Tumor model generation method and system based on adaptive edge lattice arrangement adjustment

Abstract

The application discloses a tumor model generation method and system based on adaptive edge lattice arrangement adjustment, and relates to the technical field of tumor modeling.The application comprehensively considers the influence of factors such as edge gradient, curvature and density on edge lattice points, dynamically determines the spring stiffness coefficient and damping coefficient corresponding to the virtual spring between the edge lattice points, calculates the resultant force borne by each edge lattice point, and iteratively updates the position of the edge lattice point by means of a dynamic equation, so that the edge lattice point can be stably anchored to the real edge due to the high gradient feature, that is, the micro-infiltration area is avoided to be missed, and the arrangement of the edge lattice point is more in line with the real tumor boundary.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tumor modeling, in particular to a tumor model generation method and system based on adaptive edge lattice arrangement adjustment. BACKGROUND

[0002] Radiotherapy is one of the core means of cancer treatment. Modern technologies such as intensity modulated radiotherapy (IMRT) and stereotactic radiotherapy (SRS) significantly improve the dose accuracy of tumor regions through precise dose regulation, but the toxicity of organs at risk around the tumor is still a key bottleneck. High toxicity not only limits the dose increase of radiotherapy, but also may reduce the tolerance of treatment due to normal tissue damage. Therefore, spatial fractionated radiotherapy (SFRT) is proposed as an innovative strategy. It balances efficacy and toxicity in large volume tumor treatment through non-uniform dose distribution (such as lattice dose gradient), and optimized lattice radiotherapy (LRT) as a core branch of SFRT, through the design of lattice structure in tumor to form differential dose gradient, becomes an important direction to improve treatment accuracy.

[0003] The existing LRT technology is based on the geometric contour of the tumor physical target volume (such as GTV), and generates lattice points by adopting regular initial arrangement (such as hexagonal close packing). Then, the lattice distribution is optimized by a fixed parameter spring model or empirical adjustment. However, the edge of clinical tumors (such as glioma and breast cancer) often presents "unclear" and "irregular" characteristics. The former is caused by tumor micro-infiltration (such as the presence of non-visible tumor cells within 1-3 mm of the edge), resulting in a blurred boundary. The latter is caused by morphologies such as lobulation, burr, and indentation (such as irregular edges formed by lung cancer pleural traction), resulting in complex geometric contours.

[0004] The existing LRT cannot adapt to the unclear edge characteristics: the initial lattice depends on the physical target volume contour and does not combine with biological characteristics such as edge gradient, making it difficult to identify micro-infiltration areas, resulting in sparse distribution of lattices in blurred edges (missing micro-infiltration lesions) or dense distribution (increasing the dose to normal tissues); in addition, the edge curvature difference (such as high curvature at burr requiring flexible adjustment and low curvature at smooth requiring stable constraint) is not considered, resulting in "flattening" of lattice arrangement in irregular edges (losing key morphologies such as burr) or "breaking" (lattice points are scattered between lobulation), making it difficult to fit irregular morphologies. Further, it is difficult to achieve precise differentiation between "precise coverage of micro-infiltration lesions" and "protection of normal tissues" in the existing LRT for tumors with unclear and irregular edges. SUMMARY

[0005] The present application aims to provide a tumor model generation method and system based on adaptive edge lattice arrangement adjustment to improve the above technical problems.

[0006] In order to achieve the above application purpose, the embodiments of the present application provide the following technical solutions:

[0007] A tumor model generation method based on adaptive edge lattice arrangement adjustment, comprising:

[0008] Real-time acquisition of PET-CT functional images and MRI magnetic resonance images of a patient, setting of virtual springs and generation of an initial tumor model through a HCP model, determination of edge lattice points;

[0009] Extraction of edge features of the tumor region of the PET-CT functional images and the MRI magnetic resonance images, generation of an edge probability map and determination of edge density;

[0010] Based on the edge features of the tumor region and the edge density, a spring correlation coefficient is determined, and the resultant force of each edge lattice point is calculated; the spring correlation coefficient includes a spring stiffness coefficient and a damping coefficient;

[0011] Based on the resultant force, the position information of each edge lattice point is updated; based on the updated edge lattice points, the initial tumor model is updated.

[0012] A tumor model generation system based on adaptive edge lattice arrangement adjustment, comprising:

[0013] An image acquisition module for real-time acquisition of PET-CT functional images and MRI magnetic resonance images of a patient;

[0014] An initial tumor model generation module for generating lattice target regions based on the PET-CT functional images through a hexagonal close-packed digital model; each lattice target region is used as a mass point, and a virtual spring is simulated between each two adjacent mass points to construct an initial tumor model;

[0015] An edge lattice point screening module for acquiring coordinate information of each lattice target region; based on the coordinate information, edge lattice points are determined;

[0016] A tumor region edge feature extraction module for extracting edge features of the tumor region of the PET-CT functional images and the MRI magnetic resonance images;

[0017] An edge density calculation module for obtaining an edge probability map through an edge probability generation model based on the edge features of the tumor region; based on the edge probability map, the edge density is calculated;

[0018] A spring correlation coefficient calculation module for determining a spring correlation coefficient based on the edge features of the tumor region and the edge density;

[0019] An edge lattice point resultant force calculation module for calculating the resultant force of each edge lattice point based on the spring correlation coefficient;

[0020] The tumor module updating module is configured to update the position information of each edge lattice point based on the respective force, and update the initial tumor model based on the updated edge lattice points.

[0021] The present application has the following advantages:

[0022] The present application comprehensively considers the influence of edge gradient, curvature, density and other factors on the edge lattice points, dynamically determines the spring stiffness coefficient and damping coefficient of the virtual spring between the edge lattice points, calculates the resultant force on each edge lattice point, and iteratively updates the position of the edge lattice point by means of the dynamic equation, so that the arrangement of the edge lattice points is more consistent with the real tumor boundary. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0024] Figure 1 The figure is a flow chart of the method in embodiment 1 of the present application.

[0025] Figure 2(a) is a three-dimensional distribution diagram of edge lattice points of the traditional method in embodiment 1 of the present application.

[0026] Figure 2(b) is a two-dimensional planar distribution diagram of edge lattice points of the traditional method in embodiment 1 of the present application.

[0027] Figure 2(c) is a dose distribution diagram of edge lattice points of the traditional method in embodiment 1 of the present application.

[0028] Figure 3(a) is a three-dimensional distribution diagram of updated edge lattice points in embodiment 1 of the present application.

[0029] Figure 3(b) is a two-dimensional planar distribution diagram of updated edge lattice points in embodiment 1 of the present application.

[0030] Figure 3(c) is a dose distribution diagram of updated edge lattice points in embodiment 1 of the present application.

[0031] Figure 4 The figure is a system structure diagram in embodiment 2 of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0033] Embodiment 1

[0034] Please refer to Figure 1 The embodiment provides a tumor model generation method based on adaptive edge lattice arrangement adjustment. Figure 1 The execution subject of the method can be a software and / or a hardware device. The execution subject of the present application can include but is not limited to at least one of the following: a user equipment, a network equipment, and the like. The user equipment can include but is not limited to a computer, a smart phone, a personal digital assistant (PDA), and the above-mentioned electronic devices, and the like. The network equipment can include but is not limited to a single network server, a server group composed of multiple network servers, or a cloud composed of a large number of computers or network servers based on cloud computing. The cloud computing is a kind of distributed computing, which is a super virtual computer composed of a loose coupled computer group. The embodiment is not limited in this regard.

[0035] A tumor model generation method based on adaptive edge lattice arrangement adjustment, comprising:

[0036] S1, acquiring PET-CT functional images and MRI magnetic resonance images of a patient in real time, setting a virtual spring, generating an initial tumor model through an HCP model, and determining edge lattice points;

[0037] The S1 comprises:

[0038] S1-1, acquiring PET-CT functional images through a PET-CT scanning system, and acquiring MRI magnetic resonance images through a nuclear magnetic resonance system;

[0039] Setting related parameters of the PET-CT scanning system and the nuclear magnetic resonance system, and the patient is photographed according to a standard posture to obtain corresponding PET-CT functional images and MRI magnetic resonance images.

[0040] S1-2, based on the PET-CT functional images, drawing a corresponding tumor target area and an organ at risk area;

[0041] S1-3. Based on the delineation rules, the tumor target area and the organ at risk area are delineated using automatic delineation software and the hexagonal closest packing digital model (HCP model) to obtain M lattice target areas (lattice points);

[0042] The delineation rules are:

[0043] (1) Delineate within the tumor target volume (GTV) and avoid the tumor target volume and vessels in the organ-at-risk region;

[0044] (2) The distance between the edge of the tumor target volume (GTV) and the edge of the lattice target volume (GTV-Lattice) is not less than 1 cm;

[0045] (3) The volume ratio of the tumor target volume (GTV) to the lattice target volume (GTV-Lattice) is within the range of [1%, 10%];

[0046] (4) The initial arrangement of the lattice points corresponding to the lattice target region (GTV-Lattice) is hexagonal closest packing (HCP), and the single-layer arrangement of the lattice points follows the planar hexagonal symmetry, that is:

[0047] ;

[0048] Distribution volume of the lattice target region (GTV-Lattice) for:

[0049] ;

[0050] in, 、 、 Represent column index, row index and layer index respectively, 、 、 Respectively represent the distances between the centers of two adjacent spheres in the x-axis, y-axis, and z-axis directions, Indicates the Layer Column and The position vector of the sphere center of the row, represents a constant, Indicates the absolute value, represents the number of spheres corresponding to the lattice target area, Indicates the radius of the sphere.

[0051] S1-4, obtaining coordinate information of each lattice target area; determining edge lattice points based on each coordinate information;

[0052] The coordinate information of each lattice target region is obtained by PET-CT functional image; the perpendicular straight line distance of each lattice target region and the edge obtained by delineation of the tumor region GTV is calculated; the lattice target region with the perpendicular straight line distance within the edge distance threshold range is selected as the edge lattice point.

[0053] The edge distance threshold range is formulated according to different types of tumors. For example, when the tumor is in the brain, due to the possible small infiltration (within 2mm) of benign meningioma, the risk of micro-infiltration of other brain tumors is slightly higher, but the infiltration range of other brain tumors rarely exceeds 3mm, and the surrounding normal tissue is sensitive to radiation, so as to avoid the lattice points from being excessively extended to the normal brain tissue due to the too large threshold value, and balance the coverage risk and the protection of normal tissue, the edge distance threshold range is set as (0.5mm, 3mm]. When the tumor is a gastrointestinal tumor, due to the irregularity of the gastrointestinal tumor due to the peristalsis of the digestive tract, and the existence of multiple important organs (such as liver, pancreas, mesentery, etc.) around, a larger edge distance threshold value can better adapt to the shape change of the tumor, so as to ensure the complete coverage of the tumor and the protection of the surrounding normal tissue, the edge distance threshold range is set as (1.5mm, 3.5mm].

[0054] S1-5, respectively, each lattice target region is regarded as a particle, and a virtual spring is simulated to be connected between each two adjacent particles, so as to construct and generate an initial tumor model;

[0055] According to the arrangement mode of the hexagonal close-packed digital model, a virtual spring is simulated to be connected between each two adjacent particles. The natural length of each virtual spring is the initial distance between the adjacent particle m (lattice point) and particle n (lattice point). According to the position information of the particle m and the particle n, the corresponding initial distance is calculated The corresponding formula is:

[0056] ;

[0057] Wherein, , , The position information of the particle m is represented, , , The position information of the particle n is represented.

[0058] Each lattice point represents a focus of dose pushing. Since the radiation dose will attenuate and scatter when propagating in the tissue, the dose pushing of one lattice point will not only affect the point itself, but also have an impact on the surrounding lattice points, and the virtual spring can well express the mutual influence between the lattice points, so that the virtual spring is selected to be introduced into the tumor model.

[0059] Currently, there are irregular or unclear edge tumor regions, and although the hexagonal closest packing lattice arrangement can efficiently cover the internal region, for complex and irregular tumor boundaries, the fixed arrangement of edge lattice points may not be closely fitted, and after introducing virtual springs, the edge lattice points are connected through springs, and the position of the edge lattice points can be dynamically adjusted through the elastic force of the spring, so that they can better adapt to the irregular boundary of the tumor, and in subsequent assistance to medical staff in formulating a radiation dose scheme, it can assist in optimizing the dose distribution to ensure that the high-dose region can more accurately cover the tumor, while reducing damage to the surrounding normal tissue.

[0060] S2, extract the edge features of the tumor region of the PET-CT functional image and the MRI magnetic resonance image, generate an edge probability map and determine the edge density;

[0061] The S2 comprises:

[0062] S2-1, extracting the functional metabolism features of the PET-CT functional image and the anatomical structure features of the MRI magnetic resonance image by a feature extraction algorithm; the functional metabolism features include radioactivity uptake intensity, and the anatomical structure features include texture features and boundary features in the tumor region;

[0063] S2-2, fusing the functional metabolism features and the anatomical structure features by a principal component analysis algorithm to generate a fusion feature map;

[0064] Specifically, the functional metabolism features and the anatomical structure features are fused to maximize the detail information of the PET-CT functional image and the MRI magnetic resonance image, and compared with directly fusing the PET-CT functional image and the MRI magnetic resonance image and then extracting features, the noise of the comprehensive feature image is lower, and the differentiation degree of the tumor and the normal tissue is higher, avoiding the dilution or interference of key information due to the difference in modalities (such as imaging principle, noise characteristics, feature type), which can more accurately capture the edge features and effectively identify the micro-infiltration region, thereby improving the accuracy of the edge probability, and the corresponding precision and anti-interference ability are stronger.

[0065] S2-3, edge detection of the fusion feature map by a convolutional neural network to obtain the edge features of the tumor region; since the convolutional neural network is used to extract edge information, which is prior art, no more description is given.

[0066] S2-4, inputting the edge features of the tumor region into an edge probability generation model to output an edge probability map; wherein the edge probability generation model can adopt a probabilistic neural network;

[0067] The mutual information is used to screen the edge features of the tumor region, and a fusion sensitive feature map is obtained. Since the fusion feature map is a multi-dimensional feature, including the metabolic intensity feature of the PET-CT functional image, the T2 weighted signal feature of the MRI magnetic resonance image, and the gray mean / variance of the tumor region, the fusion feature map needs to be screened. The mutual information value between the fusion feature map and the boundary information of the tumor region in S1-2 is calculated, and the features with a mutual information value less than the mutual information threshold are removed, and the features with a mutual information value greater than the mutual information threshold are retained.

[0068] The fusion sensitive feature map is normalized and input into a probabilistic neural network to obtain an edge probability map. In the probabilistic neural network, it includes convolution layers, pattern layers, summation layers and output layers connected in turn, and the normalized fusion sensitive feature map is subjected to convolution operation (convolution layer) to output an edge sensitive feature map. Each pixel value of the edge sensitive feature map represents the intensity of the position, and the higher the pixel value, the more likely it is an edge. The edge sensitive feature map is input into the pattern layer to calculate the similarity (Euclidean distance or cosine similarity) of the local feature of each pixel to the known tumor edge pattern. The similarities are input into the summation layer for weighted summation to obtain the edge score of each pixel. The edge scores are input into the output layer, and the edge scores are mapped to the range [0, 1] through the Sigmoid activation function to obtain the edge probability of each pixel. The edge probability is visualized, and the corresponding edge probability map is drawn according to the edge probability. The known tumor edge pattern refers to the typical feature combination learned from the labeled training data, which can represent the "tumor edge". The closer the edge probability is to 1, the more likely it is that the position is an edge position of the tumor region.

[0069] The edge probability map quantifies the possibility that each position is a tumor edge, providing accurate spatial quantification basis for edge lattice point adjustment for edge density calculation.

[0070] S2-5, based on the edge probability map, determine the edge density. The edge density is calculated through the edge probability map, which can quantify the possibility that each position and its surroundings are tumor edges. This edge density information can provide accurate spatial quantification basis for edge lattice point adjustment.

[0071] The edge density is the sum of the edge probability values in a unit volume or unit area. The area with high edge density indicates that the position and its surroundings are more likely to be tumor edges, while the area with low edge density indicates that the position and its surroundings are less likely to be tumor edges.

[0072] Based on the edge features of the tumor area, the edge contour of the tumor area is determined; the size of the local window is set; the edge contour of the tumor area is used as the boundary, and pixels outside the edge contour of the tumor area are not used for calculation. Each pixel in the edge probability map is used as the center of the local window to calculate the corresponding initial edge density. , the corresponding formula is:

[0073] ;

[0074] in, represents the summation function, represents the pixel at the center of the local window, represents the local window corresponding to the pixel at the center of the local window, Indicates The The edge probability of each pixel.

[0075] The size of the local window is determined according to actual application requirements. For example, when the tumor is lung cancer, the size of the local window is 5×5.

[0076] Normalize each initial edge density to obtain the corresponding edge density.

[0077] S3. Determine a spring correlation coefficient based on the edge characteristics and edge density of the tumor region, and calculate the resultant force of each edge lattice point; the spring correlation coefficient includes a spring stiffness coefficient and a damping coefficient.

[0078] At this time, only the edge lattice points in the initial tumor model are processed, and other lattice points are ignored. Only the virtual springs between the edge lattice points are processed, which can reduce processing time and facilitate the subsequent position update of the edge lattice points. Therefore, the S3 includes:

[0079] S3-1. Calculating the edge gradient and edge curvature of each edge lattice point based on the edge features of the tumor region;

[0080] The edge gradient is the rate of change in grayscale in the area surrounding the edge lattice point (the area surrounding the edge lattice point that does not extend beyond the edge contour of the tumor region). A larger edge gradient indicates a more pronounced image feature, indicating a higher likelihood that the location is a true edge. The edge curvature is the degree of curvature of the tumor region edge contour at the edge lattice point. When the edge curvature is greater than 0, the tumor region edge contour is a convex curve. A subsequent pulling force directed toward the convex vertex is required, causing the edge lattice points to stretch outward and ultimately conform to the convex contour. When the edge curvature is less than 0, the tumor region edge contour is a concave curve. A subsequent pulling force directed toward the concave vertex is required, causing the edge lattice points to contract inward and ultimately conform to the concave contour.

[0081] Wherein, the Canny operator can be used to calculate the edge gradient; specifically, first, the edge features of the tumor region are denoised by using a Gaussian filter, the image gradient amplitude and direction are calculated, non-maximum suppression is performed, the edge is refined, and finally the double-threshold detection and edge connection are performed to obtain the corresponding edge gradient.

[0082] Take the discrete edge lattice points on the edge , , , calculate and , and between the direction vectors , , and then calculate the edge curvature , and the corresponding formula is:

[0083] ;

[0084] Wherein, represents the absolute value.

[0085] S3-2, based on each edge gradient and each edge curvature, the spring stiffness coefficient of each virtual spring is calculated, and the corresponding process is:

[0086] Determine the initial spring stiffness coefficient, set the curvature attenuation coefficient and the curvature coordination weight; wherein the initial spring stiffness coefficients of each virtual spring are the same;

[0087] The edge gradient of each edge lattice point is normalized to obtain the corresponding edge gradient normalized value;

[0088] Based on the initial spring stiffness coefficient, the curvature attenuation coefficient, the curvature coordination weight and the edge gradient normalized value, the spring stiffness coefficient of each virtual spring is calculated.

[0089] Therefore, the formula corresponding to S3-2 is:

[0090] ;

[0091] Wherein, represents the spring stiffness coefficient of the virtual spring between the th edge lattice point and the adjacent th edge lattice point, represents the initial spring stiffness coefficient, , respectively represent the edge gradient normalized value corresponding to the th edge lattice point and the th edge lattice point, represents the curvature attenuation coefficient, , Respectively represent edge lattice points and the The normalized value of the edge curvature corresponding to the edge lattice point, represents a constant, represents the curvature synergy weight, .

[0092] S3-3. Based on the edge gradient, edge curvature, and edge density, the damping coefficient of each virtual spring is calculated. The corresponding process is:

[0093] Determine the initial damping coefficient, set the curvature-damping attenuation coefficient, the adjustment coefficient, and the preset maximum distance threshold; wherein, the initial damping coefficient of each virtual spring is the same;

[0094] Normalizing the edge density of each edge lattice point to obtain a corresponding normalized edge density value;

[0095] Calculate the distance between two adjacent edge lattice points;

[0096] The damping coefficient of each virtual spring is calculated based on the initial spring stiffness coefficient, the curvature-damping attenuation coefficient, the adjustment coefficient, the preset maximum distance threshold, the edge density normalization value, and the distance between two adjacent edge lattice points.

[0097] Therefore, the formula corresponding to S3-3 is:

[0098] ;

[0099] in, represents the curvature-damping attenuation coefficient, Indicates the The edge lattice point and the adjacent The damping coefficient of the virtual spring between the edge lattice points, represents the initial damping coefficient, 、 Respectively represent edge lattice points and the The normalized value of the edge density corresponding to the edge lattice point, 、 Respectively represent the preset maximum distance threshold, The edge lattice point and the adjacent The distance between the edge lattice points, represents the adjustment coefficient, .

[0100] S3-4. Calculate the spring force, damping force, and boundary guiding force on each edge lattice point based on each spring stiffness coefficient and each damping coefficient;

[0101] The formula corresponding to the spring force is:

[0102] ;

[0103] in, Indicates the The spring force on each edge lattice point is Indicates the The set of adjacent edge lattice points of an edge lattice point, Indicates the The edge lattice point and the adjacent The initial length of the virtual spring between the edge lattice points, that is, The edge lattice point and the adjacent The initial spacing between edge lattice points, Indicates the edge lattice point to the adjacent The displacement vectors of the edge lattice points, Represents the displacement vector The absolute value of .

[0104] The formula corresponding to the damping force is:

[0105] ;

[0106] in, 、 Respectively represent edge lattice points and the The velocity of the edge lattice points, Indicates the The damping force on each edge lattice point.

[0107] The formula corresponding to the boundary guiding force is:

[0108] ;

[0109] in, Indicates the The boundary normal vectors of the edge lattice points, Indicates the intensity coefficient, the value range is [0, 1], Indicates the The edge density of edge lattice points, Indicates the The boundary guiding force of the edge lattice points.

[0110] S3-5. Calculate the corresponding resultant force based on the spring force, damping force, and boundary guide force of each edge lattice point.

[0111] The formula corresponding to the resultant force is:

[0112] ;

[0113] in, Indicates the The resultant force acting on each edge lattice point.

[0114] The present invention integrates edge gradient, curvature and density into the calculation of virtual spring damping coefficient. Because gradient reflects edge realism, curvature reflects morphological flexibility and density represents spatial congestion, the three together construct an accurate mapping of damping and tumor edge characteristics. When the gradient is high, the damping is enhanced to constrain the stable movement of real edge points. When the curvature is large, the damping is reduced to allow flexible adjustment of the curved edge. When the density is high, the damping is adapted to prevent squeezing between points. Ultimately, the damping force is dynamically adapted to the needs of different areas of the tumor edge, realizing dynamic adjustment of the spring stiffness coefficient and the damping coefficient. After optimization, the movement of the edge lattice points is more reasonable, which can not only anchor the real boundary, but also fit the curved shape and avoid penetration of dense points, making the tumor edge contour more accurate.

[0115] S4. Based on the resultant forces, the position information of each edge lattice point is updated; based on the updated edge lattice points, the initial tumor model is updated.

[0116] The S4 includes:

[0117] S4-1. Initialize the initial position, initial velocity, mass and related parameters of all edge lattice points, and determine the set of adjacent edge lattice points of each edge lattice point; the related parameters include but are not limited to the curvature attenuation coefficient, the intensity coefficient, the curvature synergy weight, the curvature-damping attenuation coefficient and the adjustment coefficient; in this embodiment, since the size and shape of each edge lattice point are the same, the mass of the edge lattice point is set to 1.

[0118] S4-2. Based on the resultant forces, construct the dynamic equations of each edge lattice point; the dynamic equations are based on Newton's second law, namely:

[0119] ;

[0120] ;

[0121] in, Indicates the The mass of the edge lattice points, 、 Respectively represent edge lattice points at time ,time The position at the time, Indicates the The acceleration of the edge lattice points, Indicates time Next acceleration of the i-th edge lattice point, denotes a time step, denotes a velocity of the i-th edge lattice point.

[0122] S4-3, based on each kinetic equation, calculating the acceleration of each edge lattice point;

[0123] S4-4, based on the acceleration and the current velocity of each edge lattice point, updating the velocity of the corresponding edge lattice point, and the corresponding formula is:

[0124] ;

[0125] wherein, denotes the updated velocity of the i-th edge lattice point, denotes the current velocity of the i-th edge lattice point. S4-5, based on the current position information and the updated velocity of each edge lattice point, updating the position of the corresponding edge lattice point, obtaining the initial updated position of each edge lattice point and increasing the iteration number by 1; wherein the initial value of the iteration number is 0;

[0126] updating the position of the corresponding edge lattice point, the corresponding formula is:

[0127] ;

[0128]

[0129] wherein, , denote the position of the i-th edge lattice point before and after updating, respectively.

[0130] S4-6, based on the current position information and the initial updated position of each edge lattice point, calculating the displacement change of each edge lattice point, i.e. calculating the Euclidean distance between the current position information and the initial updated position;

[0131] S4-7, repeating S4-3 to S4-6, judging whether the iteration number reaches a preset iteration number threshold or the displacement change of each edge lattice point is less than a displacement threshold; if yes, taking the initial updated position of each edge lattice point as the final updated position, then entering S4-8; otherwise, taking the initial updated position and the updated velocity of each edge lattice point as the initial value of the next iteration, then returning to S4-3.

[0132] S4-8, based on each final updated position, adjusting the position of each edge lattice point in the initial tumor model, completing the update of the initial tumor model, and obtaining the tumor model.

[0133] ​​​​The application can more accurately simulate the dynamic change of the tumor edge by considering the interaction between the lattice points (spring force and damping force) and the boundary guiding force, and the automatic updating of the lattice point position reduces the workload of manual adjustment, which is conducive to the accuracy of accurately distinguishing normal combination and diseased tissue.

[0134] The traditional method is to construct a corresponding tumor model only through a hexagonal close-packed digital model. The same PET-CT functional image and MRI magnetic resonance image are processed by the traditional method and a tumor model generation method based on adaptive edge lattice arrangement adjustment (the present method), and the results are shown in Figures 2(a), 2(b), 2(c), 3(a), 3(b) and 3(c).

[0135] In Figures 2(a) and 3(a), the red area is the volume of the tumor (GTV), that is, the tumor area, and the yellow spherical ball is the edge lattice point. In Figure 2(b), the red area is the diseased part, the yellow area is the tumor area, and the green dot is the edge lattice point. In Figure 3(b), the red area is the diseased part, the yellow area is the tumor area, and the purple dot is the edge lattice point.

[0136] As shown in Figures 2(a) and 2(b), the edge lattice points obtained by the traditional method do not fit the tumor edge, the edge lattice points are relatively loose and weak in regularity within the tumor contour, and part of the edge lattice points deviate from the tumor edge and are in the non-tumor area. As shown in Figures 3(a) and 3(b), the edge lattice points obtained by the present method fit the irregular tumor edge better, the distribution is adaptively adjusted according to the edge morphology, the continuity is better, and the distribution is more accurate.

[0137] Based on the edge lattice points in Figures 2(a) and 2(b), dose distribution is performed on the corresponding tumor area, and the corresponding results are shown in Figure 2(c). Based on the edge lattice points in Figures 3(a) and 3(b), dose distribution is performed on the corresponding tumor area, and the corresponding results are shown in Figure 3(c). The dose distribution of the traditional method has poor matching degree with the tumor contour, the high-dose area does not accurately cover the tumor, and unnecessary irradiation is performed on the surrounding normal tissue; while the dose distribution of the present method closely fits the irregular tumor edge and internal structure, the high-dose area accurately covers the tumor area, and the dose overflow to the normal tissue is reduced, which further proves that the edge lattice points generated by the present method fit the irregular tumor edge better than the traditional method, and are conducive to distinguishing normal tissue and diseased tissue.

[0138] Example 2:

[0139] As shown in Figure 2(a) and 2(b), the edge lattice points obtained by the traditional method do not fit the tumor edge, the edge lattice points are relatively loose and weak in regularity within the tumor contour, and part of the edge lattice points deviate from the tumor edge and are in the non-tumor area. As shown in Figures 3(a) and 3(b), the edge lattice points obtained by the present method fit the irregular tumor edge better, the distribution is adaptively adjusted according to the edge morphology, the continuity is better, and the distribution is more accurate. Figure 4 The tumor model generation system based on adaptive edge lattice arrangement adjustment comprises:

[0140] An image acquisition module is configured to acquire PET-CT functional images and MRI magnetic resonance images of a patient in real time.

[0141] An initial tumor module generation module is configured to generate a lattice target region by a hexagonal close-packed digital model based on the PET-CT functional images; and to construct an initial tumor model by simulating a virtual spring between each two adjacent points.

[0142] An edge lattice point screening module is configured to acquire coordinate information of each lattice target region; and to determine edge lattice points based on the coordinate information.

[0143] A tumor region edge feature extraction module is configured to extract tumor region edge features of the PET-CT functional images and the MRI magnetic resonance images.

[0144] An edge density calculation module is configured to acquire an edge probability map by an edge probability generation model based on the tumor region edge features; and to calculate edge density based on the edge probability map.

[0145] A spring correlation coefficient calculation module is configured to determine a spring correlation coefficient based on the tumor region edge features and the edge density.

[0146] An edge lattice point resultant force calculation module is configured to calculate a resultant force of each edge lattice point based on the spring correlation coefficient.

[0147] A tumor module update module is configured to update position information of each edge lattice point based on the resultant force; and to update the initial tumor model based on the updated edge lattice points.

[0148] The system extracts edge features from PET-CT and MRI fusion images, calculates density by an edge probability map, dynamically adjusts spring coefficients, and enables the edge lattice points to be anchored to real edges due to high gradient features, to avoid missing micro-infiltration areas, and to adapt to irregular shapes due to curvature and density, thereby solving the problem of insufficient adaptation of traditional models to fuzzy and irregular edges.

[0149] It should be noted that, as for the system in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0150] The above only describes preferred embodiments of the present application and is not intended to limit the present application. The present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0151] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for generating a tumor model based on adaptive edge lattice arrangement adjustment, characterized in that: include: Acquire the patient's PET-CT functional images and MRI magnetic resonance images in real time, set virtual springs and generate an initial tumor model using the HCP model to determine the edge lattice points; Extract edge features of tumor regions from PET-CT functional images and MRI magnetic resonance images, generate edge probability maps, and determine edge density; Based on the edge characteristics and edge density of the tumor area, a spring correlation coefficient is determined, and the resultant force of each edge lattice point is calculated; the spring correlation coefficient includes a spring stiffness coefficient and a damping coefficient; Based on each resultant force, the position information of each edge lattice point is updated; Based on the updated edge lattice points, the initial tumor model is updated; The calculation of the resultant force of each edge lattice point includes: Based on the edge features of the tumor area, the edge gradient and edge curvature of each edge lattice point are calculated; Calculating the spring stiffness coefficient of each virtual spring based on the gradient of each edge and the curvature of each edge; Calculate the damping coefficient of each virtual spring based on each edge gradient, each edge curvature, and each edge density; Based on the spring stiffness coefficients and damping coefficients, the spring force, damping force and boundary guide force acting on each edge lattice point are calculated; Based on the spring force, damping force, and boundary guide force at each edge lattice point, the corresponding resultant force is calculated.

2. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: Generating an initial tumor model through the HCP model and determining edge lattice points includes: PET-CT functional images are acquired through a PET-CT scanning system, and MRI magnetic resonance images are acquired through a magnetic resonance imaging system; Based on PET-CT functional images, the corresponding tumor target area and organ-at-risk area are drawn; Based on the delineation rules, the tumor target area and the organ at risk area are delineated using the automatic delineation software and the HCP model to obtain M lattice target areas, that is, lattice points; Obtaining coordinate information of each lattice target area; determining edge lattice points based on each coordinate information; Each lattice target area is regarded as a particle, and a virtual spring is set to simulate the connection between two adjacent particles to construct an initial tumor model.

3. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: Generating an edge probability map and determining edge density include: The functional metabolic features of PET-CT functional images and the anatomical structural features of MRI magnetic resonance images are extracted through feature extraction algorithms; the functional metabolic features include the intensity of radioactive uptake, and the anatomical structural features include the texture features and boundary features within the tumor area; The principal component analysis algorithm is used to fuse functional metabolic features and anatomical structural features to generate a fusion feature map; Use convolutional neural network to perform edge detection on the fused feature map to obtain the edge features of the tumor area; The edge features of the tumor area are input into the edge probability generation model, and the edge probability map is output; Based on the edge probability map, an edge density is determined; the edge density is the sum of edge probability values ​​within a unit volume or unit area.

4. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: The formula corresponding to the spring force is: ； in, Indicates the The spring force on each edge lattice point is Indicates the The set of adjacent edge lattice points of an edge lattice point, Indicates the The edge lattice point and the adjacent The initial length of the virtual spring between the edge lattice points, that is, The edge lattice point and the adjacent The initial spacing between edge lattice points, Indicates the edge lattice point to the adjacent The displacement vectors of the edge lattice points, Represents the displacement vector The absolute value of represents the summation function, Indicates the The edge lattice point and the adjacent The spring stiffness coefficient of the virtual spring between the edge lattice points; The formula corresponding to the damping force is: ； in, 、 Respectively represent edge lattice points and the The velocity of the edge lattice points, Indicates the The damping force on each edge lattice point is Indicates the The edge lattice point and the adjacent The damping coefficient of the virtual spring between the edge lattice points; The formula corresponding to the boundary guiding force is: ； in, Indicates the The boundary normal vectors of the edge lattice points, Indicates the intensity coefficient, the value range is [0, 1], Indicates the The edge density of edge lattice points, Indicates the The boundary guiding force of the edge lattice points.

5. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: Calculating the spring stiffness coefficient of each virtual spring includes: Determine the initial spring stiffness coefficient, set the curvature attenuation coefficient and the curvature coordination weight; wherein, the initial spring stiffness coefficient of each virtual spring is the same; Normalizing the edge gradient of each edge lattice point to obtain the corresponding edge gradient normalized value; The spring stiffness coefficient of each virtual spring is calculated based on the initial spring stiffness coefficient, curvature attenuation coefficient, curvature cooperation weight and edge gradient normalization value.

6. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: The calculation of the damping coefficient of each virtual spring includes: Determine the initial damping coefficient, set the curvature-damping attenuation coefficient, the adjustment coefficient, and the preset maximum distance threshold; wherein, the initial damping coefficient of each virtual spring is the same; Normalizing the edge density of each edge lattice point to obtain a corresponding normalized edge density value; Calculate the distance between two adjacent edge lattice points; The damping coefficient of each virtual spring is calculated based on the initial spring stiffness coefficient, the curvature-damping attenuation coefficient, the adjustment coefficient, the preset maximum distance threshold, the edge density normalization value, and the distance between two adjacent edge lattice points.

7. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 1, characterized in that: The updating of the initial tumor model comprises: Initializing the initial positions, initial velocities, masses and related parameters of all edge lattice points, and determining the set of adjacent edge lattice points of each edge lattice point; Based on the resultant forces and Newton's second law, the dynamic equations of each edge lattice point are constructed; Based on the dynamic equations, the acceleration of each edge lattice point is calculated; Based on the acceleration and current velocity of each edge lattice point, update the velocity of the corresponding edge lattice point; Based on the current position information and updated speed of each edge lattice point, the position of the corresponding edge lattice point is updated to obtain the initial updated position of each edge lattice point and the number of iterations is increased by 1; wherein the initial value of the number of iterations is 0; Based on the current position information and the initial updated position of each edge lattice point, the displacement change of each edge lattice point is calculated, that is, the Euclidean distance between the current position information and the initial updated position is calculated; Repeatedly updating the position and displacement change of each edge lattice point until the number of iterations reaches a preset iteration threshold or the displacement change of each edge lattice point is less than the displacement threshold; Based on the final updated positions, the positions of the edge lattice points in the initial tumor model are adjusted to complete the update of the initial tumor model and obtain the tumor model.

8. The method for generating a tumor model based on adaptive edge lattice arrangement adjustment according to claim 7, characterized in that: The kinetic equation is: ； ； in, Indicates the The net force on each edge lattice point is, Indicates the The mass of the edge lattice points, 、 Respectively represent The edge lattice points in time ,time The position at the time, Indicates the The acceleration of the edge lattice points, Indicates time Next The acceleration of the edge lattice points, represents the time step, Indicates the The velocity of the edge lattice points, Indicates the The boundary guiding force of the edge lattice points, Indicates the The damping force on each edge lattice point is Indicates the The spring force on each edge lattice point; The formula corresponding to the speed of the edge lattice point corresponding to the update is: ； in, Indicates the The speed after the edge lattice point is updated, Indicates the The current velocity of each edge lattice point; The formula corresponding to the position of the edge lattice point corresponding to the update is: ； in, 、 Respectively represent The positions of edge lattice points before and after updating.

9. A tumor model generation system based on adaptive edge lattice arrangement adjustment, used to implement the tumor model generation method based on adaptive edge lattice arrangement adjustment according to any one of claims 1 to 8, characterized in that: include: Image acquisition module, used to obtain patients' PET-CT functional images and MRI magnetic resonance images in real time; The initial tumor module generation module is used to generate a lattice target area based on the PET-CT functional image through the hexagonal closest packing digital model. Each lattice target area is used as a particle, and a virtual spring is set to simulate the connection between two adjacent particles to construct the initial tumor model. The edge lattice point screening module is used to obtain the coordinate information of each lattice target area; based on each coordinate information, the edge lattice point is determined; Tumor region edge feature extraction module, used to extract tumor region edge features from PET-CT functional images and MRI magnetic resonance images; An edge density calculation module is used to obtain an edge probability map based on the edge features of the tumor area through an edge probability generation model; and calculate the edge density based on the edge probability map; A spring correlation coefficient calculation module, configured to determine a spring correlation coefficient based on edge features and edge density of the tumor region; the spring correlation coefficient includes a spring stiffness coefficient and a damping coefficient; The edge lattice point resultant force calculation module is used to calculate the spring force, damping force and boundary guide force of each edge lattice point based on each spring stiffness coefficient and each damping coefficient; Calculate the corresponding resultant force based on the spring force, damping force and boundary guide force of each edge lattice point; A tumor module update module is used to update the position information of each edge lattice point based on each resultant force; and update the initial tumor model based on the updated edge lattice points; The process of the spring correlation coefficient calculation module is as follows: Based on the edge characteristics of the tumor area, the edge gradient and edge curvature of each edge lattice point are calculated; based on each edge gradient and each edge curvature, the spring stiffness coefficient of each virtual spring is calculated; based on each edge gradient, each edge curvature and each edge density, the damping coefficient of each virtual spring is calculated.

Images