A quantitative reconstruction method for nuclear medicine images

Abstract

The present invention proposes a method for quantitative reconstruction of nuclear medicine images, comprising the following steps: obtaining partial information in a CT image as prior information in a reconstruction algorithm; reconstructing the prior information transformed from the CT image as input information of a statistical iterative reconstruction algorithm to obtain a SPECT reconstructed image; and transforming the SPECT reconstructed image by a quantitative conversion factor to obtain a final quantitative reconstructed image. The quantitative reconstructed image can intuitively reflect the metabolic level of the tissue to be tested, and plays an important role in clinical detection.

Description

Technical Field

[0001] The invention relates to the technical field of nuclear medicine equipment imaging, and in particular to a method for quantitative reconstruction of nuclear medicine images. Background Art

[0002] A nuclear medicine imaging system that combines functional imaging with anatomical imaging technology has shown great superiority in clinical practice. This system can not only accurately locate the anatomical position of the lesion, but also use the relevant information in the anatomical image to perform relevant corrections on the nuclear medicine image to obtain images that better meet clinical needs. However, when the rays penetrate the tissue to be tested, some of the rays will be scattered and attenuated, resulting in abnormal number of rays obtained by the detector, which ultimately causes a certain deviation between the quantitatively reconstructed image value and the true value. Therefore, scatter correction and attenuation correction are the key to quantitative reconstruction in nuclear medicine. Summary of the invention

[0003] The present invention aims to provide a nuclear medicine quantitative reconstruction method for generating an attenuation map from a CT image and obtaining characteristic information reflecting an anatomical physiological structure for correlation correction.

[0004] Specifically, a method for quantitative reconstruction of nuclear medicine images is characterized in that the quantitative reconstruction method comprises:

[0005] Obtaining partial information from the CT image as prior information in the reconstruction algorithm;

[0006] Reconstructing the prior information as input information of a statistical iterative reconstruction algorithm to obtain a reconstructed image;

[0007] The reconstructed image is transformed by a quantitative transformation factor to obtain a quantitative reconstructed image.

[0008] Furthermore, obtaining the prior information includes the following steps:

[0009] Calculating the effective energy corresponding to different high pressures of the CT image, and converting the HU value of each pixel in the CT image into the attenuation coefficient of the corresponding nuclide energy in the SPECT image through the attenuation coefficient under the corresponding effective energy;

[0010] Feature extraction is performed on CT images to obtain information reflecting the characteristics of different anatomical physiological structures, such as segmentation maps, gradient maps, and region labeling maps.

[0011] Furthermore, the reconstruction introduces the prior information of CT image transformation as input information of the statistical iterative reconstruction algorithm into the construction of the objective function, and solves the extreme value of the objective function based on the iterative algorithm designed based on the optimization transfer concept to achieve the optimization of the reconstruction result;

[0012] The objective function is constructed as follows:

[0013] x=argmax x f(x;y,u CT )

[0014] Where x represents the reconstructed tomographic image of SPECT image, y represents the acquired SPECT projection data, and u CT represents the prior information based on CT;

[0015] The objective function f(x; y, u CT )for:

[0016] f(x;y,u CT )=logP(y|x)+βR(x|u CT )

[0017] Among them, P(y|x) represents the probability of y distribution under given x conditions, R(x|u CT ) indicates that based on u CT The regularization constraint term of x, β is used to control the regularization strength.

[0018] Furthermore, the regular constraint term R(x|u CT ) is constructed based on the spatial correlation between SPECT images and CT images, and the regular constraint term is as follows:

[0019]

[0020]

[0021] Among them, δ and η are hyperparameters, represents the set of pixels in the neighborhood of pixel j, ω jk (u CT ) indicates that based on u CT The obtained correlation factor of pixel j, k.

[0022] Furthermore, an iterative algorithm designed based on the optimization transfer concept solves the extreme value of the objective function and optimizes the reconstruction result, including the following steps:

[0023] Set initial parameters, including initial image, number of subsets, number of iterations, and hyperparameters;

[0024] Based on the CT prior information, calculate all ω jk (u CT );Calculate the sensitivity p of each pixel j (the sum of all elements in the jth column of the system matrix P);

[0025] Update the image based on the MLEM algorithm as follows:

[0026]

[0027] in, represents the nth iteration image, p ij is an element of the system matrix P, representing the contribution factor of pixel j to detection unit i, represents the value of the n+1th EM iteration image at pixel j, y i represents the true count of the i-th detection unit, [Px] i represents the image front projection value of the i-th detection unit;

[0028] Based on the nth iteration image Perform smoothing as follows:

[0029]

[0030]

[0031]

[0032] in, represents the value of the smoothed image at pixel j in the n+1th iteration, represents the weight factor at pixel j calculated based on the nth iteration image, w jk represents the weight factor of the correlation between pixels j and k, Represents the value of the nth iteration result at pixel j;

[0033] Based on the MLEM iteration results and the smoothed image, the n+1th iteration image is jointly updated as shown below:

[0034]

[0035]

[0036] in, represents the value of the n+1th iteration result at pixel j, β represents the regularization strength, represents the specific impact of reaction regularization on pixel j, which is the overall regularization strength β, the sensitivity p at pixel j j and weight factor p j The comprehensive results of

[0037] Repeat image updating, smoothing, and joint updating, and after completing the specified number of iterations, output the reconstructed image.

[0038] Furthermore, the quantitative factor uses a set of cylindrical model data with known weight and filled with water in the body to mark the activity and time information of the injected nuclide, reconstructs the projection data using a statistical iterative reconstruction algorithm, and performs spatial resolution recovery correction, scattering correction, and attenuation correction.

[0039] Furthermore, the quantitative factor is used to transform the reconstructed image according to the following formula to obtain a quantitative reconstructed image:

[0040] x′=10 6 ·x / (Factor·T·PixelX·PixelY·PixelZ·0.001)

[0041] Wherein, x represents the tomographic image after SPECT image reconstruction, x′ is the quantitative reconstructed image after conversion, T is the single angle acquisition time in seconds, PixelX, PixelY, PixelZ are the pixel sizes in the x, y, and z directions respectively, and Factor is the calibration factor.

[0042] Furthermore, in step 11, the effective energy corresponding to different high voltages of the CT imaging device is obtained, a model with a known attenuation coefficient and density is subjected to CT scanning imaging, a first region of interest is outlined for the same density material area in the CT reconstructed image, the first region of interest is included in the inner side of the model data as much as possible and does not overlap with the boundary, the HU average value of all pixels in the first region of interest is calculated, and the effective energy value is obtained by using the least squares fitting method according to the HU value definition.

[0043] Furthermore, the prior information includes an attenuation coefficient value corresponding to the evaluated SPECT image and a segmentation map reflecting different anatomical physiological structure characteristics.

[0044] The beneficial effects of the present invention include:

[0045] The present invention converts the reconstructed image into a quantitative reconstructed image with a unit mark based on the quantitative calibration factor calibrated by the system. This value can intuitively reflect the new metabolic level of the tissue to be tested and plays an important role in clinical detection. At the same time, this method can greatly simplify the calculation process.

[0046] The present invention utilizes CT images to generate attenuation coefficient values ​​and characteristic information of different anatomical physiological structures, which together serve as prior information for an iterative reconstruction algorithm to obtain a reconstructed image with higher resolution. Through the calibration factor calculation method proposed by the system, the reconstructed image is converted into a quantitative image with unit information, which can greatly improve the quantitative accuracy of nuclear medicine image reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flowchart of a method for quantitative reconstruction of nuclear medicine images provided by an embodiment of the present invention;

[0048] Figure 2 It is the projection data and the generated attenuation map data in a method for quantitative reconstruction of nuclear medicine images provided by an embodiment of the present invention;

[0049] Figure 3 A schematic diagram of a convolutional neural network model in a method for quantitative reconstruction of nuclear medicine images provided in an embodiment of the present invention, wherein the model input is a CT reconstructed image and the output is a segmentation map;

[0050] Figure 4 A schematic diagram for comparing quantitative reconstruction results in a method for quantitative reconstruction of nuclear medicine images provided by an embodiment of the present invention;

[0051] Figure 5 The figure is a schematic diagram of the calculation results of the quantitative restoration coefficient in a method for quantitative reconstruction of nuclear medicine images provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following is combined with Figure 1-5 The technical solution of the present invention is described in more detail. The present invention includes but is not limited to the following embodiments.

[0053] As attached Figure 1 As shown, the present invention provides a method for quantitative reconstruction of nuclear medicine images, especially for SPECT images, i.e. single photon emission computed tomography, the quantitative reconstruction method comprises the following steps:

[0054] Step 1, obtaining part of the information in the CT image as prior information in the reconstruction algorithm;

[0055] In step 1, the prior information includes the attenuation coefficient value corresponding to the image and the information reflecting the characteristics of different anatomical physiological structures. The method of obtaining the prior information includes:

[0056] As attached Figure 2 As shown, step 11, using a measurement model with known attenuation coefficient and density to calculate the effective energy corresponding to different high voltages in the CT scanning mode, so as to convert the HU (Houß unit) value of each pixel of the CT image into the attenuation coefficient of the corresponding nuclide in the SPECT image through the attenuation coefficient under the effective energy;

[0057] The method for calculating the effective energy corresponding to different high pressures of CT is as follows: a model with known attenuation coefficient and density is scanned by CT, and a first region of interest is outlined in the CT reconstructed image. The region must contain material data of known density and not overlap with the model boundary. The HU average value in the first region of interest is calculated and substituted into the following expression:

[0058]

[0059] Among them, F i represents the fitting value under effective energy i, μ x It represents the attenuation coefficient of the material with known density under the corresponding energy, and HU represents the HU value of each pixel of the CT reconstructed image. represents the attenuation coefficient of water under the corresponding energy, μ air represents the attenuation coefficient of air under the corresponding energy, α represents the division factor, and the constant value. In the actual calculation process, all μ x , μ air By substituting the values ​​into the above expressions one by one, we can get the corresponding F i , i represents the effective energy value of the response. The least squares method is used to fit F i The effective energy value i corresponding to the minimum value is the effective energy required by this system.

[0060] As attached Figure 3 As shown in FIG. 1 , when the effective energy of the CT imaging system is known, the attenuation coefficient under the CT effective energy can be converted to the attenuation coefficient value under the corresponding nuclide energy of the SPECT image.

[0061] Step 12, using the UNet convolutional neural network to extract features from the CT image to obtain feature information representing different anatomical physiological structures, which can be a segmentation map, a gradient map, a regional labeling map, etc.

[0062] Step 2, reconstructing the SPECT reconstructed image by using the prior information of the CT image transformation as the input information of the statistical iterative reconstruction algorithm;

[0063] In step 2, the CT prior information is introduced into the construction of the objective function, and the extreme value of the objective function is solved based on the iterative algorithm designed based on the optimization transfer idea to achieve the optimization of the reconstruction result;

[0064] The objective function is constructed as follows:

[0065] x=argmax x f(x;y,u CT )

[0066] Where x represents the reconstructed tomographic image of SPECT image, y represents the acquired SPECT projection data, and u cT Represents the prior information based on CT.

[0067] Objective function f(x; y, u CT ) is a combination of the logarithmic probability function and the regularization function, as follows:

[0068] f(x;y,u CT)=logP(y|x)+βR(x|u CT )

[0069] Among them, P(y|x) represents the probability of y distribution under given x conditions, R(x|u CT ) indicates that based on u CT The regularization constraint term of x, β is used to control the regularization strength.

[0070] For SPECT image reconstruction, the acquired data y follows a Poisson distribution, and its logarithmic probability function is as follows:

[0071]

[0072] Where i is the detector unit number, P is the system matrix, and s is the estimated number of scattering events.

[0073] Regularity constraint R(x|u CT ) is constructed based on the spatial correlation between the SPECT reconstructed image and the CT reconstructed image. The regular constraints used in this scheme are as follows:

[0074]

[0075]

[0076] Among them, δ and η are hyperparameters, represents the set of pixels in the neighborhood of pixel j, ω jk (u CT ) indicates that based on u CT The obtained correlation factor of pixel j, k.

[0077] Different from the traditional conjugate gradient algorithm (CG) and the expectation maximization (EM) algorithm commonly used in SPECT imaging, this method designs a new iterative algorithm based on the optimization transfer idea to solve the objective function after introducing the prior, ensuring the monotonic convergence and non-negativity of the iterative process. Monotonic convergence means that as the number of iterations increases, the objective function f(x; y, u CT ) changes monotonically and tends to converge; non-negativity means that when the initial value of the image is non-negative, no negative value will appear during the iterative update process.

[0078] The specific steps are as follows:

[0079] Step 21: Set initial parameters, including initial image, number of subsets, number of iterations, and hyperparameters;

[0080] Step 22: Based on the CT prior information, calculate all ω jk (u CT );Calculate the sensitivity p of each pixel j (the sum of all elements in the jth column of the system matrix P);

[0081] Step 23: Update the image based on the MLEM algorithm as follows:

[0082]

[0083] in, represents the nth iteration image, p ij is an element of the system matrix P, representing the contribution factor of pixel j to detection unit i, represents the value of the n+1th EM iteration image at pixel j, y i represents the true count of the i-th detection unit, [Px] i Represents the image front projection value of the i-th detection unit.

[0084] Step 24: Based on the nth iteration image Perform smoothing as follows:

[0085]

[0086]

[0087]

[0088] in, represents the value of the smoothed image at pixel j in the n+1th iteration, represents the weight factor at pixel j calculated based on the nth iteration image, w jk represents the weight factor of the correlation between pixels j and k, Represents the value of the nth iteration result at pixel j.

[0089] Step 25: Based on the MLEM iteration result and the smoothed image, the n+1th iteration image is jointly updated as shown below:

[0090]

[0091]

[0092] in, represents the value of the n+1th iteration result at pixel j, β represents the regularization strength, represents the specific impact of reaction regularization on pixel j, which is the overall regularization strength β, the sensitivity p at pixel j j and weight factor pj The comprehensive results.

[0093] Step 26: Repeat S23-S25, and after completing the specified number of iterations, output the reconstructed image.

[0094] As attached Figure 4-5 As shown, step 3, the SPECT reconstructed image is transformed using the quantitative transformation factor calibrated by the system to obtain the final quantitative reconstructed image.

[0095] In step 3, the quantitative calibration factor calculation uses a set of cylindrical model data with known weight and filled with water in the body to mark the activity and time information of the injected nuclide, and reconstructs the projection data using a statistical iterative reconstruction algorithm. The reconstruction requires spatial resolution recovery correction, scattering correction, and attenuation correction operations. The SPECT reconstructed image is transformed using the following formula:

[0096] x′=10 6 ·x / (Factor·T·PixelX·PixelY·PixelZ·0.001)

[0097] Where x′ is the converted quantitative reconstructed image, T is the single-angle acquisition time in seconds, PixelX, PixelY, and PixelZ are the pixel sizes in the x, y, and z directions respectively, and Factor is the calibration factor. When calculating the calibration factor, it is assumed that the initial value is 1.

[0098] In the reconstructed image after the above conversion, the second region of interest is outlined in the model area. The second region of interest should be included in the inner side of the model data as much as possible and should not overlap with the model boundary data. The mean value of all pixels in the second region of interest is counted, and the calibration factor Factor is calculated according to the following formula:

[0099] Factor=C mean / C r

[0100] Among them, C mean represents the mean value in the region of interest, C r is the actual injected drug concentration.

[0101] The present invention is not limited to the above-mentioned specific implementation modes. A person skilled in the art may implement the present invention in various other specific implementation modes according to the embodiments and the disclosure of the drawings. Therefore, any design that adopts the design structure and concept of the present invention and makes some simple transformations or changes shall fall within the scope of protection of the present invention.

Claims

1. A method for quantitative reconstruction of nuclear medicine images, It is characterized in that The quantitative reconstruction method comprises: Obtaining partial information in the CT image as prior information in the reconstruction algorithm; Reconstruct the prior information as input information of a statistical iterative reconstruction algorithm to obtain a reconstructed image; The reconstructed image is transformed by a quantitative transformation factor to obtain a quantitative reconstructed image; Acquiring the prior information comprises the following steps: Calculating the effective energy corresponding to different high pressures of the CT image, and converting the HU value of each pixel in the CT image into the attenuation coefficient of the corresponding nuclide energy in the SPECT image through the attenuation coefficient under the corresponding effective energy; Extract features from CT images to obtain information reflecting the characteristics of different anatomical physiological structures, such as segmentation maps, gradient maps, and region labeling maps; The reconstruction introduces the prior information of CT image transformation as the input information of the statistical iterative reconstruction algorithm into the construction of the objective function, and solves the extreme value of the objective function based on the iterative algorithm designed based on the optimization transfer concept to achieve the optimization of the reconstruction result; The objective function is constructed as follows: x=argmax x f(x;y,u CT ) Where x represents the reconstructed tomographic image of SPECT image, y represents the acquired SPECT projection data, and u CT represents the prior information based on CT; The objective function f(x; y, u CT )for: f(x;y,u CT )=logP(y|x)+βR(x|u CT ) Among them, P(y|x) represents the probability of y distribution under given x conditions, R(x|u CT ) indicates that based on u CT The regularization constraint term of x, β is used to control the regularization strength.

2. The quantitative reconstruction method according to claim 1, It is characterized in that The regular constraint term R(x|u CT ) is constructed based on the spatial correlation between the SPECT image and the CT image, and the regular constraint term is as follows: Among them, δ and η are hyperparameters, represents the set of pixels in the neighborhood of pixel j, ω jk (u CT ) indicates that based on u CT The obtained correlation factor of pixel j, k.

3. The quantitative reconstruction method according to claim 2, It is characterized in that The iterative algorithm designed based on the optimization transfer concept solves the extreme value of the objective function and optimizes the reconstruction results, including the following steps: Set initial parameters, including initial image, number of subsets, number of iterations, and hyperparameters; Based on the CT prior information, calculate all ω jk (u CT );Calculate the sensitivity p of each pixel j That is, the sum of all elements in the jth column of the system matrix P; Update the image based on the MLEM algorithm as follows: in, represents the nth iteration image, p ij is an element of the system matrix P, representing the contribution factor of pixel j to detection unit i, represents the value of the image at pixel j in the n+1th EM iteration, y i represents the true count of the i-th detection unit, [Px] i represents the image front projection value of the i-th detection unit, s i represents the number of scattering events estimated by the i-th detection unit; Based on the nth iteration image Perform smoothing as follows: in, represents the value of the smoothed image at pixel j in the n+1th iteration, represents the weight factor at pixel j calculated based on the nth iteration image, w jk represents the weight factor of the correlation between pixels j and k, Represents the value of the nth iteration result at pixel j; Based on the MLEM iteration results and the smoothed image, the n+1th iteration image is jointly updated as shown below: in, represents the value of the n+1th iteration result at pixel j, β represents the regularization strength, represents the specific impact of reaction regularization on pixel j, which is the overall regularization strength β, the sensitivity p at pixel j j and weight factor w j The comprehensive results of Repeat image updating, smoothing, and joint updating, and after completing the specified number of iterations, output the reconstructed image.

4. The quantitative reconstruction method according to claim 1, It is characterized in that The quantitative conversion factor uses a set of cylindrical model data with known weight and filled with water in the body to mark the activity and time information of the injected nuclide, reconstructs the projection data using a statistical iterative reconstruction algorithm, and performs spatial resolution recovery correction, scattering correction, and attenuation correction.

5. The quantitative reconstruction method according to claim 4, It is characterized in that The quantitative conversion factor is used to convert the reconstructed image into a quantitative reconstructed image according to the following formula: x′=10 6 ·x / (Factor·T·PixelX·PixelY·PixelZ·0.001) Where x represents the tomographic image after SPECT image reconstruction, x′ is the quantitative reconstructed image after conversion, T is the single-angle acquisition time in seconds, PixelX, PixelY, and PixelZ are the pixel sizes in the x, y, and z directions respectively, and Factor is the calibrated quantitative conversion factor.

6. The quantitative reconstruction method according to claim 1, It is characterized in that The effective energy corresponding to different high voltages of the CT imaging device is obtained, and a model with a known attenuation coefficient and density is scanned and imaged by CT. The first region of interest is outlined for the same density material area in the CT reconstructed image. The first region of interest is included in the inner side of the model data as much as possible and does not overlap with the boundary. The HU average value of all pixels in the first region of interest is calculated, and the effective energy value is obtained by the least squares fitting method according to the HU value definition.

7. The quantitative reconstruction method according to claim 1, It is characterized in that The prior information includes an attenuation coefficient value corresponding to the evaluated SPECT image and a segmentation map reflecting different anatomical physiological structure characteristics.

Images