Image data analysis method and system for radiology department

Abstract

The invention discloses an image data analysis method and system for the radiology department, and relates to the technical field of data analysis, and the method comprises the steps: carrying out the registration and feature extraction of a PET metabolism image and an ultrasonic image, and generating a physiological metabolism feature vector; reconstructing an organ surface manifold based on the real part of the compressed composite feature, and combining Gaussian curvature detection and blood vessel continuous homology analysis to obtain a manifold feature vector; based on the physiological metabolism feature vector and the manifold feature vector, constructing a dynamic hypergraph, and generating a cross-modal fusion feature vector; inputting the cross-modal fusion feature vector into a constraint generator, solving an elastic wave equation, and generating a diagnosis report; an elastic wave equation is combined with a Monte Carlo random disturbance model, deformation abnormity is quantified through a finite element node displacement field, and the malignant lesion detection sensitivity is remarkably improved in combination with form-metabolism-topology multi-dimensional parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to a method and system for parsing radiology image data. Background Art

[0002] In recent years, multi-modal medical image fusion technology has played an important role in tumor diagnosis and vascular lesion analysis. Among them, PET-CT-ultrasound multi-modal registration based on affine transformation, vascular structure enhancement, and cross-scale feature extraction have become research hotspots. In the prior art, rigid registration or non-rigid registration algorithms based on mutual information are mostly used. However, due to the neglect of the initial parameter optimization of image metadata (such as the position of the scanning bed), the registration error accumulates, especially in the preservation of the spatial consistency of hemodynamic parameters (such as resistance index). In addition, the geometric distortion correction of CT images mostly relies on isotropic filtering or morphological processing, which is difficult to effectively distinguish between scanning bed artifacts and edge blood vessels, and the traditional Gaussian pyramid decomposition is prone to losing high-frequency details, resulting in limited cross-scale correlation of subsequent feature encoding.

[0003] The limitations of the prior art are mainly reflected in two aspects: First, the parameter optimization in the image registration process depends on global search, lacks constraints on the complexity of transformation parameters, and is prone to falling into local optima, affecting the accurate mapping of metabolic activity standardized values; Second, the vascular structure enhancement uses fixed threshold segmentation or linear interpolation, which cannot simulate the dynamic impact of periodic physiological movements on vascular deformation, resulting in the interruption of vascular continuity, and further affecting the accuracy of subsequent manifold topology analysis and biomechanical modeling. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for parsing radiology image data to solve the problems of multi-modal image registration error accumulation, insufficient geometric distortion correction of blood vessels, and insufficient effectiveness of cross-scale feature fusion.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for parsing radiology image data, which includes, Registering and extracting features from PET images and ultrasound images to generate physiological metabolic feature vectors; Performing polar coordinate sine distortion enhancement and scanning bed removal on CT images to generate geometrically corrected three-dimensional medical images; Performing multi-scale Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical images, extracting cross-scale feature tensors through encoding state construction and entanglement operation, and outputting compressed composite feature tensors; Reconstruct the organ surface manifold based on the real part of the compressed composite features, combine Gaussian curvature detection and vascular persistent homology analysis to obtain the manifold feature vector; Based on the physiological metabolism feature vector and the manifold feature vector, construct a dynamic hypergraph and generate a cross-modal fusion feature vector; Input the cross-modal fusion feature vector into a constrained generator, solve the elastic wave equation, and generate a diagnostic report.

[0007] As a preferred solution of the method for parsing radiology image data according to the present invention, wherein: the registration and feature extraction include the following steps, Based on the patient's body position and the position of the scanning bed, initialize the translation and scaling components of the affine transformation matrix; Use the quasi-Newton method to iteratively optimize the registration parameters to minimize the registration error between the PET image and the ultrasound image; Map the registered PET image to the CT space through an inverse transformation, and calculate the metabolic activity value of each voxel using the SUV normalization formula; Outline the tumor boundary in the CT space and map it to the PET image, extract the maximum SUV value and the volume of voxels above the threshold, and generate a physiological metabolism feature vector in combination with hemodynamic parameters.

[0008] As a preferred solution of the method for parsing radiology image data according to the present invention, wherein: the generation of the geometrically corrected three-dimensional medical image includes the following steps, Convert the CT voxel data to the cylindrical coordinate system and apply a periodic sine perturbation; Use trilinear interpolation to inversely map the perturbed cylindrical coordinates to the Cartesian coordinate system to generate a distortion-enhanced image; Based on the scanning bed position parameters, construct a three-dimensional binary mask, suppress the bed artifact by pixel-by-pixel multiplication, and output the geometrically corrected three-dimensional medical image.

[0009] As a preferred solution of the method for parsing radiology image data according to the present invention, wherein: the extraction of the cross-scale feature tensor includes the following steps, Perform a three-level Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical image to construct a spatial domain encoding state and a frequency domain enhanced encoding state; Fuse the spatial domain encoding state and the frequency domain enhanced encoding state through a parameterized entanglement gate to generate a composite state, and compress it into a composite feature tensor using a complex orthogonal matrix projection.

[0010] As a preferred solution of the method for parsing radiology image data according to the present invention, wherein: the reconstruction of the organ surface manifold includes the following steps, Extract the isosurface grid based on the real part of the complex number of the encoded state of the compressed composite feature tensor, calculate the Riemannian metric tensor and discrete Gaussian curvature inspection, combine the α-shape algorithm to extract the one-dimensional persistent homology features of the vascular complex, and fuse the curvature anomaly and topological invariant to generate the manifold feature vector.

[0011] As a preferred solution of the method for parsing radiology image data described in the present invention, wherein: the construction of the dynamic hypergraph includes the following steps, Concatenate the physiological metabolism feature vector and the manifold feature vector into hypergraph nodes, and dynamically generate a hyperedge association matrix based on the Pearson correlation coefficient; Aggregate cross-modal information through hypergraph convolution and backpropagation, and combine degree matrix normalization to output a fused feature vector that captures the structural-functional co-variation.

[0012] As a preferred solution of the method for parsing radiology image data described in the present invention, wherein: the generation of the diagnostic report includes the following steps, Construct a finite element model based on the fused feature vector to solve the node displacement field, quantify the deformation amount and compare it with the healthy threshold to mark the lesion area; Predict the malignancy probability through Monte Carlo sampling of the random perturbation model, and generate a multi-modal diagnostic report by combining Gaussian curvature, persistent homology and the maximum SUV value.

[0013] In a second aspect, the present invention provides a system for parsing radiology image data, including, An image registration module that registers and extracts features from PET metabolism images and ultrasound images to generate a physiological metabolism feature vector; A geometric correction module that performs polar coordinate sine distortion enhancement and scan bed removal on CT images to generate a geometrically corrected three-dimensional medical image; An encoding module that performs multi-scale Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical image, extracts cross-scale feature tensors through encoding state construction and entanglement operations, and outputs a compressed composite feature tensor; A topological analysis module that reconstructs the organ surface manifold based on the real part of the compressed composite feature, and combines Gaussian curvature detection and vascular persistent homology analysis to obtain a manifold feature vector; A hypergraph fusion module that constructs a dynamic hypergraph based on the physiological metabolism feature vector and the manifold feature vector to generate a cross-modal fused feature vector; An elastic diagnosis module that inputs the cross-modal fused feature vector into a constraint generator to solve the elastic wave equation and generate a diagnostic report.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the method for parsing radiology image data as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the method for parsing radiology image data as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: In the geometric correction of CT images, through the periodic sine perturbation in the cylindrical coordinate system and trilinear interpolation, the natural deformation of blood vessels affected by the heartbeat is simulated, and at the same time, the scanning bed mask is used to accurately suppress artifacts; in the multi-scale feature encoding, weights are dynamically allocated through the gradient magnitude, and combined with the entangled operation of the space-frequency domain coding state, the problem of high-frequency detail loss in the traditional Gaussian pyramid decomposition is solved; in the biomechanical diagnosis stage, the elastic wave equation is combined with the Monte Carlo random perturbation model, and the deformation abnormality is quantified through the finite element node displacement field. By jointly using multi-dimensional parameters of morphology-metabolism-topology, the detection sensitivity of malignant lesions is significantly improved. Through dynamic hypergraph fusion of cross-modal features, the accurate positioning of structure-function collaborative abnormalities is achieved, providing a high-confidence basis for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the method for parsing radiology image data in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of image registration and feature extraction in Embodiment 1.

[0020] Figure 3 It is a schematic diagram of feature extraction and manifold reconstruction in Embodiment 1.

[0021] Figure 4 It is a schematic diagram of dynamic hypergraph construction and elastic diagnosis in Embodiment 1 DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the drawings of the specification.

[0023] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Persons skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0024] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0025] Embodiment 1, referring to Figures 1-4 , which is the first embodiment of the present invention. This embodiment provides a method for parsing radiology image data, including the following steps: S1. Define an affine transformation matrix including translation components, rotation components, and scaling factor transformation parameters; solve the optimal transformation parameters in the affine transformation matrix through an optimization algorithm to minimize the registration error and constrain the complexity of the parameters; set the initial translation and scaling based on image metadata (such as patient position, scan bed position); use the quasi-Newton method to iteratively update the optimal transformation parameters until the registration error change rate reaches the maximum number of iterations to obtain an optimized affine transformation matrix; map the PET image and the ultrasound image to the CT space through the inverse transformation of the affine transformation matrix for alignment to obtain the registered PET metabolic image and ultrasound image.

[0026] Further explanation, when aligning the ultrasound image to the CT space, the spatial consistency of blood flow velocity and resistance index is retained.

[0027] For the registered PET image, calculate the metabolic activity normalization value for each voxel, and the expression is: ; Where represents the voxel, represents the standardized uptake value of the voxel , dimensionless, represents the radioactive count of the registered PET image at the voxel , represents the injection dose, represents the patient's weight.

[0028] Outline the tumor boundary in the CT space, map it to the registered PET image, traverse all voxels within the boundary, and record the maximum standardized metabolic activity value; set the critical threshold of the standardized metabolic activity value, screen out the volume of voxels exceeding the critical threshold, and calibrate the voxel volume according to the CT resolution to obtain the calibrated voxel volume; combine the maximum standardized metabolic activity value, voxel volume, blood flow velocity, and resistance index to form the physiological and metabolic feature vector of the PET and ultrasound images.

[0029] S2. Store the CT image as DICOM - formatted CT voxel data; for each slice in the CT voxel data, with the center of the three - dimensional image as the origin, convert from Cartesian coordinates to cylindrical coordinates while keeping the inter - slice coordinates unchanged. The cylindrical coordinate system includes radial distance, polar angle, and axial height, and the expressions are as follows: ; Among them, represents the radial distance, that is, the distance from the voxel to the image center, represents the three - dimensional voxel coordinates in the original Cartesian coordinate system, represents the image center coordinates, which are used to translate the origin of the Cartesian coordinate system to the image center, represents the polar angle (in radians), that is, the azimuth angle of the voxel relative to the center point, with a range of [0, 2π), represents the height axis, which directly corresponds to the Z - axis (number of layers or slice number) of the Cartesian coordinate system.

[0030] Further explanation: The radial distance refers to the distance to the center in the XY plane, the polar angle refers to the rotation angle in the XY plane, and the axial height refers to the layer position along the Z - axis. The height axis represents the height axis and directly corresponds to the Z - axis (number of layers or slice number) of the Cartesian coordinate system.

[0031] Apply periodic sine perturbations jointly in the radial dimension and the layer direction to enhance the continuity of blood vessels and organs. The expression is: ; Among them, represents the perturbed radial distance, represents the perturbation amplitude, which controls the intensity of blood vessel deformation, represents the angular frequency, which determines the number of sine wave cycles and simulates periodic physiological movements (such as heartbeat), represents the radial threshold, which is used to limit the perturbation to only voxels to protect the central area (such as large blood vessels) from being affected. represents the rectangle function, defined as: ; When , shield the central area and apply perturbations only to the peripheral blood vessels.

[0032] The perturbed cylindrical coordinates are mapped back to Cartesian coordinates, and a distortion-enhanced three-dimensional image is generated by trilinear interpolation. The expression is as follows: The expression for inverse mapping to the Cartesian coordinate system is: ; where, represents the perturbed Cartesian coordinates, represents the trigonometric function value of the polar angle corresponding to the original Cartesian coordinates .

[0033] The trilinear interpolation formula is: ; where, represents the voxel value of the distortion-enhanced three-dimensional image at the original Cartesian coordinates , , , represents rounding down the perturbed Cartesian coordinates to obtain the nearest integer grid point, represents the three-dimensional neighborhood index, traversing the 8 adjacent voxels around the perturbed Cartesian coordinates, represents the weight factor, which is the distance attenuation weight (linear interpolation) of the distorted coordinates and the adjacent voxels in the direction.

[0034] It should be noted that the purpose of inverse mapping to the Cartesian coordinate system is to restore the perturbed cylindrical coordinates to Cartesian coordinates and generate the distorted spatial positions. Trilinear interpolation solves the problem that the distorted coordinates may be non-integer, and smoothly distributes the contribution values of the surrounding 8 voxels through trilinear interpolation, avoiding jagged artifacts and maintaining the continuity of blood vessel edges.

[0035] For each slice, the number of slices covered by the slice in the Z-axis direction is determined according to the physical position of the scanning bed; in each Z-axis slice, the projection radius of the scanning bed is calculated from the reconstruction field of view diameter and the bed position. The expression is: ; where, represents the coverage radius of the scanning bed in the XY plane, represents the reconstruction field of view diameter, represents the position parameter of the scanning bed in the Z-axis direction.

[0036] According to the projection radius of the scanning bed, a three-dimensional binary mask is defined, marking the scanning bed area as 0 and the anatomical area as 1. The expression is: ; where, represents the three-dimensional binary mask, Indicates the start position of the slice, Indicates the end position of the slice.

[0037] Multiply each layer of the mask with the distorted and enhanced three-dimensional image pixel by pixel to suppress the scan bed artifacts and obtain the geometrically corrected three-dimensional image.

[0038] Further explanation: The voxel values in the scan bed area (Mask = 0) are set to zero to eliminate metal artifacts or bed signal interference, and the vascular signals after distortion enhancement are retained in the anatomical area (Mask = 1).

[0039] S3. Perform three-level downsampling on the geometrically corrected image. The first-level sampling is to perform 2×2 neighborhood average pooling on the geometrically corrected image and downsample it to , where and are the height and width of the geometrically corrected image; the second-level downsampling is to perform repeated average pooling on the first-layer image and downsample it to ; the third-level downsampling is to perform average pooling on the second-layer image and downsample it to the target resolution, which is .

[0040] Further explanation: The Gaussian pyramid gradually extracts the global structure information of the image and suppresses high-frequency details through multi-scale decomposition by three-level downsampling.

[0041] Calculate the image gradient of the image after multi-scale decomposition using the Sobel operator. The expression is: ; where, represents the image after three-level downsampling, and represent the gradient components at coordinates and , and represent the Sobel convolution kernels of the image at coordinates and .

[0042] Calculate the gradient magnitude (L1 norm) of each image based on the image gradient. The expression is: ; where, represents the L1 norm of the gradient.

[0043] Generate the weight coefficient based on the gradient magnitude. The expression is: ; where, represents the center coordinates of the image The weight coefficient.

[0044] Calculate the global normalization factor according to the gradient magnitude to ensure that the total energy of the encoded state is 1. The expression is: ; where represents the normalization factor.

[0045] Combine the weights with the position basis states to generate the spatial domain encoded state. The expression is: ; where represents the spatial domain encoded state, represents the two-dimensional position basis state.

[0046] Map the image after multi-scale decomposition to the frequency domain through the fast Fourier transform; design a Gaussian low-pass filter to suppress high-frequency noise in the frequency domain image and retain low-frequency textures; multiply the Gaussian filter with the frequency domain signal element by element to obtain the filtered frequency domain complex matrix and enhance the low-frequency features; calculate the amplitude spectrum of the filtered frequency domain signal and normalize it. The expression is: ; where represents the normalized frequency domain complex matrix, represents the frequency coordinate, represents the filtered frequency domain complex matrix, represents the frequency domain complex matrix of the original frequency domain signal, represents a very small constant to avoid division by zero error when the amplitude spectrum is zero.

[0047] Map the normalized frequency domain complex matrix back to the spatial domain through the inverse Fourier transform to obtain the reconstructed spatial domain signal, that is, the frequency domain enhanced encoded state.

[0048] Combine the spatial domain encoded state and the frequency domain enhanced encoded state into a high-dimensional composite state through the tensor product, where the tensor product represents the independent superposition of spatial and frequency domain features; construct a parameterized entanglement gate to force the association between the spatial coordinates and the frequency domain modes to achieve feature fusion and form a composite state; apply the entanglement gate to the composite state to generate an entangled state; define a learnable complex orthogonal matrix that satisfies the orthogonality constraint and perform a linear projection on the frequency domain mode vectors at each spatial position to obtain the compressed composite feature tensor.

[0049] S4. Input the real part of the complex number of the encoded state of the compressed composite feature tensor, and extract the isosurface grid through the improved Marching Cubes algorithm, which is defined as follows: ; where Represents a triangulated isosurface mesh, i.e., the micro-differential manifold model of blood vessels, Represents a 3D mesh vertex, Represents a vertex index, Represents a 3D real vector space, Represents the real part value of the entangled state at the vertex reflecting the feature intensity, Represents the global mean of the real part values, Represents the standard deviation of the real part values, Represents a dynamic threshold to filter low feature intensity regions (e.g., outliers or noise).

[0050] Define local parameterized coordinates at the mesh vertex and calculate the Riemannian metric tensor. The expression is: ; where, Represents the Riemannian metric tensor, describing the local geometric properties of the surface, and Represents local parameterized coordinates, Represents a vertex index.

[0051] Based on the discrete Gaussian curvature formula, calculate the curvature of the vertex. The expression is: ; where, Represents the Gaussian curvature of the vertex, Represents the set of 1-ring neighborhood vertices of the vertex, Represents the interior angle of the neighborhood triangle at the vertex, Represents the Voronoi area of the vertex (weighted sum of neighborhood triangle areas), Represents a neighborhood index.

[0052] Furthermore, the Gaussian curvature reflects the local bending degree: >0 indicates a bulge (such as a blood vessel bifurcation point); <0 indicates a depression (such as a blood vessel stenosis); =0 indicates a flat area.

[0053] Screen out the vertices with abnormal curvature from the Gaussian curvature as candidate lesion regions; apply the α-shape algorithm to construct a blood vessel complex and extract topological invariants; extract the one-dimensional persistent homology features (ring structures) of the complex; fuse the candidate lesion regions, the extracted topological invariants, and the one-dimensional persistent homology features of the complex to generate a manifold feature vector of blood vessel geometry and topology.

[0054] S5. Concatenate the manifold feature vector and the physiological metabolism feature vector along the feature dimension to generate hypergraph node features; select central nodes based on feature importance, traverse all nodes, use the Pearson correlation coefficient to calculate the correlation between the current node and the central nodes, set a correlation threshold based on experience, and filter out nodes with a correlation higher than the correlation threshold and add them to the corresponding hyperedges to form a dynamic association group matrix.

[0055] To further explain, a hyperedge represents a group of vascular regions that co-vary in the structure-function bimodal. For example, the abnormal curvature of a certain section of blood vessel and the increased metabolic activity may belong to the same hyperedge.

[0056] Combine the initial features of the hypergraph nodes and the association matrix of the hyperedges to construct the degree matrices of the nodes and hyperedges for normalizing feature propagation; transform the degree matrices of the nodes and hyperedges through the weight matrix in the hypergraph convolution formula to learn the non-linear combination of cross-modal features and enhance the feature expression ability; perform hyperedge information aggregation on the transformed degree matrices of the nodes and hyperedges, aggregate the cross-modal information received by each node through the hyperedges, convert the cross-modal information into node-hyperedge relationships, realize the reverse propagation of information from the hyperedges to the nodes, gradually increase the feature dimension, generate cross-modal fusion features, and capture global cross-modal associations.

[0057] S6. Extract the vascular elasticity parameters and ultrasonic hemodynamic data in the physiological metabolism features. Based on the biomechanical properties and clinical data of healthy blood vessels, set the elasticity parameters of the healthy reference value and the lesion correction value, as well as the tissue density and vibration frequency, set the fixed constraint of the proximal displacement of the blood vessel, and apply the dynamic load of periodic pressure to the distal end (branch end) of the blood vessel; discretize the triangular mesh into a finite element mesh through the tetrahedral element generation algorithm to generate nodes and elements, and use the linear shape function for displacement field interpolation in each element to construct the local stiffness matrix and mass matrix; combine the local stiffness matrices and mass matrices of all elements to form a global linear system; use a sparse direct solver (such as LU decomposition) to solve the linear system to obtain the node displacement field; calculate the displacement amplitude for each node in the node displacement field to quantify the total deformation; statistically calculate the mean and standard deviation of the displacement amplitudes in the healthy region; define an abnormal threshold, compare the calculated displacement amplitude with the abnormal threshold, and when the calculated displacement amplitude is greater than the abnormal threshold, mark it as an abnormal lesion area.

[0058] Load the pre-trained random perturbation model and activate the Dropout mechanism. During the test phase, retain the random neuron dropout strategy during training. Simulate the posterior distribution of the Bayesian neural network by randomly masking the weights of some neurons, convert the deterministic model into a probabilistic model, approximate the parameter posterior distribution through Monte Carlo sampling, and obtain the trained random perturbation model; use the cross-modal fusion features as input, and through multiple independent forward propagations, randomly discard 20% of the neuron weights each time (such as randomly masking part of the weight matrix in the fully connected layer). The output of the model after each perturbation predicts the malignant probability, reflecting the prediction fluctuations caused by parameter randomness.

[0059] Based on the abnormal lesion areas, extract the Gaussian curvature, persistent homology features, and maximum standardized uptake value. Screen out the lesion areas that simultaneously meet the morphological abnormality, topological abnormality, and metabolic activity abnormality to form a lesion set; based on the malignant probability prediction results, perform malignant determination and benign determination to generate a diagnosis result; if there are multiple lesions, use the lesion with the maximum standardized uptake value as the main diagnosis basis.

[0060] This embodiment also provides a radiology image data analysis system, including: An image registration module that registers and extracts features from the PET metabolic image and the ultrasound image to generate a physiological metabolic feature vector; A geometric correction module that performs polar coordinate sine distortion enhancement and gantry removal on the CT image to generate a geometrically corrected three-dimensional medical image; An encoding module that performs multi-scale Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical image, extracts cross-scale feature tensors through encoding state construction and entanglement operations, and outputs a compressed composite feature tensor; A topological analysis module that reconstructs the organ surface manifold based on the real part of the compressed composite features, combines Gaussian curvature detection and vascular persistent homology analysis to obtain a manifold feature vector; A hypergraph fusion module that constructs a dynamic hypergraph based on the physiological metabolic feature vector and the manifold feature vector to generate a cross-modal fusion feature vector; An elastic diagnosis module that inputs the cross-modal fusion feature vector into a constraint generator, solves the elastic wave equation, and generates a diagnostic report.

[0061] This embodiment also provides a computer device applicable to the radiology image data analysis method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the radiology image data analysis method proposed in the above embodiment.

[0062] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or buttons, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0063] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for parsing radiology image data proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0064] In summary, in the geometric correction of CT images, the present invention simulates the natural deformation of blood vessels affected by heartbeat through periodic sine perturbation in the cylindrical coordinate system and trilinear interpolation, and at the same time uses a scanning bed mask to accurately suppress artifacts; in multi-scale feature encoding, weights are dynamically allocated through gradient amplitude, combined with the entangled operation of the space-frequency domain coding state, solving the problem of high-frequency detail loss in traditional Gaussian pyramid decomposition; in the biomechanical diagnosis stage, the elastic wave equation is combined with the Monte Carlo random perturbation model, and the abnormal deformation is quantified through the finite element node displacement field, jointly with multi-dimensional parameters of morphology-metabolism-topology, significantly improving the detection sensitivity of malignant lesions. Through dynamic hypergraph fusion of cross-modal features, accurate positioning of structure-function collaborative abnormalities is achieved, providing a basis with high credibility for clinical decision-making.

[0065] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for analyzing radiology image data, characterized in that: include, Perform registration and feature extraction on PET images and ultrasound images to generate physiological metabolic feature vectors; Perform polar coordinate sinusoidal distortion enhancement and scanning bed removal on CT images to generate geometrically corrected three-dimensional medical images; Perform multi-scale Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical images, extract cross-scale feature tensors through coded state construction and entanglement operations, and output compressed composite feature tensors; The real part of the compressed composite feature is used to reconstruct the organ surface manifold, and the manifold feature vector is obtained by combining Gaussian curvature detection and vascular continuous homology analysis. Based on physiological metabolic feature vectors and manifold feature vectors, a dynamic hypergraph is constructed to generate cross-modal fusion feature vectors. The cross-modal fusion feature vector is input into the constraint generator to solve the elastic wave equation and generate a diagnosis report.

2. The method for analyzing radiology image data according to claim 1, characterized in that: The registration and feature extraction The following steps are included: Initialize the translation and scaling components of the affine transformation matrix based on the patient's position and the scanner bed's position; The quasi-Newton method was used to iteratively optimize the registration parameters to minimize the registration error between PET images and ultrasound images. The registered PET images were mapped to the CT space by inverse transformation, and the metabolic activity value of each voxel was calculated using the SUV standardization formula; The tumor boundary was delineated in CT space and mapped to PET images, the maximum SUV value and super-threshold voxel volume were extracted, and the physiological metabolic feature vector was generated in combination with hemodynamic parameters.

3. The method for analyzing radiology image data according to claim 1, characterized in that: The generating of geometrically corrected three-dimensional medical images comprises the following steps: The CT voxel data are transformed into a cylindrical coordinate system and a periodic sinusoidal perturbation is applied; The perturbed cylindrical coordinates are inversely mapped to the Cartesian coordinate system using trilinear interpolation to generate the distortion-enhanced image; A three-dimensional binary mask is constructed based on the position parameters of the scanning bed, and bed artifacts are suppressed by pixel-by-pixel multiplication to output a geometrically corrected three-dimensional medical image.

4. The method for analyzing radiology image data according to claim 1, characterized in that: The extracted cross-scale feature tensor The following steps are included: Perform three-level Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical images to construct spatial domain coding state and frequency domain enhanced coding state; The spatial domain coded state and the frequency domain enhanced coded state are fused through parameterized entanglement gates to generate a composite state, which is then compressed into a composite feature tensor using complex orthogonal matrix projection.

5. The method for analyzing radiology image data according to claim 1, characterized in that: The reconstructing the organ surface manifold comprises the following steps: Based on the complex real part of the encoded state of the compressed composite feature tensor, the isosurface grid is extracted, the Riemann metric tensor and discrete Gaussian curvature test are calculated, and the one-dimensional continuous homology features of the vascular complex are extracted by combining the α-shape algorithm. The curvature anomaly and topological invariant are fused to generate the manifold feature vector.

6. The method for analyzing radiology image data according to claim 1, characterized in that: The construction of the dynamic hypergraph comprises the following steps: The physiological metabolic feature vectors and the manifold feature vectors are spliced ​​into hypergraph nodes, and the hyperedge correlation matrix is ​​dynamically generated based on the Pearson correlation coefficient; Cross-modal information is aggregated through hypergraph convolution and back-propagated, combined with degree matrix normalization, to output a fused feature vector that captures structure-function co-variations.

7. The method for analyzing radiology image data according to claim 1, characterized in that: Generating a diagnostic report comprises the following steps: Based on the fused feature vector, a finite element model is constructed to solve the node displacement field, quantify the deformation and compare it with the healthy threshold to mark the diseased area; The probability of malignancy is predicted by Monte Carlo sampling of the random perturbation model, and a multimodal diagnosis report is generated by combining Gaussian curvature, continuous homology and maximum SUV value.

8. A radiology image data analysis system, based on the radiology image data analysis method according to any one of claims 1 to 7, characterized in that: include, Image registration module, which performs registration and feature extraction on PET metabolic images and ultrasound images to generate physiological metabolic feature vectors; A geometric correction module performs polar coordinate sinusoidal distortion enhancement and scanning bed removal on CT images to generate geometrically corrected three-dimensional medical images; The encoding module performs multi-scale Gaussian pyramid decomposition on the geometrically corrected three-dimensional medical images, extracts cross-scale feature tensors through encoding state construction and entanglement operations, and outputs compressed composite feature tensors; The topological analysis module reconstructs the organ surface manifold based on the real part of the compressed composite features, and obtains the manifold feature vector by combining Gaussian curvature detection and vascular continuous homology analysis; The hypergraph fusion module constructs a dynamic hypergraph based on the physiological metabolic feature vector and the manifold feature vector to generate a cross-modal fusion feature vector; The elastic diagnosis module inputs the cross-modal fusion feature vector into the constraint generator, solves the elastic wave equation, and generates a diagnosis report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for analyzing radiology image data described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for analyzing radiology image data described in any one of claims 1 to 7 are implemented.

Images