Tumor mechanics multi-parameter prognosis prediction system and method based on elastography

Abstract

The invention provides a tumor mechanics multi-parameter prognosis prediction system and method based on elastography, and relates to the field of medical imaging, oncology and biomechanics, and the system comprises an elastography acquisition module which is used for collecting elastography data of a tumor of a patient; the image preprocessing module is used for preprocessing the elasticity imaging data; the parameter extraction module is used for extracting multi-dimensional tumor mechanical parameters and carrying out feature normalization processing; the feature selection module is used for screening key prognosis related parameters; the prognosis prediction model construction module is used for constructing a prognosis prediction model based on the screened key prognosis related parameters; and the result display module is used for displaying the tumor mechanics multi-parameter prognosis prediction result in a visual mode by applying the prognosis prediction model. The problem that a traditional tumor prediction method is insufficient in accuracy is solved.

Description

Technical Field

[0001] The present invention relates to the fields of medical imaging, oncology and biomechanics, and in particular to a tumor mechanics multi-parameter prognosis prediction system and method based on elastic imaging. Background Art

[0002] The mechanical properties of tumors are closely related to their biological behavior. Studies have shown that tumor tissue is generally stiffer than normal tissue, and this stiffness difference can be detected and quantified using elastography. Currently, elastography techniques (MRI or ultrasound) used clinically are primarily based on the following principle: external stimulation is used to induce small deformations in tissues, and then the tissue response is measured to calculate its mechanical properties, such as elastic modulus and shear modulus.

[0003] However, existing elastic imaging technology has the following obvious defects:

[0004] Low parameter dimensionality: Mainly relying on single parameters such as elastic modulus and shear modulus, it cannot fully characterize the mechanical heterogeneity within the tumor;

[0005] Weak prognostic correlation: These single parameters are insufficiently correlated with important clinical prognostic indicators such as patient survival rate and recurrence rate, making it difficult to provide accurate prognostic assessment;

[0006] Lack of spatial distribution information: Failure to fully utilize the spatial distribution information of the internal mechanical parameters of the tumor and ignoring the complexity of the internal mechanical environment of the tumor.

[0007] Therefore, there is an urgent need to develop a new method that can extract high-dimensional mechanical parameters from elastic imaging and construct an accurate prognostic model. Summary of the Invention

[0008] In response to the above-mentioned deficiencies in the prior art, the present invention provides a tumor mechanics multi-parameter prognosis prediction system and method based on elastic imaging, which solves the problem of insufficient accuracy of traditional tumor prediction methods. By integrating the multi-parameter characteristics of tumor mechanics measured by elastic imaging technology, a non-invasive and accurate tumor prognosis prediction system is established, which can more accurately evaluate the patient's prognosis.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a tumor mechanics multi-parameter prognosis prediction system based on elastography, comprising:

[0010] an elastic imaging acquisition module, used to collect elastic imaging data of a patient's tumor;

[0011] An image preprocessing module, used for preprocessing the collected elastic imaging data;

[0012] Parameter extraction module, used to extract multidimensional tumor mechanical parameters from preprocessed elastic imaging data and perform feature normalization processing;

[0013] Feature selection module, used to screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters;

[0014] A prognosis prediction model building module is used to build a prognosis prediction model based on the screened key prognosis-related parameters;

[0015] The result display module is used to apply the prognostic prediction model to visually display the tumor mechanics multi-parameter prognostic prediction results.

[0016] The present invention has the following beneficial effects: It can extract high-dimensional mechanical parameters (>1000) from tumor elastography data and construct an accurate prognostic prediction model. This method, for the first time, expands 1000+ dimensional features from basic elastic parameters, capturing the spatial heterogeneity of tumor mechanical properties. The C-index was validated in predicting 5-year survival in a breast cancer dataset, significantly outperforming methods that use only clinical parameters. This method can be used in clinical applications such as tumor prognosis assessment, treatment decision-making, recurrence risk monitoring, and treatment response prediction.

[0017] Furthermore, the parameter extraction module includes:

[0018] A first processing unit is used to extract basic mechanical parameters;

[0019] A second processing unit is used to calculate mechanical statistical parameters, gradient parameters, spatial distribution parameters, topological parameters, boundary parameters and strain energy parameters based on basic mechanical parameters;

[0020] The third processing unit is used to expand the dimension of mechanical parameters through multi-scale analysis, multi-threshold analysis and feature combination to obtain multi-scale mechanical parameters, mechanical feature combinations and high-order parameters;

[0021] The fourth processing unit is used to construct a high-dimensional mechanical characteristic matrix based on the multidimensional mechanical parameters extracted by the first processing unit, the second processing unit and the third processing unit, wherein the high-dimensional mechanical characteristic matrix is ​​used to group the mechanical parameters according to categories and normalize all the grouped mechanical parameters.

[0022] The beneficial effects of the above-mentioned further scheme are: systematic and hierarchical parameter extraction: through the hierarchical design of four processing units, orderly extraction from basic parameters to complex features is achieved, ensuring the integrity and traceability of the feature extraction process, enabling the system to capture the full range of mechanical characteristics of tumors from microscopic to macroscopic;

[0023] Exponential expansion of feature dimensions: Through multi-level parameter conversion and combination, limited basic mechanical parameters are expanded to a 1000+-dimensional feature space, greatly improving the system's ability to discern subtle differences in tumor mechanical properties, far exceeding the expressive power of traditional single parameters or low-dimensional feature spaces.

[0024] Optimized computational efficiency: Through modular design, the system avoids redundant computations when extracting high-dimensional features. Data flow between processing units is efficient and orderly, significantly improving the system's ability to process large-scale clinical data.

[0025] Adaptive processing of feature normalization: The fourth processing unit realizes adaptive normalization of different categories of mechanical parameters, solving the technical problem of difficulty in comprehensive analysis of parameters of different dimensions and dimensions, and providing standardized input features for model construction;

[0026] Enhanced model interpretability: Through the systematic classification and organization of mechanical parameters, the final prognostic prediction model can be traced back to its specific physical meaning, enhancing the model's clinical interpretability and helping doctors understand the biomechanical mechanisms behind the prediction results.

[0027] Adaptability and scalability: The modular structure design enables the system to flexibly adapt to different types of elastic imaging data input and can be optimized for different types of tumors by adjusting the parameter configuration of each processing unit, thereby improving the versatility and scalability of the system.

[0028] Furthermore, the basic mechanical parameters include:

[0029] Elastic modulus E:

[0030] E=σ / ε

[0031] Among them, σ represents stress and ε represents strain;

[0032] Shear modulus G:

[0033] G = τ / γ

[0034] Where τ represents shear stress and γ represents shear strain;

[0035] Poisson's ratio ν:

[0036] ν=-εtrans / εaxial

[0037] Among them, εtrans represents the transverse strain and εaxial represents the axial strain;

[0038] Young's modulus YM:

[0039] YM=2G(1+ν);

[0040] Bulk modulus K:

[0041] K = E / (3 (1-2ν));

[0042] Compression coefficient CF:

[0043] CF=1 / K.

[0044] The beneficial effects of the above-mentioned further scheme are: full-dimensional characterization of mechanical properties: by defining six complementary mechanical parameters: elastic modulus (E), shear modulus (G), Poisson's ratio (ν), Young's modulus (YM), bulk modulus (K), and compressibility coefficient (CF), the system can comprehensively characterize the response characteristics of tumor tissue under various deformation modes, achieving a complete mechanical description from tension and shear to volume changes;

[0045] Intrinsic conversion and redundancy elimination between parameters: There is a strict mathematical relationship between these parameters. The system can derive all other parameters from a small number of measured parameters (such as E and ν), which not only ensures the completeness of the description, but also avoids measurement redundancy and optimizes the data acquisition process.

[0046] Adaptability to different elastography techniques: Different elastography techniques directly measure different mechanical parameters (e.g., ultrasound elastography primarily measures E, while MR elastography measures G). This solution enables the system to be compatible with input data from multiple imaging techniques, expanding its clinical applicability.

[0047] Multi-angle characterization of mechanical heterogeneity: Basic parameters reflect the mechanical properties of tumor tissue from different perspectives, such as E reflecting overall stiffness, G reflecting shear resistance, ν and K reflecting volume change characteristics, enabling the system to capture multiple manifestations of tumor tissue microstructural heterogeneity;

[0048] In-depth analysis of biomechanical mechanisms: Different mechanical parameters correspond to different biological significance. For example, E is related to tumor extracellular matrix density, ν is related to tissue fluid content, and K is related to tissue interstitial pressure. This provides a multi-dimensional biomechanical mechanism explanation for clinical prognostic differences.

[0049] Implementation of parameter correlation analysis: Through these interrelated parameters, the system can further analyze the ratios and correlations between parameters, revealing the inherent laws and abnormal patterns of mechanical properties in the tumor microenvironment, and providing additional prognostic information;

[0050] The system can construct a complete mechanical portrait of the tumor based on the most basic mechanical properties, laying a solid physical foundation for subsequent high-dimensional feature extraction and prognosis prediction, and improving the scientificity and reliability of the system's prediction results.

[0051] Furthermore, the mechanical statistical parameters include basic mechanical statistical parameters and quantile mechanical parameters; the basic mechanical statistical parameters include:

[0052] The mean value μ of the elastic modulus E is:

[0053] μ=(1 / N)*Σ(Ep)

[0054] Where N represents the total number of pixels, and Ep represents the elastic modulus value of the p-th pixel;

[0055] Standard deviation σ of elastic modulus E:

[0056]

[0057] Coefficient of variation CV:

[0058] CV = σ / μ;

[0059] Peak threshold ratio PTR(t):

[0060] PTR(t)=Epeak / Et

[0061] Wherein, Epeak represents the peak value of the mechanical parameter, and Et represents the value of the mechanical parameter at the threshold value t;

[0062] The quantile mechanical parameters include:

[0063] Mechanical percentile MPn: the value of the nth percentile after sorting the elastic modulus values ​​from small to large, where MPn represents the elastic modulus E of the nth percentile;

[0064] Mechanical interquartile range MIQR:

[0065] MIQR=MP75-MP25

[0066] Among them, MP75 represents the 75th percentile elastic modulus, and MP25 represents the 25th percentile elastic modulus;

[0067] Elasticity Heterogeneity Index EHI:

[0068] EHI=(MP95-MP5) / MP50

[0069] Here, MP95 represents the 95th percentile elastic modulus, MP5 represents the 5th percentile elastic modulus, and MP50 represents the median elastic modulus.

[0070] The beneficial effects of the above further solution are: the technical effects brought about by defining mechanical statistical parameters include:

[0071] Multi-level quantification of tumor heterogeneity: By combining basic statistical parameters (μ, σ, CV) and quantile parameters (MPn, MIQR, EHI), the system can quantify the heterogeneity of tumor mechanical properties at multiple levels, including central tendency, dispersion, and distribution characteristics. This can more comprehensively capture the structural complexity of tumors than a single parameter.

[0072] Effective processing of extreme values ​​and outliers: Traditional mean-variance analysis is easily affected by extreme values, while quantile statistical methods (such as MP75 and MP25) are insensitive to outliers. By using these two types of parameters in a complementary manner, the system can capture information about characteristic tumor regions without being excessively affected by imaging noise and artifacts.

[0073] Detailed characterization of tumor mechanical distribution characteristics: The mechanical interquartile range (MIQR) and elastic heterogeneity index (EHI) not only reflect the degree of data dispersion, but also characterize the shape characteristics of the distribution, such as skewness and tail characteristics, providing in-depth tumor characteristic information beyond simple statistics;

[0074] Implementation of threshold sensitivity analysis: The peak threshold ratio (PTR) parameter enables the system to analyze changes in tumor characteristics at different mechanical thresholds, providing a quantitative tool for identifying key structural areas within the tumor (such as the invasion front, necrotic areas, cystic areas, etc.);

[0075] Enhanced clinical interpretability: These statistical parameters have clear mathematical definitions and intuitive physical meanings, making them easier for clinicians to understand and accept. For example, a high EHI value intuitively indicates a high degree of heterogeneity in the mechanical properties of a tumor, which may correspond to high malignancy or treatment resistance.

[0076] Improved data robustness: By analyzing the same set of data using multiple complementary statistical methods, the system reduces the risk of bias from a single statistic and improves the robustness and reliability of feature extraction, especially when dealing with clinical imaging data of varying quality levels.

[0077] Highlighting subregional features: Quantile analysis is particularly suitable for identifying subregions within the tumor that are relatively small but have important biological significance (such as highly proliferative areas and microinvasive areas). These areas may be "swamped" in average analysis, but can be effectively highlighted in quantile analysis;

[0078] These technical effects jointly improve the system's accuracy in characterizing the mechanical properties of tumors, enabling the prognosis prediction model to be based on richer and more reliable statistical features, thereby improving the accuracy and stability of predictions.

[0079] Furthermore, the gradient parameters include:

[0080] Local mechanical gradient LMG(x,y):

[0081]

[0082] Where LMG(x,y) represents the local mechanical gradient at the position (x,y), E(x,y) represents the elastic modulus E at the position (x,y), and denote the partial derivatives of the elastic modulus E in the x and y directions, respectively. represents the gradient operator;

[0083] Mean Mechanical Gradient (MMG):

[0084] MMG=(1 / N)*ΣxΣyLMG(x,y)

[0085] Wherein, MMG represents the mean mechanical gradient of the entire tumor area;

[0086] Mechanical gradient variance MGV:

[0087] MGV=(1 / N)*ΣxΣy[LMG(x,y)-MMG]2;

[0088] Mechanical gradient direction MGD(x,y):

[0089]

[0090] Where MGD(x,y) represents the direction of the mechanical gradient at the position (x,y), and arctan[] represents the inverse tangent function;

[0091] Mechanical Directional Consistency (MDC):

[0092] MDC=|ΣxΣyexp(i”*MGD(x,y))| / N

[0093] Here, “i” represents an imaginary unit.

[0094] The beneficial effects of the above further solution are: the technical effects brought about by defining the gradient parameters include:

[0095] Accurate characterization of tumor boundaries and transition zones: Local tumor mechanical gradients (LMGs) can highlight the rate of mechanical change within and between tumor boundaries, precisely locating the interface between tumor and normal tissue, and the transition zones between different functional regions within the tumor. These regions are often active areas of tumor invasion and spread and have important prognostic significance.

[0096] Vectorized characterization of spatial heterogeneity: Traditional parameters only describe the size distribution of mechanical values, while gradient parameters (LMG and MGD) provide both size and direction information, enabling a vectorized description of the spatial variations in tumor mechanical properties and revealing deeper tissue structural information.

[0097] Quantitative assessment of mechanical barriers: Local high-gradient regions often represent biological tissue boundaries or mechanical barriers. Through quantitative analysis of MMG and MGV, the system can assess the strength and distribution characteristics of tumor mechanical barriers, providing new insights into predicting tumor invasiveness.

[0098] Capturing microstructural directionality: The mechanical gradient direction (MGD) and mechanical direction consistency (MDC) parameters can capture the directional characteristics of the internal microstructure of the tumor, such as the collagen fiber arrangement direction and the mechanical stress field distribution. These characteristics are closely related to the tumor cell migration channels and potential metastasis pathways.

[0099] Assessment of tissue anisotropy: Mechanical direction consistency (MDC) provides a quantitative indicator for assessing the degree of anisotropy in tumor tissue. A high MDC value indicates that the tissue has a consistent trend of mechanical property changes in a certain direction, which may reflect a specific biological behavior pattern, such as invasive growth along a specific direction.

[0100] Characterization of the topological structure of the mechanical field: Through comprehensive analysis of gradient parameters, the system can identify topological features of the tumor mechanical field, such as singularities, saddle points, and streamline structures. These features correspond to special structural regions in the tumor microenvironment, providing new mathematical tools for understanding tumor growth patterns.

[0101] Basic support for multi-scale analysis: The calculation and comparison of gradient parameters at different spatial scales provide the system with the basic tools for multi-scale analysis, enabling the system to simultaneously capture the macroscopic structure and microscopic change characteristics of tumors;

[0102] These technical effects enable the system to deeply analyze the complexity and heterogeneity of tumor structure from the perspective of mechanical spatial gradients, especially providing unique value in identifying potential high-risk areas, thereby improving the accuracy and clinical relevance of prognostic prediction.

[0103] Furthermore, the spatial distribution parameters include a mechanical spatial autocorrelation index, a mechanical regional clustering index, and a mechanical semivariogram function; the mechanical spatial autocorrelation index includes:

[0104] Moran Index I:

[0105] I=[N / Σi'Σj'wi'j']*[Σi'Σj'wi'j'(Ei'-μ)(Ej'-μ)] / [Σi'(Ei'-μ)2]

[0106] Where wi'j' represents the spatial weight between position i' and position j', Ei' and Ej' represent the elastic modulus E at position i' and position j', respectively;

[0107] G-Force Index C:

[0108] C=[(N-1) / 2Σi'Σj'wi'j']*[Σi'Σj'wi'j'(Ei'-Ej') 2 ] / [Σi'(Ei'-μ) 2 ];

[0109] Mechanical regional clustering index MACI(t):

[0110] MACI(t)=Σr[Ar(t) / Atotal]2

[0111] Wherein, MACI(t) represents the regional clustering index under the mechanical threshold t, Ar(t) represents the area of ​​the rth connected region under the threshold t, and Atotal represents the total tumor area;

[0112] Mechanical semivariogram function MSV(h):

[0113] MSV(h)=(1 / 2N(h))*Σi'Σj'[E(xi')-E(xj')] 2

[0114] Where MSV(h) represents the mechanical semivariance with a spatial distance of h, N(h) represents the number of sample point pairs with a spatial distance of h, ∑ and ∑ represent the double summation of all point pairs (i', j') that satisfy the distance |xi'-xj'| = h, E(xi') and E(xj') represent the elastic modulus values ​​at positions xi' and xj', respectively.

[0115] The topological parameters include mechanical contour features, mechanical connected domain features and mechanical fractal dimensions;

[0116] The mechanical contour features include:

[0117] The number of contour lines, which is the number of contour lines NI(t1,t2) in the elasticity value range [t1,t2], where t1 and t2 represent the elasticity value thresholds;

[0118] The contour length is the total length IL(t) of the contour lines with elasticity value t, where t represents the elasticity threshold;

[0119] The mechanical connectivity domain features include:

[0120] The number of connected domains NCC(t), NCC(t) represents the number of connected domains with elasticity values ​​greater than the threshold t;

[0121] Maximum connected domain area ratio:

[0122] LCCAR(t)=Amax(t) / Atotal

[0123] Among them, LCCAR(t) represents the ratio of the maximum connected domain area with elasticity value greater than threshold t to the total area, and Amax(t) represents the maximum connected domain area with elasticity value greater than threshold t.

[0124] The mechanical fractal dimension MFD(t):

[0125] MFD(t)=lim(ε→0)[log(N(ε,t)) / log(1 / ε)]

[0126] Where MFD(t) represents the mechanical fractal dimension of the region with elasticity greater than the threshold t, N(ε,t) represents the number of squares with a side length of ε required to cover the region with elasticity greater than the threshold t, and ε represents the side length of the square.

[0127] The boundary parameters include:

[0128] Boundary Mechanical Gradient:

[0129]

[0130] Where BMG represents the mean mechanical gradient at the tumor boundary, represents the mechanical gradient at the boundary point b, and Nb represents the total number of boundary points;

[0131] Boundary mechanics gradient variance:

[0132]

[0133] Among them, BMGV represents the variance of the mechanical gradient on the tumor boundary;

[0134] Boundary Mechanics Sharpness Index BMSI:

[0135] BMSI = (Ein - Eout) / d

[0136] Where Ein represents the average elastic modulus near the inner boundary of the tumor, Eout represents the average elastic modulus near the outer boundary of the tumor, and d represents the average distance between the inner and outer measurement points;

[0137] The strain energy parameters include:

[0138] Strain energy per unit volume:

[0139] SEUV(x,y)=(1 / 2)*E(x,y)*ε(x,y) 2

[0140] Where SEUV(x,y) represents the strain energy per unit volume at the position (x,y), and ε(x,y) represents the strain at the position (x,y).

[0141] Total strain energy:

[0142] TSE=ΣxΣySEUV(x,y)

[0143] Where TSE represents the total strain energy of the entire tumor area, SEUV(x,y) represents the strain energy per unit volume at position (x,y), and the strain energy density ratio SEDR is:

[0144] SEDR=SEmax / SEavg

[0145] Where SEmax represents the maximum strain energy density, and SEavg represents the average strain energy density.

[0146] The beneficial effects of the above further solution are: the technical effects brought about by defining spatial distribution parameters, topological parameters, boundary parameters and strain energy parameters include:

[0147] Comprehensive characterization of tumor internal tissue architecture: Through the combination of spatial autocorrelation index (I and C), regional clustering index (MACI), and semivariogram (MSV), the system can comprehensively characterize the spatial distribution pattern of mechanical properties within the tumor, revealing the spatial correlation and heterogeneity of the tissue architecture, providing a multi-dimensional quantitative tool for in-depth understanding of the tumor microenvironment;

[0148] Mathematical description of tumor complexity: Topological parameters such as mechanical contours, connected domains, and fractal dimension transform the complex structure of tumors into precise mathematical descriptions. In particular, fractal dimension (MFD) can quantify the self-similarity and complexity of tumor geometry, which is closely related to tumor invasiveness and prognosis.

[0149] Accurate boundary feature capture: Boundary parameters (BMG, BMGV, BMSI) provide a detailed description of the mechanical properties of the tumor boundary region, which is a key interface between tumor and host tissue. Its mechanical properties are directly related to the tumor's invasive behavior and metastatic potential.

[0150] Revealing energy storage and release characteristics: Strain energy parameters (SEUV, TSE, and SEDR) analyze the mechanical behavior of tumor tissue from an energetic perspective and quantify the distribution of the tissue's ability to store deformation energy under stress. These parameters are related to the internal stress field distribution of the tumor, the strength of intercellular interactions, and the remodeling process of the extracellular matrix.

[0151] Implementation of multi-scale spatial analysis: Through scale-dependent analysis of parameters such as MSV(h), NI(t1, t2), and MFD(t), the system enables multi-scale evaluation of tumor characteristics from micro to macro scales, capturing the pathological significance at different spatial scales;

[0152] Automatic identification of tumor heterogeneous regions: Connected domain features (NCC and LCCAR) and regional clustering index (MACI) enable the system to automatically identify heterogeneous regions within the tumor, such as necrotic areas, high-proliferation areas, and micro-invasive areas, which often have special prognostic significance;

[0153] Prediction of dynamic progression potential: Parameters such as fractal dimension (MFD) and boundary sharpness index (BMSI) are closely related to tumor growth dynamics, providing a mathematical basis for predicting future tumor progression patterns and helping to achieve the transition from static imaging to dynamic prediction;

[0154] Establishment of physical-biological correlation: Strain energy parameters (SEUV and SEDR) establish a bridge between the mechanical properties and biological behavior of tumors. Regions of high strain energy generally correspond to areas of high cell activity and rapid proliferation, providing a biophysical explanation for differences in prognosis.

[0155] Improved comprehensive diagnostic efficacy: These advanced parameters complement each other and together form a complete mechanical portrait of the tumor, significantly surpassing the characterization capabilities of traditional single or low-dimensional parameters. They provide a nearly complete tumor mechanical characteristic spectrum and significantly improve the accuracy and reliability of prognostic prediction.

[0156] These technical effects enable the system to conduct a comprehensive mechanical property analysis of tumors from multiple mathematical and physical perspectives, capturing subtle features that cannot be identified by traditional methods, and laying a solid theoretical and technical foundation for achieving high-precision individualized prognosis prediction.

[0157] Furthermore, the multi-scale mechanical parameters include:

[0158] Multi-scale regional analysis: By changing the size of the observation window, mechanical parameters are calculated from local to global scales to generate multi-scale features;

[0159] Multi-threshold mechanical analysis: For mechanical parameters that require thresholds, multiple thresholds are used to calculate features and generate multi-threshold features;

[0160] Radial analysis: Perform radial analysis from the tumor center to the periphery to calculate the distribution of mechanical parameters at different radii:

[0161] ER(r)=(1 / Nr)*ΣθE(r,θ)

[0162] Where ER(r) represents the average elastic modulus at radius r, E(r,θ) represents the elastic modulus at polar coordinates (r,θ), Nr represents the number of pixels at radius r, and θ represents the polar angle;

[0163] Layered analysis: The tumor area is divided into multiple concentric layers from the inside to the outside, and the mechanical parameters of each layer are calculated:

[0164] EL(z)=(1 / Nz)*Σp∈LzE(p)

[0165] Where EL(z) represents the average elastic modulus of the z-th layer, Nz represents the number of pixels in the z-th layer, Lz represents the set of pixels in the z-th layer, and E(p) represents the elastic modulus at pixel p.

[0166] The beneficial effects of the above further solution are: the technical effects of the multi-scale mechanical parameters include:

[0167] Hierarchical analysis of tumor internal structure: Through multi-scale regional analysis, radial analysis, and layered analysis, the system can systematically analyze the structural organization of tumors from the microscopic local to the macroscopic global, from the core to the periphery, revealing mechanical heterogeneity patterns at different spatial scales, overcoming the limitations of single-scale analysis;

[0168] Capturing scale-dependent features: Tumor tissue exhibits different mechanical properties at different observation scales. Multiscale analysis can capture this scale dependency and identify pathological features that are only evident at specific scales, such as microinvasive fronts, perivascular areas, and other key prognostic structures.

[0169] Directed assessment of tumor spatial organization: Radial analysis (ER(r)) and laminar analysis (EL(z)) provide the ability to provide a directed assessment of tumor spatial organization, quantitatively describing the mechanical gradient distribution from the center to the edge of the tumor, and revealing the spatial evolution and invasion patterns of the tumor during growth.

[0170] Expanded adaptive analysis range: Multi-threshold mechanical analysis automatically finds optimal parameter settings by using a range of thresholds, overcoming the subjectivity and limitations of traditional single-threshold methods and improving the system's adaptability to tumors of different types and imaging qualities.

[0171] Automatic identification of tissue compartments: Layered analysis can automatically divide tumors into functionally meaningful regions (such as active growth zones, intermediate transition zones, and central necrotic zones), enabling systematic analysis of internal tumor heterogeneity and thus improving the accuracy of prognostic assessment;

[0172] Scale spectrum analysis of spatial correlation: By simultaneously analyzing spatial correlation parameters at multiple scales, the system is able to construct a complete scale spectrum of spatial correlation to comprehensively characterize the spatial complexity of tumor tissue, which is not achievable with single-scale analysis.

[0173] Furthermore, the mechanical feature combination and high-order parameters include:

[0174] Mechanical ratio feature MRF(A,B):

[0175] MRF(A,B)=FeatureA / FeatureB

[0176] Among them, FeatureA and FeatureB represent two basic mechanical features;

[0177] Mechanical product characteristic MPF(A,B):

[0178] MPF(A,B)=FeatureA*FeatureB;

[0179] Mechanical weighted feature MWF(A,w):

[0180] MWF(A,w)=Σkwk*Ak

[0181] Where Ak represents the value of the mechanical characteristic A in the kth region, and wk represents the weight of the kth region;

[0182] Mechanical heterogeneity matrix:

[0183] MHM(n,m)=|En-Em| / (En+Em)

[0184] where MHM(n,m) represents the mechanical heterogeneity between regions n and m, and En and Em represent the average elastic moduli of regions n and m, respectively.

[0185] The beneficial effects of the above further solution are: the technical effects of the combination of mechanical characteristics and high-order parameters include:

[0186] In-depth exploration of the relationship between features: Mechanical ratio features (MRFs) and mechanical product features (MPFs) reveal the proportional relationship and product effect between different mechanical parameters, thereby discovering complex patterns that cannot be identified by single parameter analysis, providing a new perspective for prognosis prediction;

[0187] Nonlinear expansion of feature space: Through the mathematical combination of basic features, the system achieves a mapping from the basic feature space to a high-dimensional nonlinear feature space, significantly improving the model's ability to capture complex biological relationships without increasing the amount of original measurement data;

[0188] Accurate implementation of region-specific analysis: Mechanically weighted features (MWFs) can highlight the contribution of key prognostic regions, reduce the impact of non-critical regions, and improve the signal-to-noise ratio and predictive relevance of features by assigning appropriate weights to different regions;

[0189] Quantification of mechanical interactions within tumors: The mechanical heterogeneity matrix (MHM) has for the first time systematically quantified the mechanical differences between different regions within a tumor. This inter-regional interaction is closely related to the stress distribution and cell migration pathways within the tumor, providing a unique perspective for predicting invasiveness.

[0190] Enhanced biological interpretability: These high-order parameters not only improve prediction performance but also maintain clear physical meaning and biological interpretability. For example, MRF can represent the ratio of tissue stiffness to viscosity, and MPF can reflect the response of tissue under complex deformation modes, making the prediction results more clinically interpretable.

[0191] Deepening personalized analysis: By flexibly combining different features and adjusting their weights, the system can adaptively optimize feature sets for different tumor subtypes and different clinical scenarios, achieving more personalized prognostic predictions and improving the accuracy of clinical decision-making;

[0192] These technical effects together constitute a comprehensive system for in-depth analysis of tumor mechanical properties from multiple scales and angles, enabling the system to capture subtle characteristic patterns that cannot be identified by traditional single parameter or fixed-scale analysis, greatly improving the accuracy and reliability of tumor prognosis prediction.

[0193] Furthermore, the grouping of mechanical parameters includes basic mechanical parameter group, mechanical statistical parameter group, gradient parameter group, spatial distribution parameter group, topological parameter group, boundary parameter group, strain energy parameter group, multi-scale mechanical parameter group and mechanical feature combination group.

[0194] The beneficial effects of the above further solution are: the technical effects of the present invention dividing the mechanical parameters into nine parameter groups include:

[0195] Establishment of a systematic feature organization framework: By systematically classifying thousands of mechanical parameters into nine parameter groups, a clear and logically rigorous feature organization framework was established, making the complex high-dimensional feature space manageable and understandable, greatly improving the efficiency of feature extraction and model building;

[0196] Intelligent stratification of feature selection: This grouping structure enables the system to implement a stratification strategy during feature selection, first selecting between groups and then refining within groups. This avoids the information loss and computational burden of traditional "one-size-fits-all" feature selection methods, improving model training efficiency and generalization capabilities.

[0197] Complementary fusion of multidimensional prognostic information: Different parameter groups capture different aspects of tumor mechanical properties. For example, the basic mechanical parameter group reflects fundamental physical properties, the topological parameter group characterizes spatial structural complexity, and the boundary parameter group describes invasion characteristics. This systematic fusion of multidimensional information provides a comprehensive and complementary evidence base for prediction.

[0198] Hierarchical support for clinical interpretability: Prediction results can be traced back to specific parameter groups and parameters, enabling doctors to understand the mechanics behind the predictions (e.g., whether the predictions are primarily based on the gradient parameter group or the boundary parameter group), thus improving the clinical interpretability and acceptability of the model.

[0199] Enhanced adaptability to different tumor types: Different tumor types may exhibit different feature importance patterns. The system grouping structure enables the model to automatically adjust the weights of each parameter group according to the specific tumor type, achieving adaptive optimization for different tumor types;

[0200] Structured support for iterative optimization: Systematic grouping provides a structured framework for subsequent iterative updates of the model. This allows for targeted optimization or expansion of specific parameter groups without rebuilding the entire feature set, improving system maintainability and scalability.

[0201] A framework for integrating multidisciplinary knowledge: This grouping is not only based on mathematical and physical principles but also reflects expertise in tumor pathology and biomechanics. It provides a common language for multidisciplinary collaboration and promotes the in-depth development of medical-engineering integration.

[0202] Directed guidance of analysis paths: Clear parameter grouping provides systematic guidance for clinical analysis, enabling doctors to conduct in-depth analysis step by step along the logical path from basic parameters to high-level features, thereby improving the scientific nature and systematic nature of clinical decision-making.

[0203] Optimal allocation of computing resources: Feature importance assessment based on parameter grouping enables the system to prioritize limited computing resources to the parameter groups with the greatest predictive value, improving computing efficiency while maintaining prediction accuracy, enabling the system to be applied in routine clinical settings;

[0204] This systematic parameter grouping structure is one of the core innovations of the present invention. It not only integrates the advantages of the aforementioned various parameters, but also maximizes these advantages through structured organization, forming an efficient, reliable, and interpretable tumor mechanics multi-parameter prognostic prediction system.

[0205] The present invention provides a tumor mechanics multi-parameter prognosis prediction method based on elastography, comprising:

[0206] Acquire elastic imaging data of the patient's tumor;

[0207] Preprocessing the collected elastic imaging data;

[0208] Extract multidimensional tumor mechanical parameters from preprocessed elastography data and perform feature normalization;

[0209] Screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters;

[0210] Construct a prognostic prediction model based on the screened key prognostic-related parameters;

[0211] The prognostic prediction model is applied to new patients to predict their prognostic risk, and the tumor mechanics multi-parameter prognostic prediction results are displayed in a visual manner.

[0212] The present invention has the following beneficial effects: It can extract high-dimensional mechanical parameters (>1000) from tumor elastography data and construct an accurate prognostic prediction model. This method, for the first time, expands 1000+ dimensional features from basic elastic parameters, capturing the spatial heterogeneity of tumor mechanical properties. The C-index was validated in predicting 5-year survival in a breast cancer dataset, significantly outperforming methods that use only clinical parameters. This method can be used in clinical applications such as tumor prognosis assessment, treatment decision-making, recurrence risk monitoring, and treatment response prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0213] Figure 1 Schematic diagram of the system structure of the present invention.

[0214] Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0215] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0216] Example 1

[0217] like Figure 1 As shown, the present invention provides a tumor mechanics multi-parameter prognosis prediction system based on elastic imaging, comprising:

[0218] an elastic imaging acquisition module, used to collect elastic imaging data of a patient's tumor;

[0219] An image preprocessing module, used for preprocessing the collected elastic imaging data;

[0220] The parameter extraction module is used to extract multidimensional tumor mechanical parameters from the preprocessed elastic imaging data and perform feature normalization processing, which includes:

[0221] A first processing unit is used to extract basic mechanical parameters;

[0222] A second processing unit is used to calculate mechanical statistical parameters, gradient parameters, spatial distribution parameters, topological parameters, boundary parameters and strain energy parameters based on basic mechanical parameters;

[0223] The third processing unit is used to expand the dimension of mechanical parameters through multi-scale analysis, multi-threshold analysis and feature combination to obtain multi-scale mechanical parameters, mechanical feature combinations and high-order parameters;

[0224] a fourth processing unit, configured to construct a high-dimensional mechanical characteristic matrix based on the multidimensional mechanical parameters extracted by the first processing unit, the second processing unit, and the third processing unit, wherein the high-dimensional mechanical characteristic matrix is ​​used to group the mechanical parameters according to categories and perform normalization processing on all the grouped mechanical parameters;

[0225] Feature selection module, used to screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters;

[0226] A prognosis prediction model building module is used to build a prognosis prediction model based on the screened key prognosis-related parameters;

[0227] The result display module is used to apply the prognostic prediction model to visually display the tumor mechanics multi-parameter prognostic prediction results.

[0228] In this embodiment, the extraction of multidimensional tumor mechanical parameters includes:

[0229] Global parameters, which describe the average mechanical properties of the entire tumor region;

[0230] local parameters, which are used to describe the mechanical properties of different regions within the tumor;

[0231] Texture features, used to describe the texture structure of the internal mechanical parameters of the tumor;

[0232] Spatial distribution parameters, used to describe the spatial gradient and directionality of mechanical parameters;

[0233] Topological parameters, which describe the number, shape, and connectivity of mechanical islands within the tumor;

[0234] Boundary parameters, used to describe the mechanical properties of the tumor boundary;

[0235] Strain energy parameter, used to describe the elastic potential energy stored in tumor tissue;

[0236] Multi-scale mechanical parameters are used to describe the mechanical properties at different spatial scales.

[0237] In this embodiment, global parameters correspond to basic mechanical parameters and mechanical statistical parameters. Basic mechanical parameters include elastic modulus (E), shear modulus (G), Poisson's ratio (μ), bulk modulus (K), etc.; mechanical statistical parameters include statistical characteristics such as the coefficient of variation (CV), standard deviation (σ), and mechanical semivariogram (MSV) of these basic parameters.

[0238] Local parameters: These parameters correspond to boundary characteristic parameters, regional gradient parameters, and local mechanical inhomogeneity parameters. Specifically, they include local mechanical gradient (LMG), mechanical gradient direction (MGD), boundary mechanical gradient (BMG), boundary mechanical gradient variance (BMGV), and boundary mechanical sharpness index (BMSI), which are used to characterize the mechanical differences between different tumor regions.

[0239] Texture features: These correspond to spatial correlation parameters and fractal dimension features. These include the Moran Index (I), Girard Index (C), Mechanical Area Clustering Index (MACI), Mechanical Fractal Dimension (MFD), Number of Isolines (NI), and Isoline Length (IL). These features are used to describe the spatial pattern and complexity of the distribution of mechanical parameters within tumors. Texture features clearly indicate that they correspond to "spatial correlation parameters and fractal dimension features."

[0240] Specifically:

[0241] The Moran index (I) and the Girard index (C) are classic measures of spatial autocorrelation;

[0242] The mechanical regional clustering index (MACI) characterizes the degree of regional clustering at different thresholds;

[0243] Mechanical fractal dimension (MFD) describes the geometric complexity and self-similarity of the tumor region;

[0244] The number of contour lines (NI) and the length of contour lines (IL) characterize the distribution characteristics of mechanical parameters;

[0245] These parameters together constitute a complete texture feature system that describes the spatial distribution pattern and structural complexity of the internal mechanical parameters of the tumor, and can quantitatively characterize the heterogeneity and spatial arrangement characteristics of tumor tissue.

[0246] In this embodiment, the dimension expansion of the mechanical parameters is to derive a high-dimensional feature set from the basic mechanical parameters through multiple mathematical transformations.

[0247] In this embodiment, feature selection uses methods such as LASSO regression to screen out key prognosis-related parameters from high-dimensional features, and conducts comprehensive analysis in combination with clinical characteristics. It is not limited to the LASSO regression method, and random forest feature importance, minimum redundancy maximum correlation or elastic network can also be used.

[0248] Feature selection uses LASSO regression feature selection implementation process is as follows:

[0249] 1. Data preparation phase: All extracted 1000+ dimensional mechanical features are organized into a feature matrix X; corresponding prognostic outcome data y (including survival time and event status) are prepared; and the dataset is divided into a training set and a validation set in a 7:3 ratio.

[0250] 2. Feature preprocessing: All mechanical features are standardized to ensure that parameters of different dimensions are comparable. If clinical features need to be combined, the categorical clinical features are one-hot encoded and integrated with the standardized mechanical features.

[0251] 3. LASSO model training and parameter optimization: Establish a search grid for the regularization parameter α, usually set to 50 logarithmically divided points from 0.0001 to 1; use a 5-fold cross-validation method to evaluate model performance under different α values; use C-index as the evaluation metric, and select the α value that gives the highest C-index on the validation set as the optimal parameter.

[0252] 4. Feature screening implementation: Use the optimal α value to train the final LASSO model; observe the model coefficients, and the features corresponding to non-zero coefficients are the selected key prognostic-related parameters; generally, 20-50 key parameters with the most prognostic value are screened from 1000+ dimensional features.

[0253] 5. Evaluation of screening results: Use the selected feature subset to build a prognostic model on the validation set and calculate the C-index; compare the performance with the model built using all features to verify the effectiveness of feature screening; analyze the biological significance of the selected features to ensure the interpretability of the results.

[0254] 6. Clinical integration: Integrate the screened mechanical features with key clinical parameters (such as tumor stage, histological grade, etc.); evaluate the predictive performance of the integrated model on the training set and validation set; and determine the feature set that will ultimately enter the prognostic prediction model.

[0255] Random forest feature importance analysis process:

[0256] 1. Model building: Construct a random forest survival model containing 300-500 decision trees; set the number of feature samples per node to the square root of the total number of features; adjust parameters for survival analysis and use the log-rank splitting criterion;

[0257] 2. Feature importance calculation: The importance score of each feature is calculated by permutation importance or average impurity reduction; the higher the score, the greater the contribution of the feature to prognosis prediction; the feature importance is ranked, and the top 5% features are selected as candidates.

[0258] 3. Stability verification: Perform 100 random resampling modelings and summarize the features selected each time; calculate the frequency of each feature being selected; select features with stability exceeding 80% as the final feature set.

[0259] The minimum redundancy maximum relevance (mRMR) implementation process is as follows:

[0260] 1. Correlation calculation: Calculate the mutual information value (correlation) between each mechanical feature and the survival outcome; also calculate the mutual information value (redundancy) between features.

[0261] 2. Feature sorting: Use the mRMR criterion to sort features: max[I(fi;S)-1 / |S|∑fsI(fi;fs)], where I(fi;S) represents the mutual information between features and prognosis, and S|∑fsI(fi;fs)] represents the average redundancy of the selected feature set.

[0262] 3. Incremental feature selection: Starting from an empty set, add the features with the highest mRMR scores one by one; each time a feature is added, evaluate the predictive performance of the current feature subset; stop when the performance no longer improves significantly, and obtain the optimal feature subset.

[0263] The elastic net feature selection process is as follows:

[0264] 1. Parameter setting: Set the search grid for the regularization parameter α and the ratio of L1 and L2 penalty weights λ; α is usually set to [0.001, 0.01, 0.1, 1.0]; λ is set to 9 equally divided points from 0.1 to 0.9 (λ = 0 for ridge regression, λ = 1 for LASSO).

[0265] 2. Grid search: Perform 5-fold cross validation on each pair of (α, λ) combinations; calculate the average C-index under each parameter combination; and select the optimal parameter combination that maximizes the C-index.

[0266] 3. Feature screening: Use the optimal parameters to train the final elastic network model; analyze the model coefficients and select features whose absolute values ​​are greater than the threshold; elastic networks are more likely to retain highly correlated feature groups than LASSO and are suitable for processing highly correlated mechanical parameters.

[0267] 4. Robustness verification: Generate 100 data subsets through bootstrap sampling; repeat the feature selection process on each subset; select robust features that are selected in most subsets.

[0268] This scheme comprehensively applies multiple feature selection methods, takes the intersection or union of the selection results of each method, and finally determines the key mechanical parameter set used to construct the prognostic model. This not only ensures the high accuracy of the prediction model, but also avoids the dimensionality curse problem, and improves the generalization ability and computational efficiency of the model in clinical applications.

[0269] In this embodiment, the prognosis prediction model is constructed by: constructing a prognosis prediction model based on methods such as the Cox proportional hazards model or the random survival forest model, and using cross-validation and independent validation sets to evaluate the model performance.

[0270] The Cox proportional hazards model was constructed as follows:

[0271] 1. Model input preparation: Key mechanical parameters (approximately 30-50) selected by feature selection methods such as LASSO were used as predictor variables; patient survival time and event status (recurrence / death = 1, no recurrence / still alive = 0) were used as dependent variables; and clinicopathological characteristics (such as age, tumor stage, and histological type) were integrated as covariates.

[0272] 2. Data stratification: The patient cohort was divided into a training set and an internal validation set in a 7:3 ratio; the two groups were balanced in terms of important clinical variables (such as tumor stage and treatment regimen); continuous variables were centered and converted to a standard distribution.

[0273] 3. Model construction and parameter estimation: A multivariate Cox proportional hazards model was constructed based on the training set; the regression coefficient β was estimated using the maximum partial likelihood method; the significance of the overall model was assessed using the partial likelihood ratio test; the proportional hazards assumption was tested, and the Schoenfeld residual plot was used to confirm that the effects of each covariate remained constant over time.

[0274] 4. Variable screening and model simplification: Use the stepwise regression method to optimize the model according to the AIC / BIC criteria; use forward selection, backward elimination, or a combination of the two methods to determine the final model variables; diagnose multicollinearity between parameters and eliminate redundant variables.

[0275] 5. Model calibration and correction: Assess the accuracy of model predictions through calibration curves; perform appropriate transformations (such as logarithmic transformations) on variables based on the results of nonlinear effect tests; add interaction terms when necessary to capture the interactions between variables.

[0276] The random survival forest model construction process is as follows:

[0277] 1. Data preprocessing and segmentation: Use the same training and validation data segmentation scheme as the Cox model; no variable standardization is required, as random forests are insensitive to variable scale; prepare a bootstrap sampling dataset to construct each tree in the forest.

[0278] 2. Model parameter settings: Number of decision trees: set to 500-1000 to ensure model stability; Number of node splitting variables: set to the square root of the total number of variables (about 5-7); Minimum number of leaf node samples: set to 5% of the training sample to prevent overfitting; Splitting criterion: Use the log-rank statistic as the node splitting evaluation criterion.

[0279] 3. Model training process: Construct a survival decision tree for each bootstrap sample; at each node, randomly select mtry variables as candidate splitting variables; for each candidate variable and all possible splitting points, calculate the survival difference between child nodes; select the variable and splitting point that maximizes the log-rank statistic; recursively repeat the above steps until the stopping condition is met.

[0280] 4. Forest ensemble and prediction: For new samples, pass them to each tree in the forest; in each tree, assign them to the corresponding leaf node according to the sample characteristics; based on the survival function estimate of each leaf node, calculate the comprehensive survival prediction by averaging; calculate the cumulative hazard function and survival probability curve after integration.

[0281] 5. Feature importance assessment: Calculate the importance score of each feature through the variable permutation method; for each variable, randomly permute its value and observe the degree of model performance degradation; the greater the performance degradation, the more important the variable is to the prediction; based on the importance score ranking, determine the core predictor variables.

[0282] 6. Model validation and performance evaluation include internal validation and external independent validation:

[0283] Internal validation was performed as follows: 5-fold cross validation: the training data was divided into 5 parts, and 4 parts were used for training and 1 for validation in turn;

[0284] Bootstrap validation: 200 data subsets were generated through sampling with replacement for model training and testing;

[0285] Time segmentation validation: Data are segmented according to the time of patient enrollment, and early patients are used to predict late patients.

[0286] External independent validation was performed as follows: using independent cohort data (patients collected at different centers and at different time periods); ensuring that the independent cohort contained the same variable information as the modeling cohort; applying the model on the independent cohort without any parameter readjustment; and evaluating the generalization performance of the model in a real-world setting.

[0287] 7. Calculation of performance indicators: C-index: evaluates the model's ability to distinguish between high-risk and low-risk patients (0.5 is a random prediction, 1 is a perfect prediction); time-dependent AUC: evaluates the discriminative ability at different time points (such as 1 year, 3 years, and 5 years); calibration curve: compares the consistency of predicted survival probability with the actual observed survival rate; net reclassification improvement index (NRI): evaluates the degree of improvement in prediction compared with the baseline model; decision curve analysis (DCA): evaluates the clinical decision-making value of the model at different thresholds.

[0288] 8. Robustness test: Conduct sensitivity analysis on key variables to assess the impact of small changes in variables on predictions; evaluate model performance for patients in different subgroups (such as different tumor stages, different treatment regimens); and evaluate the impact of missing data processing methods on model performance.

[0289] Model comparison and optimization are as follows:

[0290] 1. Comparison of the performance of the Cox model and random forest: Compare the predictive performance of the two models using the same training and validation sets; compare the statistical difference in the C-index between the two models using the DeLong test; determine the optimal model or establish a model integration strategy based on comprehensive performance evaluation.

[0291] 2. Model integration and optimization: Combine the prediction results of the Cox model and random forest to improve the overall prediction accuracy through weighted averaging; the weights are determined based on the performance of each model on the validation set; and the weight parameters of the integrated model are optimized through gradient descent or grid search.

[0292] 3. Model fine-tuning and finalization: Adjust model hyperparameters (such as the number of trees in the random forest and the regularization parameter) based on the validation results; determine the complexity of the final prediction model based on clinical practicality considerations; use all available data when training the final model to maximize data utilization efficiency.

[0293] Interpretability and clinical applications are as follows:

[0294] 1. Enhanced model interpretability: For the Cox model, the hazard ratio (HR) and 95% confidence interval of each variable are provided; for the random forest, the variable importance ranking and partial dependence diagram are provided; and an interactive nomogram is developed to help clinicians intuitively understand the impact of each parameter on prognosis.

[0295] 2. Risk stratification implementation: Based on model predictions, patients were divided into three risk groups: high, intermediate, and low. The optimal cutoff method was used to determine the risk group boundaries and maximize the survival difference between groups. The log-rank test was used to verify the statistical differences in the survival curves between different risk groups.

[0296] 3. Clinical decision support system development: Encapsulate the final model into a user-friendly clinical decision support tool; develop a web or mobile application version to facilitate real-time clinical use; and design an intuitive visual interface to facilitate doctor-patient communication and treatment decision-making.

[0297] Through the above systematic process, this solution realizes the complete construction process from key mechanical parameters to accurate prognosis prediction model, ensuring the scientificity, accuracy and clinical usability of the model, and providing a decision support tool for precision tumor treatment.

[0298] In this embodiment, the tumor mechanical parameters are extracted as follows:

[0299] 1. Basic mechanical parameter extraction: As the basis for all subsequent mechanical parameter calculations, the following basic mechanical parameters are first extracted from the elastography data.

[0300] 1.1 Elastic modulus (E):

[0301] E=σ / ε

[0302] Where E is the elastic modulus in kilopascals (kPa), σ is the stress in kilopascals (kPa), and ε is the strain, which is dimensionless.

[0303] 1.2 Shear modulus (G):

[0304] G = τ / γ

[0305] Where G is the shear modulus in kilopascals (kPa), τ is the shear stress in kilopascals (kPa), and γ is the shear strain, which is dimensionless.

[0306] 1.3 Poisson's ratio (ν):

[0307] ν=-εtrans / εaxial

[0308] Where εtrans represents the transverse strain, εaxial represents the axial strain, and ν represents the Poisson's ratio, which is dimensionless.

[0309] 1.4 Young's Modulus (YM):

[0310] YM=2G(1+ν)

[0311] Where G represents the shear modulus, YM represents the Young's modulus in kilopascals (kPa), and ν represents the Poisson's ratio.

[0312] 1.5 Bulk modulus (K):

[0313] K=E / (3(1-2ν))

[0314] Where K is the bulk modulus in kilopascals (kPa), E is the elastic modulus, and ν is the Poisson's ratio.

[0315] 1.6 Compression Factor (CF):

[0316] CF=1 / K

[0317] Wherein, CF represents the compressibility coefficient, the unit is kilopascal-1 (kPa-1), and K represents the bulk modulus.

[0318] In this embodiment, the basic mechanical parameters are defined as including elastic modulus (E), shear modulus (G), Poisson's ratio (μ), bulk modulus (K) and other related parameters, which constitute the basic mechanical parameter set that the system can measure and calculate. The elastic modulus E is the most direct and commonly used parameter to characterize tissue stiffness and is the most reliable to obtain in elastic imaging. There is a mathematical relationship between other basic mechanical parameters (such as G, μ, K, etc.) and the elastic modulus E, and they can be converted into each other. In actual clinical applications, using a single elastic modulus E for subsequent calculations simplifies the complexity of the system (such as calculating mechanical statistical parameters, gradient parameters, spatial distribution parameters, topological parameters, boundary parameters and strain energy parameters based on the elastic modulus E), improves calculation efficiency and result stability. Therefore, although the system can measure and calculate a variety of basic mechanical parameters, in actual operation and calculation of derived parameters, it mainly relies on the elastic modulus E as the basic input. This design not only retains the comprehensiveness of the system (can provide a variety of basic mechanical parameters), but also ensures the efficiency and accuracy of the calculation (mainly using the most reliable elastic modulus E for subsequent parameter calculation).

[0319] 2. Mechanical statistical parameter extraction: Based on the basic mechanical parameters, the following statistical parameters are calculated:

[0320] 2.1 Basic statistical parameters

[0321] 2.1.1 Mean μ(Mean):

[0322] μ=(1 / N)*Σ(Ep)

[0323] Wherein, N represents the total number of pixels, and Ep represents the elastic modulus value of the p-th pixel.

[0324] 2.1.2 Standard Deviation:

[0325]

[0326] Where N represents the total number of pixels.

[0327] 2.1.3 Coefficient of Variation:

[0328] CV=σ / μ

[0329] Where CV represents the coefficient of variation of the elastic modulus, σ represents the standard deviation, and μ represents the mean.

[0330] 2.1.4 Peak Threshold Ratio (PTR):

[0331] PTR(t)=Epeak / Et

[0332] Wherein, Epeak represents the peak value of the mechanical parameter, Et represents the mechanical parameter value at the threshold t, and PTR(t) represents the peak threshold.

[0333] 2.2 Quantile mechanical parameters

[0334] 2.2.1 Mechanical Percentiles: Calculation formula: MPn = the value at the nth percentile of the elastic modulus values ​​sorted from smallest to largest, where MPn represents the elastic modulus E of the nth percentile and n represents a percentage (0-100).

[0335] 2.2.2 Mechanical Interquartile Range (MIQR):

[0336] MIQR=MP75-MP25

[0337] Wherein, MP75 represents the 75th percentile elastic modulus, MP25 represents the 25th percentile elastic modulus, and MIQR represents the mechanical interquartile range.

[0338] 2.2.3 Elasticity Heterogeneity Index (EHI):

[0339] EHI=(MP95-MP5) / MP50

[0340] Wherein, MP95 represents the 95th percentile elastic modulus, MP5 represents the 5th percentile elastic modulus, MP50 represents the median elastic modulus, and EHI represents the elastic heterogeneity index.

[0341] 3. Mechanical gradient parameter extraction: The following gradient parameters are extracted from the spatial variation of the basic elastic parameters:

[0342] 3.1 Local Mechanical Gradient (LMG):

[0343]

[0344] Where LMG(x,y) represents the local tumor mechanical gradient at position (x,y), E(x,y) represents the elastic modulus E at position (x,y), and denote the partial derivatives of the elastic modulus E in the x and y directions, respectively. represents the gradient operator;

[0345] It is a basic operator in vector calculus. When the gradient operator acts on a scalar field E(x,y), a vector field is generated: This vector points to the direction in which the E value increases fastest, and its magnitude represents the rate of change of E in that direction. It represents the magnitude of the gradient vector, that is, the size of the gradient, by calculating Therefore, LMG(x,y) actually represents the maximum rate of change of the elastic modulus E at position (x,y), reflecting the local spatial variation of the mechanical properties within the tumor. This parameter is important for characterizing tumor tissue heterogeneity and mechanical interfaces.

[0346] 3.2 Mean Mechanical Gradient (MMG):

[0347] MMG=(1 / N)*ΣxΣyLMG(x,y)

[0348] Where MMG represents the average mechanical gradient of the entire tumor area, and N represents the total number of pixels.

[0349] 3.3 Mechanical Gradient Variance (MGV)

[0350] MGV=(1 / N)*ΣxΣy[LMG(x,y)-MMG]2

[0351] Where MGV represents the variance of the mechanical gradient.

[0352] 3.4 Mechanical Gradient Direction (MGD):

[0353]

[0354] Among them, MGD(x,y) represents the direction of the mechanical gradient at the position (x,y), arctan[] represents the inverse tangent function, and denote the partial derivatives of the elastic modulus in the x and y directions, respectively.

[0355] 3.5 Mechanical Directional Coherence (MDC):

[0356] MDC=|ΣxΣyexp(i”*MGD(x,y))| / N

[0357] Where i'' represents the imaginary unit, MDC represents the consistency of the mechanical gradient direction, MGD(x,y) represents the mechanical gradient direction at position (x,y), and N represents the total number of pixels.

[0358] 4. Extraction of mechanical spatial distribution parameters:

[0359] 4.1 Mechanical Spatial Autocorrelation Index (MSAI)

[0360] 4.1.1 Moran's I:

[0361] I=[N / Σi'Σj'wi'j']*[Σi'Σj'wi'j'(Ei'-μ)(Ej'-μ)] / [Σi'(Ei'-μ)2]

[0362] Where wi'j' represents the spatial weight between position i' and position j', Ei' and Ej' represent the elastic modulus E at position i' and position j' respectively, N represents the total number of pixels, and μ represents the average elastic modulus.

[0363] 4.1.2 Geary's C

[0364] C=[(N-1) / 2Σi'Σj'wi'j']*[Σi'Σj'wi'j'(Ei'-Ej') 2 ] / [Σi'(Ei'-μ) 2 ];

[0365] 4.2 Mechanical Area Clustering Index (MACI):

[0366] MACI(t)=Σr[Ar(t) / Atotal]2

[0367] Wherein, MACI(t) represents the regional clustering index under the mechanical threshold t, Ar(t) represents the area of ​​the rth connected region under the threshold t, and Atotal represents the total tumor area.

[0368] 4.3 Mechanical Semivariogram (MSV):

[0369] MSV(h)=(1 / 2N(h))*Σi'Σj'[E(xi')-E(xj')] 2

[0370] Where MSV(h) represents the mechanical semivariance with a spatial distance of h, N(h) represents the number of sample point pairs with a spatial distance of h, ∑ and ∑ represent the double summation of all point pairs (i', j') that satisfy the distance |xi'-xj'|=h, and E(xi') and E(xj') represent the elastic modulus values ​​at positions xi' and xj', respectively.

[0371] 5. Mechanical topology parameter extraction:

[0372] 5.1 Mechanical Isoline Features

[0373] 5.1.1 Number of Isolines (NI): NI(t1, t2) = the number of isolines within the elasticity range [t1, t2], where NI(t1, t2) represents the number of isolines within the elasticity range [t1, t2], and t1 and t2 represent the elasticity thresholds.

[0374] 5.1.2 Isoline Length (IL): IL(t) = total length of the isolines with elasticity value t, where IL(t) represents the total length of the isolines with elasticity value t, and t represents the elasticity threshold.

[0375] 5.2 Mechanical Connected Component Features

[0376] 5.2.1 Number of Connected Components (NCC): NCC(t) = the number of connected components with elasticity greater than a threshold value t, where NCC(t) represents the number of connected components with elasticity greater than a threshold value t, and t represents the elasticity threshold.

[0377] 5.2.2 Largest Connected Component Area Ratio (LCCAR):

[0378] LCCAR(t)=Amax(t) / Atotal

[0379] Among them, LCCAR(t) represents the ratio of the maximum connected domain area with an elasticity value greater than the threshold t to the total area, Amax(t) represents the maximum connected domain area with an elasticity value greater than the threshold t, and Atotal represents the total tumor area.

[0380] 5.3 Mechanical Fractal Dimension (MFD):

[0381] MFD(t)=lim(ε→0)[log(N(ε,t)) / log(1 / ε)]

[0382] Where MFD(t) represents the mechanical fractal dimension of the region with elasticity greater than the threshold t, N(ε,t) represents the number of squares with a side length of ε required to cover the region with elasticity greater than the threshold t, and ε represents the side length of the square.

[0383] 6. Mechanical boundary parameter extraction

[0384] 6.1 Boundary Mechanical Gradient (BMG):

[0385]

[0386] Where BMG represents the mean mechanical gradient at the tumor boundary, represents the mechanical gradient at the boundary point b, and Nb represents the total number of boundary points.

[0387] 6.2 Boundary Mechanical Gradient Variance (BMGV):

[0388]

[0389] Among them, BMGV represents the variance of the mechanical gradient on the tumor boundary, represents the mechanical gradient at the boundary point b, BMG represents the average mechanical gradient of the boundary, and Nb represents the total number of boundary points.

[0390] 6.3 Boundary Mechanical Sharpness Index (BMSI):

[0391] BMSI = (Ein - Eout) / d

[0392] Wherein, Ein represents the average elastic modulus near the inner boundary of the tumor, Eout represents the average elastic modulus near the outer boundary of the tumor, d represents the average distance between the inner and outer measurement points, and BMSI represents the boundary mechanical sharpness index.

[0393] 7. Mechanical strain energy parameter extraction

[0394] 7.1 Strain Energy per Unit Volume (SEUV):

[0395] SEUV(x,y)=(1 / 2)*E(x,y)*ε(x,y) 2

[0396] Where SEUV(x,y) represents the strain energy per unit volume at the position (x,y), ε(x,y) represents the strain at the position (x,y), and E(x,y) represents the elastic modulus at the position (x,y).

[0397] 7.2 Total Strain Energy (TSE):

[0398] TSE=ΣxΣySEUV(x,y)

[0399] Where TSE represents the total strain energy of the entire tumor area, and SEUV(x,y) represents the strain energy per unit volume at position (x,y).

[0400] 7.3 Strain Energy Density Ratio (SEDR):

[0401] SEDR=SEmax / SEavg

[0402] Where SEmax represents the maximum strain energy density, SEavg represents the average strain energy density, and SEDR represents the strain energy density ratio.

[0403] 8. Multi-scale mechanical parameter expansion

[0404] 8.1 Multi-scale Region Analysis: By changing the size of the observation window, mechanical parameters are calculated from local to global to generate multi-scale features.

[0405] Window scale settings: small scale: 3×3, 5×5, 7×7 pixel windows; medium scale: 11×11, 15×15, 21×21 pixel windows; large scale: 31×31, 41×41, 51×51 pixel windows.

[0406] 8.2 Multi-threshold Mechanical Analysis: For mechanical features that require threshold settings, multiple threshold calculation features are used to generate multi-threshold features.

[0407] Threshold percentile settings: P10, P20, P30, P40, P50, P60, P70, P80, and P90.

[0408] 8.3 Radial Analysis: Perform radial analysis from the tumor center toward the periphery to calculate the distribution of mechanical parameters at different radii:

[0409] ER(r)=(1 / Nr)*ΣθE(r,θ)

[0410] Where ER(r) represents the average elastic modulus at radius r, E(r,θ) represents the elastic modulus at polar coordinates (r,θ), Nr represents the number of pixels at radius r, and θ represents the polar angle.

[0411] In this formula, θ is an angular parameter in a polar coordinate system, representing the angle measured from a reference axis (usually the horizontal). Σθ represents the sum of all points with an angle θ at a radius r. Essentially, this calculation averages the elastic modulus of all points within a circular ring with a radius r and a center of the tumor. This radial analysis can reveal how mechanical properties vary from the core to the edge of a tumor, helping to assess the heterogeneity and boundary characteristics of its internal structure.

[0412] 8.4 Layer Analysis: Divide the tumor area into multiple concentric layers from the inside to the outside, and calculate the mechanical parameters of each layer:

[0413] EL(z)=(1 / Nz)*Σp∈LzE(p)

[0414] Where EL(z) represents the average elastic modulus of the z-th layer, Nz represents the number of pixels in the z-th layer, Lz represents the set of pixels in the z-th layer, and E(p) represents the elastic modulus at pixel p.

[0415] 9. Mechanical feature combination and high-order parameters

[0416] 9.1 Mechanical Ratio Features:

[0417] MRF(A,B)=FeatureA / FeatureB

[0418] Among them, FeatureA and FeatureB represent two basic mechanical features.

[0419] 9.2 Mechanical Product Features:

[0420] MPF(A,B)=FeatureA*FeatureB.

[0421] 9.3 Mechanical Weighted Features:

[0422] MWF(A,w)=Σkwk*Ak

[0423] Where Ak represents the value of the mechanical characteristic A in the kth region, and wk represents the weight of the kth region.

[0424] 9.4 Mechanical Heterogeneity Matrix (MHM):

[0425] MHM(n,m)=|En-Em| / (En+Em)

[0426] where MHM(n,m) represents the mechanical heterogeneity between regions n and m, and En and Em represent the average elastic moduli of regions n and m, respectively.

[0427] In this embodiment, a high-dimensional mechanical feature matrix is ​​constructed: by using the above method, mechanical features with more than 1000 dimensions can be extracted from a single elastic imaging to construct a high-dimensional mechanical feature matrix.

[0428] In this embodiment, the mechanical features are grouped and organized into the following categories:

[0429] Basic mechanical parameter group: about 10-15 features; mechanical statistics parameter group: about 50-70 features; mechanical gradient parameter group: about 100-150 features; mechanical spatial distribution parameter group: about 200-250 features; mechanical topology parameter group: about 150-200 features; mechanical boundary parameter group: about 50-80 features; mechanical strain energy parameter group: about 30-50 features; multi-scale mechanical parameter group: about 200-300 features; mechanical feature combination parameter group: about 150-200 features.

[0430] In this embodiment, after grouping, all features are normalized using the following formula to eliminate dimensional differences:

[0431] x'=(x-μ) / σ

[0432] Among them, x' represents the normalized eigenvalue, x represents the original eigenvalue, μ represents the mean of the feature, and σ represents the standard deviation of the feature.

[0433] In this embodiment, for screening key prognosis-related parameters, the feature selection can adopt the LASSO regression method (not limited to), and the specific steps are as follows:

[0434] Based on the normalized multidimensional tumor mechanical parameters, the range of the regularization parameter λ was set; for each λ value, the coefficient was calculated using LASSO regression; the optimal λ value was selected through cross-validation; and the final selected feature subset was determined based on the optimal λ value.

[0435] In this embodiment, the LASSO regression feature selection objective function is as follows:

[0436] min_β{(1 / N)*Σi[yi-Σj(βj*xij)]2+λ*Σj|βj|}

[0437] Where yi represents the survival time or survival state of the i-th sample, xij represents the j-th mechanical characteristic value of the i-th sample, βj represents the regression coefficient of the j-th characteristic, λ represents the regularization parameter, and N represents the number of samples.

[0438] In this embodiment, in addition to LASSO regression, other feature selection methods may also be used, such as random forest feature importance, minimum redundancy maximum relevance (mRMR), or elastic net.

[0439] In this embodiment, the prognosis prediction model is constructed as follows:

[0440] 1. Cox proportional hazards model:

[0441] h(t)=h0(t)×exp(β1X1+β2X2+,...,+β p X p )

[0442] Among them, h(t) represents the risk function, h0(t) represents the benchmark risk function, β p represents the regression coefficient, X p Represents the mechanical characteristic parameters.

[0443] 2. Random Survival Forest: Build multiple decision trees, each based on randomly selected samples and features; combine the predictions of all decision trees into a final prediction.

[0444] 3. Deep learning survival analysis model: The survival analysis model based on neural networks can automatically learn the complex nonlinear relationship between features. The C-index calculation formula is:

[0445] C-index=P(Tj>Ti|hj>hi,δi=1)=(ΣiΣj I(Tj>Ti)*I(hj>hi)*δi) / (ΣiΣjI(Tj>Ti)*δi)

[0446] Where Tj and Ti are the survival times of patients j and i, respectively; hj and hi are the predicted risk values ​​of patients j and i, respectively; δi represents the event indicator of patient i (1 indicates an event has occurred, 0 indicates censoring); I() is an indicator function, which is 1 when the condition is met and 0 otherwise.

[0447] In this embodiment, K-fold cross validation or bootstrap validation is used to perform internal validation on the prognosis prediction model.

[0448] In this embodiment, the prognostic prediction model is externally validated by using independent cohort data to verify the model performance or multicenter validation.

[0449] In this embodiment, the C-index (consistency index), time-dependent AUC, calibration curve or net reclassification index (NRI) are used as performance evaluation indicators for the prognostic prediction model.

[0450] In this embodiment, a prognosis prediction model is constructed based on the screened key prognosis-related parameters. The construction process of the prognosis prediction model is as follows:

[0451] 1. Model selection: Based on the screened key prognostic parameters, select the appropriate prognostic prediction model type, including: Cox proportional hazard model: h(t)=h0(t)×exp(β1X1+β2X2+,...,+β p X p ); Random Survival Forest Model: constructs multiple decision trees, each based on randomly selected samples and features, and integrates the prediction results of all decision trees into the final prediction; Deep Learning Survival Analysis Model: uses neural networks to automatically learn complex nonlinear relationships between features.

[0452] 2. Parameter optimization: Determine the optimal parameter settings of the model through cross-validation:

[0453] For Cox models: optimize regularization parameters and variable selection strategies;

[0454] For random forests: adjust the number of trees, node splitting criteria, and minimum number of leaf node samples;

[0455] For deep learning models: optimize network structure, learning rate, and regularization parameters.

[0456] 3. Model training: The patient dataset was divided into a training set (70%) and a validation set (30%). The model was trained using the selected key prognostic-related parameters on the training set. For the Cox model, the regression coefficients were calculated using maximum likelihood estimation. For the random forest model, multiple decision trees were trained and the results were integrated. For the deep learning model, the network parameters were optimized using the backpropagation algorithm.

[0457] 4. Model Validation: Assess model stability through K-fold cross-validation (K=5) or bootstrap; calculate performance indicators such as consistency index (C-index), time-dependent AUC, and calibration curve; verify the generalization ability of the model using independent validation sets or multicenter data; and compare performance differences with existing clinical prediction models.

[0458] In this embodiment, the present invention is illustrated by predicting the prognosis of breast cancer patients.

[0459] Data collection: Elastography data of 80 breast cancer patients were collected; clinicopathological information and follow-up data of the patients were recorded.

[0460] Parameter extraction: 1247 mechanical parameters were extracted from elastic imaging data; the parameters included global parameters, local parameters, texture features, spatial distribution features, and topological features.

[0461] Feature selection: LASSO regression was used to select 36 key parameters from 1247 parameters, which mainly reflect different aspects of tumor internal mechanical heterogeneity.

[0462] Model construction: A Cox proportional hazards model was constructed based on the selected 36 parameters, and the model parameters were optimized.

[0463] Model validation: After combining clinical parameters, the C-index of the 5-fold cross-validation was 0.73, and the C-index on the independent validation set was 0.78, which was better than the prognostic model of a single clinical parameter (C-index = 0.68).

[0464] Clinical results: It can accurately distinguish high-risk and low-risk patients (Log-rank test p < 0.001), and the prediction accuracy is not affected by clinical factors such as tumor size, grade, and ER status.

[0465] The present invention is not limited to the following advantages:

[0466] High-dimensional parameter coverage: Through multi-scale expansion such as spatial distribution, gradient, and topology, the parameter dimensions reach over 1,000, far exceeding traditional methods (<10), significantly improving the ability to characterize tumor heterogeneity;

[0467] Improved prognostic accuracy: Validated in a breast cancer dataset (n=300) for predicting 5-year survival rates, the C-index reached 0.75 (compared to 0.68 for the traditional method), demonstrating significant improvement in prediction accuracy.

[0468] Individualized prognostic assessment: It can provide individualized prognostic assessment based on the patient's tumor mechanical properties, providing a more accurate reference for clinical decision-making.

[0469] Example 2

[0470] like Figure 2 As shown, the present invention provides a tumor mechanics multi-parameter prognosis prediction method of the tumor mechanics multi-parameter prognosis prediction system based on elastic imaging as described in Example 1, and its implementation method is as follows:

[0471] Acquire elastic imaging data of the patient's tumor;

[0472] Preprocessing the collected elastic imaging data;

[0473] Extract multidimensional tumor mechanical parameters from preprocessed elastography data and perform feature normalization;

[0474] Screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters;

[0475] Construct a prognostic prediction model based on the screened key prognostic-related parameters;

[0476] The prognostic prediction model is applied to new patients to predict their prognostic risk, and the tumor mechanics multi-parameter prognostic prediction results are displayed in a visual manner.

[0477] like Figure 2 The tumor mechanics multi-parameter prognosis prediction method provided in the illustrated embodiment can implement the technical solution shown in the tumor mechanics multi-parameter prognosis prediction system in the above-mentioned system embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0478] In this embodiment, those skilled in the art should easily realize that, in combination with the various schematic units and algorithm steps described in the embodiments disclosed in the present invention, the present invention can be implemented in the form of hardware and / or a combination of hardware and computer software. Whether a function is executed in a hardware- or computer software-driven manner depends on the specific application and design constraints of the technical solution. Different methods can be used for each specific application to implement the described function, but such implementation should not be considered to be beyond the scope of this application.

Claims

1. A multi-parameter tumor mechanics prognosis prediction system based on elastic imaging, characterized in that: include: an elastic imaging acquisition module, used to collect elastic imaging data of a patient's tumor; An image preprocessing module, used for preprocessing the collected elastic imaging data; Parameter extraction module, used to extract multidimensional tumor mechanical parameters from preprocessed elastic imaging data and perform feature normalization processing; Feature selection module, used to screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters; A prognosis prediction model building module is used to build a prognosis prediction model based on the screened key prognosis-related parameters; The result display module is used to apply the prognostic prediction model to visually display the tumor mechanics multi-parameter prognostic prediction results.

2. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 1, characterized in that: The parameter extraction module includes: A first processing unit is used to extract basic mechanical parameters; A second processing unit is used to calculate mechanical statistical parameters, gradient parameters, spatial distribution parameters, topological parameters, boundary parameters and strain energy parameters based on basic mechanical parameters; The third processing unit is used to expand the dimension of mechanical parameters through multi-scale analysis, multi-threshold analysis and feature combination to obtain multi-scale mechanical parameters, mechanical feature combinations and high-order parameters; The fourth processing unit is used to construct a high-dimensional mechanical characteristic matrix based on the multidimensional mechanical parameters extracted by the first processing unit, the second processing unit and the third processing unit, wherein the high-dimensional mechanical characteristic matrix is ​​used to group the mechanical parameters according to categories and normalize all the grouped mechanical parameters.

3. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 2, characterized in that: The basic mechanical parameters include: Elastic modulus E: E=σ / ε Among them, σ represents stress and ε represents strain; Shear modulus G: G = τ / γ Where τ represents shear stress and γ represents shear strain; Poisson's ratio ν: ν=-εtrans / εaxial Among them, εtrans represents the transverse strain and εaxial represents the axial strain; Young's modulus YM: YM=2G(1+ν); Bulk modulus K: K = E / (3 (1-2ν)); Compression coefficient CF: CF=1 / K.

4. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 3, characterized in that: The mechanical statistical parameters include basic mechanical statistical parameters and quantile mechanical parameters; the basic mechanical statistical parameters include: The mean value μ of the elastic modulus E is: μ=(1 / N)*Σ(Ep) Where N represents the total number of pixels, and Ep represents the elastic modulus value of the p-th pixel; Standard deviation σ of elastic modulus E: Coefficient of variation CV: CV = σ / μ; Peak threshold ratio PTR(t): PTR(t)=Epeak / Et Wherein, Epeak represents the peak value of the mechanical parameter, and Et represents the value of the mechanical parameter at the threshold value t; The quantile mechanical parameters include: Mechanical percentile MPn: the value of the nth percentile after sorting the elastic modulus values ​​from small to large, where MPn represents the elastic modulus E of the nth percentile; Mechanical interquartile range MIQR: MIQR=MP75-MP25 Among them, MP75 represents the 75th percentile elastic modulus, and MP25 represents the 25th percentile elastic modulus; Elasticity Heterogeneity Index EHI: EHI=(MP95-MP5) / MP50 Here, MP95 represents the 95th percentile elastic modulus, MP5 represents the 5th percentile elastic modulus, and MP50 represents the median elastic modulus.

5. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 4, characterized in that: The gradient parameters include: Local mechanical gradient LMG(x,y): Where LMG(x,y) represents the local mechanical gradient at the position (x,y), E(x,y) represents the elastic modulus E at the position (x,y), and denote the partial derivatives of the elastic modulus E in the x and y directions, respectively. represents the gradient operator; Mean Mechanical Gradient (MMG): MMG=(1 / N)*ΣxΣyLMG(x,y) Wherein, MMG represents the mean mechanical gradient of the entire tumor area; Mechanical gradient variance MGV: MGV=(1 / N)*ΣxΣy[LMG(x,y)-MMG]2; Mechanical gradient direction MGD(x,y): Where MGD(x,y) represents the direction of the mechanical gradient at the position (x,y), and arctan[] represents the inverse tangent function; Mechanical Directional Consistency (MDC): MDC=|ΣxΣyexp(i”*MGD(x,y))| / N Here, “i” represents an imaginary unit.

6. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 5, characterized in that: The spatial distribution parameters include the mechanical spatial autocorrelation index, the mechanical regional clustering index and the mechanical semivariogram function; the mechanical spatial autocorrelation index includes: Moran Index I: I=[N / Σi'Σj'wi'j']*[Σi'Σj'wi'j'(Ei'-μ)(Ej'-μ)] / [Σi'(Ei'-μ)2] Where wi'j' represents the spatial weight between position i' and position j', Ei' and Ej' represent the elastic modulus E at position i' and position j', respectively; G-Force Index C: C=[(N-1) / 2Σhead]*[Ei'-Ej') 2 ] / [Σi(Ei−µ) 2 ]: Mechanical regional clustering index MACI(t): MACI(t)=Σr[Ar(t) / Atotal]2 Wherein, MACI(t) represents the regional clustering index under the mechanical threshold t, Ar(t) represents the area of ​​the rth connected region under the threshold t, and Atotal represents the total tumor area; Mechanical semivariogram function MSV(h): MSV(h)=(1 / 2N(h))*Σi'Σj'[E(xi')-E(xj')] 2 Where MSV(h) represents the mechanical semivariance with a spatial distance of h, N(h) represents the number of sample point pairs with a spatial distance of h, ∑ and ∑ represent the double summation of all point pairs (i', j') that satisfy the distance |xi'-xj'| = h, E(xi') and E(xj') represent the elastic modulus values ​​at positions xi' and xj', respectively. The topological parameters include mechanical contour features, mechanical connected domain features and mechanical fractal dimensions; The mechanical contour features include: The number of contour lines, which is the number of contour lines NI(t1,t2) in the elasticity value range [t1,t2], where t1 and t2 represent the elasticity value thresholds; The contour length is the total length IL(t) of the contour lines with elasticity value t, where t represents the elasticity threshold; The mechanical connectivity domain features include: The number of connected domains NCC(t), NCC(t) represents the number of connected domains with elasticity values ​​greater than the threshold t; Maximum connected domain area ratio: LCCAR(t)=Amax(t) / Atotal Among them, LCCAR(t) represents the ratio of the maximum connected domain area with elasticity value greater than threshold t to the total area, and Amax(t) represents the maximum connected domain area with elasticity value greater than threshold t. The mechanical fractal dimension MFD(t): MFD(t)=lim(ε→0)[log(N(ε,t)) / log(1 / ε)] Where MFD(t) represents the mechanical fractal dimension of the region with elasticity greater than the threshold t, N(ε,t) represents the number of squares with a side length of ε required to cover the region with elasticity greater than the threshold t, and ε represents the side length of the square. The boundary parameters include: Boundary Mechanical Gradient: Where BMG represents the mean mechanical gradient at the tumor boundary, represents the mechanical gradient at the boundary point b, and Nb represents the total number of boundary points; Boundary mechanics gradient variance: Among them, BMGV represents the variance of the mechanical gradient on the tumor boundary; Boundary Mechanics Sharpness Index BMSI: BMSI = (Ein - Eout) / d Where Ein represents the average elastic modulus near the inner boundary of the tumor, Eout represents the average elastic modulus near the outer boundary of the tumor, and d represents the average distance between the inner and outer measurement points; The strain energy parameters include: Strain energy per unit volume: SEUV(x,y)=(1 / 2)*E(x,y)*ε(x,y) 2 Where SEUV(x,y) represents the strain energy per unit volume at the position (x,y), and ε(x,y) represents the strain at the position (x,y). Total strain energy: TSE=ΣxΣySEUV(x,y) Where TSE represents the total strain energy of the entire tumor area, SEUV(x,y) represents the unit volume strain energy at position (x,y); Strain energy density ratio SEDR: SEDR=SEmax / SEavg Where SEmax represents the maximum strain energy density, and SEavg represents the average strain energy density.

7. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 5, characterized in that: The multi-scale mechanical parameters include: Multi-scale regional analysis: By changing the size of the observation window, mechanical parameters are calculated from local to global scales to generate multi-scale features; Multi-threshold mechanical analysis: For mechanical parameters that require thresholds, multiple thresholds are used to calculate features and generate multi-threshold features; Radial analysis: Perform radial analysis from the tumor center to the periphery to calculate the distribution of mechanical parameters at different radii: ER(r)=(1 / Nr)*ΣθE(r,θ) Where ER(r) represents the average elastic modulus at radius r, E(r,θ) represents the elastic modulus at polar coordinates (r,θ), Nr represents the number of pixels at radius r, and θ represents the polar angle; Layered analysis: The tumor area is divided into multiple concentric layers from the inside to the outside, and the mechanical parameters of each layer are calculated: EL(z)=(1 / Nz)*Σp∈LzE(p) Where EL(z) represents the average elastic modulus of the z-th layer, Nz represents the number of pixels in the z-th layer, Lz represents the set of pixels in the z-th layer, and E(p) represents the elastic modulus at pixel p.

8. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 7, characterized in that: The mechanical feature combination and high-order parameters include: Mechanical ratio feature MRF(A,B): MRF(A,B)=FeatureA / FeatureB Among them, FeatureA and FeatureB represent two basic mechanical features; Mechanical product characteristic MPF(A,B): MPF(A,B)=FeatureA*FeatureB; Mechanical weighted feature MWF(A,w): MWF(A,w)=Σkwk*Ak Where Ak represents the value of the mechanical characteristic A in the kth region, and wk represents the weight of the kth region; Mechanical heterogeneity matrix: MHM(n,m)=|En-Em| / (En+Em) where MHM(n,m) represents the mechanical heterogeneity between regions n and m, and En and Em represent the average elastic moduli of regions n and m, respectively.

9. The tumor mechanics multi-parameter prognostic prediction system based on elastic imaging according to claim 2, characterized in that: The grouping of mechanical parameters includes basic mechanical parameter group, mechanical statistical parameter group, gradient parameter group, spatial distribution parameter group, topological parameter group, boundary parameter group, strain energy parameter group, multi-scale mechanical parameter group and mechanical feature combination group.

10. A tumor mechanics multi-parameter prognosis prediction method of the tumor mechanics multi-parameter prognosis prediction system based on elastic imaging according to any one of claims 1 to 9, characterized in that: Multi-parameter prognostic prediction methods based on tumor mechanics include: Acquire elastic imaging data of the patient's tumor; Preprocessing the collected elastic imaging data; Extract multidimensional tumor mechanical parameters from preprocessed elastography data and perform feature normalization; Screen key prognostic parameters based on normalized multidimensional tumor mechanical parameters; Construct a prognostic prediction model based on the screened key prognostic-related parameters; The prognostic prediction model is applied to new patients to predict their prognostic risk, and the tumor mechanics multi-parameter prognostic prediction results are displayed in a visual manner.

Images