from image data with high accuracy. Therefore, it is considered that highly accurate noise removal can be performed by performing multi-resolution conversion using three-dimensional data. On the other hand, when normal xyz data is used as three-dimensional data, noise is hidden in the xyz data. Therefore, even if multiresolution conversion is performed using the xyz data, the noise cannot be completely removed. Therefore, according to the first aspect of the present invention, a three-dimensional sinogram mapped in the slice direction that is each detector position in the body axis direction, in the projection direction that is each projection position, and in the channel direction that is the detector direction. Multi-resolution conversion is performed using the data as three-dimensional data. Specifically, (a) the multi-resolution sinogram dividing means decomposes the sinogram data of each two-dimensional space into two or more two-dimensional spaces among the three two-dimensional spaces of the above-described three-dimensional sinogram data. Two or more low-frequency sinogram data having a sequentially lower resolution compared to each decomposition and a plurality of high-frequency sinogram data having a sequentially lower resolution compared to each decomposition. By generating for each two-dimensional space, multi-resolution conversion is performed for each of two or more two-dimensional spaces using three-dimensional sinogram data. Then, (b) the noise removal processing means performs noise removal processing for each of the two or more two-dimensional spaces for each of the above-described low frequency sinogram data and high frequency sinogram data. Further, (c) the sinogram data synthesizing means, based on the noise-removed low-frequency sinogram data and the noise-removed high-frequency sinogram data, the noise-removed sinogram data is divided into two or more two-dimensional spaces. Synthesize each. (D) The reconstruction processing means can perform high-accuracy noise removal by performing reconstruction processing on the combined sinogram data in the two-dimensional space, and detector-dependent noise and Poisson distribution. Particle noise can be removed.
According to the second aspect of the present invention, the detector-dependent noise and the Poisson distribution particle noise can be removed by performing the steps (A) to (D).

An example of an image processing method of the above-mentioned nuclear medicine data, depending wavelet transform to decompose the sinogram data of the above two-dimensional space, the above-mentioned low frequency sinogram data above the high-frequency sinogram data by generating each (Invention of claim 3 ) Further, an example of the above-described image processing method for nuclear medicine data is the noise-removed sinogram data based on the noise-removed low-frequency sinogram data and the noise-removed high-frequency sinogram data. The above-described two or more two-dimensional spaces are combined by inverse wavelet transform (the invention according to claim 4 ).

According to the nuclear medicine diagnostic apparatus of the present invention, it is possible to remove detector-dependent noise and Poisson distribution particle noise by providing the means (a) to (d).
According to the nuclear medicine data image processing method of the present invention, the detector-dependent noise and Poisson distribution particle noise can be removed by performing the steps (A) to (D).

Embodiment 1 of the present invention will be described below with reference to the drawings.
FIG. 1 is a side view and block diagram of a PET (Positron Emission Tomography) apparatus according to the first embodiment. In addition, this Example 1 including Example 2 mentioned later demonstrates taking a PET apparatus as an example as a nuclear medicine diagnostic apparatus.

  The PET apparatus according to the first embodiment includes a top plate 1 on which a subject M is placed as shown in FIG. The top plate 1 is configured to move up and down and translate along the body axis Z of the subject M. With this configuration, the subject M placed on the top 1 is scanned from the head to the abdomen and foot sequentially through the opening 2a of the gantry 2, which will be described later. Diagnostic data such as projection data and tomographic images are obtained. This diagnostic data corresponds to nuclear medicine data in the present invention.

  In addition to the top plate 1, the PET apparatus according to the first embodiment includes a gantry 2 having an opening 2a, a plurality of scintillator blocks (not shown) and a plurality of photomultipliers (not shown). The γ-ray detector 3 is provided. The γ-ray detector 3 is arranged in a ring shape so as to surround the body axis Z of the subject M, and is embedded in the gantry 2. The photomultiplier is disposed outside the scintillator block. As a specific arrangement of the scintillator blocks, for example, two scintillator blocks are arranged in a direction parallel to the body axis Z of the subject M, and many scintillator blocks are arranged around the body axis Z of the subject M. Can be mentioned. The γ-ray detector 3 collects emission data to be described later. The γ-ray detector 3 corresponds to the detection means in this invention.

  In addition, a point line source 4 and a γ-ray detector 5 for collecting absorption correction data (also referred to as “transmission data”) to be described later are provided. The γ-ray detector 5 for absorption correction data is composed of a scintillator block (not shown) and a photomultiplier (not shown) in the same manner as the γ-ray detector 3 for collecting emission data. The dotted line source 4 is a radiation source that irradiates a radioactive drug to be administered to the subject M, that is, a radiation of the same kind as the radioisotope (RI) (in this embodiment, γ rays), and is arranged outside the subject M. It is installed. The dotted line source 4 is embedded in the gantry 2. The dotted line source 4 rotates around the body axis Z of the subject M.

  In addition, the PET apparatus according to the first embodiment includes a top plate driving unit 6, a controller 7, an input unit 8, an output unit 9, a projection data deriving unit 10, an absorption correction data deriving unit 11, an absorption correcting unit 12, and an image. A processing unit 13 and a memory unit 14 are provided. The top plate driving unit 6 is a mechanism for driving the top plate 1 so as to perform the above-described movement, and is configured by a motor or the like not shown.

  The controller 7 comprehensively controls each part constituting the PET apparatus according to the first embodiment. The controller 7 includes a central processing unit (CPU).

  The input unit 8 sends data and commands input by the operator to the controller 7. The input unit 8 includes a pointing device represented by a mouse, a keyboard, a joystick, a trackball, a touch panel, and the like. The output unit 9 includes a display unit represented by a monitor, a printer, and the like.

  The memory unit 14 is configured by a storage medium represented by ROM (Read-only Memory), RAM (Random-Access Memory), and the like. In the first embodiment, the diagnostic data processed by the projection data deriving unit 10 and the image processing unit 13 and the absorption correction data obtained by the absorption correction data deriving unit 11 are written and stored in the RAM, and if necessary. Read from RAM. The ROM stores in advance a program for performing various nuclear medicine diagnoses, and the controller 7 executes the program to perform nuclear medicine diagnosis according to the program.

  The projection data deriving unit 10, the absorption correction data deriving unit 11, the absorption correcting unit 12, and the image processing unit 13 are, for example, a program stored in a ROM of a storage medium represented by the above-described memory unit 14 or the input unit 8 or the like. This is realized by the controller 7 executing a command input by a pointing device represented by.

  The γ-rays generated from the subject M to which the radiopharmaceutical is administered are converted into light by the scintillator block of the γ-ray detector 3, and the converted light is photoelectrically converted by the photomultiplier of the γ-ray detector 3. Output to electrical signal. The electric signal is sent to the projection data deriving unit 10 as image information (pixel).

  Specifically, when a radiopharmaceutical is administered to the subject M, two γ rays are generated due to the disappearance of the positron of the positron emission type RI. The projection data deriving unit 10 checks the position of the scintillator block and the incident timing of the γ-ray, and is sent only when the γ-ray is simultaneously incident on two scintillator blocks that are opposed to each other across the subject M. The image information is determined as appropriate data. When γ rays are incident only on one scintillator block, the projection data deriving unit 10 treats the image information sent at that time as noise instead of γ rays generated by annihilation of the positron. Dismiss.

  The image information sent to the projection data deriving unit 10 is sent to the absorption correction unit 12 as projection data (also referred to as “emission data”). Absorption correction data (transmission data) sent from the absorption correction data derivation unit 11 to the absorption correction unit 12 is applied to the projection data sent to the absorption correction unit 12 to absorb γ rays in the body of the subject M. The projection data is corrected in consideration of the above.

  The point source 4 emits γ rays toward the subject M while rotating around the body axis Z of the subject M, and the irradiated γ rays are used as a scintillator block of the γ ray detector 5 for absorption correction data. (Not shown) converts it into light, and the converted light is photoelectrically converted by a photomultiplier (not shown) of the γ-ray detector 5 and output to an electrical signal. The electric signal is sent to the absorption correction data deriving unit 11 as image information (pixel).

  Absorption correction data is obtained based on the image information sent to the absorption correction data deriving unit 11. The absorption correction data deriving unit 11 uses the calculation representing the relationship between the absorption coefficient of γ-rays or X-rays and the energy, so that the projection data for CT, that is, the distribution data of the X-ray absorption coefficient is converted into the γ-ray absorption coefficient. By converting to distribution data, the distribution data of the γ-ray absorption coefficient is obtained as absorption correction data. The derived absorption correction data is sent to the absorption correction unit 12 described above.

  The corrected projection data is sent to the image processing unit 13. The image processing unit 13 reconstructs the projection data to obtain a tomographic image taking into account the absorption of γ rays in the body of the subject M. Thus, by providing the absorption correction unit 12 and the image processing unit 13, the projection data is corrected based on the absorption correction data, and the tomographic image is corrected. The corrected tomographic image is sent to the output unit 9 and the memory unit 14 via the controller 7. The image processing unit 13 corresponds to the image processing means in this invention.

  Next, specific functions of the image processing unit 13 will be described with reference to FIG. FIG. 2 is a block diagram of the image processing unit 13 and its peripheral part. The image processing unit 13 includes a wavelet transformation unit 13a, a noise removal processing unit 13b, an inverse wavelet transformation unit 13c, and a reconstruction processing unit 13d.

  The wavelet transform unit 13a performs wavelet transform on each two-dimensional space sinogram data (axial image and coronal image in the first embodiment). Specifically, with respect to each projection data absorption-corrected by the absorption correction unit 12, the slice direction z (that is, the body axis Z and the slice direction z are the same) that is each detector position in the body axis Z direction (FIG. 6). 3) are collected as three-dimensional sinogram data mapped in the projection direction s (see FIG. 6) that is each projection position and the channel direction r (see FIG. 6) that is the detector direction. Of the three two-dimensional spaces (axial, coronal, sagittal) of the three-dimensional sinogram data, the wavelet transform unit 13a has two each of two or more two-dimensional spaces (axial, coronal in the first embodiment). Wavelet transform is performed on sinogram data in the dimensional space.

  The noise removal processing unit 13b performs the noise removal processing on each of the low-frequency sinogram data and the high-frequency sinogram data wavelet transformed by the wavelet transformation unit 13a. , Each coronal). The inverse wavelet transform unit 13c performs inverse wavelet transform on the noise-removed low-frequency sinogram data and the noise-removed high-frequency sinogram data. The reconstruction processing unit 13d performs reconstruction processing on sinogram data in a two-dimensional space that has been subjected to inverse wavelet transform.

  The wavelet transform unit 13a corresponds to the multi-resolution sinogram dividing unit in the present invention, the noise removal processing unit 13b corresponds to the noise removal processing unit in the present invention, and the inverse wavelet transform unit 13c is the sinogram data synthesis unit in the present invention. The reconstruction processing unit 13d corresponds to the reconstruction processing means in this invention.

  Next, specific image processing by the image processing unit 13 will be described with reference to FIGS. FIG. 3 is a flowchart illustrating a flow of a series of image processing according to the first embodiment, FIG. 4 is a flowchart illustrating a flow of image processing with a series of axial images according to the first embodiment, and FIG. FIG. 6 is a flowchart showing a flow of image processing with a series of coronal images according to Embodiment 1, FIG. 6 is a diagram schematically showing a flow of image processing with axial images and coronal images, and FIG. FIG. 8 is a diagram showing a Daubechie sequence, FIG. 9 is a schematic image of wavelet transformation in a one-dimensional space, and FIG. FIG. 11 is a schematic image of wavelet transform in a two-dimensional space, and FIG. 11 shows a wavelet transform when the wavelet transform in a two-dimensional space is applied to the first embodiment. FIG. 12 is a schematic diagram for explaining “LL”, “HH”, and “LH”, and FIG. 13 is a diagram of noise removal processing by threshold processing. It is a schematic diagram with which it uses for description.

(Step S1) Image Processing with Axial Image Each projection data subjected to the absorption correction by the absorption correction unit 12 is converted into a three-dimensional data set by Fourier rebinning processing for converting from the projection plane format to the sinogram format. In this three-dimensional data set, a plurality of pieces of two-dimensional sinogram data mapped in the projection direction s (see FIG. 6) and the channel direction r (see FIG. 6) in the slice direction z (ie, the body axis Z). Will be collected.

  In this specification, as shown in FIG. 6A, a two-dimensional space composed of a projection direction s and a channel direction r is defined as an axial (rs plane), and is mapped to the projection direction s and the channel direction r. The obtained two-dimensional sinogram data is used as an axial image. Therefore, for each projection data subjected to the absorption correction by the absorption correction unit 12, an axial image is collected for a plurality of sheets in the slice direction z. The image processing unit 13 performs various types of image processing using the absorption-corrected axial image. This step S1 includes the following steps T1 to T4.

  Further, in this specification, as shown in FIG. 6B, a two-dimensional space constituted by a slice direction z (ie, body axis Z) and a channel direction r is defined as a coronal (rz plane), and the slice Two-dimensional sinogram data mapped in the direction z and the channel direction r is defined as a coronal image. Although not used in the first embodiment, a two-dimensional space composed of the slice direction z (that is, the body axis Z) and the projection direction s is defined as sagittal (sz plane), and the slice direction z and the projection direction s. The two-dimensional sinogram data mapped to is defined as a sagittal image.

(Step T1) Wavelet Transform The wavelet transform unit 13a performs wavelet transform on the axial image after the absorption correction for each slice direction. Specifically, the axial image in one slice direction is decomposed, and a plurality of low-frequency sinogram data (axial image in step S1) having sequentially lower resolution than each decomposition, and for each decomposition In comparison, a plurality of high-frequency sinogram data (axial images in step S1) having successively lower resolutions are generated.

  Multi-resolution transformation is applied in wavelet transformation, including the case where wavelet transformation is performed on a coronal image, which will be described later [for example, C MAGAZINE, “40th Algorithm Lab for Image Processing Extremely Compression Processing by Wavelet Transformation (1) Basics of Wavelet Transform ”, [online], Niigata University School of Medicine, Internet <URL: http://www.clg.niigata-u.ac.jp/~tsai/home-page/lecture/Wavelet-Comp1.pdf > (Non-patent literature)]. This multi-resolution conversion is as follows. Note that description of the continuous wavelet transform is omitted.

[Multi-resolution analysis]
First, multiresolution analysis will be described before describing wavelet transform. Multiresolution analysis represents a function as the sum of scaling functions φ (t). Here, as the simplest scaling function, there is a Haar scaling function φ (t) defined as the following equation (1), as shown in FIG.

The Haar scaling function φ (t) is a rectangular pulse wave, and when this scaling function φ (t) is used, an arbitrary function f (t) can be approximated as f _j (t). Such a Haar scaling function φ (t) translated and reduced / enlarged on the t-axis is expressed by the following equation (2).

In the above equation (2), 2 ^j is a scale parameter representing enlargement / reduction, and k is a shift parameter representing translation. An approximation of level j using φ _{j, k} is defined as the following equation (3) using s _k ^(j) . At this time, s _k ^(j) is called a “scaling coefficient”.

In the above equation (3), since the width of the scaling function increases as j, which is the number of levels, the approximation becomes rougher as the number of levels increases and becomes far from the original function f (t). That is, f _j (t) lacks information compared to f _j−1 (t). If this missing portion is g _j (t), the following equation (4) is established.

This g _j (t) is called a “level j wavelet component”. This g _j (t) is a rectangle that vibrates positively and negatively, and its constituent elements can be expressed by a function similar to the Haar scaling function φ (t). This function is called a “Haar wavelet function” and is represented by the following equation (5) as shown in FIG.

  This Haar wavelet function ψ (t) is also a rectangular pulse wave oscillating in positive and negative symmetry, and the Haar wavelet function ψ (t) is translated and reduced / enlarged on the t-axis. It is expressed as an expression.

Similarly to φ _{j, k} , in the above equation (6), 2 ^j is a scale parameter representing enlargement / reduction, and k is a shift parameter representing translation. When this ψ _{j, k} is used, g _j (t) described above is expressed by the following equation (7). At this time, ω _k ^{(j) is referred} to as a “level j wavelet expansion coefficient”.

  The above formulas (3), (4) and (7) can be summarized as the following formula (8).

From the above equation (8), the approximate function f _j-1 (t) of the level ( _j−1 ) is expressed as the function g _j (t) expressed by the approximate function f _j (t) of the level j and the wavelet expansion coefficient of the level j. ) (That is, wavelet components at level j). When this is used repeatedly, the approximation function f ₀ (t) at level “0” is expressed as the following equation (9) using approximation functions at levels 1, 2,.

Thus, the function f ₀ (t) can be expressed by the sum of wavelet components having different widths (that is, different resolutions). Such an analysis method is called “multi-resolution analysis”.

[Two-scale relationship]
Actually, the shape of the function may be simple and discontinuous, and the processing using the Haar function is not practical. Actually, in order to be a practical scaling function, it is necessary to satisfy a relational expression such as the following expression (10).

  The above equation (10) shows that a scaling function with a certain level can derive a scaling function with a smaller number of levels. This relationship is called “two-scale relationship”. The two-scale relationship in the Haar scaling function is shown in the following equation (11).

  However, in order to satisfy the relationship represented by the above equation (11), not only the Haar scaling function but also the Haar wavelet needs to be derived from a scaling relationship having a lower number of levels. That is, even in the case of Haar wavelets, it is necessary to establish a relationship such as the above equation (10). The two-scale relationship in the Haar wavelet is shown in the following equation (12).

[Daubechie wavelet]
As a function that satisfies the relationship expressed by the above equation (12) and is used for multiresolution analysis, there is a Daubechie wavelet. Daubechie's wavelet and scaling functions are shaped like a triangular wave, but they are very complex and cannot be expressed with known functions. In the case of using the wavelet Daubechie performs calculation using the sequence p _n, q _n of to-scale relationship as shown in FIG. In FIG. 8, N = 2 to 4 are shown, but the value of N is not particularly limited.

[Discrete wavelet expansion coefficient]
A scaling coefficient s _k ^(j) and a wavelet expansion coefficient ω _k ^(j) are _obtained using the sequences p _n and q _n of FIG. The scaling coefficient s _k ^(j) is obtained by the above equation (3). When this equation is transformed using the two-scale relationship, it is expressed as the following equation (13).

  By using the above equation (13), a scaling coefficient having a large number of levels (that is, low resolution) can be obtained from a scaling coefficient having a small number of levels (that is, high resolution). Similarly, the derivation formula for obtaining the wavelet expansion coefficient is as shown in the following formula (14).

  On the contrary, in order to obtain a scaling coefficient with a small number of levels from a scaling coefficient with a large number of levels, the following equation (15) is used. This is an inverse wavelet transform.

[High-speed wavelet transform]
When performing a wavelet transform with a large number of levels, the method of calculating according to the definition formula is the simplest. On the other hand, there is a method using the result of wavelet transform with a small number of levels. In this case, of the wavelet transform with a small number of levels, only the portion s is wavelet transformed to obtain a wavelet result with a large number of levels. Further, this conversion is repeated to increase the number of levels. An image of this processing is shown in FIG. In the case of inverse wavelet transform, processing is performed in the direction opposite to the arrow in FIG. In order to perform this processing, it is necessary that the processing result can be divided into two equal parts, and basically the data size needs to be 2 to the Nth power. Only when this condition is satisfied, the wavelet transform can be performed with a smaller amount of calculation than that calculated according to the definition formula, and this processing is called “high-speed wavelet transform”.

[Extension of discrete wavelet transform to two dimensions]
Since image data (in the first embodiment, an axial image or a coronal image described later) has a two-dimensional space spread, the wavelet transform equation shown in the above equations (13) and (14) is used to analyze this. Needs to be extended to two dimensions. That is, wavelet transformation is performed for one direction, and the result is transformed again from another direction, thereby completing a two-dimensional discrete wavelet transformation. This image is shown in FIG. First, the result of performing wavelet transform in one direction is as shown in FIG. If this is wavelet transformed from another direction, the result is as shown in FIG. In FIG. 10 (c), the upper left shows a low frequency component, the lower left shows a horizontal high frequency component, the upper right shows a vertical high frequency component, and the lower right shows a diagonal high frequency component.

  The above is an explanation of general wavelet transform. When this wavelet transform is applied to the axial image of the first embodiment, the image of FIG. 10 becomes as shown in FIG. That is, wavelet transform is performed on one direction of the axial image (for example, the projection direction s, the channel direction r, or any one direction in the rs plane), and the result is converted to another direction (for example, the channel direction r orthogonal to the projection direction s). 11 (a), the axial image shown in FIG. 11 (a) is decomposed by converting again from the projection direction s orthogonal to the channel direction r or the direction orthogonal to any one direction in the rs plane. One low-frequency sinogram data and three high-frequency sinogram data shown in FIG.

  11B and 11C, the horizontal direction is L as shown in FIG. 12, and the vertical direction is H as shown in FIG. Moreover, in FIG.11 (b), FIG.11 (c), and FIG. 12, the direction which performs wavelet transformation is shown by hatching. An image after the wavelet transform in the horizontal direction L is “LL”, and an image after the wavelet transform in the vertical direction H is “HH”. Further, an image after wavelet transform in the diagonal direction is assumed to be “LH”. Then, each image after the first wavelet transform has one low frequency sinogram data (not hatched in FIG. 11B) and high frequency sinogram data LL, HH, LH as shown in FIG. Is generated.

  Similar to the wavelet transform for the axial image shown in FIG. 11A, wavelet transform is performed on one direction of the low-frequency sinogram data shown in FIG. 11B, and the result is converted again from another direction. The low-frequency sinogram data shown in FIG. 11 (b) is decomposed to generate one low-frequency sinogram data and three high-frequency sinogram data shown in FIG. 11 (c). Similarly to FIG. 11 (b), each image after the second wavelet transform also includes one low-frequency sinogram data (no hatching in FIG. 11 (c)) and a high-frequency sinogram as shown in FIG. 11 (c). Data LL, HH, and LH are generated.

  In this way, the wavelet transform is repeated a plurality of times to decompose the axial image, and a plurality of low-frequency sinogram data having sequentially lower resolution compared to each decomposition, and sequentially compared to each decomposition. A plurality of high-frequency sinogram data having a low resolution.

(Step T2) Noise Removal Processing The noise removal processing unit 13b performs noise removal processing on each of the low-frequency sinogram data and high-frequency sinogram data that have been wavelet transformed in Step T1. Specifically, a wavelet degeneration method is used in which high-pass components (high-frequency components) of wavelet transform coefficients (scaling coefficients) are correlated by threshold processing or nonlinear threshold processing. For example, when noise removal processing is performed on the high-frequency sinogram data LL, the LL space shown in FIG. 12 is expanded one-dimensionally as shown in FIG. The noise removal processing is performed by dropping it to "". The same applies to the high-frequency sinogram data HH and the high-frequency sinogram data LH. For the purpose of reducing the noise in the diagonal direction (diagonal direction), the threshold value in the LH space is set by giving a large weight to the threshold value in the LH space.

(Step T3) Inverse Wavelet Transform The inverse wavelet transform unit 13c performs inverse wavelet transform on the low-frequency sinogram data from which noise has been removed in step T2 and the high-frequency sinogram data from which noise has also been removed. Specifically, the inverse wavelet transform is performed by synthesizing into one axial image by substituting into the above equation (15).

(Step T4) All slices?
By performing the processing in steps T1 to T3 in this way, wavelet transformation, noise removal processing, and inverse wavelet transformation are performed on an axial image in one slice direction. Then, it is determined whether or not the processing in these steps T1 to T3 has been performed on all slices. If all slices have been performed, assuming that wavelet transformation, noise removal processing, and inverse wavelet transformation are all performed on the axial image for each slice direction, step S1 including steps T1 to T4 is completed. The process proceeds to the next step S2. If not performed in all slices, the process returns to step T1 and performs the processes in steps T1 to T3 in order to perform wavelet transform, noise removal processing, and inverse wavelet transform on the axial image in the next slice direction.

  In this way, when it is determined in step T4 that the processing has been performed for all slices, an axial image subjected to noise removal processing is obtained. FIG. 6A shows an axial image before the noise removal process. As shown in FIG. 6B, the noise removal process is performed after the process of step S1 (T1, T2, T3, T4). A later axial image is obtained. The axial image after this noise removal processing is used for image processing with a coronal image in the next step S2. Therefore, FIG. 6B is a coronal image before noise removal processing using an axial image after noise removal processing.

(Step S2) Image Processing with Coronal Image The axial image for each slice direction z after the noise removal processing that has been subjected to the image processing in step S1 is made a coronal image for each projection direction s, and an image processing unit is used by using this coronal image. 13 performs various image processing. Step S2 includes the following steps U1 to U4.

(Step U1) Wavelet Transform The wavelet transform unit 13a performs wavelet transform on the coronal image for each projection direction s. Specifically, a coronal image in one projection direction is decomposed, a plurality of low frequency sinogram data (coronal image in step S2) having sequentially lower resolution than each decomposition, and each decomposition. In comparison, a plurality of high-frequency sinogram data (coronal images in step S2) having successively lower resolutions are generated. Since the specific wavelet transform has been described in step T1, the description thereof will be omitted.

(Step U2) Noise Removal Processing The noise removal processing unit 13b performs noise removal processing on each of the low frequency sinogram data and the high frequency sinogram data that have been wavelet transformed in Step U1. Since the specific noise removal processing has been described in step T2, the description thereof will be omitted.

(Step U3) Inverse Wavelet Transformation The inverse wavelet transformation unit 13c performs inverse wavelet transformation on the low-frequency sinogram data from which noise has been removed in step U2 and the high-frequency sinogram data from which noise has been removed. Since the specific inverse wavelet transform has been described in step T3, the description thereof is omitted.

(Step U4) All projections?
By performing the processing of steps U1 to U3 in this way, wavelet transformation, noise removal processing, and inverse wavelet transformation for a coronal image in one projection direction are performed. Then, it is determined whether or not the processing in these steps U1 to U3 has been performed for all projections. If all projections have been performed, assuming that wavelet transformation, noise removal processing, and inverse wavelet transformation for the coronal image have been performed for each projection, step S2 consisting of steps U1 to U4 is terminated. Proceed to the next Step S3. If not performed in all projections, the process returns to step U1 to perform the processes in steps U1 to U3 in order to perform wavelet transformation, noise removal processing, and inverse wavelet transformation on the coronal image in the next projection direction.

  As described above, when it is determined in step U4 that the processing has been performed on all slices, a coronal image subjected to noise removal processing is obtained. FIG. 6C shows an axial image after the noise removal process using the coronal image after the noise removal process (see “Denoized sinogram” in FIG. 6C), and step S2 (U1, U2, U2). Through the processing of U3, U4), as shown in FIG. 6C, a coronal image after noise removal processing is obtained.

(Step S3) Reconstruction Processing The coronal images for each projection direction s after the noise removal processing subjected to image processing in step S2 are bundled in a three-dimensional space, and the reconstruction processing unit 13d performs reconstruction processing to obtain a tomographic image. . The reconstruction process may be performed using a well-known filtered back projection (FBP) (also called “filtered back projection method”).

  In multi-resolution conversion represented by the conventional wavelet transform, etc., two-dimensional space projection data is decomposed, multi-resolution conversion is performed, noise removal processing is performed, and then reconstruction processing is performed using filtered back projection. Therefore, the noise is affected by the convolution (convolution integration) filter characteristics, making it difficult to remove noise on the image data with high accuracy. Therefore, it is considered that highly accurate noise removal can be performed by performing multi-resolution conversion using three-dimensional data. On the other hand, when normal xyz data is used as three-dimensional data, noise is hidden in the xyz data. Therefore, even if multiresolution conversion is performed using the xyz data, the noise cannot be completely removed.

  Therefore, according to the PET apparatus according to the first embodiment having the above-described configuration, the projection direction s and the detector direction are the projection positions in the slice direction z that is the detector position in the body axis Z direction. Multi-resolution conversion is performed using three-dimensional sinogram data mapped in the channel direction r as three-dimensional data.

  Specifically, (a) the multi-resolution sinogram dividing means (the wavelet transform unit 13a in the first embodiment) is 2 out of the above three two-dimensional spaces (axial, coronal, sagittal) of the three-dimensional sinogram data. A plurality of low-frequency sinograms having a resolution that is sequentially lower than each decomposition is decomposed for each two or more two-dimensional spaces (axial and coronal in the first embodiment). Two or more using three-dimensional sinogram data by generating data and a plurality of high-frequency sinogram data having successively lower resolution than each decomposition for each two or more two-dimensional spaces Multi-resolution conversion is performed for each two-dimensional space.

  Then, (b) the noise removal processing unit 13b performs noise removal processing for each of the two or more two-dimensional spaces for each of the above-described low frequency sinogram data and high frequency sinogram data. Further, (c) the sinogram data synthesizing means (inverse wavelet transform unit 13c in the first embodiment) is denoised based on the above-described low-frequency sinogram data from which noise has been removed and high-frequency sinogram data from which noise has also been removed. The sinogram data is synthesized for each of two or more two-dimensional spaces. Then, (d) the reconstruction processing unit 13d can perform highly accurate noise removal by performing reconstruction processing on the combined sinogram data of the two-dimensional space, and detector-dependent noise and Poisson. Distribution particle noise can be removed.

  In the first embodiment, the multi-resolution sinogram dividing means in this invention is wavelet transform, and the sinogram data synthesizing means in this invention is inverse wavelet transform.

Next, Embodiment 2 of the present invention will be described with reference to the drawings.
FIG. 14 is a side view and block diagram of a PET-CT apparatus including a PET apparatus and an X-ray CT apparatus according to the second embodiment.

  In the first embodiment described above, the PET apparatus includes the point source 4, the point source 4 emits the same γ-ray as the radiopharmaceutical and passes through the subject M, and the γ-ray detector 5 detects the γ-ray. Thus, transmission data was collected as morphological information based on the radiation, but in the second embodiment, CT projection data obtained from an X-ray CT apparatus is used as morphological information. The X-ray CT apparatus includes a gantry 31 having an opening 31a, an X-ray tube 32, and an X-ray detector 33. The X-ray tube 32 and the X-ray detector 33 are arranged to face each other with the subject M interposed therebetween, and are embedded in the gantry 31. A large number of detection elements constituting the X-ray detector 33 are arranged in a fan shape around the body axis Z of the subject M.

  In addition, the X-ray CT apparatus includes a gantry driving unit 34, a high voltage generation unit 35, a collimator driving unit 36, and a CT reconstruction unit 37. The CT reconstruction unit 37 is realized by the controller 7 executing, for example, a program stored in a ROM of a storage medium represented by the above-described memory unit 14 or the like or an instruction input from the input unit 8. Note that the CT projection data and the CT tomographic image processed by the CT reconstruction unit 37, which will be described later, also maintain the facing relationship between the memory unit 14 and the gantry driving unit 34 as in the first embodiment. A mechanism that drives the X-ray tube 32 and the X-ray tube detector 33 to rotate around the body axis Z of the subject M in the gantry 31 with a motor or the like not shown. .

  The high voltage generator 35 generates a tube voltage or tube current of the X-ray tube 32. The collimator driving unit 36 is a mechanism that sets an X-ray irradiation field and drives the collimator (not shown) adjacent to the X-ray tube 32 to move in the horizontal direction. It consists of

  In the case of the indirect conversion type X-ray detector 33, the X-ray irradiated from the X-ray tube 32 and transmitted through the subject M is converted into light by a scintillator (not shown) in the X-ray detector 33. The light-sensitive film (not shown) photoelectrically converts the converted light and outputs it as an electrical signal. In the case of the direct conversion type X-ray detector 33, the X-ray is directly converted into an electric signal by a radiation sensitive film (not shown) and output. The electrical signal is sent to the CT reconstruction unit 37 as image information (pixels). The image information sent to the CT reconstruction unit 37 is transmitted as projection data for CT.

  The projection data for CT has the same form information as the transmission data described in the first embodiment. In the second embodiment, the absorption correction data deriving unit is used to use the projection data for CT as transmission data. 11 and the CT reconstruction unit 37 as well.

  Image information (CT projection data) sent to the CT reconstruction unit 37 is reconstructed to obtain a CT tomographic image. This CT tomographic image is sent to the output unit 9 via the controller 7.

  The functions of the subsequent processing units (the absorption correction unit 12 and the image processing unit 13) including the absorption correction data deriving unit 11 are the same as those in the first embodiment, and thus the description thereof is omitted. The reconstructed tomographic image for PET and the CT tomographic image reconstructed by the CT reconstruction unit 37 may be superimposed and output by the output unit 9.

  As described above, in the second embodiment, the projection data for CT obtained by being detected by the X-ray detector 33 of the X-ray CT apparatus is sent to the CT reconstruction unit 37, and is also sent to the absorption correction data deriving unit 11. Infeed and CT projection data are used as absorption correction data.

  The present invention is not limited to the above-described embodiment, and can be modified as follows.

  (1) In each of the above-described embodiments, the PET apparatus has been described as an example. However, the present invention detects a single gamma ray and reconstructs a tomographic image of the subject. It can also be applied to devices.

  (2) In each of the above-described embodiments, the γ-ray detector 5 for transmission is a stationary type that detects γ-rays while still, but the γ-ray detector 5 rotates around the subject M. A rotary type that detects γ rays may be used.

  (3) In each of the above-described embodiments, the form information (transmission data in the first embodiment and X-ray CT data in the second embodiment) is also used for the absorption correction. However, it is not always necessary to perform the absorption correction. . Therefore, projection data before absorption correction may be collected as three-dimensional sinogram data mapped in the slice direction z, in the projection direction s, and in the channel direction r.

  (4) In each of the embodiments described above, after performing image processing with an axial image, image processing with a coronal image is performed using the axial image that has been subjected to noise removal processing with the image processing. The noise-removed coronal image was obtained, but conversely, after performing image processing with the coronal image, using the coronal image subjected to noise removal processing with the image processing, performing image processing with the axial image, An axial image from which noise has been removed by the image processing may be obtained. Accordingly, the order of image processing in (a) to (c) in each two-dimensional space is not particularly limited.

  (5) In each of the above-described embodiments, an axial image and a coronal image have been described as examples. However, the present invention may be applied to an axial image and a sagittal image, or a coronal image and a sagittal image. That is, after performing image processing with an axial image, using the axial image subjected to noise removal processing with the image processing, performing image processing with a sagittal image, the sagittal image subjected to noise removal processing with the image processing is processed. After performing image processing with a coronal image, the image processing with a sagittal image is performed using the coronal image that has been subjected to noise removal processing with the image processing, and noise removal processing is performed with the image processing. Sagittal images may be obtained. Conversely, after performing image processing with a sagittal image, the sagittal image that has been subjected to noise removal processing by the image processing is used to perform image processing with an axial image, and the axial image that has undergone noise removal through the image processing is processed. After performing image processing with a sagittal image, image processing with a coronal image was performed using a sagittal image that had been subjected to noise removal processing with the image processing, and noise was removed with the image processing. A coronal image may be obtained. Also in this modification (5), the order of image processing in (a) to (c) in each two-dimensional space is not particularly limited.

  (6) In each of the above-described embodiments, out of three two-dimensional spaces (axial, coronal, sagittal) of three-dimensional sinogram data, two (2) two-dimensional spaces (axial, coronal in the first embodiment) are (a) Although the image processing in (c) is performed, the image processing in (a) to (c) may be performed for each of the axial, coronal, and sagittal using all three two-dimensional spaces. Therefore, as long as the image processing in (a) to (c) is performed for each of two or more two-dimensional spaces, the combination of the two-dimensional spaces is not particularly limited. Also in the modification (6), the order of image processing in (a) to (c) in each two-dimensional space is not particularly limited.

  (7) In each of the embodiments described above, the wavelet transform is taken as an example of the multi-resolution sinogram dividing means in the present invention, and the inverse wavelet transform is taken as an example as the sinogram data synthesizing means in the present invention. The multi-resolution sinogram dividing means used in the present invention is not limited to the wavelet transform, and the sinogram data synthesizing means normally used is not limited to the inverse wavelet transform.

1 is a side view and block diagram of a PET (Positron Emission Tomography) apparatus according to Embodiment 1. FIG. It is a block diagram of an image processing part and its peripheral part. 3 is a flowchart illustrating a flow of a series of image processing according to the first embodiment. 3 is a flowchart illustrating a flow of image processing with a series of axial images according to the first embodiment. 6 is a flowchart illustrating a flow of image processing with a series of coronal images according to the first embodiment. It is the figure which showed typically the flow of the image processing in an axial image and a coronal image, (a) is an axial image before a noise removal process, (b) is the noise removal using the axial image after a noise removal process A coronal image before processing, (c) is an axial image after noise removal processing using the coronal image after noise removal processing. (A) is a Haar scaling function, and (b) is a Haar wavelet function. It is the figure which showed the Daubechie number sequence. It is a schematic image of wavelet transform in a one-dimensional space. (A)-(c) is a schematic image of the wavelet transformation in a two-dimensional space. (A)-(c) is the figure which showed typically each image in wavelet transformation when the wavelet transformation in two-dimensional space is applied to Example 1. FIG. It is a schematic diagram for description of “LL”, “HH”, and “LH”. It is a schematic diagram with which it uses for description of the noise removal process by threshold value process. It is the side view and block diagram of PET-CT apparatus provided with the PET apparatus and X-ray CT apparatus which concern on Example 2. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 3 ... Gamma ray detector 13 ... Image processing part 13a ... Wavelet transformation part 13b ... Noise removal processing part 13c ... Inverse wavelet transformation part 13d ... Reconstruction processing part z ... Slice direction s ... Projection direction r ... Channel direction M ... Subject

Claims (4)

  A nuclear medicine diagnostic apparatus for obtaining nuclear medicine data of a subject based on radiation generated from a subject to which a radiopharmaceutical is administered, each detecting and collecting the nuclear medicine data by detecting the radiation The detection means comprising the detectors and the nuclear medicine data are mapped in the slice direction, which is each detector position in the body axis direction, in the projection direction, which is each projection position, and in the channel direction, which is the detector direction. Image processing means for performing image processing using the three-dimensional sinogram data, wherein the image processing means is (a) two or more two-dimensional of the three two-dimensional spaces of the three-dimensional sinogram data. Each of the two-dimensional space sinogram data is decomposed for each space, and a plurality of low frequency sinogram data having sequentially lower resolution than each decomposition, and each Multi-resolution sinogram dividing means for generating a plurality of high-frequency sinogram data having successively lower resolutions for each solution for each of the two or more two-dimensional spaces; (b) the low-frequency sinogram data and the Noise removal processing means for performing noise removal processing on each of the two or more two-dimensional spaces with respect to each of the high-frequency sinogram data, and (c) the noise-removed low-frequency sinogram data and the noise-removed Sinogram data synthesizing means for synthesizing the noise-removed sinogram data for each of the two or more two-dimensional spaces based on the high-frequency sinogram data, and (d) for the synthesized two-dimensional space sinogram data A nuclear medicine diagnosis apparatus comprising a reconstruction processing means for performing a reconstruction process.   A method for image processing of nuclear medical data of a subject based on radiation generated from a subject to which a radiopharmaceutical is administered, wherein the nuclear medical data is sliced at each detector position in the body axis direction, When performing image processing using three-dimensional sinogram data mapped in the projection direction that is each projection position and the channel direction that is the detector direction, (A) in the three two-dimensional spaces of the three-dimensional sinogram data, Among the two or more two-dimensional spaces, each of the two-dimensional space sinogram data is decomposed, and a plurality of low-frequency sinogram data having lower resolution sequentially than each decomposition, and each decomposition A plurality of high-frequency sinogram data having successively lower resolution than each of the two or more two-dimensional spaces, (B) A noise removal process is performed on each of the two or more two-dimensional spaces for each of the wave sinogram data and the high frequency sinogram data, and (C) the low frequency sinogram data from which the noise has been removed and the noise removed Based on the high-frequency sinogram data, the noise-removed sinogram data is synthesized for each of the two or more two-dimensional spaces, and (D) reconstructing processing is performed on the synthesized two-dimensional space sinogram data. An image processing method for nuclear medicine data. Wherein in the image processing method for nuclear medicine data according to claim 2, the wavelet transform decomposes the sinogram data of the two-dimensional space, to generate the low frequency sinogram data the high-frequency sinogram data and respectively An image processing method for nuclear medicine data . 4. The image processing method for nuclear medicine data according to claim 2 , wherein noise-removed sinogram data based on the noise-removed low-frequency sinogram data and the noise-removed high-frequency sinogram data. An image processing method for nuclear medicine data , wherein each of the two or more two-dimensional spaces is synthesized by inverse wavelet transform.

Images