A method of scatter correction of a pet image and a pet system

Abstract

The application relates to a PET image scattering correction method and a PET system. The method comprises the following steps: acquiring a first PET image and a first attenuation coefficient image for the detection data of a specified axial field of view; inputting patient basic information and the first PET image and the first attenuation coefficient image into a deep learning network to acquire a second PET image and a second attenuation coefficient image output by the deep learning network; the second PET image is a simulation output image outside the axial field of view; and based on the first PET image, the second PET image, the first attenuation coefficient image, the second attenuation coefficient image and pre-established calculation information of a scattering correction SSS, scattering correction information of the detection data is acquired, and the scattering correction information is used for reconstructing a PET image. The scattering correction information acquired by the above method is more comprehensive and more accurate, and when the scattering correction information is applied to PET reconstruction, the PET image is free of artifacts and the image quality is better.

Description

Technical Field

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

[0002] Currently, during the acquisition process of positron emission tomography (PET) systems, photons may undergo Compton scattering with human tissue before reaching the detector, altering their flight direction. Due to the limited energy resolution of the detector, these scattering events are incorrectly recorded as true coincidence events, obfuscating the positional information of nuclides and thus producing scattering artifacts in the image, severely affecting image quality. Especially during 3D data acquisition, the number of scattering coincidences can reach 30%-60% of the total count, making scattering correction one of the key steps in PET reconstruction.

[0003] In commonly used PET acquisition modes, several factors often affect the accuracy of scattering distribution, thus severely impacting image quality:

[0004] First, the single-scatter simulation correction (SSS) method is widely used for scatter correction in PET reconstruction. This method simulates the scattering distribution by calculating the probability of a coincidence gamma photon experiencing a single scattering event before being detected. However, SSS typically only estimates scattering events occurring within the axial scanning field of view (FOV), failing to accurately estimate scattering events occurring outside the FOV. Currently, commonly used PET scanners typically have a limited axial FOV, thus requiring multi-bed scanning in human body scans. Therefore, scatter correction can only be performed using data from that specific bed. However, this method has limitations. Because the radiation source is continuously distributed within the body, scattering from outside the FOV enters the field of view and is collected by the detector when acquiring scan data from each bed. During SSS correction, the lack of information outside the axial FOV prevents accurate estimation of the scattering distribution outside the FOV, thus affecting the distribution of scattering events in a single-bed scan. Currently used tail fitting methods can coarsely compensate for scattering outside the FOV, but this compensation is often a simple approximation and inaccurate. In cases of low patient injection doses, short scan times, or in scans of obese patients, high noise levels can severely impact the accuracy of scatter correction.

[0005] Secondly, for multimodal acquisition modes, it is sometimes impossible to accurately obtain a linear attenuation coefficient distribution that matches the PET data, resulting in artifacts on the PET images. The image positions of different modalities may have relative deviations. Taking the PET / CT system as an example, CT scans can usually be completed in a very short time, obtaining images that are almost snapshots of a specific moment. However, PET scans are slower, typically taking several minutes per position, making it impossible to complete data acquisition while the patient is holding their breath. During long PET scans, patient movement may occur (e.g., the arm, head, etc., may move during extended scans), leading to a mismatch between PET and CT images and producing significant scattering artifacts. Significant artifacts in the attenuated image during the scan also cause significant errors in the stepwise distribution of the PET image attenuation coefficient. For example, CT images of patients with metallic substances in their bodies (such as pacemakers or metal braces) show significant bright metal artifacts, making it difficult to accurately distinguish surrounding tissues, thus causing significant artifacts in the linear attenuation coefficient distribution and severely affecting the scattering correction of the PET image. Finally, the scanning range of PET is usually larger than that of other modalities (such as CT or MRI). When scanning patients with larger body weight, other modal imaging may not provide a sufficiently large imaging range, leading to truncation of the linear attenuation coefficient distribution. This incomplete attenuation information can also produce artifacts when applied to PET reconstruction. When PET is combined with other modal imaging, the other modal images can, to some extent, limit the accuracy of PET scattering correction. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a scattering correction method and PET system for PET images, which can achieve accurate scattering correction of the scanning bed.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0010] In a first aspect, embodiments of the present invention provide a scattering correction method for PET images, comprising:

[0011] S10. For the detection data within the specified axial field of view (0,D], acquire a first PET image of the detection data and a first attenuation coefficient image corresponding to the first PET image; the first PET image is an uncorrected image reconstructed from the detection data within the axial field of view (0,D]; the first PET image and the first attenuation coefficient image can be acquired using existing methods;

[0012] S20. Based on a pre-trained deep learning network, the patient's basic information, the first PET image, and the first attenuation coefficient image are input into the deep learning network to obtain the second PET image and the second attenuation coefficient image corresponding to the second PET image output by the deep learning network; the second PET image is a simulated output image outside the axial field of view (D, D+F]; D represents the size of the axial field of view, and F represents the size of the axial extension range;

[0013] S30. Based on the first PET image, the second PET image, the first attenuation coefficient image, the second attenuation coefficient image, and the pre-established scattering correction SSS calculation information, obtain the scattering correction information of the detection data, which is used to reconstruct the PET image;

[0014] The pre-trained deep learning network is a network trained on PET images within the axial field of view and matching associated images and associated information. The network outputs PET images outside the axial field of view and the corresponding attenuation coefficient images of the PET images.

[0015] Optionally, prior to S10, the method further includes:

[0016] S00. Based on the reconstructed PET images and the matched associated images and associated information, obtain training samples for training the deep learning network.

[0017] Each training sample includes: a reconstructed first PET image / simulated first PET image, the attenuation coefficient image corresponding to the first PET image, the height, weight, radioactivity, and tracer injection interval of the user to whom the PET image belongs, and the real linear attenuation image obtained by the PET / CT or PET / MR scan used for verification.

[0018] S01. Train the deep learning network based on the training samples to obtain the trained deep learning network;

[0019] In the trained deep learning network, the network parameters θ minimize the loss function L of the optimized deep learning network, so that the unscattered PET images and attenuation coefficient images within the axial field of view are mapped to the unscattered PET images and attenuation coefficient images outside the extended axial field of view.

[0020] Optionally, S01 includes: the network parameters θ and the output of the deep learning network G are represented as follows:

[0021]

[0022] in, For the first PET image, The image shows the first attenuation coefficient. For the second PET image, 2 is the image of the second attenuation coefficient. For the user's height, For the user's weight, For radioactivity, t a The interval for injecting tracer;

[0023] Second PET image Second attenuation coefficient image As the output of G,

[0024] During the verification process, the output of G is compared with the true linear attenuation image obtained from PET / CT or PET / MR scans. and based on Reconstructed image Compare them.

[0025] Optionally, the deep learning network is one of the following: CNN network, Unet network, or GAN network.

[0026] Optionally, S10 includes:

[0027] First PET image within axial field of view Maintain linear attenuation coefficient distribution It is a constant, obtained by maximizing the log-likelihood function using the MLEM algorithm;

[0028] Linear attenuation coefficient distribution within the axial field of view Maintain the first PET image within the axial field of view The constant is used for the unknown linear decay coefficient distribution. Maximize the log-likelihood function to calculate the first attenuation coefficient image within the axial field of view using PET data. .

[0029] Optionally, S30 includes:

[0030] Acquiring a third PET image with an extended axial field of view and the third attenuation coefficient image ;

[0031] ;

[0032] Based on the third PET image and the third attenuation coefficient image The pre-established scattering correction SSS calculation information is corrected to obtain the corrected scattering correction information. ;Will As scattering correction information for the detected data;

[0033] ;

[0034] 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 μ3. 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... The linear attenuation coefficient image is μ3 ’ ; 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 μ3. 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... The linear attenuation coefficient image is μ3 ’ ;R AS , R BS denoted by and , respectively, the distances from the scattering point to detector A and detector B, and S represents the position of the scattering point; In this context, ds represents the integral over distance s;

[0035] Vs represents the total scattering volume. This represents the geometric cross-section of detector A along the γ-ray. This represents the geometric cross-section of detector B along ray γ;

[0036] Indicates the Compton scattering cross section. The differential cross section representing Compton scattering can be obtained using the Klein-Nishina formula, where Ω represents the scattering solid angle.

[0037] Optionally, the training samples further include:

[0038] Multi-bed training samples, which include: PET images of reconstructed multi-bed detection data and attenuation coefficient images of the PET images, height, weight, radioactivity, and tracer injection interval of the user to whom the PET image belongs, and true linear attenuation images obtained from PET / CT or PET / MR scans used for verification;

[0039] In use, the deep learning network outputs scattering correction information for each bed position. .

[0040] For multi-bed data, as the number of beds increases, the scanned bed data is stitched together and then input into the deep learning network. In this way, as the number of beds increases, the data input into the network becomes more complete and the information becomes richer.

[0041] Secondly, embodiments of the present invention also provide a method for reconstructing PET images, comprising,

[0042] P01. Based on the detection data within the specified axial field of view (0, D], the scattering correction information of the detection data is obtained using the scattering correction method of PET images. ;

[0043] P02, according to the above Using a pre-established log-likelihood function, PET reconstruction is performed to obtain a PET image with accurate scattering correction.

[0044] Fourthly, embodiments of the present invention also provide a PET system, comprising: a memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically performing the method described in any of the above embodiments.

[0045] (III) Beneficial Effects

[0046] The method of the present invention solves the defect in the prior art that other modal images limit the accuracy of PET scattering correction when imaging with other modalities.

[0047] In this invention, scattering outside the axial field of view is accurately estimated by using PET images and attenuation images within the axial field of view, thereby obtaining scattering correction for the region of interest. This scattering correction for the region of interest considers scattering events within the axial field of view and makes a relatively accurate estimate of scattering events outside the field of view, which greatly improves the accuracy of scattering correction.

[0048] Ultimately, this allows PET scans to no longer rely on other modalities of image, significantly reducing patient radiation dose while achieving higher image quality.

[0049] Specifically, obtaining scattering outside the axial field of view can be achieved by taking the reconstructed PET image of the current bed position, calculating the accurate attenuation coefficient image within the axial field of view, and using the patient's height, weight, injection activity, and other information as input to a deep learning network. Through deep learning, the distribution of radioactivity and linear attenuation coefficient outside the current axial field of view is estimated, so that PET scanning no longer depends on other modal images, greatly reducing the patient's radiation dose while obtaining higher image quality. Attached Figure Description

[0050] Figure 1 This is a schematic flowchart of a scattering correction method for PET images provided in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram comparing the results of PET images obtained by various algorithms. Detailed Implementation

[0052] 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.

[0053] 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.

[0054] In current PET image reconstruction, the SSS method is commonly used to estimate the scattering distribution. However, this method cannot accurately estimate scattering events outside the axial field of view, which affects image quality and physician diagnosis.

[0055] To address the problem of inaccurate scattering estimation in existing PET image reconstruction, this invention first extracts attenuation information directly from the acquired PET data without relying on other modal imaging. This effectively corrects linear attenuation coefficient distribution artifacts and broadens the application scope of PET imaging. Existing applications 201910218840.8 and 201910234260.8, both titled "Method for Obtaining Correction Information for Attenuation Correction of PET Activity Distribution Images," iteratively extract the linear attenuation coefficient image from the Time-of-Flight (TOF) information acquired by PET. These existing applications can obtain PET images and attenuation images within the axial field of view. The PET images obtained by both methods are uncorrected scattering PET images.

[0056] Next, the scattering outside the axial field of view is accurately estimated; for example, the PET image of the current bed obtained by the above-mentioned existing application is used to calculate the accurate attenuation coefficient image within the axial field of view, and together with the patient's height, weight, injection activity and other information, it is used as input to a deep neural network to estimate the distribution of radioactivity and linear attenuation coefficient outside the current axial field of view through deep learning.

[0057] Finally, the scattering distribution estimated by the SSS method was modified so that the scattering correction of the region of interest not only takes into account scattering events within the axial field of view, but also makes a relatively accurate estimate of scattering events outside the field of view, which greatly improves the accuracy of scattering correction.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment of the invention provides a scattering correction method for PET images. The execution subject of this method can be a control device / electronic device for PET image reconstruction. The control device can be integrated into the acquisition device of the PET system or a separate computer processing device. The scattering correction method for PET images includes the following steps:

[0060] S10. For the detection data within the specified axial field of view (0,D], acquire the first PET image of the detection data and the first attenuation coefficient image corresponding to the first PET image; the first PET image is an uncorrected image reconstructed from the detection data within the axial field of view (0,D].

[0061] It should be noted that the first PET image in this step can be reconstructed using existing methods, as described above, and can be obtained using the methods of the two application numbers mentioned above.

[0062] S20. Based on a pre-trained deep learning network, the patient's basic information, the first PET image, and the first attenuation coefficient image are input into the deep learning network to obtain the second PET image and the second attenuation coefficient image corresponding to the second PET image output by the deep learning network; the second PET image is a simulated output image outside the axial field of view (D, D+F); D represents the size of the axial field of view, and F represents the size of the axial extension range.

[0063] In this embodiment, the deep learning network is one of the following: CNN network, Unet network, or GAN network.

[0064] S30. Based on the first PET image, the second PET image, the first attenuation coefficient image, the second attenuation coefficient image, and the pre-established scattering correction SSS calculation information, obtain the scattering correction information of the detection data, which is used to reconstruct the PET image;

[0065] The pre-trained deep learning network is a network trained on PET images within the axial field of view and matching associated images and associated information. The network outputs PET images outside the axial field of view and the corresponding attenuation coefficient images of the PET images.

[0066] The objective of this embodiment is first to reconstruct the radioactivity distribution and linear attenuation coefficient distribution within the axial field of view (0, D) from PET data acquired within that field of view. Then, a deep learning network is used to reconstruct the radioactivity distribution and linear attenuation coefficient distribution outside the axial field of view (D, D+F). Finally, these are stitched together to form the radioactivity distribution and linear attenuation coefficient distribution within the axial field of view (0, D+F), and SSS scattering correction is performed. This allows for accurate estimation of scattering within the axial field of view (0, D). In this process, D represents the size of the axial field of view, and F represents the size of the axial extension range. This method considers the influence of scattering events outside the field of view while avoiding the influence of mismatches in other modal imaging, thus providing a more accurate estimation of scattering within the field of view.

[0067] Understandably, the value of D can be determined by the axial field of view of different machines, while F can be selected from empirical values ​​or calculated according to relevant algorithms. For example, the value of D can be 218.4 mm, and the empirical value of F can be 9.3 mm. The values ​​of D and F are not limited here and can be adjusted according to actual needs.

[0068] In practical applications, before step S10 above, Figure 1 The method shown may also include step S00, which is not shown in the figure below:

[0069] S00. Based on the reconstructed PET images and the matched associated images and associated information, obtain training samples for training the deep learning network.

[0070] Each training sample includes: a reconstructed first PET image / simulated first PET image, the attenuation coefficient image corresponding to the first PET image, the height, weight, radioactivity, and tracer injection interval of the user to whom the PET image belongs, and a true linear attenuation image obtained from the PET / CT or PET / MR scan used for verification.

[0071] S01. Train the deep learning network based on the training samples to obtain the trained deep learning network;

[0072] In the trained deep learning network, the network parameters θ minimize the loss function L of the optimized deep learning network, so that the unscattered PET images and attenuation coefficient images within the axial field of view are mapped to the unscattered PET images and attenuation coefficient images outside the extended axial field of view.

[0073] The network parameters θ and output of the deep learning network G are represented as follows:

[0074]

[0075] in, For the first PET image, The image shows the first attenuation coefficient. For the second PET image, 2 is the image of the second attenuation coefficient. For the user's height, For the user's weight, For radioactivity, t a The interval for injecting tracer;

[0076] Second PET image Second attenuation coefficient image As the output of G,

[0077] During the verification process, the output of G is compared with the true linear attenuation image obtained from PET / CT or PET / MR scans. and based on Reconstructed image Compare them.

[0078] The first PET image within the aforementioned axial field of view Maintain linear attenuation coefficient distribution It is a constant, obtained by maximizing the log-likelihood function using the MLEM algorithm;

[0079] Linear attenuation coefficient distribution within the aforementioned axial field of view Maintain the first PET image within the axial field of view The constant is used for the unknown linear decay coefficient distribution. Maximize the log-likelihood function to calculate the first attenuation coefficient image within the axial field of view using PET data. .

[0080] This embodiment first uses existing attenuation correction methods to obtain an accurate linear attenuation coefficient distribution within the axial field of view. Then, it uses deep learning methods to obtain the radioactivity distribution and linear attenuation coefficient distribution outside the axial field of view. Finally, it uniformly estimates scattering events inside and outside the axial field of view, making scattering correction more accurate, improving image quality, and avoiding the influence of other modal mismatched images on the scattering distribution.

[0081] Example 2

[0082] To better understand the method in Embodiment 1 above, the scattering correction method for PET images will be explained in detail below with the help of formulas and reasoning.

[0083] Step 1: The PET collection process can be modeled as follows:

[0084] (1)

[0085] In the formula y=[y 1t , y 2t , …,y it , …,y NT ] ' This indicates the detected data. The expression represents the average value of the probe data, N represents the size of the probe data sine curve, T represents the size of the Time-of-Flight (TOF) discrete space, i represents the index of the variable in the LOR (line of response) of the probe data sine curve, and t represents the index of the variable in the TOF discrete space. A single quote superscript indicates a matrix transpose operation. x=[x1, x2, …, x j , …, x M ] ' This represents an image of unknown radioactivity distribution, where M represents the size of the radioactivity distribution image space, j represents the variable index of the radioactivity distribution image space, and represents the point source at the corresponding spatial location. μ=[μ1,μ2, …, μ k , …,μ K ] 'This represents the linear decay coefficient image, where K represents the size of the linear decay coefficient image space, and k represents the variable index of the linear decay coefficient image space, representing the point source at the corresponding spatial location. A=[A ijt [l] is the system matrix, which mathematically expresses the probability that a spatial point source j in the PET system is detected by the response line LORi with a time-of-flight (TOF) of t. It reflects the physical characteristics of the system. ik [r1, r2, ..., r] is the linear attenuation coefficient matrix, representing the trajectory intersection length when LOR i crosses the spatial location source k. N Let ' represent the average value of random noise, and S = [s1, s2, …, s]. N ]' represents the average value of the scattering noise.

[0086] Step 2: The commonly used scattering estimation method in clinical practice, SSS (Equation (2)), uses PET reconstructed images and attenuation coefficient images to calculate the scattering distribution based on analytical methods. As can be seen from Equation (2), SSS requires integration of PET reconstructed images and attenuation coefficient images, which has high robustness and thus reduces the detail requirements of PET and attenuation coefficient images.

[0087] The SSS formula is as follows:

[0088] (2)

[0089] 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 .

[0090] 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 .

[0091] Vs represents the total scattering volume. This represents the geometric cross-section of detector A along the γ-ray. This represents the geometric cross-section of detector B along ray γ.

[0092] Indicates the Compton scattering cross section. The differential cross section representing Compton scattering can be obtained using the Klein-Nishina formula, where Ω represents the scattering solid angle.

[0093] R AS , R BS These represent the distances from the scattering point to detector A and detector B, respectively.

[0094] S represents the location of the scattering point.

[0095] Step 3: According to the SSS formula (2), the estimation of the scattering distribution requires known images of the radioactivity distribution and the attenuation coefficient distribution. The radioactivity images and attenuation coefficient images within the axial field of view can be obtained directly from the PET detection data through alternating iterations.

[0096] First PET image within axial field of view Maintain linear attenuation coefficient distribution Since it is a constant, it can be obtained by maximizing the log-likelihood function, which is the general MLEM algorithm for PET image reconstruction:

[0097] (3)

[0098] n represents the number of iterations. The initial value of x1 in formula (3) can be set to a constant or an image obtained by other reconstruction methods or an uncorrected PET image. In this embodiment, it is set to a constant distribution within the imaging field of view, and the constant is selected as 1.

[0099] Step 4: Linear attenuation coefficient distribution within the axial field of view Then, the calculation is performed using formula (4), keeping the first PET image within the axial field of view. The constant is used for the unknown linear decay coefficient distribution. Maximizing the log-likelihood function, the first linear attenuation coefficient image within the axial field of view is directly calculated using PET data. :

[0100] (4)

[0101] n represents the number of iterations. The initial value of μ1 in formula (4) can be set to a constant or to the distribution of the linear attenuation coefficient that is not perfectly matched in other modes. In this embodiment, it is set to 0 for the distribution within the imaging field of view.

[0102] After obtaining the radioactivity distribution and linear attenuation coefficient distribution within the axial field of view in step 4, the radioactivity and linear attenuation coefficient outside the axial field of view are not entirely unknown. The radioactivity distribution and linear attenuation coefficient distribution within a living organism are continuous, and the similarity of activity or tissue distribution within the human body is significant. Therefore, in the following steps, deep learning can be used to extract the empirical distribution of radioactivity and attenuation within the human body, enabling accurate extrapolation. This allows for appropriate extrapolation from the radioactivity distribution and linear attenuation coefficient distribution within the field of view, resulting in accurate estimates of the radioactivity distribution and linear attenuation coefficient distribution within a limited area outside the field of view.

[0103] The aforementioned attenuation coefficient image distribution is used for integration operations, therefore the requirements for image detail are not high, making deep learning extrapolation feasible. Furthermore, due to the influence of the scattering angle, only scattering from a limited distance outside the field of view affects the scattering correction within the field of view, further reducing the estimation difficulty. 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 within the patient's body, this information is used as prior knowledge, along with the image, as input to the deep network. Real-world or simulated extended axial field-view PET images and attenuation coefficient images are used as output to train the deep learning network model and optimize network parameters, thereby obtaining a second axially extended PET image and a second attenuation coefficient image.

[0104] Step 5: Deep learning network G achieves first PET image First attenuation coefficient image To the second PET image Second attenuation coefficient image Mapping of 2, ;

[0105] That is, the first PET image First attenuation coefficient image and the patient's height ,weight Radioactivity t interval with the injection tracer a As input, the second PET image Second attenuation coefficient image As output.

[0106] During training, the output of G is also used to compare with the true linear attenuation image obtained from PET / CT or PET / MR scans. and reconstructed images By comparing and optimizing the training network parameters θ to minimize the loss function L, it is possible to map the uncorrected PET images and attenuation coefficient images within the FOV to PET images and attenuation coefficient images after expanding the axial FOV, i.e.:

[0107] (5)

[0108] The training dataset can come from simulation or actual data acquisition. The training dataset needs to be preprocessed and filtered to ensure that the attenuation coefficient images outside the field of view are completely matched with the unattenuated PET images, and that there are no truncation or motion artifacts.

[0109] Without loss of generality, the training network G can be selected from CNN networks, Unet networks, GAN networks, or other networks, and all of these should be within the scope of patent protection.

[0110] Step 6: To maximize the accuracy of scattering correction, the PET image and attenuation coefficient image within the axial field of view are obtained using formulas 3 and 4; for the area outside the FOV, the PET image and attenuation coefficient image are obtained using deep learning. Thus, the third PET image and third attenuation coefficient image with extended axial field of view are obtained as shown in formula (6):

[0111] (6)

[0112] Step 7: Through steps 4-6, the SSS formula (2) can be transformed into:

[0113] (7)

[0114] The third PET image and the third attenuation coefficient image in formula (7) both take into account the distribution within and outside the field of view, so that the scattering The estimate is more accurate. R AS , R BS denoted by and , respectively, the distances from the scattering point to detector A and detector B, and S represents the position of the scattering point.

[0115] It should be noted that for actual multi-bed clinical scans, to fully utilize the scan data, steps 3 and 4 can be used for iterative calculation. For the m-th bed, firstly, a first PET image and a first attenuation coefficient image are generated by stitching together the images from the m beds; then, a second PET image and a second attenuation coefficient image with an expanded field of view are generated using the above method; next, a third PET image and a third attenuation coefficient image are generated for calculating the scattering distribution; finally, the accurate scattering distribution, including the scattering distribution outside the field of view, is estimated. This iterative calculation is performed gradually as the number of beds increases. With the increase in the number of beds, the stitched images become more complete, and the information within the axial field of view of the current bed gradually becomes richer, making the estimation of the scattering distribution outside the axial field of view more accurate.

[0116] What is meant here is that, during the training of G, the training samples may also include: multi-bed training samples, which include: PET images of the reconstructed multi-bed detection data and the attenuation coefficient image of the PET image, the height, weight, radioactivity, and tracer injection interval of the user to whom the PET image belongs, and the true linear attenuation image obtained by the PET / CT or PET / MR scan used for verification.

[0117] Therefore, in use, the deep learning network outputs scattering correction information for each bed position. For multi-bed datasets, as the number of beds increases, the scanned bed data is stitched together and then input into the deep learning network. In this way, as the number of beds increases, the data input into the network becomes more complete and the information is richer, resulting in a more accurate training algorithm (G).

[0118] The above method combines existing attenuation correction patented algorithms and deep learning methods to estimate the activity distribution and linear attenuation coefficient distribution of objects outside the axial field of view, and then uniformly estimates scattering events inside and outside the axial field of view, thereby performing accurate scattering correction on the scanning bed, improving image quality, and making PET scattering correction no longer dependent on other modal images, thus significantly reducing the radiation dose received by the patient.

[0119] Step 8: Based on the detection data within the specified axial field of view (0, D], the scattering correction information of the detection data is obtained using the scattering correction methods described in steps 1 to 7 above. ;

[0120] P02, according to the above Using a pre-established log-likelihood function, PET reconstruction is performed to obtain a PET image with accurate scattering correction.

[0121] Experiments have shown that the PET images in this embodiment are more accurate and free of artifacts. Figure 2(a) The PET reconstructed image obtained by calculating the scattering distribution using the traditional algorithm shows obvious uptake artifacts around the bladder because the scattering distribution outside the axial field of view is not fully considered. Figure 2 (b) To calculate the scattering distribution using the method of this embodiment, the PET image obtained under the same reconstruction method and parameters has a more accurate scattering distribution, no obvious artifacts, and better image quality.

[0122] The above method overcomes the shortcomings of traditional scattering calculation methods that cannot consider scattering distribution outside the axial field of view, and also eliminates the dependence on other modal images. It obtains a more accurate scattering distribution even when other modal images are absent or contain significant artifacts, resulting in higher quality images while significantly reducing patient radiation dose. Therefore, it has wider applicability. Compared with traditional scattering correction algorithms, it has higher correction accuracy, contributing to improved image quality.

[0123] In addition, embodiments of the present invention also provide a PET system, which includes: a memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically performing the aforementioned PET image reconstruction method or PET image scattering correction method, etc. It is understood that the PET system can be an operating console communicatively connected to a detector.

[0124] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0125] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0126] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0127] 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.

Claims

1. A method of scatter correction of a PET image, characterized by, The method comprises: S10, for the specified axial field of view range (0, D], obtain the first PET image of the detection data and the first attenuation coefficient image corresponding to the first PET image; the first PET image is an unscattered corrected image reconstructed by the detection data in the axial field of view range (0, D]; S20, based on the pre-trained deep learning network, input the patient basic information and the first PET image and the first attenuation coefficient image into the deep learning network, obtain the second PET image output by the deep learning network and the second attenuation coefficient image corresponding to the second PET image; the second PET image is a simulation output image outside the axial field of view range (D, D+F]; D represents the size of the axial field of view range, and F represents the size of the axial extension range; S30, based on the first PET image, the second PET image, the first attenuation coefficient image, the second attenuation coefficient image, and the calculation information of the pre-established scatter correction SSS, obtain the scatter correction information of the detection data, which is used for reconstructing the PET image; The pre-trained deep learning network is a network obtained by training a deep learning network based on the PET image in the axial field of view range and the matching associated image and associated information, which outputs the PET image outside the axial field of view range and the attenuation coefficient image corresponding to the PET image; The S30 comprises: Third pet image with extended axial field of view and third attenuation coefficient image ; ; then based on the third PET image and the third attenuation coefficient image , the pre-established calculation information of scatter correction SSS is corrected to obtain modified scatter correction information ; the scatter correction information of the detection data; ； wherein, represents positron annihilation at any point on S1 emitting a pair of photons, one photon without scattering, moving along path S1, detected by detector A, photon energy 511 keV, detection efficiency , linear attenuation coefficient image μ3, the other photon scattering at S point and moving along path S2, detected by detector B, photon energy less than 511 keV, detection efficiency , linear attenuation coefficient image μ3 ’ ; represents positron annihilation at any point on S2 emitting a pair of photons, one photon without scattering, moving along path S2, detected by detector B, photon energy 511 keV, detection efficiency , linear attenuation coefficient image μ3, the other photon scattering at S point and moving along path S1, detected by detector A, photon energy less than 511 keV, detection efficiency , linear attenuation coefficient image μ3 ’ ;R AS , R BS respectively represent the distance from the scattering point to detector A and detector B, and S represents the scattering point position; wherein ds represents the integral with respect to distance s; Vs represents the total scattering volume, SAB represents the geometric cross section of the detector A along the gamma ray, SBB represents the geometric cross section of the detector B along the gamma ray, denotes the Compton scattering cross section, denotes the differential cross section for Compton scattering, which can be obtained by the Klein-Nishina formula, and Ω denotes the scattering solid angle.

2. The method of claim 1, wherein, Before S10, the method further comprises: S00, based on the reconstructed PET image and the matching associated image and associated information, obtain a training sample for training the deep learning network; Wherein, each training sample comprises: a reconstructed first PET image / simulated first PET image, an attenuation coefficient image corresponding to the first PET image, a height, a weight, a radioactivity, an injection tracer interval time of a user to which the PET image belongs, and a real linear attenuation image obtained by verifying a PET / CT or PET / MR scan; S01, training the deep learning network based on the training sample to obtain the trained deep learning network; The network parameter θ in the trained deep learning network makes the loss function L of optimizing the training deep learning network minimum, so that the unscattered corrected PET image and the attenuation coefficient image in the axial field of view range are mapped to the unscattered corrected PET image and the attenuation coefficient image outside the extended axial field of view range.

3. The method of claim 2, wherein, S01 comprises: the network parameter θ of the deep learning network G and the output of G are represented as follows: ； wherein, is a first PET image, is a first attenuation coefficient image, is a second PET image, 2is a second attenuation coefficient image, is a height of the user, is a weight of the user, is a radioactivity, t a is an injection tracer interval; second PET image , second attenuation coefficient image as output to G, During the validation process, the output of G is compared to the true linear attenuation image obtained from a PET / CT or PET / MR scan , and based on the reconstructed image .

4. The method of claim 1, wherein, The deep learning network is one of the following: CNN network, Unet network, GAN network.

5. The method of claim 3, wherein, The S10 comprises: first pet image within an axial field of view , the linear attenuation coefficient distribution is kept constant is constant, obtained by maximizing the log-likelihood function by means of the MLEM algorithm; Linear attenuation coefficient distribution within an axial field of view , maintaining a first PET image within the axial field of view is constant for the unknown linear attenuation coefficient distribution maximizing a log-likelihood function, calculating a first attenuation coefficient image within the axial field of view from the PET data .

6. The method of claim 2, wherein, The training sample further comprises: A multi-bed training sample, which comprises: a PET image of the reconstructed multi-bed detection data and an attenuation coefficient image of the PET image, a height, a weight, a radioactivity, an injection tracer interval time of a user to which the PET image belongs, and a real linear attenuation image obtained by verifying a PET / CT or PET / MR scan; In use, the deep learning network outputs scatter correction information for each bed position .

7. A method of reconstruction of a PET image, characterized in that, The method comprises: P01. Based on the specified axial field of view range (0, D], the scatter correction information of the detection data is obtained by using the PET image scatter correction method of any one of claims 1 to 5. ; P02. The method of claim 1, wherein the PET image is reconstructed using a pre-established log-likelihood function. and the pre-established log-likelihood function to perform PET reconstruction to obtain a scatter correction accurate PET image.

8. A PET system characterized by, The method comprises: A memory and a processor; the memory has stored computer program instructions, and the processor executes the computer program instructions stored in the memory, and specifically executes the method of any one of claims 1 to 7.

Images