A method and system for scatter correction of a pet image

Abstract

The application relates to a PET image scattering correction method, which comprises the following steps: optimizing and training a double-domain deep learning network G by taking a first PET reconstruction image x1, first detection data y1, a pre-acquired large-view PET reconstruction image xTure and pre-acquired large-view detection data yTure as a data set, so as to obtain a second PET reconstruction image x2, second detection data y2 and a mask matrix D; determining a third PET reconstruction image x3 of an extended view according to the scanning view field ranges of x1 and x2; determining third detection data y3 of the extended view according to the scanning view field ranges of y1 and y2; determining a linear attenuation coefficient image mu of the extended view according to x3 and y3; determining a large-view scattering distribution according to x3, mu and D; obtaining corrected x1 based on the large-view scattering distribution; and alternately iterating the linear attenuation coefficient image in the scanning view field and the corrected x1 to obtain a PET image after scattering correction. The PET image scattering correction method has the beneficial effect that, compared with a traditional scattering correction algorithm, the correction accuracy is higher, and the image quality is improved.

Description

Technical Field

[0001] This invention relates to the field of medical imaging technology, and in particular to a scattering correction method and system for PET images. Background Technology

[0002] Positron emission tomography (PET) is a high-end nuclear medicine imaging diagnostic device. In practice, it utilizes radioactive isotopes (such as...) 18 F, 11 (C) This involves labeling metabolic substances and injecting radionuclides into the body, then using a PET system to perform functional metabolic imaging to reflect the status of vital metabolic activities, thereby achieving diagnostic purposes. Currently, commercially available PET scans are typically integrated with other modal imaging systems, such as computed tomography (CT) or magnetic resonance imaging (MRI), to simultaneously image the patient's anatomical structures. This allows for precise localization of PET radionuclide distribution, improving the accuracy of lesion localization. Ultimately, the fusion of functional and anatomical imaging in a single machine combines the advantages of dual-modal imaging, providing a clear overview of the entire body's condition, enabling early detection of lesions and diagnosis of diseases, and offering significant advantages in guiding the diagnosis and treatment of tumors, heart diseases, and brain diseases.

[0003] During positron emission tomography (PET) data acquisition, photons may undergo Compton scattering inside the body before reaching the detector, altering their flight path. Due to the limited energy resolution of the detector, these scattering events are incorrectly recorded as true coincidence events, obfuscating the annihilation location information of nuclides and thus producing scattering artifacts in the image, severely impacting image quality. Especially during 3D data acquisition, the number of scattering coincidences can reach 30%-60% of the total count, making scattering correction a crucial step in PET reconstruction.

[0004] Currently used PET acquisition modes often contain factors that affect the accuracy of scattering distribution and thus seriously affect image quality:

[0005] Single-scatter simulation correction (SSS) is widely used for scatter correction in PET reconstruction. This method simulates the scattering distribution by calculating the probability of a coincident gamma photon undergoing a single scattering event before being detected. However, SSS typically only estimates scattering events occurring within the scanned field of view (FOV), failing to accurately estimate scattering events occurring outside the FOV. Firstly, commonly used PET scanners often have a limited field of view, leading to truncation in human scans for heavier patients or those who cannot meet positioning requirements. The lack of information outside the FOV prevents SSS from accurately estimating the scattering distribution there, thus affecting the distribution of scattering events within the FOV and resulting in significant scattering artifacts. Secondly, for multimodal acquisition modes, images from different modalities often provide the attenuation information needed for scatter correction. In clinical practice, relative deviations in the positions of multimodal images can cause scattering artifacts on PET images. For example, in PET / CT systems, the scanning range of PET is typically larger than that of other modalities (such as CT or MRI). When scanning patients with larger body weight, other modal imaging may not be able to provide a sufficiently large imaging range, which can lead to truncation of the linear attenuation coefficient image and a mismatch in imaging range between PET and CT images.

[0006] Furthermore, CT scans can usually be completed in a very short time, obtaining images that are almost a snapshot of a single moment. However, PET scans are slower, typically taking several minutes per patient position, making it impossible to acquire data while the patient is holding their breath. Therefore, respiratory movements can cause positional mismatches between PET and CT images. Additionally, during prolonged PET scans, patient movement (such as movement of the arm or head during extended scans) can also lead to phase mismatches between PET and CT images. Finally, CT images of patients with metallic substances in their bodies (such as pacemakers or metal braces) exhibit significant bright metallic artifacts, resulting in substantial errors in the CT images.

[0007] In summary, under clinical conditions, PET truncation or mismatch between PET and other modal images can produce erroneous linear attenuation information, leading to incorrect scattering correction and even severe scattering artifacts on PET images. The accuracy of scattering correction needs further improvement. Summary of the Invention

[0008] (a) Technical problems to be solved

[0009] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a scattering correction method and system for PET images, which solves the technical problem of how to further improve the accuracy of scattering correction of PET images.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0012] In a first aspect, the present invention provides a scattering correction method for PET images, comprising:

[0013] The PET acquisition process is modeled to obtain the first PET reconstructed image x1 within the scanning field of view;

[0014] Using x1, the first detection data y1, and the pre-acquired large-field PET reconstructed image x Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G was optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second detection data y2, and the mask matrix D.

[0015] The third PET reconstructed image x3 with an extended field of view is determined based on the scanning field of view of x1 and x2;

[0016] The third detection data y3 for the extended field of view is determined based on the scanning field of view of y1 and y2;

[0017] The linear attenuation coefficient image μ for the expanded field of view is determined based on x3 and y3;

[0018] Determine the large field-of-view scattering distribution based on x3, μ, and D;

[0019] The corrected x1 is obtained based on the large field-of-view scattering distribution;

[0020] The scattering-corrected PET image is obtained by iteratively alternating between the linear attenuation coefficient image within the scanned field of view and the corrected x1.

[0021] Optionally, modeling the PET acquisition process to obtain the first PET reconstructed image within the field of view includes:

[0022] The PET acquisition process is modeled using the following formula to obtain the first PET reconstructed image x1 within the scanning field of view:

[0023] in, The mean value of the detected data; j is the variable index of the radioactivity distribution image space; M is the size of the radioactivity distribution image space; A ijt Here, represents the system matrix; i is the variable index of the LOR (Lead-of-Right) response line of the sinusoidal plot of the probe data; t is the variable index of the time-of-flight (TOF) discrete space; x j Reconstructing images from unknown PET scans; ikHere, k is the linear attenuation coefficient matrix; k is the variable index of the linear attenuation coefficient image space; K is the size of the linear attenuation coefficient image space; μ k Image of linear attenuation coefficient; R i S is the average value of random noise; i denoted as the average value of the scattering noise; N is the size of the sine wave of the detected data; and T is the size of the time-of-flight (TOF) discrete space.

[0024] Optionally, the method uses x1, first detection data y1, and pre-acquired large-field-of-view PET reconstructed image x. Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G is optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second probe data y2, and the mask matrix D, including:

[0025] Obtain the patient's height h, weight w, radioactivity a, and the interval t between the injection of the tracer. a ;

[0026] Using x1, y1, h, w, a, and t a As input, x Ture and y Ture To achieve the learning objective, the network parameters θ of the dual-domain deep learning network G are optimized and trained to minimize the loss function, outputting x², y², and D, as shown in the following formula:

[0027]

[0028]

[0029] in, The optimized network parameters are represented by the mask matrix D, which represents the effective area of ​​the scanned object.

[0030] Optionally, determining the extended field of view of the third PET reconstructed image x3 based on the scanning field of view of x1 and x2 includes:

[0031] Determine x3 using the following formula:

[0032]

[0033] Where xF1 is the actual scanning field of view and xF2 is the extended scanning field of view.

[0034] Optionally, determining the third detection data y3 for the extended field of view based on the scanning field of view ranges of y1 and y2 includes:

[0035] y3 is determined using the following formula:

[0036]

[0037] Where yF1 is the actual scanning field of view and yF2 is the extended scanning field of view.

[0038] Optionally, determining the linear attenuation coefficient image μ of the expanded field of view based on x3 and y3 includes:

[0039] The linear attenuation coefficient image μ for the extended field of view is determined according to the following formula:

[0040]

[0041] Where n is the number of iterations.

[0042] Optionally, determining the large field-of-view scattering distribution based on x3, μ, and D includes:

[0043] The large field-of-view scattering distribution is determined using the following formula.

[0044]

[0045]

[0046]

[0047] Among them, I A This means that a positron annihilates at any point on S1, emitting a pair of gamma photons. One photon is not scattered and travels along path S1, where it is detected by detector A. The photon energy is 511 keV, and the detection efficiency is ε. AS The linear attenuation coefficient is μ. Another photon, scattered at point S, moves along path S2 and is detected by detector B. The photon energy is less than 511 keV, and the detection efficiency is ε′. BS The linear attenuation coefficient image is μ′;

[0048] I B This indicates that a positive and negative electron annihilate at any point on S2, emitting a pair of gamma photons. One photon, unscattered, travels along path S2 and is detected by detector B. The photon energy is 511 keV, and the detection efficiency is ε. BS The linear attenuation coefficient is μ. Another photon, scattered at point S, moves along path S1 and is detected by detector A. The photon energy is less than 511 keV, and the detection efficiency is ε. A ′ B The linear attenuation coefficient image is μ′;

[0049] Vs is the total scattering volume, σ AS Let σ be the geometric cross-section of detector A along the γ-ray. BS Let B be the geometric cross section along ray γ;

[0050] σ c The Compton scattering cross section, Let Ω be the differential cross section of the Compton scattering, and let Ω represent the scattering solid angle.

[0051] Alternatively, the corrected x1 can be obtained according to the following formula:

[0052] Optionally, the linear attenuation coefficient image within the scanned field of view is obtained using the following formula:

[0053]

[0054] In a second aspect, the present invention provides a scattering correction system for PET images, comprising:

[0055] The modeling module models the PET acquisition process to obtain the first PET reconstructed image x1 within the scanning field of view;

[0056] The training module is optimized using x1, the first probe data y1, and the pre-acquired large-field PET reconstructed image x. Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G was optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second detection data y2, and the mask matrix D.

[0057] The image field of view expansion module determines the third PET reconstructed image x3 with expanded field of view based on the scanning field of view ranges of x1 and x2;

[0058] The data field of view expansion module determines the third detection data y3 of the expanded field of view based on the scanning field of view ranges of y1 and y2;

[0059] The first determining module determines the linear attenuation coefficient image μ of the extended field of view based on x3 and y3;

[0060] The second determining module determines the large field-of-view scattering distribution based on x3, μ, and D;

[0061] The preliminary correction module obtains the corrected x1 based on the large field-of-view scattering distribution.

[0062] The scattering correction module obtains a scattering-corrected PET image by iteratively alternating between the linear attenuation coefficient image within the scanned field of view and the corrected x1.

[0063] (III) Beneficial Effects

[0064] The beneficial effects of this invention are as follows: The scattering correction method for PET images of this invention first utilizes deep learning to obtain the radioactivity distribution outside the field of view, a sinusoidal graph, and the object mask matrix required for estimating scattering events. Then, combined with the linear attenuation coefficient distribution, the distribution of scattering events beyond the field of view is finally estimated, making scattering correction more accurate and improving image quality. It also avoids the influence of other modal mismatched images on the scattering distribution. Compared to related technologies, this invention provides a more accurate scattering correction method, overcoming the shortcomings of traditional scattering calculation methods that cannot consider the scattering distribution in the field, and eliminating dependence on other modal images. It obtains a more accurate scattering distribution even when other modal images are absent or contain significant artifacts, significantly reducing patient radiation dose while obtaining higher quality images, thus making it more applicable. Compared to traditional scattering correction algorithms, it has higher correction accuracy, contributing to improved image quality. Attached Figure Description

[0065] Figure 1 A schematic flowchart of the scattering correction method for PET images provided in an embodiment of the present invention;

[0066] Figure 2 This invention provides a PET image obtained by calculating the scattering distribution using a conventional algorithm, as shown in the embodiments of the present invention.

[0067] Figure 3 The PET image obtained by calculating the scattering distribution using the scattering correction method of the PET image provided in this embodiment of the invention;

[0068] Figure 4 A block diagram of a scattering correction system for PET images provided in an embodiment of the present invention.

[0069] [Explanation of Labels in the Attached Image]

[0070] 400: Scattering correction system for PET images;

[0071] 401: Modeling module;

[0072] 402: Optimize the training module;

[0073] 403: Image field of view expansion module;

[0074] 404: Data Vision Expansion Module;

[0075] 405: First Determined Module;

[0076] 406: Second Determination Module;

[0077] 407: Preliminary calibration module;

[0078] 408: Scattering correction module. Detailed Implementation

[0079] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0080] In PET image reconstruction, the SSS method is typically used to estimate scattering distribution. However, in cases of multimodal image mismatch, this method can produce incorrect scattering event estimations and image artifacts, affecting the doctor's diagnosis. This invention applies attenuation correction and deep learning methods to estimate object activity distribution and conform to a sine curve, no longer limited by the traditional scanning field of view. This allows for accurate scattering correction of the scan data, improving image quality and eliminating PET scattering correction from dependence on other modal images, thus eliminating mismatch artifacts.

[0081] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0082] Firstly, referring to Figure 1 This embodiment provides a scattering correction method for PET images, including:

[0083] S101, Model the PET acquisition process to obtain the first PET reconstructed image x1 within the scanning field of view.

[0084] Optionally, modeling the PET acquisition process to obtain the first PET reconstructed image within the field of view includes:

[0085] The PET acquisition process is modeled using the following formula to obtain the first PET reconstructed image x1 within the scanning field of view:

[0086] in, The mean value of the detected data; j is the variable index of the radioactivity distribution image space; M is the size of the radioactivity distribution image space; A ijt Here, represents the system matrix; i is the variable index of the LOR (Lead-of-Right) response line of the sinusoidal plot of the probe data; t is the variable index of the time-of-flight (TOF) discrete space; x j Reconstructing images from unknown PET scans; ik Here, k is the linear attenuation coefficient matrix; k is the variable index of the linear attenuation coefficient image space; K is the size of the linear attenuation coefficient image space; μ k Image of linear attenuation coefficient; R iS is the average value of random noise; i denoted as the average value of the scattering noise; N is the size of the sine wave of the detected data; and T is the size of the time-of-flight (TOF) discrete space.

[0087] S102, using x1, first detection data y1, and pre-acquired large-field PET reconstructed image x... Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G was optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second detection data y2, and the mask matrix D.

[0088] Optionally, the method uses x1, first detection data y1, and pre-acquired large-field-of-view PET reconstructed image x. Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G is optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second probe data y2, and the mask matrix D, including:

[0089] Obtain the patient's height h, weight w, radioactivity a, and the interval t between the injection of the tracer. a ;

[0090] Using x1, y1, h, w, a, and t a As input, x Ture and y Ture To achieve the learning objective, the network parameters θ of the dual-domain deep learning network G are optimized and trained to minimize the loss function, outputting x², y², and D, as shown in the following formula:

[0091]

[0092]

[0093] in, The optimized network parameters are represented by the mask matrix D, which represents the effective area of ​​the scanned object.

[0094] The first PET image x1 and detection data y1, patient height h, weight w, radioactivity a, and injection tracer interval t are compared. a As input, the second PET image x2 and the expanded detection data y2 are used as output, along with the large-field-of-view reconstructed image x. Ture and y Ture By comparing and optimizing the training parameters θ of the dual-domain network to minimize the loss function L, it is possible to map the PET image within the FOV (field of view) to the PET image after expanding the FOV, and at the same time obtain the expanded symbol detection data y2.

[0095] The dataset can come from simulations or actual data collection. The training dataset needs to be preprocessed and filtered to ensure that the unattenuated PET images are free of truncation or motion artifacts.

[0096] The dual-domain deep learning network G can be a CNN network, a Unet network, a GAN network, or another network.

[0097] If the field of view (FOV) of PET data is directly enlarged, the noise at the FOV edges will increase due to the larger oblique incidence angle. Therefore, to meet the actual clinical data volume requirements, large-field-of-view PET acquisition data often requires a longer acquisition time. In addition, large-field-of-view PET reconstruction time is longer and the calibration process is more complex. Therefore, the method of this invention is practically feasible.

[0098] The calculation of scattering requires integrating the distribution of radioactivity and the attenuation coefficient, thus the requirements for image detail are relatively low, making deep learning extrapolation methods feasible. On the other hand, due to the scattering angle, only a limited distance of scattering outside the field of view affects the scattering correction within the field of view. The radioactivity information outside the imaging field of view is not entirely unknown; the distribution of radioactivity within a living organism is continuous, and the similarity of activity or tissue distribution within the human body is significant. In actual scanning, the scanning posture is basically fixed. Therefore, using deep learning networks to extract radioactivity from within the human body can achieve accurate extrapolation. This allows for the appropriate extrapolation of the radioactivity distribution within the field of view and the corresponding data to obtain the radioactivity distribution within a limited range outside the field of view, thus reducing the difficulty of estimation. Since the patient's height and weight affect the distribution of their radioactivity and attenuation coefficient, and the injection activity and injection interval affect the distribution of radioactivity in the patient's body, this information is used as prior knowledge, along with the image domain and sine curve, as input to the deep network. The expanded field-of-view PET images and sine curves, either real-world acquisitions or simulations, are used as outputs to train the deep learning network model and optimize the network parameters, thereby obtaining the second PET image and the corresponding data sine curve after expanding the field of view.

[0099] S103, determine the third PET reconstructed image x3 with an extended field of view based on the scanning field of view of x1 and x2.

[0100] Optionally, determining the extended field of view of the third PET reconstructed image x3 based on the scanning field of view of x1 and x2 includes:

[0101] Determine x3 using the following formula:

[0102]

[0103] Where xF1 is the actual scanning field of view and xF2 is the extended scanning field of view.

[0104] S104, determine the third detection data y3 for the extended field of view based on the scanning field of view of y1 and y2.

[0105] Optionally, determining the third detection data y3 for the extended field of view based on the scanning field of view ranges of y1 and y2 includes:

[0106] y3 is determined using the following formula:

[0107]

[0108] Where yF1 is the actual scanning field of view and yF2 is the extended scanning field of view.

[0109] S105, determine the linear attenuation coefficient image μ of the extended field of view based on x3 and y3.

[0110] Optionally, determining the linear attenuation coefficient image μ of the expanded field of view based on x3 and y3 includes:

[0111] The linear attenuation coefficient image μ for the extended field of view is determined according to the following formula:

[0112]

[0113] Where n is the number of iterations.

[0114] The linear attenuation coefficient distribution μ is calculated using the above formula. Keeping the second PET reconstructed image x2 as a constant, the log-likelihood function is maximized for the unknown linear attenuation coefficient distribution μ, and the linear attenuation coefficient image μ of the extended field of view is directly calculated using the extended PET data.

[0115] S106, determine the large field-of-view scattering distribution based on x3, μ and D.

[0116] Optionally, determining the large field-of-view scattering distribution based on x3, μ, and D includes:

[0117] The large field-of-view scattering distribution is determined using the following formula.

[0118]

[0119]

[0120]

[0121] Among them, I A This means that a positron annihilates at any point on S1, emitting a pair of gamma photons. One photon is not scattered and travels along path S1, where it is detected by detector A. The photon energy is 511 keV, and the detection efficiency is ε. ASThe linear attenuation coefficient is μ. Another photon, scattered at point S, moves along path S2 and is detected by detector B. The photon energy is less than 511 keV, and the detection efficiency is ε′. BS The linear attenuation coefficient image is μ′;

[0122] I B This indicates that a positive and negative electron annihilate at any point on S2, emitting a pair of gamma photons. One photon, unscattered, travels along path S2 and is detected by detector B. The photon energy is 511 keV, and the detection efficiency is ε. BS The linear attenuation coefficient is μ. Another photon, scattered at point S, moves along path S1 and is detected by detector A. The photon energy is less than 511 keV, and the detection efficiency is ε. A ′ B The linear attenuation coefficient image is μ′;

[0123] Vs is the total scattering volume, σ AS Let σ be the geometric cross-section of detector A along the γ-ray. BS Let B be the geometric cross section along ray γ;

[0124] σ c The Compton scattering cross section, The differential cross section for Compton scattering can be obtained using the Klein-Nishina formula, where Ω represents the scattering solid angle.

[0125] SSS requires integration operations on PET reconstructed images and attenuation coefficient images, which is highly robust and thus reduces the detail requirements of PET and attenuation coefficient images.

[0126] The third PET reconstructed image and attenuation coefficient image in the formula both take into account the distribution within and outside the field of view, thus facilitating scattering. The estimate is more accurate.

[0127] S107, the corrected x1 is obtained based on the large field-of-view scattering distribution.

[0128] Alternatively, the corrected x1 can be obtained according to the following formula:

[0129] Keeping the linear attenuation coefficient distribution μ constant, the corrected first PET reconstructed image x1 can be obtained by maximizing the log-likelihood function, i.e., the MLEM algorithm commonly used for PET image reconstruction. The initial value of x1 can be set to a constant, or it can be an image obtained by other reconstruction methods or an uncorrected PET image. In this patent, it is set to a constant distribution within the imaging field of view, and the constant is selected as 1000.

[0130] S108, based on the linear attenuation coefficient image within the scanning field of view and the corrected x1, alternately iterates to obtain the scatter-corrected PET image.

[0131] Optionally, the linear attenuation coefficient image within the scanned field of view is obtained using the following formula:

[0132]

[0133] The linear attenuation coefficient distribution μ1 within the field of view is calculated using the above formula. Keeping the first PET reconstructed image x1 within the axial field of view constant, the log-likelihood function is maximized for the unknown linear attenuation coefficient distribution μ1, and the linear attenuation coefficient image μ1 within the field of view is directly calculated using PET data.

[0134] The scattering correction method for PET images provided in this embodiment first applies a traditional reconstruction method to obtain the first PET radioactivity distribution within the field of view (i.e., the first PET reconstructed image); then, using a deep learning network and information such as the patient's height, weight, injection activity, and interval time, it reconstructs the second radioactivity distribution within the expanded field of view (i.e., the second PET reconstructed image), conforming to the data sine curve and the mask matrix of the scanned object range; subsequently, it obtains the attenuation coefficient distribution after expanding the field of view; finally, it uses the object mask matrix for SSS to make the scattering point range more accurate and the tail fitting range more precise, no longer limited by the inaccurate scattering values ​​caused by the traditional threshold selection error, thus obtaining an accurate scattering distribution over a large field of view; finally, it iteratively alternatingly, it obtains an accurately corrected PET image. This method considers the influence of scattering events outside the field of view, avoids the influence of other modal imaging mismatches, and overcomes the problem of inaccurate scaling parameter factors caused by traditional thresholding methods, thus providing more accurate scattering estimation.

[0135] The scattering correction method for PET images provided in this embodiment will be further described below with reference to specific examples:

[0136] The patient underwent a whole-body scan using a PET scanner. The PET scan field of view was 600 mm, while the CT scan field of view was 500 mm. The PET images were reconstructed using the ordered subsets expectation-maximization (OSEM) method, with 17 subsets and 2 iterations. The image matrix was 192x192, and the Gaussian filter had a half-width at half-maximum (FWHM) of 4 mm.

[0137] A CNN network was selected to learn from the reconstructed images and acquired chord graph data. The training dataset consisted of 300 sets of data, and the test dataset consisted of 20 sets. The CNN network used one hidden layer, two convolutional-hybrid layers, containing 300 hidden neurons. The input layer had 784 neurons, a 5×5 convolutional kernel, and 20 feature maps. The output layer used one-hot encoding, and the final layer was a softmax layer. The training process employed a log-likelihood cost function, 60 iterations, a learning rate of 0.1, and a mini-batch size of 10 for stochastic gradient descent.

[0138] The expanded field of view was 700 mm, the number of alternating iteration subsets was 17, the number of PET image iterations was 2, and the number of linear attenuation coefficient iterations was 3. The reconstruction results are shown below. Figure 2 and Figure 3 . Figure 2 In PET reconstructed images obtained by calculating scattering distribution using traditional algorithms, significant artifacts are present on the patient's body because the scattering distribution outside the field of view is not fully considered. Figure 3 To calculate the scattering distribution using the method of this invention, the PET images obtained under the same reconstruction method and parameters have a more accurate scattering distribution, no obvious artifacts, and better image quality.

[0139] Secondly, such as Figure 4 As shown, this embodiment provides a scattering correction system 400 for PET images, including: a modeling module 401, an optimization training module 402, an image field-of-view expansion module 403, a data field-of-view expansion module 404, a first determination module 405, a second determination module 406, a preliminary correction module 407, and a scattering correction module 408. The modeling module 401 models the PET acquisition process to obtain a first PET reconstructed image x1 within the scanning field of view. The optimization training module 402 uses x1, first detection data y1, and the pre-acquired large-field-of-view PET reconstructed image x1 as inputs. Ture and pre-acquired wide-field detection data y TureThe dual-domain deep learning network G is optimized and trained using the dataset to obtain a second PET reconstructed image x2, second probe data y2, and a mask matrix D. The image field-of-view expansion module 403 determines a third PET reconstructed image x3 with an expanded field of view based on the scanning field of view ranges of x1 and x2. The data field-of-view expansion module 404 determines the third probe data y3 with an expanded field of view based on the scanning field of view ranges of y1 and y2. The first determination module 405 determines the linear attenuation coefficient image μ of the expanded field of view based on x3 and y3. The second determination module 406 determines the large field-of-view scattering distribution based on x3, μ, and D. The preliminary correction module 407 obtains the corrected x1 based on the large field-of-view scattering distribution. The scattering correction module 408 performs alternating iterations based on the linear attenuation coefficient image within the scanning field of view and the corrected x1 to obtain a scattering-corrected PET image. The scattering correction system for PET images provided in this embodiment, since it is used to implement the steps of the scattering correction method for PET images provided in the first aspect embodiment of the present invention, possesses all the technical effects of the scattering correction method for PET images, which will not be elaborated further here.

[0140] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the scattering correction method for PET images described in any of the first aspects above.

[0141] Fourthly, embodiments of the present invention provide a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the scattering correction method for PET images described in any of the first aspects above.

[0142] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

[0144] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A scattering correction method for PET images, characterized in that, include: The PET acquisition process is modeled to obtain the first PET reconstructed image x1 within the scanning field of view; Using x1, the first detection data y1, and the pre-acquired large-field PET reconstructed image x Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G was optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second detection data y2, and the mask matrix D. The third PET reconstructed image x3 with an extended field of view is determined based on the scanning field of view of x1 and x2; The third detection data y3 for the extended field of view is determined based on the scanning field of view of y1 and y2; The linear attenuation coefficient image μ for the expanded field of view is determined based on x3 and y3; Determine the large field-of-view scattering distribution based on x3, μ, and D; The corrected x1 is obtained based on the large field-of-view scattering distribution; The scatter-corrected PET image is obtained by alternating the linear attenuation coefficient image within the scanned field of view and the corrected x1. The determination of the large field-of-view scattering distribution based on x3, μ, and D includes: The large field-of-view scattering distribution is determined using the following formula. : ； in, This indicates that a positron annihilates at any point on S1, emitting a pair of electrons. A photon, unscattered, travels along path S1 and is detected by detector A. The photon energy is 511 keV, and the detection efficiency is... The linear attenuation coefficient image is Another photon, scattered at point S, travels along path S2 and is detected by detector B. The photon energy is less than 511 keV, and the detection efficiency is... The linear attenuation coefficient image is ; This means that a positive and a negative electron annihilate at any point on S2, emitting a pair of electrons. A photon, unscattered, travels along path S2 and is detected by detector B. The photon energy is 511 keV, and the detection efficiency is... The linear attenuation coefficient image is Another photon, scattered at point S, travels along path S1 and is detected by detector A. The photon energy is less than 511 keV, and the detection efficiency is... The linear attenuation coefficient image is ; Vs is the total scattering volume. Let A be the geometric cross-section of detector A along the γ-ray beam. Let B be the geometric cross section along ray γ; The Compton scattering cross section, Let Ω be the differential cross section for Compton scattering, and let Ω represent the scattering solid angle. The corrected x1 obtained based on the large field-of-view scattering distribution includes: The corrected x1 is obtained using the following formula: ； The linear attenuation coefficient image within the scanned field of view is obtained using the following formula: 。 2. The scattering correction method for PET images according to claim 1, characterized in that, The process of modeling the PET acquisition process to obtain the first PET reconstructed image within the field of view includes: The PET acquisition process is modeled using the following formula to obtain the first PET reconstructed image x1 within the scanning field of view: ,in, is the average value of the detected data; j is the variable index of the radioactivity distribution image space; M is the size of the radioactivity distribution image space; is the system matrix; i is the variable index of the LOR (Lead-of-Right) response line of the sinusoidal plot of the probe data; t is the variable index of the time-of-flight (TOF) discrete space; Reconstructing images from unknown PET scans; is the linear attenuation coefficient matrix; k is the variable index of the linear attenuation coefficient image space; K is the size of the linear attenuation coefficient image space; Image of linear attenuation coefficient; This represents the average value of the random noise. denoted as the average value of the scattering noise; N is the size of the sine wave of the detected data; and T is the size of the time-of-flight (TOF) discrete space.

3. The scattering correction method for PET images according to claim 2, characterized in that, The data includes x1, the first detection data y1, and the pre-acquired large-field PET reconstructed image x. Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G is optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second probe data y2, and the mask matrix D, including: Obtain the patient's height h, weight w, radioactivity a, and the interval t between the injection of the tracer. a ; Using x1, y1, h, w, a, and t a As input, x Ture and y Ture To achieve the learning objective, the network parameters θ of the dual-domain deep learning network G are optimized and trained to improve the loss function. To minimize the output, we need to find x², y², and D, as shown in the following formula: ； in, The optimized network parameters are represented by the mask matrix D, which represents the effective area of ​​the scanned object.

4. The scattering correction method for PET images according to claim 3, characterized in that, The determination of the extended field of view of the third PET reconstructed image x3 based on the scanning field of view of x1 and x2 includes: Determine x3 using the following formula: ； Where xF1 is the actual scanning field of view and xF2 is the extended scanning field of view.

5. The scattering correction method for PET images according to claim 4, characterized in that, The determination of the third detection data y3, which extends the field of view based on the scanning field of view ranges of y1 and y2, includes: y3 is determined using the following formula: ； Where yF1 is the actual scanning field of view and yF2 is the extended scanning field of view.

6. The scattering correction method for PET images according to claim 5, characterized in that, The step of determining the linear attenuation coefficient image μ of the expanded field of view based on x3 and y3 includes: The linear attenuation coefficient image μ for the extended field of view is determined according to the following formula: ； Where n is the number of iterations.

7. A scattering correction system for PET images, characterized in that, A scattering correction method for performing any one of claims 1 to 6 on a PET image includes: The modeling module models the PET acquisition process to obtain the first PET reconstructed image x1 within the scanning field of view; The training module is optimized using x1, the first probe data y1, and the pre-acquired large-field PET reconstructed image x. Ture and pre-acquired wide-field detection data y Ture The dual-domain deep learning network G was optimized and trained using the dataset to obtain the second PET reconstructed image x2, the second detection data y2, and the mask matrix D. The image field of view expansion module determines the third PET reconstructed image x3 with expanded field of view based on the scanning field of view ranges of x1 and x2; The data field of view expansion module determines the third detection data y3 of the expanded field of view based on the scanning field of view ranges of y1 and y2; The first determining module determines the linear attenuation coefficient image μ of the extended field of view based on x3 and y3; The second determining module determines the large field-of-view scattering distribution based on x3, μ, and D; The preliminary correction module obtains the corrected x1 based on the large field-of-view scattering distribution. The scattering correction module obtains a scattering-corrected PET image by iteratively alternating between the linear attenuation coefficient image within the scanned field of view and the corrected x1.

Images