Real-time positioning method for tumor target area and gold marker implant

Abstract

The application discloses a real-time positioning method of a tumor target area and a gold mark implant, and belongs to the technical field of tumor positioning, and specifically comprises the following steps: implanting a gold mark in a tumor area, constructing a gold mark-tumor dynamic model through CT and acquiring a respiratory phase feature vector; a two-dimensional projection is matched with a three-dimensional model in space by using an improved hybrid registration algorithm, and a tumor composite displacement vector is acquired; a respiratory-displacement correlation model is established through time sequence alignment, a tumor motion atlas containing a real-time position and a predicted trajectory is generated, and a reference coordinate system is updated; a rigid safety boundary is established based on the gold mark, an elastic treatment boundary is generated by combining convolutional neural network analysis of soft tissue texture and the predicted trajectory, and a non-marked area dose weight is dynamically adjusted; and finally, sub-millimeter dynamic tracking is executed through a multi-leaf collimator; the problems of tumor motion uncertainty and individual deformation difference are solved, and the target area coverage precision and dose distribution rationality are improved.

Description

Technical Field

[0001] This invention belongs to the field of tumor localization technology, specifically a real-time localization method for tumor target areas and a gold label implant. Background Technology

[0002] In cancer treatment, precise localization of the tumor target area is crucial for improving treatment efficacy and reducing damage to surrounding normal tissues. Currently, gold-labeled implants, as commonly used tumor markers, are widely applied in cancer radiotherapy and other treatment methods. However, existing tumor localization and treatment methods based on gold-labeled implants have certain limitations. On the one hand, after locating the gold-labeled implant, the definition of the tumor target area is often based solely on the location of the gold marker, without fully considering situations where the gold marker cannot completely cover the tumor area. This leads to inaccurate assessment of the tumor tissue surrounding the gold marker, potentially resulting in tumor residue or overtreatment. On the other hand, current technologies struggle to achieve real-time dynamic localization of the tumor target area during treatment, making it impossible to adjust the treatment plan promptly based on changes in tumor location, thus affecting treatment effectiveness. Therefore, a more precise, real-time method for locating the tumor target area is urgently needed to improve the accuracy and effectiveness of cancer treatment. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a real-time tumor target localization method and a gold label implant. A gold label is implanted in the tumor region, and a dynamic model of the gold label and tumor is constructed using CT scans to obtain respiratory phase feature vectors. An improved hybrid registration algorithm is used to achieve spatial matching between the two-dimensional projection and the three-dimensional model, obtaining the tumor composite displacement vector. A respiratory-displacement correlation model is established through time-series alignment, generating a tumor motion atlas containing real-time position and predicted trajectory, and updating the baseline coordinate system. A rigid safety boundary is established based on the gold label, and an elastic treatment boundary is generated by combining convolutional neural network analysis of soft tissue texture and predicted trajectory, dynamically adjusting the dose weight of the unlabeled area. Finally, sub-millimeter-level dynamic tracking is performed using a multi-leaf collimator. This method solves the problems of tumor motion uncertainty and individual deformation differences, improving target coverage accuracy and dose distribution rationality.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for real-time localization of tumor target areas includes:

[0006] S1: Implant at least 4 gold markers inside and at the edge of the tumor, simultaneously acquire 4D CT image data and patient respiratory signals, construct a gold marker-tumor dynamic model based on the 4D CT image data, and establish a tumor reference coordinate system. At the same time, the patient's respiratory signals are preprocessed to generate respiratory phase feature vectors.

[0007] S2: The improved Demons-Powell hybrid registration algorithm is used to spatially match the two-dimensional gold standard projection image captured in real time by dual-plane X-ray fluoroscopy with the gold standard-tumor dynamic model. The six-degree-of-freedom displacement parameters of the gold standard group are obtained through iterative optimization, and the tumor composite displacement vector is calculated.

[0008] S3: Align the tumor composite displacement vector with the respiratory phase feature vector over time to establish a respiratory-displacement correlation model, generate a tumor motion atlas, and update the tumor baseline coordinate system based on the motion atlas; the tumor motion atlas includes the real-time position of the tumor and the predicted trajectory of the tumor.

[0009] S4: Establish a dual safety boundary based on the tumor motion map and automatically adjust the treatment dose weight of the unlabeled area according to the gold standard interval;

[0010] S5: The multi-leaf collimator performs sub-millimeter-level dynamic tracking based on the tumor motion map and safety boundary parameters.

[0011] Specifically, the steps of S1 include:

[0012] S1.1: At least four gold markers are implanted inside and at the edge of the tumor, and respiratory signals are simultaneously acquired using an optical surface imaging system; the gold markers are non-coplanarly distributed.

[0013] S1.2: Based on the acquired respiratory signals and respiratory signals after Hilbert transformation Calculate the instantaneous phase The respiratory cycle is divided into 10 phases based on instantaneous phase. A CT scan is triggered at a specific time, and a timestamp is recorded. , where i represents the phase index value or time point index value after partitioning;

[0014] S1.3: After completing the CT scan within each respiratory window, obtain four-dimensional CT image data. ;

[0015] S1.4: For each gold standard k, extract the location sequence from the four-dimensional CT image data. Where T represents transpose and k represents the gold standard index. This indicates the position of the k-th gold marker in the i-th phase. , and These represent the X, Y, and Z coordinates of the k-th gold marker in the i-th phase, respectively.

[0016] Specifically, the specific steps of S1 further include:

[0017] S1.5: Continuous motion trajectory is obtained using a cubic spline interpolation algorithm. Simultaneously, the tumor motion displacement field is generated through difference calculation. The cubic spline interpolation algorithm is implemented by multiplying the extracted position sequence with B-spline basis functions; the number of B-spline basis functions is equal to the number of phase divisions.

[0018] S1.6: Average the position sequence of all gold markers at the resting phase to obtain the centroid O of the gold marker group; the resting phase is... ;

[0019] S1.7: Based on and global average position Calculate the motion covariance matrix A of the gold standard group; the global average position is The mean across all gold standards and phases;

[0020] S1.8: Perform singular value decomposition on the motion covariance matrix A of the gold standard group, and take the eigenvector corresponding to the largest eigenvalue after singular value decomposition as the Z-axis orthogonal basis. ;

[0021] S1.9: Based on the determined Z-axis, generate an orthogonal basis for the Y-axis through orthogonalization. orthogonal basis of the X-axis ;

[0022] S1.10: Based on the orthogonal basis of the X-axis Y-axis orthogonal basis Z-axis orthogonal basis Establish a tumor baseline coordinate system using the gold standard group centroid O.

[0023] Specifically, the specific steps of S1 further include:

[0024] S1.11: Based on respiratory signals and respiratory signals after Hilbert transformation Calculate the amplitude envelope ;

[0025] S1.12: Combine the instantaneous phase, amplitude envelope, and instantaneous phase derivative to form a respiratory phase feature vector. ,in, express The derivative of .

[0026] Specifically, the steps of S2 include:

[0027] S2.1: Real-time acquisition of two-dimensional gold standard projection images via dual-plane X-ray fluoroscopy and The two-dimensional gold standard projection image is an orthogonal projection and is a grayscale image; the dual planes are two-dimensional detector planes, where x represents the corresponding X-axis coordinate and y represents the corresponding Y-axis coordinate.

[0028] S2.2: Projecting the two-dimensional gold standard image and Gold standard enhancement is performed, and candidate points are selected by detecting the principal axis direction of the gold standard projection through random transformation.

[0029] S2.3: Define the X-ray source-detector geometric parameters and set the initial six degrees of freedom parameters; the X-ray source-detector geometric parameters include the distance d from the X-ray to the detector, the rotation angle, etc. and The six degrees of freedom parameters include translation parameters. and rotation parameters ,in, , , These represent the translation amounts along the X-axis, Y-axis, and Z-axis, respectively. , , These represent the rotation angles around the X-axis, Y-axis, and Z-axis, respectively.

[0030] Specifically, the steps of S2 further include:

[0031] S2.4: Obtain the gold standard-tumor dynamic model and extract the gold standard 3D coordinates from the gold standard-tumor dynamic model. ;

[0032] S2.5: Combining the geometric parameters of the X-ray source-detector and the six-degree-of-freedom parameters, the three-dimensional coordinates of the gold standard are... Projected onto the plane of the two-dimensional detector;

[0033] S2.6: By calculating the actual X-ray image at pixel points grayscale value at Simulated X-ray images generated by projection of gold standard-tumor dynamic model in grayscale value at The difference is used to obtain the grayscale difference value. ;

[0034] S2.7: Based on the projected coordinates of the k-th gold standard in the dual planes and the two-dimensional coordinates of the k-th gold standard detected in the actual X-ray image, construct an error function. ;

[0035] S2.8: Based on error function Selecting the set of conjugate directions in the six-dimensional parameter space And along each conjugate direction, by adding the six degrees of freedom parameters to the preset optimal step size and the current conjugate direction. A one-dimensional search is performed using the product of these parameters to obtain the updated six degrees of freedom parameters, where r represents the conjugate direction index;

[0036] S2.9: Based on the updated six-degree-of-freedom parameters, combined with the tumor motion displacement field in S1.5 The tumor composite displacement vector was obtained. .

[0037] Specifically, the steps of S3 include:

[0038] S3.1: Obtain the tumor composite displacement vector and respiratory phase eigenvectors And perform timestamp alignment; the tumor composite displacement vector and respiratory phase feature vector are 3-dimensional vectors;

[0039] S3.2: Combine the tumor composite displacement vector and respiratory phase eigenvectors Merged into a 6-dimensional spacetime tensor ;

[0040] S3.3: Set up a sliding window to display the spacetime tensor. Data segmentation is performed to obtain spatiotemporal feature data;

[0041] S3.4: Load the pre-trained Long Short-Term Memory (LSTM) network model, input the spatiotemporal feature data into the pre-trained LSM network model, and output the real-time location. and predicted phase ;

[0042] S3.5: Based on the obtained real-time location and predicted phase By combining the mean square error and the phase cosine similarity, the composite loss value is obtained;

[0043] S3.6: Obtain the current tumor center coordinates from the real-time location output by the Long Short-Term Memory network model. Combined with instantaneous phase The system predicts the trajectory for the next second and plots the real-time position in red in the tumor baseline coordinate system, while plotting the predicted trajectory using a blue gradient color band, thus generating a tumor motion atlas.

[0044] S3.7: Based on the tumor composite displacement vector Displacement data over the past 5 seconds Incremental principal component analysis is performed to obtain the updated covariance matrix; the covariance matrix is ​​a square matrix and its dimension is equal to the dimension of the tumor composite displacement vector.

[0045] S3.8: The updated X-axis direction vector in the tumor reference coordinate system is obtained by calculating the vector that maximizes the product of the covariance matrix and the unit vector.

[0046] Specifically, the establishment of a dual security boundary in S4 includes:

[0047] The first layer of rigid safety boundary is established based on the gold standard.

[0048] By analyzing the soft tissue texture features around the gold marker in the two-dimensional gold marker projection image using a convolutional neural network, and combining this with the tumor prediction trajectory, a second layer of elastic treatment boundary is generated.

[0049] Specifically, a real-time tumor target localization method further includes:

[0050] During each treatment, gold standard displacement data and soft tissue texture features are collected simultaneously. An online adaptive mechanism is established to feed back the deviation between real-time gold standard displacement data and tumor prediction trajectory to the respiration-displacement correlation model.

[0051] After each treatment, the actual radiation dose distribution was verified by cone-beam CT, and the error analysis results were fed back to the gold standard-tumor dynamic model for parameter correction.

[0052] A gold-labeled implant for a tumor target area includes a gold-labeled body made of pure gold and in the form of a cylinder; the gold-labeled body includes multiple spheres and connecting segments; the multiple spheres and the connecting segments are sequentially fixed together in a top-to-bottom manner.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. This invention proposes a real-time tumor target localization method. By constructing a dynamic model of tumor based on gold standard implantation and 4D CT, combined with respiratory signal feature extraction, the spatiotemporal modeling of tumor movement is accurately modeled. An improved registration algorithm and time series alignment technology ensure the dynamic matching accuracy between the 2D projection and the 3D model, thereby generating a tumor motion atlas containing real-time position and predicted trajectory. Combined with the predicted trajectory, an elastic treatment boundary is generated, solving the problems of tumor movement uncertainty and individual deformation differences. This allows the treatment boundary to dynamically adapt to the real-time position and potential range of tumor movement, improving the accuracy of target coverage while reducing the radiation dose to normal tissues.

[0055] 2. This invention proposes a real-time tumor target localization method. During treatment, an online adaptive mechanism feeds back the deviation between the real-time displacement of the gold standard and the predicted trajectory to the respiratory-displacement correlation model, enabling real-time correction of the dynamic prediction model. After treatment, based on the dose verification results of cone-beam CT, the parameters of the gold standard-tumor dynamic model are further corrected, forming a complete closed loop of modeling-tracking-verification-optimization. This dual feedback mechanism improves the system's adaptability to dynamic changes such as respiratory motion and tumor regression, ensuring the consistency of dose deposition between fractionated treatments. At the same time, by automatically adjusting the dose weight of the unlabeled area through the gold standard spacing, the dynamic optimization allocation of the target dose is achieved, maximizing the protection of surrounding normal tissues while ensuring tumor control rate, demonstrating the intelligent and individualized advantages of precision radiotherapy. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of a real-time tumor target region localization method according to the present invention;

[0057] Figure 2 This is a flowchart illustrating the principle of a real-time tumor target region localization method according to the present invention.

[0058] Figure 3 This is a flowchart illustrating the implementation of a real-time tumor target region localization method based on the composite displacement vector of the tumor, as described in this invention. Detailed Implementation

[0059] Example 1

[0060] Please see Figures 1-3 The present invention provides an embodiment of a method for real-time localization of tumor target areas, comprising the following steps:

[0061] S1: Implant at least 4 gold markers inside and at the edge of the tumor, simultaneously acquire 4D CT image data and patient respiratory signals, construct a gold marker-tumor dynamic model, establish a tumor baseline coordinate system, and at the same time, generate respiratory phase feature vectors after preprocessing the patient's respiratory signals.

[0062] Among them, three-dimensional dynamic image data is four-dimensional data, which is a sequence of three-dimensional spatial data in the time dimension, and therefore is four-dimensional data.

[0063] Furthermore, gold-label implantation includes:

[0064] (1) Guidance method: CT guidance or ultrasound guidance is used;

[0065] (2) Implantation route: Percutaneous puncture is used, avoiding important organs;

[0066] (3) Distribution strategy: Tetrahedral structure distribution is adopted to ensure spatial uniqueness, including:

[0067] At least three gold markers were located at the tumor margin to detect deformation;

[0068] At least one gold marker is located inside the tumor to track the movement of the center of mass.

[0069] It is important to emphasize that verification is still required after the gold standard is implanted. The verification includes:

[0070] (1) The gold mark spacing is greater than or equal to 5 mm;

[0071] (2) The positional deviation of the gold mark is less than 2 mm.

[0072] S2: The improved Demons-Powell hybrid registration algorithm is used to spatially match the two-dimensional gold standard projection image captured in real time by dual-plane X-ray fluoroscopy with the gold standard-tumor dynamic model. The six-degree-of-freedom displacement parameters of the gold standard group are obtained through iterative optimization, and the tumor composite displacement vector is calculated.

[0073] S3: Align the tumor composite displacement vector with the respiratory phase feature vector over time to establish a respiratory-displacement correlation model, obtain a tumor motion atlas, and update the tumor baseline coordinate system; the tumor motion atlas includes the real-time position of the tumor and the predicted trajectory of the tumor.

[0074] S4: Establish a dual safety boundary based on the tumor motion map and automatically adjust the treatment dose weight of the unlabeled area according to the gold standard interval;

[0075] The establishment of dual security boundaries in S4 includes:

[0076] The first layer of rigid safety boundary is established based on the gold standard.

[0077] By analyzing the soft tissue texture features around the gold marker in the two-dimensional gold marker projection image using a convolutional neural network, and combining this with the tumor prediction trajectory, a second layer of elastic treatment boundary is generated.

[0078] Furthermore, by analyzing the soft tissue texture features around the gold marker in the two-dimensional gold marker projection image using a convolutional neural network, and combining this with the tumor prediction trajectory, a second layer of elastic treatment boundary is generated, including:

[0079] (1) Obtain a two-dimensional X-ray projection image, including a dual-plane grayscale image of the gold standard and its surrounding soft tissue;

[0080] (2) Obtaining the tumor prediction trajectory is the real-time position sequence of the tumor in the future time window output by the respiratory-displacement correlation model;

[0081] (3) Extract the gold label region by threshold segmentation and generate a binary mask. The threshold segmentation method is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0082] (4) In the gold standard area, a circular area with a radius of 20mm is selected as the region of interest, i.e., the ROI area, with the gold standard as the center;

[0083] (5) Perform contrast-limited histogram equalization on the ROI region to obtain the ROI value after contrast-limited histogram equalization. The contrast-limited histogram equalization is a prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0084] (6) Load the trained convolutional neural network model, input the ROI value after contrast-limited histogram equalization into the trained convolutional neural network model, and predict the probability of soft tissue deformation. The convolutional neural network is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0085] (7) The difference between the real-time tumor location and the predicted real-time tumor location output by the respiratory-displacement association model is multiplied by the predicted soft tissue deformation probability after passing through an exponential function to obtain the dynamic weight.

[0086] (8) Add the product of the dynamic weight and the maximum elastic extension to the first rigid safety boundary to obtain the second elastic treatment boundary.

[0087] S5: The multi-leaf collimator performs sub-millimeter-level dynamic tracking based on the tumor motion map and safety boundary parameters.

[0088] Furthermore, the specific steps of S5 include:

[0089] (1) Obtain tumor motion atlases, including real-time tumor location and predicted trajectory;

[0090] (2) Obtain the first rigid safety boundary and the second elastic treatment boundary;

[0091] (3) Based on the real-time location of the tumor, dynamic programming is used to optimize the leaf movement path and minimize the leaf adjustment time. The dynamic programming optimization method is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0092] Furthermore, a real-time localization method for tumor target areas also includes:

[0093] S6: During each treatment, gold standard displacement data and soft tissue texture features are collected simultaneously to establish an online adaptive mechanism that feeds back the deviation between real-time gold standard displacement data and tumor prediction trajectory to the respiration-displacement correlation model.

[0094] S7: After each treatment, the actual radiation dose distribution is verified by cone-beam CT, and the error analysis results are fed back to the gold standard-tumor dynamic model for parameter correction.

[0095] Furthermore, the specific steps of S7 include:

[0096] (1) Immediately after treatment, cone-beam CT scan was performed to obtain three-dimensional anatomical images and simultaneously record the multi-leaf collimator log file during treatment, including parameters such as leaf position, dose rate, and gantry angle;

[0097] (2) The Demons algorithm is used to perform deformation registration between the three-dimensional anatomical images and the planned CT images to generate a displacement field. The Demons algorithm is a prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0098] (3) Based on the electron density image of three-dimensional anatomical images and the multi-leaf collimator log file, the actual dose distribution is reconstructed using the Monte Carlo algorithm. The Monte Carlo algorithm is existing technology in this field and is not an inventive solution of this application. It will not be described in detail here.

[0099] (4) The actual dose distribution of the reconstruction is mapped to the planned CT coordinate system through the displacement field to obtain the actual dose;

[0100] (5) The difference between the actual position of the gold standard extracted from the three-dimensional anatomical image and the predicted position of the gold standard-tumor dynamic model is calculated to obtain the gold standard error value;

[0101] (6) Based on the error value of the gold standard and the parameters of the gold standard-tumor dynamic model, a Jacobi matrix is ​​constructed, and the parameters of the gold standard-tumor dynamic model are updated using Bayesian theory according to the Jacobi matrix. The Jacobi matrix and Bayesian theory are existing technologies in this field and are not the inventive solutions of this application, and will not be elaborated here.

[0102] (7) Retrain using the updated gold standard-tumor dynamic model parameters.

[0103] The specific steps of S1 include:

[0104] S1.1: At least four gold markers are implanted inside and at the edge of the tumor, and respiratory signals are simultaneously acquired using an optical surface imaging system; the gold markers are non-coplanarly distributed.

[0105] S1.2: Based on the acquired respiratory signals and respiratory signals after Hilbert transformation Calculate the instantaneous phase The respiratory cycle is divided into 10 phases based on instantaneous phase. A CT scan is triggered at a specific time, and a timestamp is recorded. And satisfy: Where i represents the phase index value or time point index value after partitioning. Represents the arctangent function. The Hilbert transform representing the respiratory signal converts the respiratory signal into the imaginary part of the analytic signal, which is used to extract the instantaneous amplitude and phase of the signal. The specific formula of the Hilbert transform is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0106] S1.3: After completing the CT scan within each respiratory window, obtain four-dimensional CT image data. ;

[0107] It should be noted that the set of motion image sequences of the tumor and gold standard in 3D space is four-dimensional CT image data with a slice thickness of 1 mm.

[0108] S1.4: For each gold standard k, extract the location sequence from the four-dimensional CT image data. Where T represents transpose and k represents the gold standard index. This indicates the position of the k-th gold marker in the i-th phase. , and These represent the X, Y, and Z coordinates of the k-th gold marker in the i-th phase, respectively.

[0109] S1.5: Continuous motion trajectory is obtained using a cubic spline interpolation algorithm. Simultaneously, the tumor motion displacement field is generated through difference calculation. ,in, The B-spline basis functions are represented; the cubic spline interpolation algorithm is implemented by multiplying the extracted position sequence with the B-spline basis functions; the number of B-spline basis functions is equal to the number of phase divisions;

[0110] Furthermore, the calculation process for the tumor motion displacement field includes:

[0111] (1) Using the formula Align the CT images of each phase, where, Represents the three-dimensional deformation field corresponding to the i-th phase or time point, and represents the spatial point. The transformed coordinates, This represents the three-dimensional deformation field corresponding to the previous phase, where l, m, and n represent the indices of the control points in the X, Y, and Z axes, respectively. Describes the cubic B-spline basis functions. , , Represent the coordinate positions of the l-th, m-th, and n-th control points respectively, according to the spacing. Uniformly distributed three-dimensional grid points Represents the control point in the i-th phase The deformation coefficient determines the displacement contribution of the control point to the surrounding area in the current phase. Indicates the spacing between control points;

[0112] (2) Based on Deformation field of reference phase The difference is used to obtain the tumor motion displacement field. , where t represents the current time.

[0113] As shown in S1.2, in this invention, the respiratory cycle is divided into 10 phases based on instantaneous phase. Therefore, the phases and time are synchronized. and Equivalent.

[0114] S1.6: Take the average of the position sequences of all gold markers in the resting phase to obtain the mass center of the gold marker group. Where N represents the number of implanted gold markers; the resting phase is ;

[0115] In this invention, .

[0116] S1.7: Based on and global average position Calculate the motion covariance matrix of the gold standard group. The global average position is The mean across all gold standards and phases;

[0117] in, satisfy: .

[0118] S1.8: Perform singular value decomposition on the motion covariance matrix A of the gold standard group, and take the eigenvector corresponding to the largest eigenvalue after singular value decomposition as the Z-axis orthogonal basis. Singular value decomposition is a prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0119] S1.9: Based on the defined Z-axis, generate the Y-axis orthogonal basis through Gram-Schmidt orthogonalization. orthogonal basis of the X-axis ;

[0120] Furthermore, the Y-axis orthogonal basis orthogonal basis of the X-axis satisfy:

[0121] ; ;

[0122] in, This indicates the position of the first gold marker in the resting phase. This represents the Euclidean norm.

[0123] S1.10: Based on the orthogonal basis of the X-axis Y-axis orthogonal basis Z-axis orthogonal basis Establish a tumor baseline coordinate system using the gold standard group centroid O;

[0124] S1.11: Based on respiratory signals and respiratory signals after Hilbert transformation Calculate the amplitude envelope ;

[0125] S1.12: Combine the instantaneous phase, amplitude envelope, and instantaneous phase derivative to form a respiratory phase feature vector. ,in, express The derivative of .

[0126] The specific steps of S2 include:

[0127] S2.1: Real-time acquisition of two-dimensional gold standard projection images via dual-plane X-ray fluoroscopy and The two-dimensional gold standard projection image is an orthogonal projection and is a grayscale image; the dual planes are two-dimensional detector planes, where x represents the corresponding X-axis coordinate and y represents the corresponding Y-axis coordinate.

[0128] S2.2: Projecting the two-dimensional gold standard image and Gold standard enhancement is performed, and candidate points are selected by detecting the principal axis direction of the gold standard projection through random transformation.

[0129] It should be noted that in this invention, the gold standard enhancement is performed by extracting the highlight area using the morphological Top-Hat transformation method. The morphological Top-Hat transformation is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0130] S2.3: Define the X-ray source-detector geometric parameters and set the initial six degrees of freedom parameters; the X-ray source-detector geometric parameters include the distance d from the X-ray to the detector, the rotation angle, etc. and The six degrees of freedom parameters include translation parameters. and rotation parameters ,in, , , These represent the translation amounts along the X-axis, Y-axis, and Z-axis, respectively. , , These represent the rotation angles around the X-axis, Y-axis, and Z-axis, respectively.

[0131] Furthermore, in the translation parameters, the translation along the X-axis is in the left-right direction and is the coronal plane; the translation along the Y-axis is in the front-back direction and is the sagittal plane; and the translation along the Z-axis is in the up-down direction and is the cross-section.

[0132] In the rotation parameters, the rotation angle around the X-axis is the forward or backward tilt angle, the rotation angle around the Y-axis is the side tilt angle, and the rotation angle around the Z-axis is the rotation angle.

[0133] S2.4: Obtain the gold standard-tumor dynamic model and extract the gold standard 3D coordinates from the gold standard-tumor dynamic model. ;

[0134] S2.5: Combining the geometric parameters of the X-ray source-detector and the six-degree-of-freedom parameters, the three-dimensional coordinates of the gold standard are... Projected onto the plane of the two-dimensional detector;

[0135] Furthermore, the three-dimensional coordinates of the gold standard The specific formula for projecting onto the plane of the two-dimensional detector is as follows: ,in, This represents the projected coordinates of the k-th gold standard. , , These represent the X-axis, Y-axis, and Z-axis coordinates after gold standard transformation.

[0136] S2.6: By calculating the actual X-ray image at pixel points grayscale value at Simulated X-ray images generated by projection of gold standard-tumor dynamic model in grayscale value at The difference is used to obtain the grayscale difference value. ;

[0137] S2.7: Based on the projected coordinates of the k-th gold standard in the two planes , and the two-dimensional coordinates of the k-th gold standard detected in the actual X-ray image , Construct the error function ;

[0138] S2.8: Based on error function Selecting the set of conjugate directions in the six-dimensional parameter space And along each conjugate direction, by adding the six degrees of freedom parameters to the preset optimal step size and the current conjugate direction. A one-dimensional search is performed using the product of these parameters to obtain the updated six degrees of freedom parameters, where r represents the conjugate direction index;

[0139] S2.9: Based on the updated six-degree-of-freedom parameters, combined with the tumor motion displacement field in S1.5 The tumor composite displacement vector was obtained. .

[0140] The specific steps of step S3 include:

[0141] S3.1: Obtain the tumor composite displacement vector and respiratory phase eigenvectors And perform timestamp alignment; the tumor composite displacement vector and respiratory phase feature vector are 3-dimensional vectors;

[0142] S3.2: Combine the tumor composite displacement vector and respiratory phase eigenvectors Merged into a 6-dimensional spacetime tensor ;

[0143] S3.3: Set up a sliding window to display the spacetime tensor. Data segmentation is performed to obtain spatiotemporal feature data;

[0144] S3.4: Load the pre-trained Long Short-Term Memory (LSTM) network model, input the spatiotemporal feature data into the pre-trained LSM network model, and output the predicted real-time location. and predicted phase ;

[0145] S3.5: Based on the obtained predicted real-time location and predicted phase By combining the mean square error and the phase cosine similarity, the composite loss value is obtained;

[0146] The composite loss value is the mean of the weighted sum of comprehensive errors; the weighted sum of comprehensive errors is the sum of the first variable and the second variable; the first variable is the Euclidean norm of the difference between the set of true positions of all gold targets and the predicted real-time positions in the i-th phase; the second variable is the product of the weight coefficient and the third variable; the third variable is the difference between 1 and the fourth variable; the fourth variable is the cosine of the true breathing phase and the predicted phase in the i-th phase.

[0147] S3.6: Obtain the current tumor center coordinates from the real-time location output by the Long Short-Term Memory network model. Combined with instantaneous phase It predicts the trajectory in the next second, and uses red to draw the real-time position in the tumor reference coordinate system, and uses blue gradient color band to draw the predicted trajectory to generate a tumor motion atlas;

[0148] S3.7: Based on the tumor composite displacement vector Displacement data over the past 5 seconds Incremental principal component analysis was performed to obtain the updated covariance matrix. ,in, Let b represent the covariance matrix estimate from the previous iteration, and b represent the forgetting factor, which is usually 0. In this invention, the selected The covariance matrix is ​​a square matrix, and its dimension is equal to the dimension of the tumor composite displacement vector.

[0149] S3.8: The updated X-axis direction vector in the tumor reference coordinate system is obtained by calculating the vector that maximizes the product of the covariance matrix and the unit vector E. , where arg min represents finding the maximum value of the objective function.

[0150] Example 2

[0151] Another embodiment of the present invention provides: a gold label implant for a tumor target area, comprising: a gold label body, the gold label body being made of pure gold and being cylindrical; the gold label body comprising a plurality of spheres and connecting segments; the plurality of spheres and the connecting segments being sequentially fixed together in a top-to-bottom manner.

[0152] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A method for real-time localization of tumor target areas, characterized in that, include: S1: Implant at least 4 gold markers inside and at the edge of the tumor, simultaneously acquire 4D CT image data and patient respiratory signals, construct a gold marker-tumor dynamic model based on the 4D CT image data, and establish a tumor reference coordinate system. At the same time, the patient's respiratory signals are preprocessed to generate respiratory phase feature vectors. S2: The improved Demons-Powell hybrid registration algorithm is used to spatially match the two-dimensional gold standard projection image captured in real time by dual-plane X-ray fluoroscopy with the gold standard-tumor dynamic model. The six-degree-of-freedom displacement parameters of the gold standard group are obtained through iterative optimization, and the tumor composite displacement vector is calculated. S3: Align the tumor composite displacement vector with the respiratory phase feature vector over time to establish a respiratory-displacement correlation model, generate a tumor motion atlas, and update the tumor baseline coordinate system based on the motion atlas; the tumor motion atlas includes the real-time position of the tumor and the predicted trajectory of the tumor. S4: Establish a dual safety boundary based on the tumor motion map and automatically adjust the treatment dose weight of the unlabeled area according to the gold standard interval; S5: The multi-leaf collimator performs sub-millimeter-level dynamic tracking based on tumor motion maps and safety boundary parameters; The specific steps of S1 include: S1.1: At least four gold markers are implanted inside and at the edge of the tumor, and respiratory signals are simultaneously acquired using an optical surface imaging system; the gold markers are non-coplanarly distributed. S1.2: Based on the acquired respiratory signals and respiratory signals after Hilbert transformation Calculate the instantaneous phase The respiratory cycle is divided into 10 phases based on instantaneous phase. A CT scan is triggered at a specific time, and a timestamp is recorded. ,in, i This represents the phase index value or time point index value after partitioning; S1.3: After completing the CT scan within each respiratory window, obtain four-dimensional CT image data. ; S1.4: For each gold standard k Extracting location sequences from 4D CT image data Where T represents transpose. k Indicates the gold standard index. Indicates the first k The gold medal in the first i The position of each phase , and They represent the first k The gold medal in the first i Each phase X Y Z Axis coordinates; S1.5: Continuous motion trajectory is obtained using a cubic spline interpolation algorithm. Simultaneously, the tumor motion displacement field is generated through difference calculation. The cubic spline interpolation algorithm is implemented by multiplying the extracted position sequence with B-spline basis functions; the number of B-spline basis functions is equal to the number of phase divisions. S1.6: Average the position sequence of all gold markers at the resting phase to obtain the centroid O of the gold marker group; the resting phase is... ; S1.7: Based on and global average position Calculate the motion covariance matrix A of the gold standard group; the global average position is The mean across all gold standards and phases; S1.8: Perform singular value decomposition on the motion covariance matrix A of the gold standard group, and take the eigenvector corresponding to the largest eigenvalue after singular value decomposition as the Z-axis orthogonal basis. ; S1.9: Based on deterministic Z Axis, generated through orthogonalization Y Axiorthogonal base and X Axiorthogonal base ; S1.10: According to X Axiorthogonal base , Y Axiorthogonal base , Z Axiorthogonal base Establish a tumor baseline coordinate system using the gold standard group centroid O; S1.11: Based on respiratory signals and respiratory signals after Hilbert transformation Calculate the amplitude envelope ; S1.12: Combine the instantaneous phase, amplitude envelope, and instantaneous phase derivative to form a respiratory phase feature vector. ,in, express The derivative; The specific steps of S2 include: S2.1: Real-time acquisition of two-dimensional gold standard projection images via dual-plane X-ray fluoroscopy and The two-dimensional gold standard projection image is an orthogonal projection and is a grayscale image; the dual planes are two-dimensional detector planes, wherein... x Indicates the corresponding X Axis coordinates y Indicates the corresponding Y Axis coordinates; S2.2: Projecting the two-dimensional gold standard image and Gold standard enhancement is performed, and candidate points are selected by detecting the principal axis direction of the gold standard projection through random transformation. S2.3: Define the X-ray source-detector geometric parameters and set the initial six degrees of freedom parameters; the X-ray source-detector geometric parameters include the distance from the X-ray to the detector. d Rotation angle and The six degrees of freedom parameters include translation parameters. and rotation parameters ,in, , , They represent along X axis, Y axis, Z Translation of the axis , , They represent circumference respectively. X axis, Y axis, Z The rotation angle of the shaft; S2.4: Obtain the gold standard-tumor dynamic model and extract the gold standard 3D coordinates from the gold standard-tumor dynamic model. ; S2.5: Combining the geometric parameters of the X-ray source-detector and the six-degree-of-freedom parameters, the three-dimensional coordinates of the gold standard are... Projected onto the plane of the two-dimensional detector; S2.6: By calculating the actual X-ray image at pixel points grayscale value at Simulated X-ray images generated by projection of gold standard-tumor dynamic model in grayscale value at The difference is used to obtain the grayscale difference value. ; S2.7: Based on the first k The first gold standard detected in the dual-plane projection coordinates and the actual X-ray image. k Using two-dimensional coordinates of a gold standard, an error function is constructed. ; S2.8: Based on error function Selecting the set of conjugate directions in the six-dimensional parameter space And along each conjugate direction, by adding the six degrees of freedom parameters to the preset optimal step size and the current conjugate direction. A one-dimensional search is performed using the product method to obtain the updated six-degree-of-freedom parameters, where, r Indicates the conjugate direction index; S2.9: Based on the updated six-degree-of-freedom parameters, combined with the tumor motion displacement field in S1.5 The tumor composite displacement vector was obtained. ; The specific steps of S3 include: S3.1: Obtain the tumor composite displacement vector and respiratory phase eigenvectors And perform timestamp alignment; the tumor composite displacement vector and respiratory phase feature vector are 3-dimensional vectors; S3.2: Combine the tumor composite displacement vector and respiratory phase eigenvectors Merged into a 6-dimensional spacetime tensor ; S3.3: Set up a sliding window to display the spacetime tensor. Data segmentation is performed to obtain spatiotemporal feature data; S3.4: Load the pre-trained Long Short-Term Memory (LSTM) network model, input the spatiotemporal feature data into the pre-trained LSM network model, and output the real-time location. and predicted phase ; S3.5: Based on the obtained real-time location and predicted phase By combining the mean square error and the phase cosine similarity, the composite loss value is obtained; S3.6: Obtain the current tumor center coordinates from the real-time location output by the Long Short-Term Memory network model. Combined with instantaneous phase The system predicts the trajectory for the next second and plots the real-time position in red in the tumor baseline coordinate system, while plotting the predicted trajectory using a blue gradient color band, thus generating a tumor motion atlas. S3.7: Based on the tumor composite displacement vector Displacement data over the past 5 seconds Incremental principal component analysis is performed to obtain the updated covariance matrix; the covariance matrix is ​​a square matrix and its dimension is equal to the dimension of the tumor composite displacement vector. S3.8: By calculating the vector that maximizes the product of the covariance matrix and the unit vector, the updated coordinates in the tumor baseline are obtained. X Axial direction vector.

2. The real-time localization method for tumor target area as described in claim 1, characterized in that, The establishment of dual security boundaries in S4 includes: The first layer of rigid safety boundary is established based on the gold standard. By analyzing the soft tissue texture features around the gold marker in the two-dimensional gold marker projection image using a convolutional neural network, and combining this with the tumor prediction trajectory, a second layer of elastic treatment boundary is generated.

3. The real-time localization method for tumor target area as described in claim 2, characterized in that, Also includes: During each treatment, gold standard displacement data and soft tissue texture features are collected simultaneously. An online adaptive mechanism is established to feed back the deviation between real-time gold standard displacement data and tumor prediction trajectory to the respiration-displacement correlation model. After each treatment, the actual radiation dose distribution was verified by cone-beam CT, and the error analysis results were fed back to the gold standard-tumor dynamic model for parameter correction.

4. A gold-labeled implant for a tumor target area, used to implement a real-time localization method for a tumor target area as described in any one of claims 1-3, characterized in that, The device includes a gold label body, which is made of pure gold and is cylindrical; the gold label body includes multiple spheres and connecting segments; the multiple spheres and the connecting segments are sequentially fixed together as one unit from top to bottom.

Images