Method for identifying and processing artifact region in nuclear medicine imaging image

Abstract

The invention discloses a method for identifying and processing an artifact region in a nuclear medicine imaging image, and relates to the technical field of medical image registration. Comprising the following steps: S1, acquiring a PET dynamic image sequence, a CT dual-energy spectrum image and a respiratory movement monitoring signal, and generating a registration quality map; s2, the PET dynamic image sequence and the CT image are combined, statistical dependence between the PET dynamic image sequence and the CT image is obtained, and a focus of an artifact area in the CT image is positioned or reconstructed and compensated; and S3, carrying out artifact reconstruction on the optimized CT image, and processing a reconstructed artifact region through a double-branch network model. According to the method, the instantaneous phase of the respiratory signal is extracted through Hilbert transform, the PET dynamic image sequence is grouped according to the respiratory phase, accurate matching of same-phase frames can be achieved, meanwhile, the registration quality map is generated through the B-spline Demons algorithm, and organ deformation artifacts caused by respiratory movement can be remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image registration, and in particular to a method for identifying and processing artifact regions in nuclear medicine imaging images. Background Art

[0002] Nuclear medicine imaging (such as PET / CT, SPECT, etc.) is an important medical imaging technology that dynamically observes the metabolic and functional status of the human body through the distribution of tracers. It has a wide range of applications in tumor diagnosis, cardiovascular disease assessment, and neuroscience research. However, due to interference from various factors during the imaging process, artifacts often appear in the image, which significantly affects image quality and diagnostic accuracy. The sources of artifacts are complex and mainly include: equipment-related artifacts (such as uneven detector sensitivity, system noise, photon statistical fluctuations, etc.), motion artifacts (image blur or misalignment caused by patient breathing, heartbeat, or autonomic activity), attenuation artifacts (signal attenuation distortion caused by differences in tissue density), scattering artifacts (inelastic scattering of gamma-ray or positron annihilation photons), and metal implant artifacts (beam hardening effects caused by high-density materials such as titanium alloys).

[0003] Chinese invention patent publication number CN110163897A discloses a multimodal image registration method based on synthetic ultrasound images. Based on magnetic resonance images and real ultrasound images, a generative adversarial network consisting of a generator and a discriminator is constructed to generate a realistic synthetic ultrasound image. The simulated synthetic ultrasound image and the real ultrasound image are then registered to obtain registration parameters. The registration parameters are then applied to the magnetic resonance image to complete the final registration and fusion of the magnetic resonance image and ultrasound image. This method can synthesize ultrasound images from magnetic resonance images in real time, meeting the requirements of real-time image-guided surgery. The synthesized realistic ultrasound images are closer to real ultrasound images, have higher image quality, and better preserve important details. Even when the magnetic resonance image contains a tumor, the realistic ultrasound image can still be accurately synthesized. The final registration technology does not require a complex registration algorithm, but only requires a traditional simple registration algorithm to achieve good registration results.

[0004] However, in nuclear medicine multimodal imaging, high-uptake lesions (such as inflammation or tumors) and metal implant artifacts both appear as localized abnormalities in radioactive uptake in PET / SPECT images, and both have similar grayscale distributions and texture characteristics. Single-modality imaging cannot provide sufficient anatomical information to distinguish between the two, so most rely on multimodal data such as CT / MRI for auxiliary judgment. However, due to factors such as resolution differences between modalities and organ deformation caused by respiratory movement, the cross-modality registration process introduces new spatial misalignment and noise interference, which may not only obscure the true lesion but also generate false signals during artifact correction, which in turn affects the precise localization of the lesion and artifact suppression. Summary of the Invention

[0005] The object of the present invention is to provide a method for identifying and processing artifact regions in nuclear medicine imaging images, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for identifying and processing artifact areas in nuclear medicine imaging images, comprising:

[0007] S1: Data processing: Acquire PET dynamic image sequences, CT dual-energy spectral images, and respiratory motion monitoring signals, and generate a registration quality map using the B-spline Demons algorithm, including:

[0008] S1.1: Acquire image data: Acquire PET dynamic image sequences, CT dual-energy spectral images, and respiratory waveform signals through the PET detector, CT scanner, and respiratory monitoring system, respectively;

[0009] S1.2: performing phase grouping: obtaining the instantaneous phase of the respiratory waveform signal by Hilbert transform, and grouping the PET dynamic image sequence according to the instantaneous phase;

[0010] S1.3: CT parameter optimization: Based on the CT dual-energy spectrum image, obtain the linear attenuation coefficient. Simultaneously, based on the linear attenuation coefficient and the mass attenuation coefficient data of the titanium alloy, construct a titanium alloy energy spectrum fitting formula, determine the photoelectric effect weight coefficient and Compton scattering weight coefficient of titanium, and adjust the CT scanner to obtain an optimized CT image.

[0011] S2: Obtaining a generated quality map: combining the grouped PET dynamic image sequence and the optimized CT image using a B-spline Demons algorithm to obtain a statistical dependency between the PET dynamic image sequence and the CT image, and locating or reconstructing a lesion in an artifact area in the CT image based on the statistical dependency;

[0012] S3: Image reconstruction: performing artifact reconstruction on the optimized CT image, and processing the reconstructed artifact area through a dual-branch network model to obtain a reconstructed CT image.

[0013] Furthermore, the PET dynamic image sequence is grouped, including:

[0014] S1.2.1: Obtaining an analytical signal: Performing a Hilbert transform on the respiratory waveform signal to obtain an analytical signal, specifically:

[0015]

[0016] in: To analyze the signal, is the original respiratory signal, is the imaginary unit, is the Hilbert transform of the respiratory signal, is the Hilbert transform operator;

[0017] S1.2.2: Phase extraction: According to the analytical signal, obtain the corresponding instantaneous phase, and normalize the instantaneous phase to obtain the normalized instantaneous phase, specifically:

[0018]

[0019] in: is the instantaneous phase, is the original respiratory signal, is the Hilbert transform of the respiratory signal, is pi, is the normalized instantaneous phase;

[0020] S1.2.3: Phase grouping: grouping the normalized instantaneous phases according to a preset separation interval to obtain a normalized instantaneous phase grouping table;

[0021] S1.2.4: PET image matching: Based on the acquisition time of each frame of PET image, determine the displacement size monitored by the respiratory monitoring system corresponding to the acquisition time, and repeat steps S1.2.1-step S1.2.3 to obtain the normalized instantaneous phase corresponding to the displacement size, and match the PET image corresponding to the normalized instantaneous phase with the normalized instantaneous phase grouping table to obtain the grouped PET image.

[0022] Furthermore, the grouped PET images were verified, including:

[0023] W1: Perform intra-group consistency verification: Obtain the SUVmax values ​​corresponding to all frames of PET images within the normalized instantaneous phase group, determine the standard deviation of the SUVmax corresponding to the normalized instantaneous phase group, and determine the consistency quality result corresponding to the normalized instantaneous phase group based on the comparison result of the standard deviation of the SUVmax and the preset standard deviation threshold;

[0024] W2: Perform respiratory cycle stability verification: obtain the standard deviation of the respiratory cycle duration based on the corresponding number of consecutive complete respiratory cycles in the normalized instantaneous phase group, and determine the cycle stability quality result corresponding to the normalized instantaneous phase group based on the comparison result of the standard deviation of the respiratory cycle duration and the preset duration standard deviation threshold.

[0025] Furthermore, based on the comparison result of the standard deviation of the SUVmax and the preset standard deviation threshold, the consistency quality result corresponding to the normalized instantaneous phase group is determined, specifically:

[0026] When the standard deviation of the SUVmax is less than the preset standard deviation threshold, the normalized instantaneous phase group corresponding to the standard deviation of the SUVmax passes the consistency quality test; otherwise, the normalized instantaneous phase group corresponding to the standard deviation of the SUVmax fails the consistency quality test, and steps S1.1 to S1.2.4 are repeated to reacquire the respiratory waveform signal and perform grouping processing, and step W1 is repeated until the standard deviation of the SUVmax is less than the preset standard deviation threshold;

[0027] The formula for obtaining the standard deviation of SUVmax is specifically:

[0028]

[0029] in: is the standard deviation of SUVmax, is the total number of PET image frames in the current normalized instantaneous phase group, is the SUVmax value of the i-th frame PET image, is the average value of SUVmax of all frames of PET images in the current normalized instantaneous phase group, is the frame index of the PET image within the current normalized instantaneous phase group.

[0030] Furthermore, based on the comparison result of the standard deviation of the respiratory cycle duration and the preset duration standard deviation threshold, the cyclic stability quality result corresponding to the normalized instantaneous phase group is determined, specifically:

[0031] When the standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold, the normalized instantaneous phase group corresponding to the standard deviation of the respiratory cycle duration passes the cyclic stability quality test; otherwise, the normalized instantaneous phase group corresponding to the standard deviation of the respiratory cycle duration fails the cyclic stability quality test, and steps S1.1 to S1.2.4 are repeated to adjust the normalized instantaneous phase grouping table, and step W2 is repeated until the standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold;

[0032] The formula for obtaining the standard deviation of the respiratory cycle duration is specifically:

[0033]

[0034] in: is the standard deviation of the respiratory cycle length, is the total number of respiratory cycles, is the duration of the jth respiratory cycle, is the average respiratory cycle duration, is the index of the respiratory cycle.

[0035] Furthermore, the photoelectric effect weight coefficient and Compton scattering weight coefficient of titanium are corrected, including:

[0036] M1: Periodic verification: According to the set detection cycle, the photoelectric effect weight coefficient and the Compton scattering weight coefficient are regularly compared with the preset photoelectric effect weight threshold and the preset Compton scattering weight threshold respectively, and the final weight coefficient is determined based on the comparison results;

[0037] M2: Coefficient adjustment: The actual X-ray photon energy is obtained through a high-voltage sensor, and the photoelectric effect weight coefficient is corrected. Specifically:

[0038]

[0039] in: is the corrected photoelectric effect weight coefficient of titanium, is the photoelectric effect weight coefficient of titanium, is the previously measured X-ray photon energy, is the actual measured X-ray photon energy;

[0040] The real-time working temperature of the detector is obtained through the temperature sensor, and the Compton scattering weight coefficient is corrected as follows:

[0041]

[0042] in: is the modified Compton scattering weight coefficient of titanium, is the Compton scattering weight coefficient of titanium, is the temperature compensation coefficient, is the working temperature actually measured by the detector, is the previous operating temperature of the detector.

[0043] Furthermore, the error ratio between the photoelectric effect weight coefficient and the preset photoelectric effect weight threshold, and the error ratio between the Compton scattering weight coefficient and the preset Compton scattering weight threshold are obtained, and the error ratio is compared with the preset ratio threshold. The final weight coefficient is determined based on the comparison result, specifically:

[0044] When the error ratio is less than the preset ratio threshold, the photoelectric effect weight coefficient and Compton scattering weight coefficient corresponding to the error ratio are the final photoelectric effect weight coefficient and Compton scattering weight coefficient. Otherwise, the next step M2 is executed to correct the photoelectric effect weight coefficient and Compton scattering weight coefficient, and step M1 is repeated according to the corrected weight coefficient until the error ratio is less than the preset ratio threshold.

[0045] Furthermore, the lesion in the artifact area in the CT image is located or reconstructed and compensated, including:

[0046] S2.1: Obtaining the statistical dependence between images: Based on the deformation field constructed in the B-spline Demons algorithm, the grayscale values ​​of the PET image, and the grayscale values ​​of the CT image, obtain the statistical dependence between the PET image and the CT image. Specifically:

[0047]

[0048] in: is the statistical dependence between PET images and CT images, is the total gray value of the CT image, is the total grayscale value of the PET image, is the grayscale value of the CT image, is the grayscale value of the PET image, is the grayscale joint probability of the CT image and PET image at the same spatial position, is the gray value in the CT image The probability of occurrence, is the gray value of the PET image Probability of occurrence;

[0049] S2.2: Determine the final optimized CT image: Compare the statistical dependence with a preset dependence threshold, and determine the final optimized CT image based on the comparison result, specifically:

[0050] When the statistical dependence is greater than a preset dependence threshold, the optimized CT image corresponding to the statistical dependence is the final optimized CT image. Otherwise, steps S1.3 to S2.2 are repeated to reacquire the optimized CT image until the statistical dependence is greater than the preset dependence threshold.

[0051] S2.3: J value comparison: The final optimized CT image and the grouped PET dynamic image sequence are used as inputs to the B-spline Demons algorithm model, and the J value of each position point between the CT image and the PET image is output, specifically:

[0052]

[0053] in: is the Jacobian determinant, The spatial coordinates in the image are The displacement vector corresponding to the control point of is the Jacobian matrix of the displacement vector, is the weight coefficient of the kth control point, is the B-spline basis function of the k-th control point, is the number of control points, is the spatial coordinate in the image, is the determinant of the matrix;

[0054] The J value is compared with a preset J threshold range, and image fusion / reconstruction / labeling is performed based on the comparison result, specifically:

[0055] When the J value is not less than the upper limit of the preset J threshold range, step S2.4 is executed to fuse the image area corresponding to the J value. When the J value is within the preset J threshold range, step S3 is executed to reconstruct the image area corresponding to the J value. When the J value is less than the lower limit of the preset J threshold range, the image area corresponding to the J value is manually marked, deleted, and re-acquired.

[0056] S2.4: Image fusion: Fuse the grayscale values ​​corresponding to the fused image region in the PET image with the corresponding grayscale values ​​in the CT image to obtain the grayscale values ​​of the region in the fused image, specifically:

[0057]

[0058] in: The spatial coordinates of the fused image are The grayscale value corresponding to the control point of is the weight coefficient of the PET image, The spatial coordinates in the PET image are The grayscale value corresponding to the control point of The spatial coordinates in the CT image are The grayscale value corresponding to the control point of are the spatial coordinates in the image.

[0059] Furthermore, the reconstructed CT image is obtained, including:

[0060] S3.1: Artifact reconstruction: Perform artifact quantitative correction based on the point spread function of titanium alloy, specifically:

[0061]

[0062] in: is the point spread function, is the standard deviation of the Gaussian distribution, is the base of natural logarithms, is the influence distance of metal artifacts, is the inverse operation of convolution operation, is the corrected standardized uptake value, is the SUV value measured by the original PET;

[0063] S3.2: Obtaining feature data: The artifact-corrected CT image is used as the input of the anatomical branch of the two-branch network model, and the time-activity curve corresponding to the PET image is used as the input of the functional branch of the two-branch network model. The proliferation and inflammation features are obtained as outputs, specifically:

[0064]

[0065] in: is the time activity curve, is the initial activity of the first component, is the initial activity of the second component, is the decay rate constant of the first component, is the decay rate constant of the second component, For time, is the base of natural logarithms;

[0066] S3.3: Obtaining attention weights: Determine the cross-modal attention weights based on the eigenvectors of the CT image and the eigenvectors of the PET image. Specifically:

[0067]

[0068] in: is the correlation between PET features and CT features at position m, n, is the query vector of the mth PET feature, is the inversion of the key vector of the nth CT feature, is the dimension of the feature vector, 、 is a trainable linear transformation weight matrix, is the feature vector corresponding to the PET image, is the feature vector corresponding to the CT image;

[0069] S3.4: Image reconstruction: Based on the proliferation characteristics, inflammation characteristics and cross-modal attention weights, the optimized CT image is used as the input of the dual-branch network model, and the reconstructed CT image is obtained as the output.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] First, the present invention extracts the instantaneous phase of the respiratory signal through Hilbert transform and groups the PET dynamic image sequence according to the respiratory phase, which can achieve accurate matching of the same-phase frames. At the same time, non-rigid registration is performed through the B-spline Demons algorithm to generate a registration quality map, which can significantly reduce organ deformation artifacts caused by respiratory motion. That is, through dynamic group registration, the spatial consistency of PET images and CT images is improved, avoiding the obscuration or dislocation of lesions due to respiratory motion.

[0072] Second, the present invention uses dual-energy CT data to construct a photoelectric effect and Compton scattering weight coefficient model of titanium alloys and dynamically adjusts CT scanning parameters. Combined with real-time tube voltage feedback and temperature drift compensation, this method can improve attenuation correction accuracy, effectively suppress metal artifacts, avoid confusion between high-density artifacts and high-uptake lesions, and enhance the accuracy of tumor or inflammation localization.

[0073] Third: The present invention quantifies the statistical dependence between PET images and CT images through mutual information, and screens out the optimal registration results. At the same time, it combines the Jacobian determinant to analyze the local characteristics of the deformation field, so that it can adaptively distinguish artifacts from lesions, avoid erroneous fusion or reconstruction, and improve the reliability of multimodal images. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Schematic diagram of the flow of the artifact area identification and processing method of the present invention;

[0075] Figure 2 Schematic diagram of the process of obtaining grouped PET images in the present invention;

[0076] Figure 3 A schematic diagram of a flow chart for determining an image processing method in the present invention;

[0077] Figure 4 This is a diagram analyzing the respiratory cycle stability in the present invention;

[0078] Figure 5 This is the iterative calibration diagram of the photoelectric effect weight coefficient in the present invention. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] In nuclear medicine multimodal imaging, high-uptake lesions (such as inflammation or tumors) and metal implant artifacts both appear as localized abnormalities in radioactive uptake in PET / SPECT images, and both have similar grayscale distributions and texture characteristics. Single-modality imaging cannot provide sufficient anatomical information to distinguish between the two, so most rely on multimodal data such as CT / MRI for auxiliary judgment. However, due to factors such as resolution differences between modalities and organ deformation caused by respiratory motion, the cross-modality registration process introduces new spatial misalignment and noise interference, which may not only obscure the true lesion but also generate false signals during artifact correction, affecting the precise localization of the lesion and artifact suppression. The technical solution of the present application realizes non-rigid registration in the same phase and eliminates motion artifacts by processing the correlation of PET dynamic image sequences, CT dual-energy spectrum images and respiratory motion monitoring signals. At the same time, a titanium alloy attenuation coefficient model is constructed based on CT dual-energy spectrum data, the weight coefficient is optimized, and the optimized CT image is obtained. The statistical dependence between PET images and CT images is evaluated through mutual information, and the image areas are divided into three categories: fusion, reconstruction and label deletion, and corresponding processing is performed, so as to effectively suppress metal artifacts, avoid confusion between high-density artifacts and high-uptake lesions, and improve the accuracy of tumor or inflammation positioning.

[0081] Example 1

[0082] refer to Figure 1-Figure 3 This embodiment provides a method for identifying and processing artifact regions in nuclear medicine imaging images. The method specifically includes the following steps:

[0083] Step S1: Data processing. This involves acquiring PET dynamic image sequences, CT dual-energy spectral images, and respiratory motion monitoring signals through timestamp synchronization. PET images are grouped based on respiratory phases, and the B-spline Demons algorithm is used to perform frame-by-frame non-rigid registration of the PET and CT images to obtain a registration quality map. The details are as follows:

[0084] Step S1.1: Acquire image data. This involves acquiring dynamic PET image sequences using a PET detector, acquiring dual-energy CT images using a CT scanner, and acquiring displacements of patient surface markers (such as the sternal notch and xiphoid process) using a respiratory monitoring system (e.g., the C-RAD Catalyst optical surface monitoring system). These are then converted into corresponding respiratory waveform signals.

[0085] Furthermore, during the data acquisition process, nanosecond-level time synchronization is established between the PET detector, CT scanner, and respiratory monitoring system through the IEEE 1588 precision time protocol, thereby enabling the acquisition of PET dynamic image sequences, CT dual-energy spectral images, and respiratory motion monitoring signals through timestamp synchronization.

[0086] Step S1.2: Phase grouping. Specifically, based on the respiratory waveform signal acquired in step S1.1, a Hilbert transform is performed on the respiratory waveform signal, and the corresponding instantaneous phase is determined based on the Hilbert transform result. Simultaneously, the instantaneous phase is evenly divided according to the preset group size, and the phase interval corresponding to the instantaneous phase of each group is obtained. PET images are then grouped based on the phase interval. The details are as follows:

[0087] Step S1.2.1: Obtain analytical signal. The respiratory waveform signal obtained in step S1.1 is subjected to Hilbert transform to obtain the corresponding analytical signal, specifically:

[0088]

[0089] in: To analyze the signal, is the original respiratory signal, is the imaginary unit, is the Hilbert transform of the respiratory signal, is the Hilbert transform operator.

[0090] Step S1.2.2: Phase extraction. That is, based on the analytical signal obtained in step S1.2.1, determine its corresponding instantaneous phase, specifically:

[0091]

[0092] in: is the instantaneous phase, is the original respiratory signal, is the Hilbert transform of the respiratory signal, is pi.

[0093] Furthermore, the instantaneous phase obtained is normalized so that its phase size is mapped to the range of 0%-100%, specifically:

[0094]

[0095] in: is the normalized instantaneous phase, is the instantaneous phase, is pi.

[0096] Step S1.2.3: Phase grouping. That is, the normalized instantaneous phases obtained in step S1.2.2 are grouped according to a preset separation interval (which can be specifically set according to actual needs, so it is not specifically explained in this embodiment) to obtain a normalized instantaneous phase grouping table.

[0097] In the specific implementation process, the preset separation interval is set to 12.5%, so the normalized instantaneous phase in the range of 0%-100% can be divided into 8 groups, as shown in the following Table 1:

[0098] Table 1: Normalized instantaneous phase grouping table

[0099]

[0100] Step S1.2.4: PET Image Matching. This involves determining the displacement recorded by the respiratory monitoring system corresponding to the acquisition time of each PET image frame. Repeat steps S1.2.1 through S1.2.3 to obtain the normalized instantaneous phase corresponding to the displacement. This normalized instantaneous phase is the normalized instantaneous phase corresponding to the PET image frame.

[0101] Furthermore, the normalized instantaneous phase corresponding to each frame of the PET image is matched with the phase interval grouped in step S1.2.3, thereby matching the PET image with the normalized instantaneous phase grouping table in step S1.2.3 to obtain the grouped PET image.

[0102] During the implementation, the acquisition time for the 15th PET image was 70 seconds, and its corresponding normalized instantaneous phase was 43.3%. Based on the phase intervals in Table 1, this 15th PET image matches group 4 in the normalized instantaneous phase grouping table.

[0103] Step S1.3: CT parameter optimization. That is, based on the CT dual-energy spectrum image obtained in step S1.1, it is decomposed into a linear combination of the photoelectric effect and Compton scattering, and the linear attenuation coefficient corresponding to the CT dual-energy spectrum image is obtained, specifically:

[0104]

[0105] in: The energy of X-ray photons in matter The linear attenuation coefficient under is the weight coefficient of the photoelectric effect term, The energy of X-ray photons in matter The photoelectric effect attenuation coefficient under is the weight coefficient of the Compton scattering term, The energy of X-ray photons in matter The Compton scattering attenuation coefficient under is the energy of X-ray photons.

[0106] Furthermore, a titanium alloy energy spectrum fitting formula was constructed based on the linear attenuation coefficient calculation formula and titanium alloy mass attenuation coefficient data. The titanium alloy energy spectrum fitting formula was also used to determine the photoelectric effect weight coefficient and Compton scattering weight coefficient of titanium. Based on these determined photoelectric effect weight coefficients and Compton scattering weight coefficients, adjustments were made to the X-ray tube energy spectrum filtering, detector response curve, and high-voltage generator stability obtained by the CT scanner. Optimized CT images were then obtained using the adjusted CT scanner.

[0107] In this embodiment, the titanium alloy energy spectrum fitting formula is specifically:

[0108]

[0109] in: is the linear attenuation coefficient of titanium at 80 kVp energy, is the photoelectric effect weight coefficient of titanium, is the linear attenuation coefficient of titanium at 140 kVp energy, is the Compton scattering weight coefficient of titanium, is the photoelectric effect attenuation coefficient of titanium at 80kVp energy, is the photoelectric effect attenuation coefficient of titanium at 140 kVp energy, is the Compton scattering attenuation coefficient of titanium at 80 kVp energy, is the Compton scattering attenuation coefficient of titanium at 140 kVp energy.

[0110] Step S2: Obtain the generated quality map. That is, the grouped PET images obtained in step S1.2.4 are processed using the B-spline Demons algorithm to obtain the in-phase frames in the PET images. Simultaneously, the CT images optimized in step S1.3 are processed to suppress the metal artifacts in the CT images. The in-phase frames in the PET images and the CT images after metal artifact suppression are combined to obtain the statistical dependence between the two. At the same time, based on the obtained statistical dependence, the lesions in the artifact area are located or reconstructed and compensated. The details are as follows:

[0111] Step S2.1: Obtain the statistical dependence between the images. Specifically, based on the deformation field constructed in the B-spline Demons algorithm and the grayscale values ​​of the PET image and the CT image in the deformation field, the statistical dependence between the PET image and the CT image is determined. Specifically,

[0112]

[0113] in: is the statistical dependence between PET images and CT images, is the total gray value of the CT image, is the total gray value of the PET image, is the grayscale value of the CT image, is the grayscale value of the PET image, is the grayscale joint probability of the CT image and PET image at the same spatial position, is the gray value in the CT image The probability of occurrence, is the gray value of the PET image Probability of occurrence.

[0114] Step S2.2: Determine the final optimized CT image. The statistical dependence obtained in step S2.1 is compared with a preset dependence threshold, and the final optimized CT image is determined based on the comparison result. Specifically:

[0115] If the obtained statistical dependence is greater than the preset dependence threshold, the optimized CT image corresponding to the obtained statistical dependence is the final optimized CT image. Otherwise, steps S1.3 through S2.2 are repeated to obtain a new optimized CT image until the obtained statistical dependence is greater than the preset dependence threshold.

[0116] Step S2.3: J value comparison. The final optimized CT image determined in step S2.2 and the grouped PET image obtained in step S1.2.4 are used as inputs to the B-spline Demons algorithm model. The J value of each position point between the CT image and the PET image is output as follows:

[0117]

[0118] in: is the Jacobian determinant, The spatial coordinates in the image are The displacement vector corresponding to the control point of is the Jacobian matrix of the displacement vector, is the weight coefficient of the kth control point, is the B-spline basis function of the k-th control point, is the number of control points, is the spatial coordinate in the image, is the determinant of the matrix.

[0119] Furthermore, the obtained J value is compared with the preset J threshold range, and image fusion / reconstruction / labeling is performed based on the comparison result. Specifically:

[0120] When the obtained J value is not less than the upper limit of the preset J threshold range, step S2.4 is executed to fuse the image area corresponding to the J value. When the obtained J value is within the preset J threshold range, step S3 is executed to reconstruct the image area corresponding to the J value. When the obtained J value is less than the lower limit of the preset J threshold range, the image area corresponding to the J value is manually marked, deleted, and re-acquired.

[0121] Step S2.4: Image fusion. That is, based on the image region that can be fused in step S2.3, the grayscale value corresponding to the fused image region in the PET image is fused with the grayscale value corresponding to the CT image, and the grayscale value of the region in the fused image is obtained, specifically:

[0122]

[0123] in: The spatial coordinates of the fused image are The grayscale value corresponding to the control point of is the weight coefficient of the PET image, The spatial coordinates in the PET image are The grayscale value corresponding to the control point of The spatial coordinates in the CT image are The grayscale value corresponding to the control point of are the spatial coordinates in the image.

[0124] Step S3: Image reconstruction. This involves performing artifact reconstruction on the CT image region corresponding to the J value within the preset J threshold range, and processing the reconstructed artifact region using a dual-branch network model to obtain a reconstructed CT image. This is specifically as follows:

[0125] Step S3.1: Artifact reconstruction. This involves obtaining the corresponding point spread function based on the physical parameters of the titanium alloy, namely its density and photon ray energy. Based on this point spread function, artifact quantization correction is performed on the marked artifact areas that require reconstruction through point spread function deconvolution. Specifically, the following steps are performed:

[0126]

[0127] in: is the point spread function, is the standard deviation of the Gaussian distribution, is the base of natural logarithms, is the influence distance of metal artifacts, is the inverse operation of convolution operation, is the corrected standardized uptake value, SUV value measured by original PET.

[0128] Step S3.2: Acquire feature data. This involves using the anatomical branch and the functional branch of the constructed two-branch network model to obtain a 64-dimensional anatomical feature vector, proliferation features, and inflammatory features. Specifically, the anatomical branch uses the artifact-corrected CT image from step S3.1 as input, and outputs the corresponding bone texture gradient and metal-tissue interface sharpness, i.e., a 64-dimensional anatomical feature vector, to determine the location of metal in the CT image.

[0129] Furthermore, based on the determined metal location in the CT image, the time activity curve corresponding to the PET image, that is, the rate of change of the SUVmax value in the PET image over time, is used as the input of the functional branch, and the corresponding proliferation and inflammation characteristics are output, specifically:

[0130]

[0131] in: is the time activity curve, is the initial activity of the first component, is the initial activity of the second component, is the decay rate constant of the first component, is the decay rate constant of the second component, For time, is the base of natural logarithms.

[0132] Step S3.3: Obtaining attention weights. Specifically, based on the feature vectors corresponding to the CT image (e.g., bone density, metal artifact contours) and the feature vectors corresponding to the PET image (e.g., metabolic activity, proliferation features, and inflammation features obtained in step S3.2), the cross-modal attention weights are determined as follows:

[0133]

[0134] in: is the correlation between PET features and CT features at position m, n, is the query vector of the mth PET feature, is the inversion of the key vector of the nth CT feature, is the dimension of the feature vector, 、 is a trainable linear transformation weight matrix, is the feature vector corresponding to the PET image, is the feature vector corresponding to the CT image.

[0135] Step S3.4: Image reconstruction. Based on the proliferation and inflammation features determined in step S3.2 and the cross-modal attention weights determined in step S3.3, the CT image region to be reconstructed, determined in step S2.3, is used as input to the dual-branch network model in this embodiment, and the corresponding reconstructed CT image is obtained as output.

[0136] Example 2

[0137] This embodiment provides a method for identifying and processing artifact regions in nuclear medicine imaging images. The specific implementation method is the same as in Example 1, except that, during the PET image matching process in step S1.2.4, the post-matching grouping effect is verified. The present invention is described below with reference to specific implementations of this embodiment.

[0138] In this embodiment, the grouping effect of the PET image matching in step S1.2.4 is verified by intra-group consistency verification and respiratory cycle stability verification, as follows:

[0139] Step W1: Perform intra-group consistency verification. That is, based on all PET image frames corresponding to the same normalized instantaneous phase group, obtain the SUVmax value corresponding to each PET image frame. Based on this SUVmax value, obtain the standard deviation of all PET image frames in the same normalized instantaneous phase group. Specifically,

[0140]

[0141] in: is the standard deviation of SUVmax, is the total number of PET image frames in the current normalized instantaneous phase group, is the SUVmax value of the i-th frame PET image, is the average value of SUVmax of all frames of PET images in the current normalized instantaneous phase group, is the frame index of the PET image within the current normalized instantaneous phase group.

[0142] Furthermore, the obtained standard deviation of the SUVmax corresponding to the current normalized instantaneous phase group is compared with the preset standard deviation threshold, and the consistency quality result corresponding to the current normalized instantaneous phase group is determined based on the comparison result. Specifically:

[0143] If the standard deviation of the SUVmax corresponding to the current normalized instantaneous phase group is less than the preset standard deviation threshold, the current normalized instantaneous phase group passes the consistency quality test. Otherwise, the current normalized instantaneous phase group fails the consistency quality test and steps S1.1 through S1.2.4 must be repeated to reacquire the respiratory waveform signals, perform grouping processing, and repeat step W1 until the standard deviation of the SUVmax corresponding to the current normalized instantaneous phase group is less than the preset standard deviation threshold, and the current normalized instantaneous phase group passes the consistency quality test.

[0144] In the specific implementation process, the consistency quality test data table shown in Table 2 below is set up, specifically:

[0145] Table 2: Consistency quality test data table

[0146]

[0147] According to the data in Table 2 above, the average SUVmax value corresponding to the PET images in this group of frames is 2.45, and the corresponding standard deviation of SUVmax is 0.13. Furthermore, the preset standard deviation threshold in this embodiment is set to 0.5, which means that this normalized instantaneous phase group passes the consistency quality test.

[0148] Step W2: Verify the respiratory cycle stability. This involves obtaining a continuous number of complete respiratory cycles from each normalized instantaneous phase group, obtaining the time differences between adjacent inhalation start points, and determining the standard deviation of the respiratory cycle duration using these time differences. Specifically, the standard deviation is:

[0149]

[0150] in: is the standard deviation of the respiratory cycle length, is the total number of respiratory cycles, is the duration of the j-th respiratory cycle, is the average respiratory cycle duration, is the index of the respiratory cycle.

[0151] It is worth noting that the number of complete breathing cycles can be specifically set according to actual needs, so it is not explained in this embodiment.

[0152] Furthermore, the obtained standard deviation of the respiratory cycle duration is compared with a preset duration standard deviation threshold, and based on the comparison result, the cycle stability quality result corresponding to the current normalized instantaneous phase group is determined. Specifically:

[0153] If the obtained standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold, the current normalized instantaneous phase group passes the cyclic stability quality test. Otherwise, the current normalized instantaneous phase group fails the cyclic stability quality test, and steps S1.2.1 through S1.2.4 must be repeated to adjust the current normalized instantaneous phase grouping table. Step W2 must also be repeated until the obtained standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold, and the current normalized instantaneous phase group passes the cyclic stability quality test.

[0154] In the specific implementation process, a periodic stability quality test data table as shown in Table 3 below is provided, specifically:

[0155] Table 3: Cycle stability quality test data

[0156]

[0157] According to the data in Table 3, the average respiratory cycle duration within the normalized instantaneous phase group is 5 seconds, and the corresponding standard deviation of the respiratory cycle duration is approximately 0.15 seconds. Furthermore, the preset duration standard deviation threshold in this embodiment is set to 0.3 seconds. In other words, the normalized instantaneous phase group passes the cyclic stability quality test.

[0158] refer to Figure 4 , Figure 4 is a diagram for analyzing the stability of the respiratory cycle in this embodiment, Figure 4 It can be seen that when the standard deviation of the respiratory cycle length is 0.13 seconds, it is significantly lower than the preset threshold of 0.3 seconds. At this time, the respiratory signal is highly stable, with a fluctuation range of 4.8-5.2 seconds. This ensures that dynamic PET image registration is not affected by respiratory cycle fluctuations, providing a stable basis for motion artifact correction.

[0159] Example 3

[0160] This embodiment provides a method for identifying and processing artifact regions in nuclear medicine imaging images. The specific implementation method is the same as in Example 1, except that, in step S1.3, the determined photoelectric effect weight coefficient and Compton scattering weight coefficient are corrected based on real-time tube voltage feedback and temperature drift compensation. The present invention will be described below with reference to specific implementations of this embodiment.

[0161] In this embodiment, the photoelectric effect weight coefficient is corrected by real-time tube voltage feedback, and the Compton scattering weight coefficient is corrected by temperature drift compensation. The details are as follows:

[0162] Step M1: Periodic Verification. This involves verifying the photoelectric effect weight coefficient and Compton scattering weight coefficient determined in step S1.3 according to the set detection period. Specifically, the determined photoelectric effect weight coefficient is compared with a preset photoelectric effect weight threshold, and the determined Compton scattering weight coefficient is compared with a preset Compton scattering weight threshold. The final weight coefficients are determined based on the comparison results.

[0163] Furthermore, the error ratio between the photoelectric effect weight coefficient and the preset photoelectric effect weight threshold, and the error ratio between the Compton scattering weight coefficient and the preset Compton scattering weight threshold are obtained, and the obtained error ratio is compared with the preset ratio threshold, and the next step M2 is executed based on the comparison result. Specifically,

[0164] If the obtained error ratio is less than the preset ratio threshold, the determined photoelectric effect weight coefficient and Compton scattering weight coefficient are the final photoelectric effect weight coefficient and Compton scattering weight coefficient. Otherwise, the next step M2 is executed to correct the photoelectric effect weight coefficient and Compton scattering weight coefficient. Step M1 is repeated based on the corrected weight coefficients until the obtained error ratio is less than the preset ratio threshold.

[0165] In this embodiment, the formula for obtaining the error ratio is specifically:

[0166]

[0167] in: is the error ratio, is the actual weight coefficient obtained, is the preset weight threshold.

[0168] Step M2: Coefficient adjustment. This involves obtaining the actual X-ray photon energy through a high-voltage sensor, and correcting the photoelectric effect weight coefficient based on the obtained photon energy to obtain the corrected photoelectric effect weight coefficient, specifically:

[0169]

[0170] in: is the corrected photoelectric effect weight coefficient of titanium, is the photoelectric effect weight coefficient of titanium, is the previously measured X-ray photon energy, is the actual measured X-ray photon energy.

[0171] Furthermore, the real-time operating temperature of the detector is obtained through a temperature sensor, and the Compton scattering weight coefficient is corrected according to the real-time operating temperature of the detector to obtain the corrected Compton scattering weight coefficient, which is specifically:

[0172]

[0173] in: is the modified Compton scattering weight coefficient of titanium, is the Compton scattering weight coefficient of titanium, is the temperature compensation coefficient, is the working temperature actually measured by the detector, is the previous operating temperature of the detector.

[0174] refer to Figure 5 , Figure 5 is the iterative calibration diagram of the photoelectric effect weight coefficient in this embodiment, Figure 5 It can be seen that after iterative calibration, the photoelectric effect weight coefficient converged from an initial value of 1.1 to 1.01, reaching the target threshold. This means that the closed-loop feedback mechanism (such as tube voltage and temperature compensation) achieves precise optimization of dual-spectral CT parameters, ensuring accurate metal artifact correction and avoiding over- or under-correction.

[0175] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is limited by the accompanying embodiments and their equivalents.

Claims

1. A method for identifying and processing artifact areas in nuclear medicine imaging images, characterized in that: Includes: S1: Data processing: Acquire PET dynamic image sequences, CT dual-energy spectral images, and respiratory motion monitoring signals, and generate a registration quality map using the B-spline Demons algorithm, including: S1.1: Acquire image data: Acquire PET dynamic image sequences, CT dual-energy spectral images, and respiratory waveform signals through the PET detector, CT scanner, and respiratory monitoring system, respectively; S1.2: performing phase grouping: obtaining the instantaneous phase of the respiratory waveform signal by Hilbert transform, and grouping the PET dynamic image sequence according to the instantaneous phase; S1.3: CT parameter optimization: Based on the CT dual-energy spectrum image, obtain the linear attenuation coefficient. Simultaneously, based on the linear attenuation coefficient and the mass attenuation coefficient data of the titanium alloy, construct a titanium alloy energy spectrum fitting formula, determine the photoelectric effect weight coefficient and Compton scattering weight coefficient of titanium, and adjust the CT scanner to obtain an optimized CT image. S2: Obtaining a generated quality map: combining the grouped PET dynamic image sequence and the optimized CT image using a B-spline Demons algorithm to obtain statistical dependencies between the PET dynamic image sequence and the CT image, and locating or reconstructing lesions in artifact areas of the CT image based on the statistical dependencies; S3: Image reconstruction: performing artifact reconstruction on the optimized CT image, and processing the reconstructed artifact area through a dual-branch network model to obtain a reconstructed CT image.

2. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 1, characterized in that: The PET dynamic image sequence is grouped, including: S1.2.1: Obtaining an analytical signal: Performing a Hilbert transform on the respiratory waveform signal to obtain an analytical signal, specifically: ； in: To analyze the signal, is the original respiratory signal, is the imaginary unit, is the Hilbert transform of the respiratory signal, is the Hilbert transform operator; S1.2.2: Phase extraction: According to the analytical signal, obtain the corresponding instantaneous phase, and normalize the instantaneous phase to obtain the normalized instantaneous phase, specifically: ； in: is the instantaneous phase, is the original respiratory signal, is the Hilbert transform of the respiratory signal, is pi, is the normalized instantaneous phase; S1.2.3: Phase grouping: grouping the normalized instantaneous phases according to a preset separation interval to obtain a normalized instantaneous phase grouping table; S1.2.4: PET image matching: Based on the acquisition time of each frame of PET image, determine the displacement size monitored by the respiratory monitoring system corresponding to the acquisition time, and repeat steps S1.2.1-step S1.2.3 to obtain the normalized instantaneous phase corresponding to the displacement size, and match the PET image corresponding to the normalized instantaneous phase with the normalized instantaneous phase grouping table to obtain the grouped PET image.

3. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 2, characterized in that: Verification of the grouped PET images includes: W1: Perform intra-group consistency verification: Obtain the SUVmax values ​​corresponding to all frames of PET images within the normalized instantaneous phase group, determine the standard deviation of the SUVmax corresponding to the normalized instantaneous phase group, and determine the consistency quality result corresponding to the normalized instantaneous phase group based on the comparison result of the standard deviation of the SUVmax and the preset standard deviation threshold; W2: Perform respiratory cycle stability verification: obtain the standard deviation of the respiratory cycle duration based on the corresponding number of consecutive complete respiratory cycles in the normalized instantaneous phase group, and determine the cycle stability quality result corresponding to the normalized instantaneous phase group based on the comparison result of the standard deviation of the respiratory cycle duration and the preset duration standard deviation threshold.

4. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 3, characterized in that: According to the comparison result of the standard deviation of the SUVmax and the preset standard deviation threshold, the consistency quality result corresponding to the normalized instantaneous phase group is determined, specifically: When the standard deviation of the SUVmax is less than the preset standard deviation threshold, the normalized instantaneous phase group corresponding to the standard deviation of the SUVmax passes the consistency quality test; otherwise, the normalized instantaneous phase group corresponding to the standard deviation of the SUVmax fails the consistency quality test, and steps S1.1 to S1.2.4 are repeated to reacquire the respiratory waveform signal and perform grouping processing, and step W1 is repeated until the standard deviation of the SUVmax is less than the preset standard deviation threshold; The formula for obtaining the standard deviation of SUVmax is specifically: ； in: is the standard deviation of SUVmax, is the total number of PET image frames in the current normalized instantaneous phase group, is the SUVmax value of the i-th frame PET image, is the average value of SUVmax of all frames of PET images in the current normalized instantaneous phase group, is the frame index of the PET image within the current normalized instantaneous phase group.

5. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 3, characterized in that: According to the comparison result of the standard deviation of the respiratory cycle duration and the preset duration standard deviation threshold, the cyclic stability quality result corresponding to the normalized instantaneous phase group is determined, specifically: When the standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold, the normalized instantaneous phase group corresponding to the standard deviation of the respiratory cycle duration passes the cyclic stability quality test; otherwise, the normalized instantaneous phase group corresponding to the standard deviation of the respiratory cycle duration fails the cyclic stability quality test, and steps S1.1 to S1.2.4 are repeated to adjust the normalized instantaneous phase grouping table, and step W2 is repeated until the standard deviation of the respiratory cycle duration is less than the preset duration standard deviation threshold; The formula for obtaining the standard deviation of the respiratory cycle duration is specifically: ； in: is the standard deviation of the respiratory cycle length, is the total number of respiratory cycles, is the duration of the jth respiratory cycle, is the average respiratory cycle duration, is the index of the respiratory cycle.

6. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 1, characterized in that: Corrections are made to the photoelectric effect weight coefficient and Compton scattering weight coefficient of titanium, including: M1: Periodic verification: According to the set detection cycle, the photoelectric effect weight coefficient and the Compton scattering weight coefficient are regularly compared with the preset photoelectric effect weight threshold and the preset Compton scattering weight threshold respectively, and the final weight coefficient is determined based on the comparison results; M2: Coefficient adjustment: The actual X-ray photon energy is obtained through a high-voltage sensor, and the photoelectric effect weight coefficient is corrected. Specifically: ； in: is the corrected photoelectric effect weight coefficient of titanium, is the photoelectric effect weight coefficient of titanium, is the previously measured X-ray photon energy, is the actual measured X-ray photon energy; The real-time working temperature of the detector is obtained through the temperature sensor, and the Compton scattering weight coefficient is corrected as follows: ； in: is the modified Compton scattering weight coefficient of titanium, is the Compton scattering weight coefficient of titanium, is the temperature compensation coefficient, is the working temperature actually measured by the detector, is the previous operating temperature of the detector.

7. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 6, characterized in that: Obtain the error ratio between the photoelectric effect weight coefficient and the preset photoelectric effect weight threshold, and the error ratio between the Compton scattering weight coefficient and the preset Compton scattering weight threshold, and compare the error ratio with the preset ratio threshold. Based on the comparison result, determine the final weight coefficient, specifically: When the error ratio is less than the preset ratio threshold, the photoelectric effect weight coefficient and Compton scattering weight coefficient corresponding to the error ratio are the final photoelectric effect weight coefficient and Compton scattering weight coefficient. Otherwise, the next step M2 is executed to correct the photoelectric effect weight coefficient and Compton scattering weight coefficient, and step M1 is repeated according to the corrected weight coefficient until the error ratio is less than the preset ratio threshold.

8. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 1, characterized in that: The lesion location or reconstruction compensation in the artifact area of ​​the CT image includes: S2.1: Obtaining the statistical dependence between images: Based on the deformation field constructed in the B-spline Demons algorithm, the grayscale values ​​of the PET image, and the grayscale values ​​of the CT image, obtain the statistical dependence between the PET image and the CT image. Specifically: ； in: is the statistical dependence between PET images and CT images, is the total gray value of the CT image, is the total gray value of the PET image, is the grayscale value of the CT image, is the grayscale value of the PET image, is the grayscale joint probability of the CT image and PET image at the same spatial position, is the gray value in the CT image The probability of occurrence, is the gray value of the PET image Probability of occurrence; S2.2: Determine the final optimized CT image: Compare the statistical dependence with a preset dependence threshold, and determine the final optimized CT image based on the comparison result, specifically: When the statistical dependence is greater than a preset dependence threshold, the optimized CT image corresponding to the statistical dependence is the final optimized CT image. Otherwise, steps S1.3 to S2.2 are repeated to reacquire the optimized CT image until the statistical dependence is greater than the preset dependence threshold. S2.3: J value comparison: The final optimized CT image and the grouped PET dynamic image sequence are used as inputs to the B-spline Demons algorithm model, and the J value of each position point between the CT image and the PET image is output, specifically: ； in: is the Jacobian determinant, The spatial coordinates in the image are The displacement vector corresponding to the control point of is the Jacobian matrix of the displacement vector, is the weight coefficient of the kth control point, is the B-spline basis function of the k-th control point, is the number of control points, is the spatial coordinate in the image, is the determinant of the matrix; The J value is compared with a preset J threshold range, and image fusion / reconstruction / labeling is performed based on the comparison result, specifically: When the J value is not less than the upper limit of the preset J threshold range, step S2.4 is executed to fuse the image area corresponding to the J value. When the J value is within the preset J threshold range, step S3 is executed to reconstruct the image area corresponding to the J value. When the J value is less than the lower limit of the preset J threshold range, the image area corresponding to the J value is manually marked, deleted, and re-acquired. S2.4: Image fusion: Fuse the grayscale values ​​corresponding to the fused image region in the PET image with the corresponding grayscale values ​​in the CT image to obtain the grayscale values ​​of the region in the fused image, specifically: ； in: The spatial coordinates of the fused image are The grayscale value corresponding to the control point of is the weight coefficient of the PET image, The spatial coordinates in the PET image are The grayscale value corresponding to the control point of The spatial coordinates in the CT image are The grayscale value corresponding to the control point of are the spatial coordinates in the image.

9. The method for identifying and processing artifact areas in nuclear medicine imaging images according to claim 1, characterized in that: Obtain reconstructed CT images, including: S3.1: Artifact reconstruction: Perform artifact quantitative correction based on the point spread function of titanium alloy, specifically: ； in: is the point spread function, is the standard deviation of the Gaussian distribution, is the base of natural logarithms, is the influence distance of metal artifacts, is the inverse operation of convolution operation, is the corrected standardized uptake value, is the SUV value measured by the original PET; S3.2: Obtaining feature data: The artifact-corrected CT image is used as the input of the anatomical branch of the two-branch network model, and the time-activity curve corresponding to the PET image is used as the input of the functional branch of the two-branch network model. The proliferation and inflammation features are obtained as outputs, specifically: ； in: is the time activity curve, is the initial activity of the first component, is the initial activity of the second component, is the decay rate constant of the first component, is the decay rate constant of the second component, For time, is the base of natural logarithms; S3.3: Obtaining attention weights: Determine the cross-modal attention weights based on the eigenvectors of the CT image and the eigenvectors of the PET image. Specifically: ； in: is the correlation between PET features and CT features at position m, n, is the query vector of the mth PET feature, is the inversion of the key vector of the nth CT feature, is the dimension of the feature vector, 、 is a trainable linear transformation weight matrix, is the feature vector corresponding to the PET image, is the feature vector corresponding to the CT image; S3.4: Image reconstruction: Based on the proliferation characteristics, inflammation characteristics and cross-modal attention weights, the optimized CT image is used as the input of the dual-branch network model, and the reconstructed CT image is obtained as the output.

Images