A method and system for predicting neoadjuvant chemotherapy efficacy of breast cancer based on ultrasound contrast

Abstract

This invention discloses a method and system for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging, belonging to the field of medical image data processing technology. It includes: acquiring and registering ultrasound image data before and after drug administration to generate a spatiotemporally aligned sequence; tracking microbubble trajectories to delineate functional sub-regions and construct a dynamic functional connectivity network, quantifying microcirculation functional evolution parameters; constructing a fluid-structure interaction transport model using tissue biophysical mapping parameters to calculate the spatial concentration distribution map of the drug; identifying spatial clusters of low concentration and low response to locate characteristic regions with evolutionary risks; and using a heterogeneity assessment model to fuse the topological features of risk regions with clinical phenotypic data to output quantitative reference parameters to assist in assessing chemotherapy responsiveness. This invention employs a collaborative architecture of physical perception simulation and spatiotemporal network analysis, which can capture the dynamic evolutionary differences of the microenvironment, improving the precision and objectivity of extracting response features of neoadjuvant chemotherapy for breast cancer.

Description

Technical Field

[0001] This invention relates to the field of medical image-assisted diagnosis and information processing, and in particular to a method and system for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging. Background Technology

[0002] Neoadjuvant chemotherapy, a crucial treatment for breast cancer, aims to reduce tumor size and pathological stage through preoperative administration. Real-time monitoring and quantification of chemotherapy response characteristics are of significant value in guiding subsequent treatment plans and analyzing patient status. Contrast-enhanced ultrasound, a non-invasive microcirculation perfusion imaging technique, utilizes the real-time reflection echoes of microbubble contrast agents within tumor vessels to acquire digital flow field information reflecting the tumor microenvironment. Analysis of the contrast-enhanced sequences using medical information processing technology has become an important research direction in modern smart healthcare for assisting in the assessment of tumor responsiveness.

[0003] Currently, in clinical practice, the evaluation of neoadjuvant chemotherapy efficacy using medical imaging typically employs tumor response assessment criteria. Common methods include measuring changes in the maximum tumor diameter using routine ultrasound imaging at baseline and after drug administration, or extracting global time-to-intensity curve parameters such as peak intensity and area under the curve within the region of interest using contrast-enhanced imaging. These methods attempt to indirectly infer the drug's cytotoxic effect on tissue cells by observing a reduction in the macroscopic tumor volume or a decrease in mean perfusion brightness.

[0004] However, existing technologies exhibit significant lag and limitations in practical applications. On one hand, physical changes in tumor volume often lag significantly behind metabolic changes in cellular function, making it difficult for data processing to identify low-response regions in the early stages. On the other hand, simple global perfusion parameters ignore the significant spatial heterogeneity within the tumor, failing to quantify the biophysical barriers that prevent chemotherapeutic drugs from being delivered through microvascular walls into the interstitial space. Especially when dealing with complex microcirculatory topological evolution, existing methods struggle to model the dynamic relationships between different functional zones before and after treatment, resulting in inaccurate extraction of localization features for potential drug resistance risk areas and an inability to provide refined quantitative parameters with auxiliary reference capabilities. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method and system for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging. It employs a combination of microcirculation dynamic functional connectivity network construction and fluid-structure interaction drug delivery simulation, enabling multidimensional quantitative analysis of chemotherapy response heterogeneity by identifying low-response-risk feature clusters and fusing multidimensional clinical phenotypic data.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging is provided, comprising: acquiring multi-temporal ultrasound images of subjects before and after administration of neoadjuvant chemotherapy, and performing three-dimensional image registration and motion compensation processing to generate a spatiotemporally aligned contrast imaging sequence with voxel-level spatial alignment features; tracking the motion displacement of contrast agent microbubbles from the spatiotemporally aligned contrast imaging sequence, extracting microbubble motion trajectories reflecting hemodynamic characteristics, and dividing functional sub-regions with physiological functional differences by performing unsupervised clustering processing on the microbubble motion trajectories; constructing a dynamic functional connectivity network for characterizing the microcirculatory structural state based on the topological connectivity relationships and correlation strength between the functional sub-regions, and quantifying microcirculatory functional evolution parameters describing the response intensity of the microcirculatory system by comparing the topological differences of the dynamic functional connectivity network before and after drug administration; and obtaining tissue biophysical mapping parameters reflecting tissue mechanical properties. The microcirculation function evolution parameters and the tissue biophysical mapping parameters are used as boundary constraints input into the fluid-structure interaction transport model. By performing an inverse solution operation, the spatial concentration of chemotherapy drugs in the interstitial space is calculated, generating a predicted distribution concentration map. The predicted distribution concentration map and the microcirculation function evolution parameters are spatially mapped and feature superimposed to identify spatial clusters that simultaneously possess low concentration distribution and low response activity, locating and generating low-response risk feature regions. The spatial topological features of the low-response risk feature regions, the drug coverage index of the predicted distribution concentration map, and the microcirculation function evolution parameters are extracted. The extracted features are input into a preset heterogeneity assessment model for feature fusion. The heterogeneity assessment model is then adaptively adjusted according to the clinical phenotypic data of the subjects, and outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotypic data includes serological indicators and biopsy histological features.

[0007] Based on the above technical solutions, in the method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging provided in this application, the technical means of combining the construction of microcirculation dynamic functional connectivity network with fluid-structure interaction drug delivery simulation can be used to identify low response risk feature spatial clusters and integrate multidimensional clinical phenotypic data to achieve multidimensional quantitative analysis of chemotherapy response heterogeneity.

[0008] In conjunction with the first aspect above, in one possible implementation, acquiring multi-temporal ultrasound images of the subject before and after neoadjuvant chemotherapy administration, and performing three-dimensional image registration and motion compensation processing includes: acquiring dynamic contrast-enhanced videos at baseline and after the first dose; using a linearization transformation algorithm to map pixel brightness to acoustic intensity values ​​reflecting microbubble concentration to generate a standardized grayscale matrix; extracting feature points of anatomical structures from the standardized grayscale matrix; constructing a non-rigid deformation field model by calculating the displacement vectors of feature points between adjacent frames; applying the non-rigid deformation field model to perform geometric transformation on the grayscale matrix to eliminate anatomical displacement caused by the respiratory cycle, thereby obtaining a motion-compensated image sequence; calculating the mutual information measure between the baseline and post-administration image sequences; adjusting coordinate transformation parameters according to the principle of maximizing the measure to achieve voxel matching of multi-temporal images; and outputting a spatiotemporally aligned contrast-enhanced sequence.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the motion displacement of contrast agent microbubbles is tracked from the spatiotemporally aligned contrast imaging sequence, and microbubble motion trajectories reflecting hemodynamic characteristics are extracted. Unsupervised clustering of the microbubble motion trajectories is then performed, including: locking the microbubble centroid within the search neighborhood of the spatiotemporally aligned contrast imaging sequence, recording the spatial coordinates of the microbubbles in consecutive time frames, and generating an initial microbubble displacement vector; performing temporal smoothing on the initial microbubble displacement vector to eliminate observation noise, and calculating the rate of change of displacement on the time axis to extract microbubble motion trajectories containing flow velocity and motion direction; measuring the geometric distance of each microbubble motion trajectory in the feature space, and aggregating trajectory clusters with feature distances within a preset similarity threshold range into a unified spatial entity, thus dividing functional sub-regions.

[0010] In conjunction with the first aspect above, in one possible implementation, a dynamic functional connectivity network for characterizing the microcirculation structural state is constructed based on the topological connectivity and correlation strength between the functional sub-regions. Microcirculation functional evolution parameters describing the response intensity of the microcirculation system are quantified by comparing the topological differences of the dynamic functional connectivity network before and after drug administration. This includes: calculating the cross-correlation coefficient of the average flow velocity signal within each functional sub-region, and using the cross-correlation coefficient as edge weights to connect adjacent sub-regions to construct the dynamic functional connectivity network; statistically analyzing the reduction in the overall network connection weights after drug administration to calculate the overall weight decay rate describing the intensity of macroscopic blood flow reduction; identifying key nodes in the dynamic functional connectivity network whose centrality exceeds a preset proportion, and statistically analyzing the number of failed connection edges of the key nodes to obtain the connection interruption proportion reflecting local structural damage; and combining the overall weight decay rate with the connection interruption proportion to generate microcirculation functional evolution parameters.

[0011] In conjunction with the first aspect above, in one possible implementation, the step of obtaining tissue biophysical mapping parameters reflecting tissue mechanical properties, and inputting the microcirculation functional evolution parameters and the tissue biophysical mapping parameters as boundary constraints into a fluid-structure interaction transport model, and calculating the spatial concentration of chemotherapy drugs in the interstitial space by performing an inverse solution operation, includes: matching permeability reference values ​​and interstitial pressure ranges corresponding to the current subject subtype from a preset tumor database to generate tissue biophysical mapping parameters; simplifying the functional subregion into grid cells with uniform medium characteristics, and extracting the microcirculation functional evolution parameters as the initial flux load of the grid cells; constructing a flow resistance matrix between grid cells using the tissue biophysical mapping parameters to establish a fluid-structure interaction transport model describing the delivery of drug solution from the vascular system to the interstitial space; using the pressure difference between each grid cell as the driving force, applying the finite element method to perform a mass conservation iterative solution on the fluid-structure interaction transport model to obtain the drug accumulation amount of each grid cell within a preset drug administration time, and outputting a predicted distribution concentration map after performing voxelization smoothing processing.

[0012] In conjunction with the first aspect above, in one possible implementation, spatial mapping and feature overlay are performed on the predicted distribution concentration map and the microcirculation function evolution parameters to identify spatial clusters that simultaneously possess low concentration distribution and low response activity. This includes: setting an effective drug action threshold; extracting a set of voxels with concentration values ​​lower than the effective action threshold from the predicted distribution concentration map and marking them as low drug distribution feature area masks; extracting the blood flow fluctuation amplitude of local sub-regions from the microcirculation function evolution parameters; identifying tissues whose function has not significantly deteriorated by comparing them with a preset system stability range and marking them as inert function masks; performing a logical intersection operation between the low drug distribution feature area mask and the inert function mask; locating the spatial coordinate range where the two overlap; and generating a low response risk feature region.

[0013] In conjunction with the first aspect above, in one possible implementation, the step of inputting the extracted features into a preset heterogeneity assessment model for feature fusion, and then having the heterogeneity assessment model perform adaptive weight adjustment based on the subject's clinical phenotype data to output a quantitative reference score for assisting in the assessment of chemotherapy responsiveness, includes: the heterogeneity assessment model calculating the feature vector centrality of the low-response-risk feature region in the dynamic functional connectivity network to obtain the risk region topological weight as a spatial topological feature; the heterogeneity assessment model calling a preset weight allocation function based on the molecular subtype category in the clinical phenotype data, performing a weight enhancement operation for the preset refractory subtype to increase the proportion of the risk region topological weight in the fusion calculation, and generating an adaptive decision operator; the heterogeneity assessment model applying the adaptive decision operator to perform a weighted aggregation operation on the risk region topological weight, the average drug coverage in the predicted distribution concentration map, and the microcirculation function evolution parameters to obtain a value representing the tendency of tissue response intensity, and outputting the value as a quantitative reference score.

[0014] In conjunction with the first aspect above, in one possible implementation, the method further includes: using the low-response-risk feature region as a simulation source point, retrieving associated paths in the dynamic functional connectivity network, simulating the diffusion trajectory of drug resistance features using a random walk algorithm, and calculating the risk evolution probability; mapping the risk evolution probability to the original ultrasound spatial coordinate system, coloring the probability values ​​using a color coding function, and generating a response risk evolution heatmap; and overlaying the response risk evolution heatmap with the baseline ultrasound structure map using transparency to output a spatial risk map indicating the distribution of response anomaly regions.

[0015] In conjunction with the first aspect above, in one possible implementation, the generation of the adaptive decision operator includes: acquiring the subject's serological indicators and biopsy histological characteristics, identifying the corresponding chemotherapy-sensitive subtype or low-response tendency group, and generating a subject classification feature vector; for subjects labeled as specific response characteristic groups, reducing the weighting coefficient of the global indicators in the microcirculation function evolution parameters, while performing nonlinear enhancement processing on the characteristics of the low-response risk feature region, optimizing decision sensitivity, and generating an adaptive decision operator that adapts to the differences in different subtypes.

[0016] Secondly, a system for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging is provided, comprising: an image registration and compensation module for acquiring multi-temporal ultrasound images of the subject before and after administration of neoadjuvant chemotherapy, and performing three-dimensional image registration and motion compensation processing to generate a spatiotemporally aligned contrast imaging sequence with voxel-level spatial alignment features; a kinetic feature extraction module for tracking the motion displacement of contrast agent microbubbles from the spatiotemporally aligned contrast imaging sequence, extracting microbubble motion trajectories reflecting hemodynamic characteristics, and dividing functional sub-regions with physiological functional differences by performing unsupervised clustering processing on the microbubble motion trajectories; a microcirculation evolution modeling module for constructing a dynamic functional connectivity network to characterize the microcirculation structural state based on the topological connectivity and correlation strength between the functional sub-regions, and quantifying microcirculation functional evolution parameters describing the response intensity of the microcirculation system by comparing the topological differences of the dynamic functional connectivity network before and after drug administration; and a drug transport simulation module for acquiring tissue biomechanical properties reflecting tissue biological characteristics. The physical mapping parameters are input into the fluid-structure interaction transport model as boundary constraints, using the microcirculation function evolution parameters and the tissue biophysical mapping parameters. By performing an inverse solution operation, the spatial concentration of chemotherapy drugs in the interstitial space is calculated, generating a predicted distribution concentration map. The risk feature region localization module is used to perform spatial mapping and feature superposition on the predicted distribution concentration map and the microcirculation function evolution parameters to identify spatial clusters that simultaneously possess low concentration distribution and low response activity, and to locate and generate low-response risk feature regions. The response heterogeneity assessment module is used to extract the spatial topological features of the low-response risk feature regions, the drug coverage index of the predicted distribution concentration map, and the microcirculation function evolution parameters. The extracted features are input into a preset heterogeneity assessment model for feature fusion. The heterogeneity assessment model performs weight adaptive adjustment based on the subject's clinical phenotypic data and outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotypic data includes serological indicators and biopsy histological features.

[0017] Compared with the prior art, the present invention has the following advantages: This invention achieves precise voxel-level alignment of multi-temporal contrast imaging data before and after neoadjuvant chemotherapy by constructing a three-dimensional image registration and motion compensation mechanism. This effectively eliminates the interference of respiratory motion and tissue deformation on the quantitative analysis of microcirculation, and provides a standardized spatial base map for the subsequent extraction of high-confidence hemodynamic features.

[0018] This invention introduces a dynamic functional connectivity network and a fluid-structure interaction transport model, overcoming the limitations of traditional ultrasound contrast imaging that only focuses on surface perfusion parameters. It enables in-depth modeling from two dimensions: the evolutionary trend of the microcirculation topology and the physical diffusion law of the drug solution in the interstitial space. This cross-scale simulation method can identify low-concentration characteristic spatial clusters caused by differences in interstitial pressure distribution or vascular function, thereby locating characteristic regions at risk of evolution before morphological changes occur.

[0019] This invention establishes an assessment framework that integrates multidimensional heterogeneity features, adaptively invoking clinical phenotypic data of subjects to adjust the decision weights of the computational model in real time. This computational approach, which combines radiomics dynamics features with physiological indicators and histological features, fully considers the heterogeneity differences among different molecular subtypes of breast cancer, improving the accuracy and robustness of calculations for quantitative parameters such as microcirculation response intensity.

[0020] This invention provides a response risk evolution heatmap and spatial distribution map. It simulates the potential evolutionary trajectory of response characteristics using a random walk algorithm, transforming the results of complex mathematical modeling into visual charts with anatomical structural references. The aim is to intuitively present the spatial heterogeneity distribution of response intensity within a target area, providing multi-dimensional objective auxiliary reference data for the formulation of individualized solutions.

[0021] Other features and advantages of the invention will be set forth in the following description and will be apparent from the description, or may be learned by practicing the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures pointed out in the description, claims, and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A structural architecture diagram of a system for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging, provided in an embodiment of this application; Figure 2 A flowchart illustrating a method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging, provided as an embodiment of this application; Figure 3 This is a comparison diagram of the time-intensity curves of contrast agents in different functional sub-regions provided in the embodiments of this application; Figure 4This is a spatial thermal distribution map of the evolution of drug resistance risk within tumors provided in the embodiments of this application. Detailed Implementation

[0024] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging provided in this application embodiment can be applied to, for example... Figure 1 In a system 100 for predicting the efficacy of neoadjuvant chemotherapy in breast cancer based on ultrasound contrast imaging, as shown, Figure 1 As shown, the system includes: an image registration and compensation module, used to acquire multi-temporal ultrasound images of the subject before and after neoadjuvant chemotherapy, and perform three-dimensional image registration and motion compensation processing to generate a spatiotemporally aligned contrast imaging sequence with voxel-level spatial alignment features; a kinetic feature extraction module, used to track the motion displacement of contrast agent microbubbles from the spatiotemporally aligned contrast imaging sequence, extract microbubble motion trajectories reflecting hemodynamic characteristics, and divide functional sub-regions with physiological functional differences by performing unsupervised clustering processing on the microbubble motion trajectories; a microcirculation evolution modeling module, used to construct a dynamic functional connectivity network to characterize the microcirculation structural state based on the topological connectivity and correlation strength between the functional sub-regions, and quantify and generate microcirculation functional evolution parameters describing the response intensity of the microcirculation system by comparing the topological differences of the dynamic functional connectivity network before and after drug administration; and a drug transport simulation module, used to acquire tissue biophysical mapping parameters reflecting tissue mechanical properties, and to simulate the microcirculation process. The circulatory function evolution parameters and the tissue biophysical mapping parameters are used as boundary constraints input into the fluid-structure interaction transport model. By performing an inverse solution operation, the spatial concentration of chemotherapy drugs in the interstitial space is calculated, generating a predicted distribution concentration map. The risk feature region localization module is used to perform spatial mapping and feature superposition of the predicted distribution concentration map and the microcirculatory function evolution parameters to identify spatial clusters that simultaneously possess low concentration distribution and low response activity, and to locate and generate low-response risk feature regions. The response heterogeneity assessment module is used to extract the spatial topological features of the low-response risk feature regions, the drug coverage index of the predicted distribution concentration map, and the microcirculatory function evolution parameters. The extracted features are input into a preset heterogeneity assessment model for feature fusion. After the heterogeneity assessment model performs weight adaptive adjustment based on the clinical phenotypic data of the subjects, it outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotypic data includes serological indicators and biopsy histological features.

[0026] like Figure 2 As shown in the embodiments of this application, a method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging is provided, including: Multi-temporal ultrasound images of subjects before and after neoadjuvant chemotherapy were acquired, and three-dimensional image registration and motion compensation were performed to generate spatiotemporally aligned contrast imaging sequences with voxel-level spatial alignment features. The movement and displacement of contrast agent microbubbles are tracked from the spatiotemporally aligned contrast sequence, the microbubble movement trajectory reflecting hemodynamic characteristics is extracted, and functional sub-regions with physiological differences are divided by unsupervised clustering of the microbubble movement trajectory. Based on the topological connectivity and association strength between functional sub-regions, a dynamic functional connectivity network is constructed to characterize the state of the microcirculation structure. By comparing the topological differences of the dynamic functional connectivity network before and after drug administration, microcirculation functional evolution parameters describing the response intensity of the microcirculation system are quantified. Obtain tissue biophysical mapping parameters that reflect tissue mechanical properties, input the microcirculation function evolution parameters and tissue biophysical mapping parameters as boundary constraints into the fluid-structure interaction transport model, and calculate the spatial concentration of chemotherapy drugs in the interstitial space by performing inverse solution operations to generate a predicted distribution concentration map. The predicted distribution concentration map is spatially mapped and feature overlaid with the microcirculation function evolution parameters to identify spatial clusters that simultaneously possess low concentration distribution and low response activity, thereby locating and generating low response risk feature regions. The spatial topological features of the low-response-risk region, the drug coverage index of the predicted distribution concentration map, and the microcirculation function evolution parameters are extracted. The extracted features are input into a preset heterogeneity assessment model for feature fusion. The heterogeneity assessment model performs weight adaptive adjustment based on the clinical phenotype data of the subjects and outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotype data includes serological indicators and biopsy histological features.

[0027] It should be noted that the method disclosed in this invention for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging is based on the principle of transforming multi-temporal dynamic contrast-enhanced ultrasound images into multi-dimensional physical and dynamic descriptors, and achieving prediction by establishing a coupling relationship between the microcirculation system and the drug delivery model. First, using three-dimensional image registration and motion compensation technology, a high degree of alignment of anatomical structures before and after treatment is achieved in the voxel dimension, providing a spatial benchmark for evolutionary analysis. Second, by tracking the movement trajectory of microbubbles and performing unsupervised clustering, complex hemodynamic features are extracted into sub-regions with physiological functional differences, and a dynamic functional connectivity network is further constructed to characterize the stability of the microcirculation topology. On this basis, tissue biophysical mapping parameters are introduced as constraints, and a fluid-structure interaction transport model is applied to perform inverse solving, mapping macroscopic flow field information to drug concentration distribution at the microscopic interstitial level. Finally, by superimposing spatial features, low-response-risk regions with insufficient physical drug coverage and delayed physiological response are identified, and heterogeneous weighted fusion is performed in combination with clinical phenotypic data to construct a logical chain from microscopic transport to macroscopic efficacy assessment.

[0028] In one possible implementation of the embodiments of this application, combined with Figure 2 The acquisition of multi-temporal ultrasound images of the subject before and after neoadjuvant chemotherapy administration, and the performance of three-dimensional image registration and motion compensation processing, includes: Dynamic contrast-enhanced videos were acquired at the baseline period and after the first dose. A linearization transformation algorithm was used to map pixel brightness to acoustic intensity values ​​reflecting microbubble concentration, generating a standardized grayscale matrix. Feature points of anatomical structures are extracted from the standardized grayscale matrix. A non-rigid deformation field model is constructed by calculating the displacement vectors of feature points between adjacent frames. The non-rigid deformation field model is then applied to perform a geometric transformation on the grayscale matrix to eliminate anatomical displacement caused by the respiratory cycle, resulting in a motion-compensated image sequence. The mutual information measure between the baseline and post-drug administration image sequences is calculated. The coordinate transformation parameters are adjusted according to the principle of maximizing the measure to achieve voxel matching of multi-temporal images and output spatiotemporally aligned imaging sequences.

[0029] In some implementations, dynamic contrast-enhanced videos of the subject at baseline and after the first dose are first acquired. To eliminate the interference of nonlinear mapping in ultrasound imaging on hemodynamic quantification, a linearization transformation algorithm is employed. The standardized grayscale matrix is ​​processed using the following calculation formula: ; in, The acoustic intensity value represents the concentration of microbubbles. Here, 'a' represents the original pixel brightness value in the image sequence, 'a' is the device gain compensation coefficient, and 'b' is the inverse mapping exponent of the logarithmic compression mapping, typically ranging from 1.5 to 2.2. Qualitative grayscale visual features are transformed into physical concentration measures with additivity and proportionality, thus ensuring the accuracy of subsequent hemodynamic calculations. Feature extraction algorithms are used to identify feature points of anatomical structures such as ductal wall boundaries and glandular edges from the standardized grayscale matrix. A non-rigid deformation field model is constructed to address the flexible compressibility of breast tissue. By calculating the displacement vectors of feature points between adjacent frames. : ; in, The coordinates of the feature points in the current frame. Using the coordinates of feature points in the reference frame, apply Spline interpolation functions generate continuous, non-rigid deformation fields. The deformation field is then used to calculate the corrected coordinates for each pixel via a geometric transformation function. The calculation formula is: ; in, The original pixel space vector, The displacement compensation is derived from the non-rigid deformation field. This action compensates for anatomical shifts of 2 to 5 millimeters caused by the respiratory cycle, generating motion-compensated image sequences. Mutual information is introduced as a similarity evaluation criterion for multi-temporal alignment. The mutual information of two temporal image sequences is calculated using a formula. : ; in, The marginal entropy representing the baseline sequence, The marginal entropy represents the sequence after drug administration. The joint entropy represents both. Both marginal entropy and joint entropy are obtained by calculating the probability density function of the voxel intensity distribution of the corresponding temporal phase. During the iterative search of coordinate transformation parameters, the optimal spatial rotation and translation matrices are determined through the principle of maximizing measure. When the mutual information reaches a preset convergence threshold of 0.85 or higher, voxel matching of multi-temporal images is completed. The tumor spatial structure of different temporal phases is reconstructed to a unified Cartesian coordinate system, outputting a spatiotemporally aligned imaging sequence with three-dimensional spatial feature alignment capabilities.

[0030] For example, the original image is standardized using a linearization transformation algorithm, with a device gain compensation coefficient 'a' of 1.2 and an inverse mapping exponent 'b' of logarithmic compression mapping of 2.0. If the original brightness value of a certain key pixel... Substitute 10 into the formula The acoustic intensity value was calculated. This transforms qualitative grayscale values ​​into a quantitative concentration matrix. To eliminate displacement caused by respiration, feature points at the gland edges are identified, and a non-rigid deformation field model is constructed. If the coordinates of the feature points in the current frame for Reference frame coordinates for The displacement vector is calculated. Substitute into the geometric transformation function Coordinate correction for each pixel successfully compensated for approximately 2 millimeters of anatomical offset. A mutual information metric was introduced. Perform voxel alignment and set the baseline sequence margin entropy. The post-drug administration sequence marginal entropy is 4.5. The joint entropy is 4.6. The value is 8.3, which is then substituted into the formula. The calculated mutual information value is 0.8. The coordinate transformation parameters are adjusted using an iterative optimizer to... Increased to above the convergence threshold of 0.88.

[0031] In one possible implementation, combining Figure 2 The movement and displacement of contrast agent microbubbles are tracked from the spatiotemporally aligned contrast imaging sequence, and the microbubble movement trajectories reflecting hemodynamic characteristics are extracted. Unsupervised clustering of these microbubble movement trajectories is then performed, including: Within the search neighborhood of the spatiotemporally aligned imaging sequence, the centroid of the microbubble is locked, the spatial coordinates of the microbubble in consecutive time frames are recorded, and the initial microbubble displacement vector is generated. The initial microbubble displacement vector is subjected to time smoothing to eliminate observation noise, and the rate of change of displacement on the time axis is calculated to extract the microbubble motion trajectory containing flow velocity and motion direction. The geometric distance of each microbubble motion trajectory in the feature space is measured, and the trajectory clusters with feature distances within a preset similarity threshold range are aggregated into a unified spatial entity, and functional sub-regions are divided.

[0032] In some implementations, mask matching is first performed within a pre-defined search neighborhood of the spatiotemporally aligned imaging sequence to locate the microbubble centroid. The search neighborhood refers to a search window within a certain pixel diameter surrounding the current pixel to detect high-contrast response signals. The microbubble centroid is calculated using the following formula: ; in, This represents the centroid vector of the microbubble in a three-dimensional coordinate system. This represents the weighted sum of the spatial coordinates of locally highlighted pixels. This represents the total number of effective pixels covered by the microbubble mask. The microbubble is recorded in consecutive time frames using a macro correlation algorithm. The resulting spatial displacement generates the initial microbubble displacement vector. : ; in, The centroid vector at the current moment. This represents the centroid vector corresponding to the previous moment. This process transforms the dynamic jumps in image grayscale into directional and displacement attributes with spatial geometric meaning, laying the data foundation for describing the microscopic flow field. Temporal smoothing is performed on the initial microbubble displacement vector, eliminating displacement offsets caused by observation noise or system flicker by using a filter with a time window step size of 3 to 5 frames. The smoothed microbubble path is defined as... Motion features are extracted by calculating the rate of change of the path over time. The calculation formula is as follows: ; in, Represents an instantaneous vector containing information about the flow velocity scalar and the direction of motion. This represents the differential increment of the smooth displacement path in a very short time interval. The acquisition time step represents the time step between adjacent image frames, typically ranging from 0.02 seconds to 0.05 seconds. By analyzing the displacement gradient, microbubble motion trajectories containing perfusion velocity, vessel curvature, and flow stability characteristics are extracted. The similarity of each microbubble motion trajectory in the feature space is quantified, and the feature space distance is calculated using the following formula: in, Let be the characteristic Euclidean distance between the two motion trajectories. and These represent the feature vectors of different microbubble motion trajectories at the th... Components of each dimension Let be the total dimension of the feature space. This formula achieves a quantitative assessment of hemodynamic similarity by calculating the geometric deviation between two instantaneous vector sequences in multidimensional space. The feature distance is then used. Within the preset similarity threshold range Trajectory clusters within the same feature space are merged into the same feature space label. Similarity threshold range. Typically, the standard deviation of the mean flow velocity within the tumor is dynamically defined, ranging from 0.15 to 0.25. Unsupervised clustering is then performed to divide physically adjacent sets of pixels with consistent kinematic characteristics into functional subregions with physiological differences. These functional subregions represent anatomical units within the tumor that exhibit consistent perfusion levels and similar microenvironmental pressures.

[0033] For example, a microbubble mask is located within the search neighborhood, and the weighted sum of the coordinates of its locally highlighted pixels is used to determine the mask. for Total number of effective pixels covered Substitute 10 into the formula Calculate the center of mass of the microbubble for If the current centroid is in a continuous time frame for The center of mass at the previous moment for Then the initial displacement vector The path is obtained by smoothing the displacement vector using a filter. If the smooth displacement increment is within a time step of 0.02s between adjacent frames, It is 0.08mm, substituting it into the formula. Instantaneous flow velocity was extracted The speed is 4 mm / s. Two trajectories are quantized. and To assess the similarity between them, let the sum of squares of the differences in the velocity and curvature components of their eigenvectors be 0.01, and substitute this into the formula. Calculate the feature space distance The value is 0.1. Since this value is within the preset similarity threshold range of 0.15 to 0.25, the trajectory cluster is aggregated to divide functional sub-regions with consistent perfusion levels.

[0034] In one possible implementation, combining Figure 2 Based on the topological connectivity and association strength between the functional sub-regions, a dynamic functional connectivity network is constructed to characterize the microcirculation structural state. By comparing the topological differences of this dynamic functional connectivity network before and after drug administration, microcirculation functional evolution parameters describing the response intensity of the microcirculation system are quantified, including: Calculate the cross-correlation coefficient of the average flow velocity signal in each of the functional sub-regions, and use the cross-correlation coefficient as the edge weight to connect adjacent sub-regions to construct a dynamic functional connection network; The reduction in the overall network connection weights in the dynamic functional connection network after drug administration was statistically analyzed, and the overall weight decay rate describing the intensity of macroscopic blood flow reduction was calculated. Key nodes in the dynamic functional connection network whose centrality exceeds a preset ratio are identified, and the number of failed connection edges of the key nodes is counted to obtain the connection interruption ratio reflecting local structural damage. The overall weight decay rate is combined with the connection interruption ratio to generate micro-circulation functional evolution parameters.

[0035] In some implementations, the cross-correlation coefficients of the average flow velocity signals within each functional sub-region are calculated over an observation period, typically covering 120 to 300 image frames covering the contrast agent's entry, peak, and decay. The cross-correlation coefficients are calculated using the following formula: ; in, For correlation scalar, The covariance of the velocity time series of the two sub-regions. This is the product of the standard deviations of the two series. The calculated cross-correlation coefficient is then used. As edge weights connecting adjacent spatial sub-regions, only connections with correlations greater than a preset threshold (e.g., 0.65 to 0.8) are retained, constructing a dynamic functional connectivity network characterizing the microcirculatory structural state. This process transforms the originally isolated hemodynamic features into a complex network topology with spatial correlations. The reduction in the overall network connection weights before and after drug administration is statistically analyzed. The overall weight decay rate is calculated. ; Where A is the attenuation rate describing the intensity of macroscopic blood flow reduction. This represents the sum of the edge weights of the dynamic functional connectivity network during the baseline period. This represents the sum of all network edge weights after the first dose. (Calculated...) and The ratio of the difference to the baseline value can quantitatively identify the decrease in global blood flow caused by drug-induced changes in vascular permeability and microthrombus formation, outputting an attenuation index representing the macroscopic response. The eigenvector centrality of each node in the dynamic functional connectivity network is calculated, and nodes ranking in the top 15% to 25% of centrality values ​​are identified as critical nodes. Critical nodes refer to sub-regions in the microcirculation network that undertake the main flow distribution function. The connection interruption ratio is calculated using the formula: ; in, This represents a dimensionless quantity that reflects local structural damage. This represents the total number of critical nodes that fail after drug administration. This represents the initial total number of edges possessed by the critical node at the baseline. Micro-loop functional evolution parameters are generated by performing a weighted fusion operation: ; in, These are the final evolution parameters, used to quantify the microcirculation function evolution parameters. To transform the boundary conditions into physical models, this invention establishes a set of mapping functions. Traffic source item mapping: [This will be done / implemented] Mapped to the initial mass flux load of the grid cells. Using the formula The calculation yielded, where This is the flow conversion factor determined by the baseline average flow velocity. Pressure gradient correction: effective pressure gradient. by Nonlinear correction, i.e. ,in This is a vascular permeability attenuation factor, used to characterize the decrease in effective transport driving force due to microcirculatory topological disruption. Through the above mapping, the evolutionary characteristic values ​​reflecting tissue activity are transformed into momentum and mass source terms in the fluid-structure interaction transport model, thereby achieving quantitative estimation from kinetic characteristics to physical concentration distribution. and These are empirical weighting constants determined by the pathological type, typically The value ranges from 0.4 to 0.6. This parameter is determined by a combination of macroscopic weight decay and local topology disruption. For example... Figure 3 As shown, the evolution of acoustic intensity over time in sensitive functional subregions and low-response-risk regions within the tumor during contrast imaging demonstrates the characteristics of slow perfusion and delayed resolution in drug-resistant regions due to increased interstitial pressure.

[0036] For example, the cross-correlation coefficient of the average flow velocity signals between sub-regions is calculated, assuming the covariance of the flow velocity time series of two adjacent sub-regions is... The product of standard deviations is 0.12. The value is 0.15, which is then substituted into the formula. Calculate the correlation scalar It is 0.8. Because... The weight is greater than the preset threshold of 0.7, so it is used as the edge weight to connect these two sub-regions, constructing a dynamic functional connectivity network. For the assessment of blood supply capacity after the first dose of chemotherapy, the total edge weight of the entire network at the baseline is set. The weighted total after drug administration is 150.0. The value is 97.5. Substitute it into the formula. The overall weight decay rate was calculated. The value is 0.35. To identify local structural collapses, the top 15% of key nodes in the centrality ranking are identified. The total number of associated edges of a node that fail after drug administration is considered. The initial total number of edges during the baseline period is 6. Substitute 10 into the formula Calculated connection interruption rate The value is 0.6. A weighted fusion operation is performed, with the empirical weight constant determined by the pathological type. It is 0.4. The value is 0.6, so we substitute it into the formula. Calculation of microcirculation functional evolution parameters .

[0037] In one possible implementation, combining Figure 2 The step involves obtaining tissue biophysical mapping parameters that reflect tissue mechanical properties, using the microcirculation functional evolution parameters and the tissue biophysical mapping parameters as boundary constraints input into the fluid-structure interaction transport model, and calculating the spatial concentration of chemotherapy drugs in the interstitial space by performing an inverse solution operation. The permeability reference value and interstitial pressure range corresponding to the current subject subtype are matched from the preset tumor database to generate tissue biophysical mapping parameters; The functional subregion is simplified into grid cells with uniform medium properties, and the microcirculation function evolution parameters are extracted as the initial flux load of the grid cells. The flow resistance matrix between grid cells is constructed using the tissue biophysical mapping parameters, and a fluid-structure interaction transport model describing the delivery of drug solution from the vascular system to the interstitial space is established. Using the pressure difference between each grid cell as the driving force, the finite element method is applied to solve the fluid-structure interaction transport model by mass conservation iteration to obtain the drug accumulation amount of each grid cell within a preset drug administration time. After performing voxelization smoothing, the predicted distribution concentration map is output.

[0038] In some implementations, tissue biophysical mapping parameters are generated by matching permeability reference values ​​and interstitial pressure ranges corresponding to the current subject's subtype from a pre-defined tumor database. For different molecular subtypes such as Luminal A, Luminal B, HER2-positive, or triple-negative, the corresponding matrix permeability coefficients are retrieved. and initial interstitial fluid pressure . This represents the physical ability of biological tissue to allow fluid to pass through, and its value is typically in the range of... arrive The parameters range from square meters per pascal-second. The mechanical boundary between tumor parenchyma and stroma is defined using prior parameters, providing a spatial resistance background for drug molecule diffusion simulations. Functional subregions are simplified into grid cells with homogeneous medium properties; each grid cell represents an independently calculated finite element volume, with side lengths typically set between 0.5 and 1 millimeter. Microcirculation functional evolution parameters are extracted as the initial flux load for the grid cells, characterizing the source term intensity of drug release from microvessels into the interstitial space. Simultaneously, the obtained tissue biophysical mapping parameters are used to construct the flow resistance matrix between grid cells. Flow resistance matrix This describes the viscous forces and path tortuosity experienced by the drug solution as it permeates between grid cells. This process establishes a mathematical connection between macroscopic hemodynamic evolution and microscopic interstitial transport behavior through a fluid-structure interaction (FSI) transport model. Driven by the pressure difference between grid cells, the FSI model is solved iteratively using the finite element method to maintain mass conservation. Single-step concentration update is then performed. ; in, This represents the amount of drug accumulated in a grid cell after a preset dosing time. This represents the initial drug concentration determined by clinical prescription. Represents the grid permeability coefficient determined by tissue biophysical mapping parameters. This represents the effective pressure gradient after correction by microcirculation functional evolution parameters. (The text then abruptly shifts to a seemingly unrelated topic: establishing microcirculation functional evolution parameters.) The mapping relationship with the fluid dynamics equations of the interstitial space is obtained; using The changes in microvascular permeability were characterized, and the pressure difference distribution between grid cells was derived by reverse derivation using Darcy's law. This represents the duration of the simulated drug delivery. The effective permeable surface area of ​​the grid cell is defined. Through multi-step iterations, the total drug accumulation in each grid cell within a preset dosing duration is obtained. Voxelization smoothing is performed on the discrete grid data, and three-dimensional Gaussian filtering is used to eliminate numerical artifacts caused by grid division, outputting a predicted distribution concentration map with spatial continuity.

[0039] For example, the tissue biophysical mapping parameters corresponding to the subtype are matched from a preset tumor database, and its matrix permeability coefficient is set. for The tumor region is divided into sections with sides of length [missing information]. The homogeneous medium lattice unit was used, and the microcirculation functional evolution parameters obtained from previous calculations were extracted. The initial flux load of the grid cell. The effective permeable surface area of ​​the grid cell. Based on the grid point side length, set as ,Right now Driven by the pressure difference between grid points, the finite element method is used to perform a mass conservation iterative solution. Let the initial drug concentration determined by the clinical prescription be... for ,Right now Effective pressure gradient after correction of microcirculation function evolution parameters for Simulated drug delivery duration for Substitute into the calculation formula Perform a single-step update; the calculation process is as follows: The amount of drug accumulated at this grid point after a preset time period was calculated. .

[0040] In one possible implementation, combining Figure 2 By performing spatial mapping and feature overlay on the predicted distribution concentration map and the microcirculation function evolution parameters, spatial clusters that simultaneously possess low concentration distribution and low response activity are identified, including: A drug effective action threshold is set, and a set of voxels with concentration values ​​lower than the effective action threshold is extracted from the predicted distribution concentration map and marked as a low drug distribution feature region mask. The blood flow fluctuation amplitude of local sub-regions is extracted from the microcirculation function evolution parameters. By comparing it with the preset system stability range, tissues whose function has not significantly deteriorated are identified and marked as inert function masks. Perform a logical intersection operation between the low drug distribution feature region mask and the inert function mask to locate the spatial coordinate range where the two overlap, and generate a low response risk feature region.

[0041] In some implementations, based on the specific chemotherapy regimen used, the median effective concentration (MIC) of the drug for killing breast tumor cells is extracted from a pharmacokinetic database. This concentration is set as the effective drug action threshold, and a traversal search is performed on the predicted concentration distribution map. Mask extraction is then performed. ; in, To create a mask for the low drug distribution feature region, It is a matrix of all ones containing the concentration of the entire spatial domain. As an indicator matrix, when the concentration value of the corresponding voxel in the predicted distribution concentration map is greater than or equal to the effective action threshold, The corresponding element is set to 1, otherwise it is set to 0. The continuous concentration distribution is transformed into a binary spatial geometric constraint, accurately locating the physical regions where drugs cannot adequately cover the tissue due to tumor vascular bed occlusion or excessive interstitial pressure. Blood flow fluctuation amplitudes of each local sub-region are extracted from microcirculation functional evolution parameters to describe the tissue's sensitivity to functional changes before and after drug administration. The extracted fluctuation amplitudes are compared with a preset stable range, where the stable range represents the background perturbation deviation of non-responsive tissues under physiological conditions. The functional degradation rate is calculated. ; in, For functional degradation scalar, The standard deviation of the mean flow rate time series in a local sub-region after the first dose is given. This represents the corresponding standard deviation for this region during the baseline period. If If the value is within a stable range of 0.85 to 1.15, the region is considered to have no significant pharmacological decline in circulating activity and is marked as an inert functional mask. The mask indicates refractory tissue clusters that maintain their original hemodynamic characteristics after chemotherapy. The logical intersection operation between the low drug distribution feature region mask and the inert functional mask is performed in a unified anatomical coordinate system. Spatial mapping and feature overlay are then performed. ; in, This represents the final low-response-risk feature region mask. The binary logic state representing the corresponding voxel in the mask of the low drug distribution feature region. This represents the binary logic state of the corresponding voxel in the inert functional mask. Through this logic operation, false positive regions that, despite insufficient drug coverage, have already undergone tissue degeneration and necrosis are excluded, as are hysteretic regions where drug coverage is sufficient but activity has not yet decreased in the short term. The resulting low-response-risk feature region represents the set of spatial coordinates within the tumor that simultaneously possess low drug coverage and low blood flow response intensity.

[0042] For example, the half-maximal effective concentration threshold corresponding to a specific chemotherapy regimen is set as follows: Traverse the predicted distribution concentration map and perform mask extraction on voxels with concentration values ​​below a certain threshold. Let the concentration of a certain voxel be... Through formula Mark it as a mask for low drug distribution feature regions. The region with a logic state of 1. Local sub-region blood flow fluctuation amplitudes are extracted to assess functional sensitivity, assuming the standard deviation of the mean flow velocity after the first dose. The baseline standard deviation is 4.1. The value is 4.2, so we substitute it into the formula. The functional degradation rate was calculated. Since this value falls within the preset stable range of 0.85 to 1.15, it is classified as refractory tissue with no significant functional degradation and marked as an inert functional mask with a logical state of 1. This is then analyzed using the formula within a unified anatomical coordinate system. Perform a logical intersection operation to generate a low-response-risk feature region mask with a logical value of 1 for spatial coordinates that simultaneously satisfy the conditions of insufficient drug coverage and stable cyclic activity. .

[0043] In one possible implementation, combining Figure 2 The extracted features are input into a preset heterogeneity assessment model for feature fusion. After the heterogeneity assessment model adaptively adjusts the weights based on the subjects' clinical phenotype data, it outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness, including: The heterogeneity assessment model calculates the feature vector centrality of the low-response risk feature region in the dynamic functional connectivity network, and obtains the risk region topological weight as a spatial topological feature. The heterogeneity assessment model calls a preset weight allocation function based on the molecular subtype categories in the clinical phenotype data, performs a weight boosting operation for the preset refractory subtype, increases the proportion of the risk region topological weight in the fusion calculation, and generates an adaptive decision operator. The heterogeneity assessment model applies the adaptive decision operator to perform a weighted aggregation operation on the topological weights of the risk region, the average drug coverage in the predicted distribution concentration map, and the microcirculation function evolution parameters to obtain a value representing the tendency of tissue response intensity, and outputs the value as a quantitative reference score.

[0044] In some implementations, the eigenvector centrality of low-response-risk regions in a dynamically connected network is calculated to obtain the topological weights of these risk regions as spatial topological features. Eigenvector centrality is a metric used to evaluate the importance of nodes in complex networks; a node has a higher centrality score when it is connected to multiple nodes with similarly high influence. Perform matrix operations: ; in, For the generated risk region topology weights, To describe the adjacency matrix of a dynamic functional connectivity network, This is the network principal feature vector obtained through power iteration. Static spatial distribution information is transformed into a dynamic network performance description, identifying highly stable drug-resistant regions located in key microcirculatory pathways. The heterogeneity assessment model calls a pre-defined weight allocation function based on the acquired clinical phenotypic data. Clinical phenotypic data includes tumor marker concentrations in serological indicators and molecular subtype classifications in biopsy histological features, such as triple-negative breast cancer or HER-2 positivity. For subjects with pre-defined refractory subtypes or high marker concentrations, an enhancement operation is performed on the topological weights of risk regions to generate an adaptive decision operator. The value of the decision operator is determined by the following formula: ; in, For adaptive decision operators, The gain coefficient is assigned based on the molecular subtype category, and typically ranges from 1.2 to 1.5. This is a dimensionless correction scalar obtained by mapping serological indicators. This operator is used to increase the weight of refractory features in the model decision-making process, enhancing the sensitivity of identifying high-risk pathological features. The heterogeneity assessment model applies the generated adaptive decision operator to perform weighted aggregation operations on the extracted features of each dimension. The efficacy prediction assessment results are calculated using the following formula: ; in, This is a numerical value representing the probability of complete pathological remission, ranging from 0 to 1. For the generated adaptive decision operator, For risk region topological weights, The average drug coverage rate is calculated from the predicted distribution concentration map. The microcirculation function evolution parameter represents the overall response intensity. The training process ensures that the contribution rate coefficients of the pre-defined features are always equal to 1. The calculation formula guarantees that the evaluation results simultaneously consider physical drug dosage attainment, changes in microcirculation function, and the topological stubbornness of drug-resistant lesions, and translates the numerical values... The output is a quantitative prediction of the efficacy of neoadjuvant chemotherapy.

[0045] For example, the heterogeneity assessment model first calculates the eigenvector centrality of low-response-risk feature regions in a dynamic functional connectivity network. Let the adjacency matrix describing this network be... The network principal feature vector obtained by power iteration method Perform matrix operations using formulas Calculate the topological weights of the risk region The value is 0.72. Subsequently, the model incorporates the individual molecular biological background of the subjects, calling a weighted allocation function based on their triple-negative breast cancer subtype and high tumor marker concentrations. Let the allocation gain coefficient be... The value is 1.35, a correction scalar obtained by mapping serological indicators. The value is 0.10, which is then substituted into the formula. Determine the adaptive decision operator The value is 1.45. Finally, the model applies this operator to perform a weighted aggregation operation on the extracted features of each dimension, assuming the average drug coverage calculated from the predicted distribution concentration map is... The value is 0.58, representing the microcirculation functional evolution parameter of the overall response intensity. The preset feature contribution rate coefficient is 0.51. The values ​​are 0.4, 0.3, and 0.3 respectively. Substitute them into the formula. The efficacy prediction assessment results were calculated. This value indicates a high probability of complete pathological remission.

[0046] In one possible implementation, combining Figure 2 The method further includes: Using low-response-risk characteristic regions as simulation source points, we retrieved associated paths in the dynamic functional connectivity network, simulated the diffusion trajectory of drug resistance characteristics using a random walk algorithm, and calculated the risk evolution probability. The risk evolution probability is mapped to the original ultrasound spatial coordinate system, and the probability value is colored using a color coding function to generate a response risk evolution heatmap. The response risk evolution heatmap is overlaid with the baseline ultrasonic structure map with transparency to output a spatial risk map indicating the distribution of response anomaly regions.

[0047] In some implementations, a low-response-risk region is first identified as a simulated source point, and all physical paths associated with that source point in the dynamic functional connectivity network are retrieved. A random walk algorithm is used to simulate the diffusion process of drug resistance features in the network topology. This simulation is based on the biophysical assumption that drug-resistant cancer cells or resistance factors tend to spatially infiltrate along low-resistance blood flow perfusion pathways within the tumor; the transition probability matrix is ​​then used to simulate this diffusion. Quantify this cascading risk along the microcirculation network. Calculate the probability of risk evolution using the formula: ; in, Let be a vector representing the probability of each sub-region being affected by drug resistance features. This is the transition probability matrix of the dynamic functional connectivity network, which is obtained by normalizing the edge weights between nodes. To simulate the walking step size for diffusion, a value typically ranges from 10 to 50 steps. This is a binary source vector characterizing the initial distribution state of low-response risk regions. It quantifies spatial locations that, while not yet exhibiting drug resistance, possess high translocation risk due to their downstream location in the microcirculation or strong correlation nodes. The calculated risk evolution probabilities are mapped back to the original ultrasound spatial coordinate system. A preset color coding function is used to color the probability values, and color mapping is performed using the following formula: ; in, For the output voxel color vector, This is a pre-defined pseudo-color lookup table. This represents the risk evolution probability corresponding to each voxel. The lookup table maps high-probability intervals to high-alert colors like dark red and low-probability intervals to cool colors like dark blue, thus generating a response risk evolution heatmap. This process visualizes the complex network dynamics calculations, clearly showing the spatial trajectory of drug resistance characteristics that may spread through the blood perfusion pathway. The response risk evolution heatmap is then overlaid with the baseline ultrasound structural image using transparency processing. Voxel fusion calculations are performed using the following formula: ; in, The pixel value corresponding to the coordinates in the generated spatial risk map. In response to the color components of the corresponding coordinates in the risk evolution heatmap, This represents the brightness components of anatomical structures in the baseline ultrasound structural map. This is the weighting coefficient for the transparency of the heatmap. For the background weight coefficients of the structure diagram, and and The sum of all equals 1, usually The value ranges from 0.3 to 0.6. This spatial overlay display technology allows clinicians to accurately identify potential drug resistance risk areas within lesions while observing their anatomical contours. The final output is a spatial risk map indicating the distribution of abnormal response areas. Figure 4 As shown, the thermal distribution visually presents the risk probability of low response characteristics evolving from the core risk area along the microcirculation network to the surrounding tissues. The dark areas indicate the spatial locations of potential relapse and drug resistance high-risk areas that require clinical attention.

[0048] For example, the identified low-response risk characteristic regions are used as simulation source points, and all physical paths associated with them in the Dynamic Functional Link Network (DFLNet) are retrieved. In the risk evolution probability calculation phase, the transition probability matrix of the DFLNet is set as follows: The transition probability characteristic value of a certain local region is 0.9, representing the binary source vector characterizing the initial distribution state of the low response risk characteristic region. Let the vector consist of 1s and 0s. Set the walk step size for the simulated diffusion. There are 30 steps; substitute them into the formula. Calculations are performed. If the calculation is simplified using this local eigenvalue, the probability of risk evolution is... This vector quantifies the likelihood that each sub-region downstream of the microcirculation will be affected by drug resistance characteristics. It represents the probability of risk evolution. Map back to the original ultrasonic spatial coordinate system and call the color encoding function to perform color mapping. Assume the preset pseudo-color lookup table is... This lookup table maps probability intervals to color components, and if a certain voxel corresponds to a risk evolution probability... The value is 0.85, which is then substituted into the formula. Perform the calculation. If If the color vector weight in this interval is 255, then the corresponding voxel color vector is calculated. for The corresponding color is dark red, generating a heatmap that visually displays the evolution of response risk and the trajectory of drug resistance spread. This heatmap is then overlaid with the baseline ultrasound structure image using transparency overlay. A transparency weighting coefficient is set for the heatmap. The background weight coefficient of the structure diagram is 0.5. The sum of the two is always equal to 1, which is 0.5. If the color component of the corresponding coordinate in the response risk evolution heatmap... The value is 216.75 for deep red, representing the brightness component of the baseline structure plot. Given a grayscale value of 120, substitute it into the formula. Perform voxel fusion operation to calculate the fused pixel values. Based on this, a spatial risk map is output.

[0049] In one possible implementation, combining Figure 2The adaptive decision generator includes: The serological indicators and biopsy histological characteristics of the subjects were obtained to identify the corresponding chemotherapy-sensitive subtypes or low-response tendency groups and generate subject classification feature vectors. For subjects categorized into specific response characteristic groups, the weighting coefficient of global indicators in the microcirculation function evolution parameters is reduced, while nonlinear enhancement processing is performed on the characteristics of low response risk regions to optimize decision sensitivity and generate adaptive decision operators that adapt to the differences between different subtypes.

[0050] In some implementations, serological indicators and biopsy histological characteristics of the subjects are retrieved from the electronic medical record system. Serological indicators typically include the concentrations of tumor markers such as carcinoembryonic antigen (CEA) or carbohydrate antigens, while biopsy histological characteristics cover molecular subtype-related data such as estrogen receptor, progesterone receptor, and cell proliferation index. A subject classification feature vector is generated using a calculation formula. ; in, Represents the overall pathology score. Represents serological marker measurements. This represents the clinical threshold for this indicator. This represents the proportion of key proteins expressed in the biopsy tissue. This represents the criterion value for that proportion. (Through...) The range determination of values ​​can identify the corresponding chemotherapy-sensitive subtypes or low-response tendency groups, and define them as subject classification feature vectors. For subjects labeled with specific response feature groups, nonlinear enhancement processing is performed on the characteristics of low-response risk feature regions. Nonlinear enhancement processing refers to using power functions or exponential functions to numerically stretch the topological weights of specific regions to highlight their negative contribution to the overall prognosis. The enhancement operation is performed using the calculation formula: ; in, To enhance the scalar value of drug resistance characteristics, To extract the original spatial topological feature values ​​from low-response-risk feature regions, As an enhancing factor, it is determined by the subject classification feature vector and is typically found in drug-resistant subtypes. The value ranges from 1.5 to 2.5. This improves the sensitivity of the decision-making system to small drug-resistant lesions. Weight allocation logic is executed based on the subject classification feature vector, reducing the weighting coefficient of global indicators in the microcirculation function evolution parameters while increasing the contribution of local drug resistance features. This optimizes decision sensitivity and generates an adaptive decision operator. ; in, The final output is the adaptive decision operator. For the global components of the microcirculation function evolution parameters, This is a scalar value representing the enhanced drug resistance characteristics. and These are complementary weighting coefficients, whose values ​​are dynamically adjusted based on the subject's classification feature vector. In the low-response propensity group, they are automatically weighted... Reduced to between 0.2 and 0.4, and The efficiency is increased to between 0.6 and 0.8. This operator is directly applied to the final feature fusion layer, enabling the model to generate customized decision-weighted outputs that adapt to the differences between different subtypes.

[0051] For example, retrieve the subject's pathological label data and set its serological marker measurements. for Clinical threshold for The proportion of key protein expression in biopsy tissue The standard value is 0.85. The value is 0.30. Substitute it into the formula. Calculate the overall pathology score Due to its high score, it was classified as a low-response tendency group and a corresponding pathological label was generated. Nonlinear enhancement processing was performed on this drug resistance tendency feature, assuming the original spatial topological feature value extracted from the low-response risk feature region... The enhancement factor is 0.65. The value is 2.0. Substitute it into the formula. The enhanced drug resistance scalar was calculated. Weights are assigned based on drug resistance tendency labels, and the global components of microcirculation function evolution parameters are then integrated. Weighting coefficients Reduced to 0.3, the weighted coefficient of the contribution of drug resistance characteristics Increased to 0.7. Let the global score be... The value is 0.45, which is then substituted into the formula. The final adaptive decision operator is calculated. .

[0052] It should be noted that all equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for predicting the efficacy of neoadjuvant chemotherapy in breast cancer based on ultrasound contrast imaging, characterized in that, The method includes: Multi-temporal ultrasound images of subjects before and after neoadjuvant chemotherapy were acquired, and three-dimensional image registration and motion compensation were performed to generate spatiotemporally aligned contrast imaging sequences with voxel-level spatial alignment features. The movement and displacement of contrast agent microbubbles are tracked from the spatiotemporally aligned contrast sequence, the microbubble movement trajectory reflecting hemodynamic characteristics is extracted, and functional sub-regions with physiological differences are divided by unsupervised clustering of the microbubble movement trajectory. Based on the topological connectivity and association strength between the functional sub-regions, a dynamic functional connectivity network is constructed to characterize the state of the microcirculation structure. By comparing the topological differences of the dynamic functional connectivity network before and after drug administration, microcirculation functional evolution parameters describing the response intensity of the microcirculation system are quantified and generated. Obtain tissue biophysical mapping parameters that reflect tissue mechanical properties, input the microcirculation function evolution parameters and tissue biophysical mapping parameters as boundary constraints into the fluid-structure interaction transport model, and calculate the spatial concentration of chemotherapy drugs in the interstitial space by performing inverse solution operations to generate a predicted distribution concentration map. The predicted distribution concentration map is spatially mapped and feature overlaid with the microcirculation function evolution parameters to identify spatial clusters that simultaneously possess low concentration distribution and low response activity, thereby locating and generating low response risk feature regions. The spatial topological features of the low-response-risk region, the drug coverage index of the predicted distribution concentration map, and the microcirculation function evolution parameters are extracted. The extracted features are input into a preset heterogeneity assessment model for feature fusion. The heterogeneity assessment model performs weight adaptive adjustment based on the clinical phenotype data of the subjects and outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotype data includes serological indicators and biopsy histological features.

2. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, The acquisition of multi-temporal ultrasound images of the subject before and after neoadjuvant chemotherapy administration, and the performance of three-dimensional image registration and motion compensation processing, includes: Dynamic contrast-enhanced videos were acquired at the baseline period and after the first dose. A linearization transformation algorithm was used to map pixel brightness to acoustic intensity values ​​that reflect microbubble concentration, generating a standardized grayscale matrix. Feature points of anatomical structures are extracted from the standardized grayscale matrix. A non-rigid deformation field model is constructed by calculating the displacement vectors of feature points between adjacent frames. The non-rigid deformation field model is then applied to perform a geometric transformation on the grayscale matrix to eliminate anatomical displacement caused by the respiratory cycle, resulting in a motion-compensated image sequence. The mutual information measure between the baseline and post-drug administration image sequences is calculated. The coordinate transformation parameters are adjusted according to the principle of maximizing the measure to achieve voxel matching of multi-temporal images and output spatiotemporally aligned imaging sequences.

3. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, The movement and displacement of contrast agent microbubbles are tracked from the spatiotemporally aligned contrast imaging sequence. Microbubble trajectories reflecting hemodynamic characteristics are extracted, and unsupervised clustering of the microbubble trajectories is performed, including: Within the search neighborhood of the spatiotemporally aligned imaging sequence, the centroid of the microbubble is locked, the spatial coordinates of the microbubble in consecutive time frames are recorded, and the initial microbubble displacement vector is generated. The initial microbubble displacement vector is subjected to time smoothing to eliminate observation noise, and the rate of change of displacement on the time axis is calculated to extract the microbubble motion trajectory containing flow velocity and motion direction. The geometric distance of each microbubble motion trajectory in the feature space is measured, and the trajectory clusters with feature distances within a preset similarity threshold range are aggregated into a unified spatial entity, and functional sub-regions are divided.

4. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, Based on the topological connectivity and association strength among the functional sub-regions, a dynamic functional connectivity network is constructed to characterize the state of the microcirculation structure. By comparing the topological differences of the dynamic functional connectivity network before and after drug administration, microcirculation functional evolution parameters describing the response intensity of the microcirculation system are quantified, including: Calculate the cross-correlation coefficient of the average flow velocity signal in each of the functional sub-regions, and use the cross-correlation coefficient as the edge weight to connect adjacent sub-regions to construct a dynamic functional connection network; The reduction in the overall network connection weights in the dynamic functional connection network after drug administration was statistically analyzed, and the overall weight decay rate describing the intensity of macroscopic blood flow reduction was calculated. Key nodes in the dynamic functional connection network whose centrality exceeds a preset ratio are identified, and the number of failed connection edges of the key nodes is counted to obtain the connection interruption ratio reflecting local structural damage. The overall weight decay rate is combined with the connection interruption ratio to generate micro-circulation functional evolution parameters.

5. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, The process involves obtaining tissue biophysical mapping parameters that reflect tissue mechanical properties, inputting the microcirculation functional evolution parameters and the tissue biophysical mapping parameters as boundary constraints into the fluid-structure interaction transport model, and calculating the spatial concentration of chemotherapy drugs in the interstitial space by performing an inverse solution operation. The permeability reference value and interstitial pressure range corresponding to the current subject subtype are matched from the preset tumor database to generate tissue biophysical mapping parameters; The functional subregion is simplified into grid cells with uniform medium properties, and the microcirculation function evolution parameters are extracted as the initial flux load of the grid cells. The flow resistance matrix between grid cells is constructed using the tissue biophysical mapping parameters, and a fluid-structure interaction transport model describing the delivery of drug solution from the vascular system to the interstitial space is established. Using the pressure difference between each grid cell as the driving force, the finite element method is applied to solve the fluid-structure interaction transport model by mass conservation iteration to obtain the drug accumulation amount of each grid cell within a preset drug administration time. After performing voxelization smoothing, the predicted distribution concentration map is output.

6. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, By performing spatial mapping and feature overlay on the predicted concentration distribution map and the microcirculation function evolution parameters, spatial clusters that simultaneously possess low concentration distribution and low response activity are identified, including: A drug effective action threshold is set, and a set of voxels with concentration values ​​lower than the effective action threshold is extracted from the predicted distribution concentration map and marked as a low drug distribution feature region mask. The blood flow fluctuation amplitude of local sub-regions is extracted from the microcirculation function evolution parameters. By comparing it with the preset system stability range, tissues whose function has not significantly deteriorated are identified and marked as inert function masks. Perform a logical intersection operation between the low drug distribution feature region mask and the inert function mask to locate the spatial coordinate range where the two overlap, and generate a low response risk feature region.

7. The method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 1, characterized in that, The extracted features are input into a preset heterogeneity assessment model for feature fusion. After the heterogeneity assessment model performs weight adaptive adjustment based on the subject's clinical phenotype data, it outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness, including: The heterogeneity assessment model calculates the feature vector centrality of the low-response risk feature region in the dynamic functional connectivity network, and obtains the risk region topological weight as a spatial topological feature. The heterogeneity assessment model calls a preset weight allocation function based on the molecular subtype categories in the clinical phenotype data, performs a weight boosting operation for the preset refractory subtype, increases the proportion of the risk region topological weight in the fusion calculation, and generates an adaptive decision operator. The heterogeneity assessment model applies the adaptive decision operator to perform a weighted aggregation operation on the topological weights of the risk region, the average drug coverage in the predicted distribution concentration map, and the microcirculation function evolution parameters to obtain a value representing the tendency of tissue response intensity, and outputs the value as a quantitative reference score.

8. A method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 7, characterized in that, The method further includes: Using the low-response-risk characteristic region as the simulation source point, the associated paths in the dynamic functional connection network are retrieved, and the diffusion trajectory of drug resistance characteristics is simulated using a random walk algorithm to calculate the risk evolution probability. The risk evolution probability is mapped to the original ultrasound spatial coordinate system, and the probability value is colored using a color coding function to generate a response risk evolution heatmap. The response risk evolution heatmap is overlaid with the baseline ultrasonic structure map with transparency to output a spatial risk map indicating the distribution of response anomalies.

9. A method for predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging according to claim 7, characterized in that, The adaptive decision generator includes: The serological indicators and biopsy histological characteristics of the subjects were obtained to identify the corresponding chemotherapy-sensitive subtypes or low-response tendency groups and generate subject classification feature vectors. For subjects categorized as having specific response characteristics, the weighting coefficient of the global index in the microcirculation function evolution parameters is reduced, while nonlinear enhancement processing is performed on the characteristics of the low response risk feature region to optimize decision sensitivity and generate an adaptive decision operator that adapts to the differences between different subtypes.

10. A system for predicting the efficacy of neoadjuvant chemotherapy in breast cancer based on ultrasound contrast imaging, characterized in that, The system is used in any one of claims 1-9 for a method of predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on ultrasound contrast imaging, the system comprising: The image registration and compensation module is used to acquire multi-temporal ultrasound images of subjects before and after neoadjuvant chemotherapy, and to perform three-dimensional image registration and motion compensation processing to generate a spatiotemporally aligned contrast imaging sequence with voxel-level spatial alignment features. The dynamic feature extraction module is used to track the motion displacement of contrast agent microbubbles from the spatiotemporally aligned contrast sequence, extract the microbubble motion trajectory reflecting hemodynamic characteristics, and divide the microbubble motion trajectory into functional sub-regions with physiological differences by performing unsupervised clustering processing on the microbubble motion trajectory. The microcirculation evolution modeling module is used to construct a dynamic functional connection network to characterize the state of the microcirculation structure based on the topological connection relationship and correlation strength between the functional sub-regions, and to quantify and generate microcirculation functional evolution parameters describing the response intensity of the microcirculation system by comparing the topological differences of the dynamic functional connection network before and after drug administration. The drug transport simulation module is used to obtain tissue biophysical mapping parameters that reflect tissue mechanical properties. The microcirculation function evolution parameters and the tissue biophysical mapping parameters are used as boundary constraints to input the fluid-structure interaction transport model. By performing the inverse solution operation, the spatial concentration of chemotherapy drugs in the interstitial space is calculated, and a predicted distribution concentration map is generated. The risk feature region localization module is used to perform spatial mapping and feature overlay between the predicted distribution concentration map and the microcirculation function evolution parameters, identify spatial clusters that simultaneously have low concentration distribution and low response activity, and locate and generate low response risk feature regions. The response heterogeneity assessment module is used to extract the spatial topological features of the low response risk characteristic region, the drug coverage index of the predicted distribution concentration map, and the microcirculation function evolution parameters. The extracted features are input into a preset heterogeneity assessment model for feature fusion. The heterogeneity assessment model performs weight adaptive adjustment based on the clinical phenotype data of the subjects and outputs a quantitative reference score for assisting in the assessment of chemotherapy responsiveness. The clinical phenotype data includes serological indicators and biopsy histological features.

Images