Data-driven PET imaging motion correction method, storage medium and equipment
Through the data-driven PET imaging motion correction method, the motion signals are directly extracted from the PET data and the acquisition window is automatically selected, which solves the problem of motion artifacts in PET imaging, improves image quality and analysis accuracy, simplifies operations, and improves the Patient comfort and scanning efficiency.
Patent Information
- Application Number
- CN202411062243.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-29
AI Technical Summary
Inadvertent motion of patients in PET imaging results in image artifacts, reducing imaging quality and the accuracy of quantitative analysis. The existing technology relies on external devices for motion monitoring, which has safety and efficiency problems, and drug sedation also brings side effects.
Using the data-driven PET imaging motion correction method, the motion signal is extracted from the PET data through the PCA algorithm, and the optimal data acquisition window is automatically selected to reconstruct PET image to reduce motion artifacts and reduce dependence on external monitoring devices.
It significantly reduces motion artifacts, improves image clarity and quantitative analysis accuracy, enhances the applicability and computing efficiency of the method, avoids the use of external devices, and improves patient comfort and scanning efficiency.
Smart Images

Figure CN119048622A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical imaging technology, and more particularly to a data-driven motion correction method, storage medium, and device for PET imaging. Background Art
[0002] PET (Positron Emission Tomography) represents a cutting-edge clinical imaging diagnostic technology in the field of nuclear medicine. Its core mechanism can be summarized as follows: using radioactive nuclides that release positrons upon decay (e.g., [missing information]). 18 F. 11 (e.g., C) label key molecules involved in vital metabolic activities, forming tracers which are then injected into the human body. This technology maps the metabolic activity state based on the degree of tracer uptake by different tissues and lesion sites. It has shown extremely high value, especially in tumor detection, early identification, diagnosis and staging, treatment stage assessment, recurrence monitoring, primary lesion localization, metastasis tracking, and treatment response monitoring of cardiovascular diseases. It is an indispensable medical imaging tool in the fight against major diseases.
[0003] PET scans are time-consuming, with each posture typically requiring several minutes. During this time, patients often struggle to remain completely still, introducing motion errors. In chest and abdominal PET scans, respiratory activity is one of the primary sources of interference; during quiet breathing, the diaphragm rises and falls by approximately 20 millimeters, and the liver shifts by about 11 millimeters. Cardiac PET scans are affected by the heartbeat; the coronary arteries can move 8 to 23 millimeters with each heartbeat, and the range of cardiac and coronary artery movement in conjunction with respiration is between 5 and 9 millimeters, posing a challenge to accurately analyzing cardiac structure and function. During whole-body PET scans, the lack of immobilization measures can easily lead to unconscious body movements during prolonged scanning, such as hand swings and chest rises and falls. Although the head is stabilized with a headrest during brain scans, slight head swaying may still occur during prolonged scanning. If a patient experiences a seizure during the scan, it will cause violent and uncontrollable body movements. For children or patients with illnesses that make it difficult to remain still, the range of motion may be even greater, significantly impacting image quality.
[0004] Due to patient movement during scanning, PET data can be viewed as a collection of photons emitted by organs or tumors at different moving positions. This results in image reconstruction as a stacking of images under various motion states, which profoundly affects image clarity. Breathing, heartbeat, and subtle body movements not only shift the spatial distribution of the captured radioactive signals but also make PET images appear hazy. This blurring effect weakens the spatial resolution of the image, increasing the difficulty for clinicians to distinguish closely adjacent structures or subtle pathological changes. Simultaneously, motion can induce non-physiological PET image abnormalities—highlights or shadows, i.e., motion artifacts—which may conceal real lesions or fabricate pathological appearances, misleading diagnostic conclusions. In combined PET / CT scans, patient movement causes misalignment between the PET image and the corresponding CT anatomical layer, hindering accurate anatomical localization and attenuation correction. Most importantly, given the widespread use of PET scans for quantitative assessment, such as measuring metabolic levels in specific areas, motion interference can lead to inaccurate region of interest (ROI) delineation, thereby affecting the accuracy and reliability of quantitative analysis.
[0005] To address the interference caused by motion in PET imaging, previous strategies often relied on external devices to track and analyze the subject's movement in real time during the scan, integrating this motion information into the PET image reconstruction process. Examples include using respiratory or cardiac gating devices to acquire respiratory or cardiac cycle motion signals, and using visual tracking systems such as cameras to capture body movements. Furthermore, visual tracking systems like cameras are used to capture a wider range of body movements, ensuring accurate image correspondence. However, the integration and operation of this external hardware is not only technically demanding, but these devices also require rapid operation, maintenance, and calibration in a radioactive environment, posing a dual challenge to the safety and efficiency of technicians. In addition, while physical restraint devices, such as straps used during whole-body scans to reduce involuntary body swaying, can mitigate the impact of motion to some extent, they sacrifice comfort and may trigger or exacerbate patient anxiety and discomfort, sometimes even leading to scan interruptions or the need for re-scanning, thus prolonging the overall examination time and consuming resources. For children and some adults who have difficulty maintaining stillness, sedation has become a necessary adjunct. Appropriate doses of sedatives can help them relax, reduce movement caused by tension or anxiety, and thus improve image clarity. However, this practice is not without risks and side effects. Drug sedation may affect an individual's metabolism, particularly the absorption and distribution of fluorodeoxyglucose (FDG), which is the most commonly used tracer in PET scans. Changes in its physiological uptake may interfere with the assessment of disease status. Summary of the Invention
[0006] Technical problems to be solved
[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a data-driven PET imaging motion correction method, storage medium and device, which solves the technical problem of how to effectively deal with image artifacts caused by unintentional patient movements during PET scans.
[0008] Technical Solution
[0009] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:
[0010] In a first aspect, the present invention provides a data-driven motion correction method for PET imaging, comprising: acquiring PET data X∈R N×S Where S is the number of frames in the dynamic acquisition time, and N is the dimension of the probe data;
[0011] based on Perform centralized preprocessing on data X;
[0012] Apply the PCA algorithm to the data X to find the feature space P∈R S×N The data X is linearly projected onto the feature space to obtain the weight factor P·X∈R. S×S ;
[0013] Select the motion signal F = p from the weight factor matrix. k ·X∈R 1×S ,k∈{1,2};
[0014] Based on optimization model Select time frames, where M = [m1, m2, ..., m S ]∈R 1×S A mask vector representing the selected time frame, used to mark the selected time frame, c i The total number of matching events in the i-th time frame;
[0015] The acquired data of the selected time frame is retained and PET reconstruction is performed to complete motion correction for PET imaging.
[0016] Optionally, the PCA algorithm is applied to the data X to find the feature space P∈R. S×N ,include:
[0017] Applying the Lagrange multiplier method to the optimization formula, constructing the Lagrange function, and then solving for the derivative with respect to P, yields the characteristic matrix P = [p1, p2, ..., p] of the characteristic space basis vectors. S ]∈R S×n The corresponding eigenvalues are λ1≥λ2≥…≥λ S .
[0018] Optionally, the optimized formula is: Take the derivative of P using the following formula: Right now
[0019] Where Λ represents the covariance matrix of X.
[0020] Optionally, a constant value is added to or subtracted from the motion signal F to ensure that F i ≠0, i=1,…,S.
[0021] Optionally, based on the optimization model Select time frames, including:
[0022] Construct the Lagrangian function based on the optimization model;
[0023] The simultaneous equations are obtained by finding the extreme value of the gradient of the Lagrange function;
[0024] The pseudo-inverse matrix of the simultaneous equations is obtained by singular value decomposition.
[0025] Based on the pseudo-inverse matrix
[0026] exist If the value is greater than the preset threshold, select this. The corresponding time frame.
[0027] Optionally, the Lagrange function is:
[0028] Where α is a Lagrange multiplier.
[0029] Optionally, the simultaneous equations are: Among them, F d ∈R S×S Let F be the diagonal matrix of the motion signal vector. It is a positive definite invertible matrix;
[0030] And thus obtain
[0031]
[0032] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements a data-driven PET imaging motion correction method as described in any of the first aspects above.
[0033] Thirdly, the present invention provides a storage device including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements a data-driven PET imaging motion correction method as described in any one of the first aspects above.
[0034] Beneficial effects
[0035] The beneficial effects of this invention are as follows: This invention proposes a data-driven motion correction method for PET imaging, which adaptively extracts patient motion information directly from PET-acquired data and automatically determines the optimal data acquisition window, ensuring sufficient data within the acquisition cycle while minimizing motion amplitude. This method is applicable not only to elastic deformation correction caused by respiration and cardiac motion but also to rigid motion correction in brain imaging. It corrects image artifacts caused by patient motion in PET scans and reduces reliance on external monitoring equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 A schematic flowchart of a data-driven motion correction method for PET imaging provided in an embodiment of the present invention;
[0037] Figure 2 A schematic diagram of respiratory motion waveforms extracted from data and selected data points provided in an embodiment of the present invention;
[0038] Figure 3 This is a PET reconstructed image without motion correction provided in an embodiment of the present invention;
[0039] Figure 4 The PET reconstructed image with motion correction provided in the embodiments of the present invention. Detailed Implementation
[0040] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.
[0041] To correct image artifacts caused by patient movement in PET scans and reduce reliance on external monitoring equipment, this invention proposes a data-driven motion correction method for PET imaging, achieving motion correction within the PET imaging process. Traditional data-driven elastic motion correction algorithms rely on estimating the motion field that changes over time in the imaging space, describing the displacement path of each pixel or voxel from one time point to another, reflecting the elastic deformation of the tissue, and then using the motion field for motion correction. This process encompasses several strategies: readjusting the original line-of-response (LOR) coordinates; introducing and reconstructing the motion field in the modeling of the acquisition process; and registering and fusing all image frames to the same coordinate system. This elastic motion correction method is typically used for respiratory or cardiac motion correction, but the process is not only cumbersome and time-consuming but also often lacks stability, limiting its widespread application. Another type of data-driven rigid motion correction algorithm relies on intermediate images at each time point to estimate rigid body motion parameters and applies rigid body transformations in the data or image domain to reduce motion-induced artifacts. This rigid motion correction method is typically used for brain motion correction. However, the accuracy of rigid motion parameters is limited by intermediate image quality and is susceptible to image noise. In contrast, this invention adaptively extracts patient motion information directly from PET data and automatically determines the optimal data acquisition window. This ensures sufficient data within the acquisition period while minimizing motion amplitude, achieving a near-static state of the scanned object. Reconstruction is performed only on PET data within the optimized timeframe, significantly suppressing motion artifacts and enhancing image clarity and the accuracy of quantitative analysis. This method exhibits excellent adaptive characteristics, applicable not only to elastic deformation correction caused by respiration and cardiac motion but also to rigid motion correction in brain imaging. It avoids complex elastic transformations or noise-sensitive rigid transformations in the imaging space, improving its applicability and computational efficiency.
[0042] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.
[0043] Firstly, referring to Figure 1 This embodiment provides a data-driven motion correction method for PET imaging, including:
[0044] S1, collect PET data X.
[0045] X∈R N×SWhere S is the number of frames in the dynamic acquisition time, and N is the dimension of the probe data.
[0046] During the total acquisition time, PET data is divided into S frames according to a fixed unit duration. Data framing needs to balance the impact of statistical noise and the preservation of temporal information. C = [c1, c2, ..., c S ]∈R 1×S This represents the total number of matching instances per frame. To reduce the impact of noise and alleviate computational burden, downsampling is typically performed in the spatial domain.
[0047] S2 performs centralized preprocessing on data X.
[0048] based on Perform centralized preprocessing on data X.
[0049] S3. Apply the PCA algorithm to the data X to find the feature space P, and linearly project the data X onto the feature space to obtain the weight factors.
[0050] P∈R S×N Weighting factor P·X∈R S×S .
[0051] P represents the basis functions or principal components of the feature space. Principal component analysis (PCA) aims to minimize information loss by transforming a set of correlated variables into a set of linearly uncorrelated variables through orthogonal transformations. The basis vectors of the transformed feature space are called principal components. Eigenvectors corresponding to the smallest eigenvalues are often related to noise; discarding them can improve noise reduction and signal extraction to some extent. Therefore, to solve for the feature space, it is required that the covariance between the transformed feature variables be zero and the internal variance be as large as possible.
[0052] Optionally, the PCA algorithm is applied to the data X to find the feature space P∈R. S×N ,include:
[0053] Applying the Lagrange multiplier method to the optimization formula, constructing the Lagrange function, and then solving for the derivative with respect to P, yields the characteristic matrix P = [p1, p2, ..., p] of the characteristic space basis vectors. S ]∈R S×N The corresponding eigenvalues are λ1≥λ2≥…≥λ S .
[0054] Optionally, the optimized formula is: 1 S Let P be an S-dimensional identity matrix. In the formula, T denotes the matrix transpose operation. Differentiate P with respect to the following formula: Right now Where Λ represents the covariance matrix of X.
[0055] Eigenvalues represent the magnitude of data variation, and eigenvectors are the principal components. The eigenvector corresponding to the largest eigenvalue is the first principal component, and so on. The weighting factor for projecting the collected data onto the feature space is P·X∈R. S×S In practical applications, the weighting factors corresponding to the first or second principal component are usually selected as the extracted motion signal. The weighting factors corresponding to other principal components can be regarded as noise signals and discarded.
[0056] S4, select the motion signal F from the weight factor matrix.
[0057] F = p k ·X∈R 1×S ,k∈{1,2}.
[0058] S5, selects time frames based on the optimization model.
[0059] Optimization Model Where M = [m1, m2, ..., m S ]∈R 1×S A mask vector representing the selected time frame, used to mark the selected time frame, c i The total number of matching events in the i-th time frame.
[0060] The motion signal F characterizes the motion amplitude of each frame. When selecting time frames, the core objective is to ensure that the number of events within the acquisition period reaches a preset minimum requirement, while minimizing the variation in motion amplitude between these time frames to maintain motion consistency. Therefore, the selection process of time frames transforms into finding an optimal solution that satisfies the statistical requirement for the number of events while minimizing the variation in motion amplitude. To this end, the variance of the motion signal is used as a metric for motion amplitude, transforming the above objectives into a clear mathematical optimization model.
[0061] Optionally, based on the optimization model Select time frames, including:
[0062] Construct the Lagrangian function based on the optimization model;
[0063] The simultaneous equations are obtained by finding the extreme value of the gradient of the Lagrange function;
[0064] The pseudo-inverse matrix of the simultaneous equations is obtained by singular value decomposition.
[0065] Based on the pseudo-inverse matrix
[0066] exist If the value is greater than the preset threshold, select this. The corresponding time frame.
[0067] Optionally, the Lagrange function is:
[0068] Where α is a Lagrange multiplier.
[0069] The above formula can be written in matrix form: Among them, F d ∈R S ×S Let F be the diagonal matrix of the motion signal vector F, i.e.
[0070] Optionally, a constant value is added to or subtracted from the motion signal F to ensure that F i ≠0, i=1,…,S.
[0071] R is set as a threshold, for example, 50% of the total number of events, to limit the minimum number of events that must be included in the selected time frame. Since the optimization function relies solely on variance calculations, baseline calibration is performed on the motion signal F to ensure the stability and accuracy of the optimization process. Without loss of generality, a small constant value is added to or subtracted from the motion signal F to ensure that the motion amplitude is non-zero in all time frames.
[0072] Optionally, the simultaneous equations are: The motion signal F does not pass through zero, and the diagonal matrix It is a positive definite invertible matrix; when S is large enough, Reversible; even It is not invertible, but its pseudo-inverse matrix can be obtained through singular value decomposition, thus approximating it as follows:
[0073]
[0074] because Since it is a scalar and not equal to 0, we get:
[0075]
[0076]
[0077] The number of time frames S, as the dimension of the matrix in the above formula, has a relatively small value, thus reducing the computational burden, and the mask vector... It can achieve real-time calculation.
[0078] S6 retains the acquired data of the selected time frame and performs PET reconstruction to complete PET imaging motion correction.
[0079] right An empirical threshold is set; when the vector value exceeds this threshold, the data for the corresponding frame is retained, and otherwise discarded. Finally, only the retained data is subjected to regular PET reconstruction. In this way, motion-induced image artifacts are significantly reduced because the motion amplitude is kept to a minimum, while ensuring that the signal-to-noise ratio of the reconstructed image meets the requirements.
[0080] This embodiment presents a data-driven motion correction method for PET imaging, which exhibits excellent adaptive characteristics. It is applicable not only to elastic deformation correction caused by respiration and cardiac motion but also to rigid motion correction in brain imaging. This avoids complex elastic transformations or rigid transformations susceptible to noise in the imaging space, thereby significantly enhancing computational efficiency while improving the method's versatility. Motion information is directly correlated with the distribution of PET radioactivity within the patient's body, theoretically ensuring higher accuracy in motion signal detection, surpassing traditional methods that rely on external monitoring equipment. The absence of external monitoring equipment simplifies operation and maintenance, significantly reducing the risk of radiation exposure. During the scan, the reliance on physical restraint devices is eliminated, greatly improving patient comfort. Patients no longer require sedation during the scan and can undergo PET imaging in a natural state, avoiding any drug intervention.
[0081] Figure 2 This refers to the respiratory motion waveform extracted from the data and the selected data points. Figure 3 For PET reconstructed images without motion correction, Figure 4 The image shown is a motion-corrected PET reconstruction. It is clearly visible that the lung tumor outline is clear after motion correction, and motion artifacts are eliminated. This invention does not require cumbersome external control equipment operation, nor does it require elastic or rigid transformations.
[0082] The following example, using a whole-body PET scan to extract respiratory signals and reconstruct images after motion correction, further illustrates the implementation steps of this invention:
[0083] A PET scan is performed on the patient, and respiratory motion signals are obtained from the data collected from the lungs. These signals are manifested as... Figure 2 The waveforms in the image are shown. Each bed scan lasted 2 minutes. At the lung scanning bed, data was acquired in 0.5-second frames, totaling 240 frames. Half of the data was used for motion correction, and the optimized frame positions are marked as follows: Figure 2 The point marked by the middle circle.
[0084] The same reconstruction algorithm and parameters were used for image reconstruction of each bed in the whole-body PET scan: Ordered Subsets Expectation-Maximization (OSEM), 17 subsets, 2 iterations, image matrix 192x192, Gaussian post-filter, and a half-width at half-height of 4mm.
[0085] PET reconstructed images without motion correction, such as Figure 3 As shown; motion-corrected PET reconstructed image as follows Figure 4 As shown, it is clear that the lung tumor outline is sharp after motion correction, and motion artifacts are eliminated. The signal-to-noise ratio of the lung reconstruction image is similar to that of other bed reconstruction images; although the data volume is reduced by half, no significant difference is observed in the images.
[0086] In a second aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements a data-driven PET imaging motion correction method as described in any of the first aspects above.
[0087] Thirdly, embodiments of the present invention provide a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements a data-driven PET imaging motion correction method as described in any of the first aspects above.
[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0089] 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.
[0090] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A data-driven PET imaging motion correction method, characterized in that: include: Collect PET data X∈R N×S , where S is the number of dynamic acquisition time frames and N is the dimension of the detection data; based on Centralize and preprocess data X; Apply the PCA algorithm to the data X to find the feature space P∈R S×N , and linearly project the data X into the feature space to obtain the weight factor P·X∈R S×S ; Select the motion signal F = p from the weight factor matrix k ·X∈R 1×S , k∈{1,2}; Based on optimization model Select a time frame, where M = [m1, m2, ..., m S ]∈R 1×S The mask vector representing the time frame selection is used to mark the selected time frame, c i is the total number of matching cases in the i-th time frame; The acquisition data of the selected time frame is retained and PET reconstruction is performed to complete PET imaging motion correction.
2. A data-driven PET imaging motion correction method according to claim 1, characterized in that: Apply the PCA algorithm to the data X to find the feature space P∈R S×N ,include: Apply the Lagrange multiplier method to the optimization formula, construct the Lagrange function and solve it to get the feature space basis vector feature matrix P = [p1, p2, ..., p S ]∈R S×N , corresponding to the eigenvalues λ1≥λ2≥…≥λ S .
3. A data-driven PET imaging motion correction method according to claim 2, characterized in that: The optimization formula is Derivative P according to the following formula: Right now where Λ represents the covariance matrix of X.
4. A data-driven PET imaging motion correction method according to claim 3, characterized in that: A constant value is added or subtracted from the motion signal F to ensure that F i ≠0, i=1,…,S.
5. A data-driven PET imaging motion correction method according to claim 4, characterized in that: Based on optimization model Select a time frame, including: Construct Lagrangian function based on optimization model; The simultaneous equations are obtained by taking the gradient of the Lagrangian function and finding the extreme value; The pseudo-inverse matrix of the simultaneous equations is obtained by singular value decomposition method; Based on the pseudo-inverse matrix exist If it is greater than the preset threshold, select The corresponding time frame.
6. A data-driven PET imaging motion correction method according to claim 5, characterized in that: The Lagrangian function is: Where α is the Lagrange multiplier.
7. A data-driven PET imaging motion correction method according to claim 6, characterized in that: The simultaneous equations are Among them, F d ∈R S×S is the diagonal matrix of the motion signal vector F, the diagonal matrix is a positive definite invertible matrix; Then get 8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the data-driven PET imaging motion correction method according to any one of claims 1 to 7 is implemented.
9. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, wherein: When the processor executes the computer program, the data-driven PET imaging motion correction method described in any one of claims 1 to 7 is implemented.
