Meniscus internal tearing quantitative evaluation method and system based on ultrasonic shear wave elastography

Abstract

The invention discloses an ultrasonic shear wave elastography-based intrameniscus tearing quantitative evaluation method and system, and relates to the technical field of computer-aided diagnosis. The method is used for solving the problem of insufficient quantification of meniscus tearing detection. The method comprises the following steps: firstly, fixing a knee joint at a preset buckling angle, exciting shear waves along an annular fiber direction and a vertical direction, and collecting ultrasonic radio-frequency signals of an affected side meniscus and an uninjured side meniscus in different load states; then, a three-dimensional anisotropic elastic atlas is constructed based on the shear wave propagation velocity field, and spatial elastic distribution characteristics of the tissue are obtained; thirdly, calculating an elastic continuity index, a spatial variation coefficient and an abnormal region morphological parameter, and quantitatively representing the continuity and local tearing characteristics of the tissue structure; and finally, inputting the multi-dimensional features into a multi-branch neural network, fusing the features through a cross-branch attention mechanism, and outputting a meniscus tearing quantitative evaluation value, thereby realizing intelligent and objective evaluation of the tearing degree, and providing a basis for clinical diagnosis and postoperative rehabilitation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer-aided diagnosis, in particular to a meniscus internal tear quantitative evaluation method and system based on ultrasonic shear wave elastography. BACKGROUND

[0002] Meniscus tear, as one of the most common types of knee joint injury, has become an important problem in clinical diagnosis and rehabilitation evaluation. Meniscus plays a key role in the stability and cushioning of the knee joint, and its structural damage not only causes pain and movement disorders, but also accelerates cartilage degeneration and osteoarthritis progression. Modern clinical diagnosis has gradually shifted from static morphological recognition to dynamic functional imaging, and doctors need to understand not only the changes in meniscus morphology, but also its mechanical response and tissue continuity characteristics under stress. How to accurately and quantitatively evaluate the biomechanical changes inside the meniscus under non-invasive conditions has become an important research direction in sports medicine and imaging.

[0003] The current clinical imaging diagnosis methods mainly include magnetic resonance imaging and traditional ultrasonic structural imaging. Magnetic resonance imaging can show meniscus morphology and signal abnormalities, but its image features mainly come from tissue water and signal intensity changes, which cannot truly reflect the mechanical state and fiber structure integrity of the tissue, and have low sensitivity to the biomechanical response under dynamic load. Traditional ultrasound has real-time and convenient operation, but the imaging results mainly depend on the acoustic impedance difference, and the mechanical abnormalities inside the meniscus tissue are not fully reflected, which is easily affected by the probe angle and operator experience, resulting in poor diagnostic consistency. The existing ultrasonic elastography technology often only considers one-way shear wave propagation, ignoring the complex anisotropic fiber arrangement characteristics of the meniscus, which makes the elastic distribution results deviate in direction and be discontinuous in space, making it difficult to accurately quantify the tear range, direction and mechanical abnormality degree, thereby limiting its clinical application value in meniscus function evaluation. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a meniscus internal tear quantitative evaluation method and system based on ultrasonic shear wave elastography, which solves the problems in the above background art.

[0005] To achieve the above object, the application is implemented by the following technical solutions: A meniscus internal tear quantitative evaluation method based on ultrasonic shear wave elastography, comprising the following steps: S1. Fix the knee joint at a preset flexion angle, excite shear waves along the circumferential fiber direction and the vertical direction of the meniscus by a high-frequency linear array ultrasonic probe, and collect ultrasonic radio frequency signals of the affected side and the healthy side meniscus under the conditions of no load and standardized axial load; S2. According to the ultrasonic radio frequency signals, the propagation velocity field of the shear wave in the meniscus tissue is calculated by time domain analysis, and the elastic modulus distribution maps along the circumferential fiber direction and the vertical direction are established respectively, the anisotropic elastic ratio is calculated by comparing the elastic modulus values in the two directions, and the three-dimensional anisotropic elastic atlas is constructed by combining the elastic modulus changes under the load state; S3. Based on the three-dimensional anisotropic elastic atlas, the elastic continuity index along the fiber direction is obtained by calculating the elastic modulus gradient change of adjacent regions along the fiber direction, the spatial variation coefficient of the anisotropic elastic ratio is obtained by statistically analyzing the dispersion degree of the anisotropic elastic ratio in the whole meniscus region, and the morphological parameters including the boundary curvature and the shape irregularity are obtained by analyzing the geometric characteristics of the abnormal elastic region; S4. The elastic continuity index, the spatial variation coefficient and the morphological parameters are respectively input into three feature processing branches of a multi-branch neural network for feature processing, the feature representations output by each branch are weighted and fused through a cross-branch attention mechanism, and a meniscus tear quantitative evaluation value is output.

[0006] Further, step S1 comprises the following steps: fixing the knee joint at a preset flexion angle, exposing the meniscus body to the effective detection range of the ultrasonic probe, positioning the probe direction along the circumferential fiber direction and the vertical direction of the meniscus by the high-frequency linear array ultrasonic probe, ensuring that the shear wave excitation direction has a specific spatial relationship with the fiber structure, collecting reference ultrasonic radio frequency signals under the condition of no load, recording the initial elastic properties of the meniscus, applying a preset standardized axial load to simulate the weight-bearing state of the knee joint, and collecting ultrasonic radio frequency signals under the load state; the above measurement process is repeated for the affected side and the healthy side meniscus to ensure the consistency of the measurement conditions.

[0007] Further, according to the ultrasonic radio frequency signals, the propagation velocity field of the shear wave in the meniscus tissue is calculated by time domain analysis, and the specific process of establishing the elastic modulus distribution maps along the circumferential fiber direction and the vertical direction is as follows: the collected ultrasonic radio frequency signals are subjected to beam synthesis and signal enhancement processing, the time domain characteristics of the shear wave propagation are extracted, the propagation process of the shear wave in the meniscus tissue is tracked by a motion estimation algorithm, and the propagation velocity of each point is calculated; according to the constitutive relationship between the shear wave propagation velocity and the elastic modulus, the elastic modulus values along the circumferential fiber direction and the vertical direction are calculated respectively, and the elastic modulus distribution maps along the circumferential fiber direction and the vertical direction are synchronously generated through spatial interpolation and gridding processing.

[0008] Further, the specific process of calculating the anisotropic elasticity ratio by comparing the elastic modulus values in two directions is as follows: the anisotropic elasticity ratio matrix is established by calculating the elastic modulus ratio of the annular fiber direction to the vertical direction point by point, the change characteristics of the elastic modulus under the load state are analyzed, and the load response coefficient is calculated; the anisotropic elasticity ratio matrix and the load response coefficient are spatially registered and feature fused, and the anisotropic elasticity distribution atlas containing spatial position information is constructed through a three-dimensional reconstruction algorithm; the elastic characteristics under different load states are integrated to form a three-dimensional anisotropic elasticity atlas of dynamic load response.

[0009] Further, the specific process of obtaining the elastic continuity index of the annular fiber direction by calculating the elastic modulus gradient change between adjacent regions along the fiber direction is as follows: based on the three-dimensional anisotropic elasticity atlas, the reference path of the fiber direction is determined, the continuous elastic modulus sampling point sequence is extracted along the path, the elastic modulus gradient change between adjacent sampling points is calculated, and the distribution curve of the elastic modulus along the fiber direction is established; the gradient fluctuation characteristics of each section in the distribution curve are statistically analyzed through a sliding window, the standard deviation and coefficient of variation of the gradient in the window are calculated, the gradient statistical characteristics of each window are integrated, and the elastic continuity index reflecting the integrity of the fiber structure is constructed.

[0010] Further, the specific process of obtaining the spatial variation coefficient of the anisotropic elasticity ratio by statistically analyzing the dispersion degree of the anisotropic elasticity ratio in the entire meniscus region is as follows: based on the three-dimensional anisotropic elasticity atlas, the anisotropic elasticity ratio of each spatial position in the meniscus region is extracted, the spatial distribution matrix of the anisotropic elasticity ratio is constructed, the statistical characteristics of the spatial distribution matrix are analyzed, and the local range and spatial autocorrelation index of the anisotropic elasticity ratio are calculated; the dispersion degree of the anisotropic elasticity ratio in the entire meniscus region is quantified through variogram analysis, and the spatial variation coefficient of the anisotropic elasticity ratio is constructed in combination with the local statistical characteristics and the overall dispersion degree.

[0011] Further, the specific process of simultaneously analyzing the geometric characteristics of the abnormal elasticity region to obtain the morphological parameters including the boundary curvature and shape irregularity is as follows: based on the three-dimensional anisotropic elasticity atlas, the abnormal elasticity region is identified through a region growing algorithm, the complete boundary profile of the region is extracted, the local curvature of each point on the boundary profile is calculated, the irregular characteristics of the boundary are analyzed, and the curvature distribution atlas is established; the complexity of the region shape is quantified through polygon approximation, the shape irregularity index is calculated, and the complete set of morphological parameters is constructed by integrating the boundary curvature distribution characteristics and the shape irregularity index.

[0012] Further, the specific process of inputting the elastic continuity index, the spatial variation coefficient, and the morphological parameter into three feature processing branches of the multi-branch neural network for feature processing is as follows: the elastic continuity index is input into the first feature processing branch, the local sequence pattern feature of the fiber structure is extracted through a one-dimensional convolution layer, and the elastic continuity change law along the fiber direction is captured through a long short-term memory network; the spatial variation coefficient is input into the second feature processing branch, the spatial topological relationship of the anisotropic elastic ratio is constructed through a graph convolution network, and the variation features of the key region are aggregated by using an attention pooling layer; the morphological parameter is input into the third feature processing branch, the geometric feature of the boundary contour is extracted through a two-dimensional convolution layer, and the multi-scale shape irregularity feature is captured by using a spatial pyramid pooling; the fiber continuity feature, the spatial heterogeneity feature, and the geometric morphological feature extracted by the deep network are output by the respective feature processing branches.

[0013] Further, the specific process of outputting the meniscus tear quantitative evaluation value by weighting and fusing the feature representations output by the branches through the cross-branch attention mechanism is as follows: the fiber continuity feature, the spatial heterogeneity feature, and the geometric morphological feature output by the three feature processing branches are spliced to construct a multi-dimensional feature tensor, the importance weight of each feature channel is calculated through the cross-branch attention mechanism, wherein the query vector comes from the fiber continuity feature, and the key-value pair comes from the spatial heterogeneity feature and the geometric morphological feature; the attention weight is normalized by using a softmax function to realize the enhancement of important features and the suppression of unimportant features; the weighted feature representation is input into a fully connected layer for dimension reduction fusion, and a sigmoid activation function is used to output a meniscus tear quantitative evaluation value in the range of 0 to 1, which comprehensively reflects the fiber structure integrity, the tissue heterogeneity, and the lesion morphological feature.

[0014] The application discloses a kind of based on ultrasonic shear wave elastography meniscus internal tear quantitative evaluation system, including the following modules: data acquisition module, for fixing knee joint at preset flexion angle, by high-frequency linear array ultrasonic probe respectively along the direction of meniscus annular fiber and vertical direction excitation shear wave, collection side and healthy side meniscus under the condition of no load and standard axial load state ultrasonic radio frequency signal;Elasticity atlas construction module, for according to ultrasonic radio frequency signal, by time domain analysis calculation shear wave in meniscus tissue propagation velocity field, respectively establish along the direction of annular fiber and vertical direction elasticity modulus distribution graph, by comparing the elasticity modulus value of two directions calculation anisotropic elasticity ratio, combined with the elasticity modulus change under load state constructs three-dimensional anisotropic elasticity atlas;Characteristic parameter extraction module, for based on three-dimensional anisotropic elasticity atlas, by calculating the elasticity modulus gradient change of adjacent area along fiber direction obtains the elasticity continuity index of annular fiber direction, by statistics the discrete degree of anisotropic elasticity ratio in whole meniscus region obtains the spatial variation coefficient of anisotropic elasticity ratio, simultaneously by analyzing the geometric characteristics of abnormal elastic region obtains including boundary curvature and shape irregularity in morphological parameter;Quantitative evaluation module, for elasticity continuity index, spatial variation coefficient, morphological parameter are respectively input to the three feature processing branches of multi-branch neural network and are carried out feature processing, by cross-branch attention mechanism to the feature representation output by each branch is weighted fusion, output meniscus tear quantitative evaluation value.

[0015] The application has the following beneficial effects:

[0016] (1) A kind of based on ultrasonic shear wave elastography meniscus internal tear quantitative evaluation method, by respectively exciting shear wave in annular fiber direction and vertical direction, and under the condition of no load and standard axial load state synchronous acquisition ultrasonic radio frequency signal, realize the mechanical response detection of meniscus tissue in multiple directions, multiple states.This mode breaks through the limitation of traditional one-way shear wave measurement, can comprehensively reflect the complex anisotropic structural characteristics of meniscus.By time domain analysis calculation shear wave propagation velocity field, and establish the elasticity modulus distribution graph of different directions, the application can accurately quantify the directional elasticity difference of meniscus tissue.Further combined with the elasticity modulus change under load state, three-dimensional anisotropic elasticity atlas is constructed, so that the visualization and fine characterization of meniscus elasticity distribution is realized in space, effectively improve the detection sensitivity of early micro tear and mechanical abnormal region.

[0017] (2) An intra-meniscus tear quantitative evaluation system based on ultrasonic shear wave elastography, on the basis of a three-dimensional anisotropic elasticity map, an elastic continuity index of the annular fiber direction is obtained by calculating the elastic modulus gradient change, and the uniformity of the elastic distribution of the tissue is characterized by the spatial variation coefficient of the anisotropic elasticity ratio, and the boundary curvature and morphological irregularity of the abnormal area are extracted as geometric features, which provides a basis for accurately depicting the damage range and morphological abnormalities of the tissue. The above elastic continuity, spatial variation and morphological parameters are input into a multi-branch neural network for feature fusion analysis, and a cross-branch attention mechanism is used to realize the weighted integration of elastic features of different dimensions, and output a quantitative tear evaluation value. The present application can simultaneously capture the coupling changes of the structure and mechanical characteristics of the meniscus, improve the objectivity and consistency of the tear degree evaluation, and provide a quantitative diagnostic index with more physical meaning and functional correlation for clinical use.

[0018] Of course, implementing any product of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of a meniscus tear quantitative evaluation method based on ultrasonic shear wave elastography according to the present application.

[0020] Figure 2 A flowchart of a meniscus tear quantitative evaluation system based on ultrasonic shear wave elastography according to the present application. DETAILED DESCRIPTION

[0021] The present application embodiment solves the problems of difficulty in quantitatively identifying small tears in the meniscus, inability to accurately characterize anisotropic mechanical characteristics, and poor direction sensitivity of the results of elastic map analysis in the prior art by a meniscus tear quantitative evaluation method and system based on ultrasonic shear wave elastography.

[0022] The general idea of the scheme in the present application embodiment is as follows:

[0023] A high-frequency linear array ultrasonic probe is used to excite shear waves in the annular fiber direction and the vertical direction, ultrasonic radio frequency signals of the meniscus under no load and standard axial load are collected, and the shear wave propagation velocity field is calculated by time domain analysis to establish the elastic modulus distribution in different directions. Further, anisotropic elasticity ratios are calculated to construct a three-dimensional anisotropic elasticity map, realizing multi-dimensional visualization of the spatial distribution of the meniscus tissue elasticity. Based on the map, an elastic continuity index reflecting fiber continuity, a spatial variation coefficient reflecting tissue non-uniformity, and geometric feature parameters reflecting tear morphology are extracted. Finally, the above multi-source elastic and morphological features are input into a multi-branch neural network, and a cross-branch attention mechanism is used for feature fusion to output a quantitative evaluation value of the meniscus tear.

[0024] Please refer to Figure 1 The embodiment of the present application provides a technical scheme: a meniscus internal tear quantitative evaluation method based on ultrasonic shear wave elastography, comprising the following steps: S1. fixing the knee joint at a preset flexion angle, exciting shear waves along the meniscus annular fiber direction and the vertical direction through a high-frequency linear array ultrasonic probe, and collecting ultrasonic radio frequency signals of the affected side and the healthy side meniscus under the no-load state and the standardized axial load state; S2. according to the ultrasonic radio frequency signals, calculating the shear wave propagation velocity field in the meniscus tissue through time domain analysis, establishing the elastic modulus distribution map along the annular fiber direction and the vertical direction respectively, calculating the anisotropic elastic ratio value by comparing the elastic modulus values of the two directions, and constructing a three-dimensional anisotropic elastic atlas combined with the elastic modulus change under the load state; S3. based on the three-dimensional anisotropic elastic atlas, obtaining the elastic continuity index of the annular fiber direction by calculating the elastic modulus gradient change along the fiber direction of the adjacent region, obtaining the spatial variation coefficient of the anisotropic elastic ratio value by statistically analyzing the dispersion degree of the anisotropic elastic ratio value in the whole meniscus region, and simultaneously obtaining the morphological parameters including the boundary curvature and the shape irregularity by analyzing the geometric characteristics of the abnormal elastic region; S4. inputting the elastic continuity index, the spatial variation coefficient and the morphological parameters into three feature processing branches of a multi-branch neural network respectively for feature processing, weighting and fusing the feature representations output by each branch through a cross-branch attention mechanism, and outputting the meniscus tear quantitative evaluation value.

[0025] In this embodiment, S1: Obtain the original physical quantity (ultrasound radio frequency signal) reflecting the meniscus under different mechanical directions and mechanical states, to provide reliable basic data for subsequent wave speed-based elastic inversion and anisotropy modeling. Collecting both the affected side and the healthy side is conducive to individualized comparative analysis. Fix the knee joint at a predetermined flexion angle to standardize the detection geometry; use a high-frequency linear array ultrasound probe to excite shear waves along two sets of directions: one is along the ring-shaped fiber direction of the meniscus (tangential direction), and the other is nearly perpendicular to the ring-shaped fiber direction (radial direction). Collect data under two mechanical states: no load state and standardized axial load state (the latter is realized by a predetermined loading device or body position control), and record the ultrasound radio frequency (RF) signals of the affected side and the healthy side under each state. Output: Time-domain radio frequency signal sequence of each direction, each mechanical state, and each side (contains original amplitude and phase information for time-domain wave speed extraction). High-frequency linear array ultrasound probe: refers to a linear array crystal, a probe with a relatively high working frequency, which provides high axial and lateral resolution, facilitating the excitation of shear waves in the superficial structure of the meniscus. Ultrasound radio frequency signal (RF): original echo signal without envelope processing, containing phase information, suitable for time-domain or phase-related speed estimation, containing more physical quantities than conventional B-mode grayscale images, conducive to accurate wave speed measurement. Standardized axial load: in order to make the measurements comparable, the axial pressure or displacement is uniformly set or calibrated under loading conditions, to ensure the repeatability of the elastic changes induced by the load. S2: Convert the radio frequency signal into physically interpretable elastic parameters, obtain the elastic modulus distribution in the ring direction and the vertical direction respectively, and fuse the load information to construct a three-dimensional anisotropic elastic atlas to reflect the mechanical differences in space and state. Perform time-domain processing (such as group delay estimation, cross-correlation or phase decomposition) on the radio frequency signal obtained in S1 to calculate the propagation time or speed field of the shear wave in the local tissue. Convert the local shear wave speed to elastic modulus (elastic modulus as a direct scalar representation of mechanical properties) through the relationship between material mechanics to obtain two-dimensional or voxelized elastic distribution maps along the fiber direction and the vertical direction. Combine the elastic distribution under the no-load and axial load states to construct a three-dimensional anisotropic elastic atlas: three-dimensional here refers to the distribution in the spatial three-dimensional coordinates, and also includes the coupling of the two dimensions of direction and mechanical state. Time-domain analysis of shear waves: a method of calculating the wave speed distribution by analyzing the time difference and phase change of the shear wave arriving at different receiving positions over time. Compared with only using frequency domain methods, time domain can directly use the radio frequency phase information to improve the local resolution. Elastic modulus: a physical quantity that measures the material's resistance to deformation, and the local elastic modulus of the meniscus reflects the tissue stiffness; in this method, the wave speed-elasticity relationship is used to map from speed to modulus. Anisotropy: the property of a material that exhibits different mechanical properties in different directions. The fiber arrangement of the meniscus leads to differences in elasticity along the fiber and perpendicular to the fiber directions, and this invention explicitly measures and models this difference.S3: Extract physical and geometric indices from the three-dimensional anisotropic elasticity map that can be used to quantify the tear, which not only reflects the mechanical continuity of the fiber direction, but also describes the spatial non-uniformity and morphological abnormalities, thereby converting micro-mechanical heterogeneity into descriptive features for machine learning discrimination. Elastic continuity index: Calculate the gradient or difference of elastic modulus between adjacent voxels / regions along the circular fiber direction, and construct an index to characterize local continuity (or fracture) by counting the distribution of gradient amplitude. This index is highly sensitive to sudden elastic jumps along the fiber direction and can indicate fiber integrity damage. Spatial variation coefficient of anisotropic elastic ratio: Calculate the distribution of the ratio of elastic modulus along two directions for the entire meniscus region, and evaluate the degree of fluctuation of the ratio in space using the variation coefficient or dispersion measure, reflecting the damage degree of tissue mechanical consistency. Morphological parameters (boundary curvature, shape irregularity, etc.): Perform connected component analysis on the detected abnormal elastic regions, and extract their boundary curvature, perimeter / area ratio, shape irregularity, etc. These features help to distinguish between long strip-shaped abnormalities and diffuse degenerative abnormalities. S4: Efficiently fuse multi-modal features obtained from different physical dimensions (elastic continuity, spatial statistics, geometric morphology) through learning models, and finally output meniscus tear quantitative evaluation values (quantitative scores or classification probabilities, etc.) that can be directly used for clinical interpretation. This step realizes the mapping from features to decisions and improves the robustness and interpretability of diagnosis. Use a multi-branch neural network structure: Each branch is dedicated to a type of feature (elastic continuity index branch, spatial variation branch, morphological branch) for specialized feature transformation and coding (e.g. multi-layer perceptron, one-dimensional / two-dimensional convolution or graph neural network, depending on the feature structure). Introduce cross-branch attention mechanism: Attention-weight the output representations of the three branches to adaptively emphasize important modalities or regions related to the task, thereby suppressing noisy features and improving fusion results. After fusion, output the final tear quantitative evaluation value through several fully connected layers or regression / classification heads, while retaining intermediate features for interpretability analysis (e.g. feature importance mapping). Multi-branch neural network: refers to the parallel existence of multiple sub-network branches in the network structure, each branch focusing on processing a modality or feature set, which helps to preserve the specialized representation within the modality. Cross-branch attention mechanism: a learned weight distribution strategy that allows the model to automatically adjust the contribution of each branch output to the final decision based on the input instance, thereby achieving adaptive collaboration between modalities.

[0026] Specifically, step S1 includes the following steps: fixing the knee joint at a preset flexion angle, exposing the meniscus body to the effective detection range of the ultrasound probe, positioning the probe direction along the annular fiber direction and vertical direction of the meniscus using a high-frequency linear array ultrasound probe to ensure that the shear wave excitation direction has a specific spatial relationship with the fiber structure; acquiring a reference ultrasound radio frequency signal under no-load conditions, recording the initial elastic characteristics of the meniscus, applying a preset standardized axial load to simulate the knee joint's load-bearing state, and acquiring ultrasound radio frequency signals under load conditions; repeating the above measurement process for the affected and healthy menisci to ensure consistency of measurement conditions.

[0027] In this implementation scheme, the knee joint is initially fixed at a preset flexion angle to position the meniscus body within the optimal acoustic window of the ultrasound probe, thereby reducing acoustic interference between soft tissue and bony structures. The flexion angle is typically determined based on biomechanical studies and generally falls within the range of 20°–30° to ensure visualization of the annular fibrous structure and stable stress state. A high-frequency linear array ultrasound probe (typically at 10–15 MHz) is used, whose high-resolution imaging capability can distinguish the fine structures within the meniscus. By arranging the probes along the annular fiber direction (i.e., the direction of the main collagen fibers of the meniscus) and perpendicularly, shear waves are excited in two orthogonal directions. The propagation speed of this lateral wave in the tissue is significantly affected by the material stiffness, thus reflecting local elastic differences in the tissue. Shear wave elastography is a technique that uses ultrasound to excite local mechanical disturbances and tracks the propagation of shear waves in real time. Its basic principle is: the faster the wave speed, the "harder" the tissue; the slower the wave speed, the "softer" the tissue. In meniscus tissue, the fiber orientation exhibits significant anisotropy, meaning that the mechanical response differs in different directions. Therefore, bidirectional excitation is required to capture the complete elastic properties. Next, a baseline ultrasound radio frequency signal (RF signal) is acquired under no-load conditions to record the original elastic properties of the meniscus. The RF signal is the most raw time-domain waveform data received by the ultrasound transducer. Compared to conventional grayscale images, it retains the amplitude and phase information of sound wave propagation, which can be used for subsequent precise wave velocity calculations and tissue elasticity inversion. A standardized axial load is then applied to simulate the physiological and mechanical conditions of the knee joint under load. Ultrasound radio frequency signals are acquired again at this time to observe the elastic response to load changes. This dynamic load-response comparison reveals the elastic adaptability of the meniscus under functional stress, which is particularly sensitive for early detection of microtears. Simultaneously, performing the above process on the affected and healthy sides separately provides control data, eliminating the influence of individual differences and ensuring the statistical reliability and clinical interpretability of the subsequently calculated elasticity difference indices. The purpose of step S1 is to construct a bidirectional, multi-state, repeatable ultrasonic shear wave excitation and signal acquisition system, providing high-quality input data for subsequent elastic modulus calculation and three-dimensional anisotropy modeling.

[0028] Specifically, according to the ultrasonic radio frequency signal, the specific process of calculating the shear wave propagation velocity field in the meniscus tissue by time domain analysis, and establishing the elastic modulus distribution map along the annular fiber direction and the vertical direction respectively is as follows: the collected ultrasonic radio frequency signal is subjected to beam synthesis and signal enhancement processing, the time domain characteristics of shear wave propagation are extracted, the propagation process of shear wave in the meniscus tissue is tracked by a motion estimation algorithm, and the propagation velocity of each point is calculated; according to the constitutive relationship between the shear wave propagation velocity and the elastic modulus, the elastic modulus values along the annular fiber direction and the vertical direction are calculated respectively, and the elastic modulus distribution maps along the annular fiber direction and the vertical direction are generated synchronously through spatial interpolation and gridding processing.

[0029] In the embodiment, first, the collected ultrasonic radio frequency signal is subjected to beam synthesis and signal enhancement processing to improve the signal-to-noise ratio and suppress multi-path interference. The essence of beam synthesis is to coherently superimpose the signals of different receiving channels according to the sound velocity delay, form a beam with direction selectivity, and thus enhance the time resolution of the shear wave front. Subsequently, a motion estimation algorithm (least mean square error tracking method) is used to track the propagation trajectory of the shear wave front in consecutive ultrasound frames. Let the arrival time of the wave at the spatial coordinate point be , then the propagation velocity in the local region can be expressed as: ; wherein, is the spatial distance between adjacent sampling points; is the time delay corresponding to the wave front propagation. The calculation is performed point by point in the entire meniscus region to form a propagation velocity field. According to the constitutive relationship between the shear wave propagation velocity and the material elastic modulus, the local elastic modulus can be calculated, and the expression is: ; wherein, is the equivalent density parameter of the meniscus tissue; is the shear wave propagation velocity. The relationship is based on the small deformation elastic theory, which assumes that the local tissue is an isotropic element. In order to simultaneously depict the mechanical properties of the meniscus in different directions, the elastic modulus distribution maps along the annular fiber direction and the vertical direction are established respectively. Through spatial interpolation and gridding processing, the elastic modulus values of discrete sampling points can be mapped to continuous two-dimensional distribution functions and , which respectively represent the elastic modulus distribution along the fiber direction and the vertical direction. The gridding adopts a weighted average interpolation algorithm, and the weight coefficient is determined according to the point distance: ; wherein, is the Euclidean distance between the target point and the sampling point; is the distance attenuation constant, which is determined by the cross-validation method of minimizing the interpolation error. Through this process, the distribution map of the biaxial elastic modulus of the meniscus tissue is obtained, providing a basis for the subsequent calculation of the anisotropic elastic ratio.

[0030] Specifically, the anisotropic elastic ratio is calculated by comparing the elastic modulus values in two directions, and the specific process of constructing a three-dimensional anisotropic elastic atlas combining the elastic modulus changes under the load state is as follows: the elastic modulus ratio of the circular fiber direction to the vertical direction is calculated point by point, an anisotropic elastic ratio matrix is established, the change characteristics of the elastic modulus under the load state are analyzed, and the load response coefficient is calculated; the anisotropic elastic ratio matrix and the load response coefficient are spatially registered and feature fused, and a three-dimensional reconstruction algorithm is used to construct an anisotropic elastic distribution atlas containing spatial position information; the elastic characteristics under different load states are integrated to form a three-dimensional anisotropic elastic atlas of dynamic load response.

[0031] In this embodiment, the anisotropic elastic ratio is calculated by comparing the elastic modulus values in two directions, and a three-dimensional anisotropic elastic atlas is constructed combining the elastic modulus changes under the load state, and the technical core lies in quantifying the mechanical differences related to fiber orientation and load response. In this step, the elastic modulus ratio of the circular fiber direction to the vertical direction is calculated point by point, and the anisotropic elastic ratio : is defined as , wherein is the elastic modulus along the fiber direction at position is the elastic modulus in the vertical direction; is a small regularization constant to prevent the denominator from being zero. The larger this ratio is, the stronger the dominant elasticity in the fiber direction of the region is. Then, by comparing the elastic modulus changes under the load state and the no-load state, the load response coefficient : is calculated, wherein and represent the elastic modulus under the load state and the no-load state, respectively; is a regularization constant. This coefficient reflects the elastic change rate of the tissue under axial stress, and can sensitively capture the elastic nonlinear response caused by local micro-tears in the meniscus. Then, and are spatially registered and feature fused to reconstruct a three-dimensional distribution in a unified coordinate frame. The three-dimensional reconstruction algorithm uses a voxel-level weighted average re-projection method to fuse the weights According to the local signal consistency, it is determined that ; wherein represents the correlation coefficient of the neighborhood elastic ratio; The weight smoothing coefficient is selected by traversing and verifying to minimize the reconstruction error. Finally, the fused three-dimensional anisotropic elasticity atlas The annular fiber orientation, vertical direction elasticity difference and load response characteristics are integrated to form an elasticity spatial distribution model that can be used for quantitative analysis. The atlas not only reflects the mechanical integrity of the meniscus as a whole, but also reveals the area of internal microstructure continuity destruction in the early stage, thereby providing reliable input features for subsequent fracture degree quantification.

[0032] Specifically, the specific process of obtaining the elasticity continuity index of the annular fiber direction by calculating the elasticity modulus gradient change of adjacent regions along the fiber direction is as follows: based on the three-dimensional anisotropic elasticity atlas, a reference path of the fiber direction is determined, a continuous elasticity modulus sampling point sequence is extracted along the path, the elasticity modulus gradient change between adjacent sampling points is calculated, and a distribution curve of the elasticity modulus along the fiber direction is established; the gradient fluctuation characteristics of each section in the distribution curve are analyzed by a sliding window, the standard deviation and the coefficient of variation of the gradients in the window are calculated, the gradient statistical characteristics of each window are integrated, and the elasticity continuity index reflecting the fiber structure integrity is constructed.

[0033] In the embodiment, specifically, first, in the three-dimensional anisotropic elasticity atlas, a reference path is determined according to the dominant direction of the annular fiber of the meniscus , a uniform interval elasticity modulus sampling point sequence is extracted along the path . For adjacent sampling points and , the elasticity modulus gradient change rate is calculated: ; wherein, is the elasticity modulus of the th sampling point on the path; is the arc length interval between adjacent sampling points. The gradient reflects the change rate of the local elasticity modulus along the fiber direction. Subsequently, a sliding window (the window length is ) is moved on the path, the standard deviation and the coefficient of variation of the gradient value sequence in each window are calculated, and the formulas are as follows: ; wherein, is the average value of the gradients in the window; is a small regularization constant to prevent the denominator from being zero. The smaller the standard deviation and the lower the coefficient of variation, the more continuous the elasticity distribution in the region and the more complete the fiber structure. Finally, the calculated from each window is integrated along the path to construct the overall elasticity continuity index : ; wherein, is the total number of windows; is the The weight coefficient of a window. The weight coefficient is determined by the gradient fluctuation energy: ; wherein, represents the logarithmic value of the gradient variance of the window; is an adjustment parameter, which is determined by maximizing the correlation between the continuity index and the actual rupture degree. The higher the elastic continuity index, the more stable the mechanical conduction of the meniscal fiber network and the lower the possibility of rupture.

[0034] Specifically, the specific process of obtaining the spatial variation coefficient of the anisotropic elastic ratio by statistically analyzing the dispersion degree of the anisotropic elastic ratio in the entire meniscus region is as follows: based on the three-dimensional anisotropic elasticity atlas, the anisotropic elastic ratio of each spatial position in the meniscus region is extracted, a spatial distribution matrix of the anisotropic elastic ratio is constructed, the statistical characteristics of the spatial distribution matrix are analyzed, and the local range and spatial autocorrelation index of the anisotropic elastic ratio are calculated; the dispersion degree of the anisotropic elastic ratio in the entire meniscus region is quantified by variogram analysis, and the spatial variation coefficient of the anisotropic elastic ratio is constructed by combining the local statistical characteristics and the overall dispersion degree.

[0035] In the embodiment, the process of obtaining the spatial variation coefficient of the anisotropic elastic ratio by statistically analyzing the dispersion degree of the anisotropic elastic ratio in the entire meniscus region is to measure the spatial consistency and local difference of the meniscus in different regions in the mechanical directionality. First, in the three-dimensional anisotropic elasticity atlas, the anisotropic elastic ratio of each spatial position is extracted to form a spatial distribution matrix of the anisotropic elastic ratio . Subsequently, the local range and the spatial autocorrelation coefficient in the region are calculated by statistical analysis: ; ; wherein, represents the neighborhood set of point ; is the covariance of the ratio in the neighborhood; is the standard deviation in the neighborhood; is a stabilization constant. To further depict the global dispersion degree, the variogram function is introduced: ; wherein, is the number of point pairs with a distance of ; is the anisotropic elastic ratio of the corresponding point. This function represents the similarity of the anisotropic elastic ratio at different spatial distances, and the more dramatic the change, the stronger the spatial heterogeneity. By combining the local characteristics and the global statistical results, the spatial variation coefficient is constructed: ; wherein, The integral normalization constant. The greater the spatial variation coefficient, the more uneven the anisotropic distribution of the meniscus, indicating that the local structure may have micro-tears or fiber arrangement disorder.

[0036] Specifically, the specific process of acquiring morphological parameters including boundary curvature and shape irregularity by analyzing the geometric characteristics of the abnormal elasticity region is as follows: based on the three-dimensional anisotropic elasticity atlas, the abnormal elasticity region is identified by the region growing algorithm, the complete boundary contour of the region is extracted, the local curvature of each point on the boundary contour is calculated, the irregular features of the boundary are analyzed, and the curvature distribution atlas is established; the complexity of the shape of the region is quantified by polygon approximation, the shape irregularity index is calculated, and the complete set of morphological parameters is constructed by comprehensively considering the boundary curvature distribution characteristics and the shape irregularity index.

[0037] In the embodiment, first, a seed point is determined in the elasticity atlas and region growing is performed to segment out the abnormal elasticity region. Let the overall mean of the elasticity modulus of the whole atlas be , and the overall standard deviation be . The threshold value is used as the seed selection standard: ; parameter description: ; overall mean of the elasticity modulus of the whole atlas; ; overall standard deviation of the elasticity modulus of the whole atlas; ; region growing threshold coefficient, used to control the sensitivity of the seed threshold value, the value is determined by the method of maximizing the segmentation accuracy or Dice coefficient on the labeled data; ; threshold value for judging the initial seed point. After the initial seed point is selected by , region growing (iterative expansion based on threshold and connectivity constraint) is used to incorporate adjacent voxels that satisfy the similarity condition into the region until convergence, obtaining a set of several connected abnormal regions. Boundary extraction is performed on each connected region to obtain the parametric representation of the boundary contour. A certain boundary is represented as a planar parametric curve , where , , are the components of the boundary in the local two-dimensional slice coordinates. The smooth derivative can be obtained by spline interpolation for the discrete sampling points. 3) Local boundary curvature calculation After the contour parameterization, the local curvature of each point on the contour is calculated , and the parametric form formula is: ; parameter description: ; first component function of the boundary curve at parameter ; ; second component function of the boundary curve at parameter ; ; Regarding first derivative of the function ; first derivative of the function ; ; second derivative of the function ; second derivative of the function ; the local curvature value at the point, the larger the curvature, the more obvious the boundary bending. After calculating the curvature distribution of the entire boundary, several statistics of the distribution can be calculated as morphological features, such as the mean curvature , the curvature variance , and the high-order quantile of the curvature, to represent the overall degree and local mutation of the boundary bending. The shape irregularity is used to quantify the overall shape complexity of the region. First, the boundary is approximated as a polygon with a specified approximation tolerance , and then the perimeter and area of the approximated polygon are obtained. The shape irregularity index is defined as: ; parameter description: : the area of the region (the area of the region measured in the slice scale); : the perimeter of the region (the perimeter of the approximated polygon); : a smoothing stabilization constant, used to avoid division by zero or numerical instability, the value is determined by cross-validation or stability analysis; : the shape irregularity index, usually ranging from 0 to close to 1, the larger the value, the more irregular and complex the shape. The comprehensive morphological parameter set combines the boundary curvature statistics and the shape irregularity to form the morphological parameter vector of the region, such as: ; parameter description: : the morphological parameter vector of the region ; : the mean of the boundary curvature; : the standard deviation of the boundary curvature; : the shape irregularity. The vector can be used as one of the input features of the subsequent classification or regression model (such as the morphological branch of a multi-branch neural network). (region growing threshold coefficient): on the labeled training set, the Dice coefficient or IoU of the region segmentation is maximized by grid search or Bayesian optimization to determine; : according to the boundary smoothing requirement and the trade-off between calculation efficiency, the sensitivity of the morphological features to the stability of the segmentation result is minimized on the training set to select; (smoothing stabilization constant): take a small positive value, or select through cross-validation to make not sensitive to small measurement noise.

[0038] Specifically, the specific process of inputting the elastic continuity index, the spatial variation coefficient, and the morphological parameter into three feature processing branches of the multi-branch neural network for feature processing is as follows: the elastic continuity index is input into the first feature processing branch, the local sequence pattern feature of the fiber structure is extracted through a one-dimensional convolution layer, and the elastic continuity change rule along the fiber direction is captured through a long short-term memory network; the spatial variation coefficient is input into the second feature processing branch, the spatial topological relationship of the anisotropic elastic ratio is constructed through a graph convolution network, and the variation features of key regions are aggregated by using an attention pooling layer; the morphological parameter is input into the third feature processing branch, the geometric feature of the boundary contour is extracted through a two-dimensional convolution layer, and the multi-scale shape irregularity feature is captured by using a spatial pyramid pooling.

[0039] In the embodiment, the elastic continuity index vector is input into the first feature processing branch (fiber continuity branch). The branch first extracts the local sequence pattern (such as short-term mutation, periodic fluctuation, etc.) along the fiber direction through a one-dimensional convolution layer, then transmits the convolution output into a long short-term memory network to capture the time sequence dependence and long-range association of the sequence, and finally outputs the fiber continuity feature vector. Let the input sequence be , the one-dimensional convolution obtains the intermediate representation : ; the parameters are as follows: : the sampling sequence of the elastic continuity index; : the convolution kernel width; : the weight parameter of the th convolution kernel; : the convolution bias term; : the activation value at position after convolution. The is taken as input to obtain the output sequence, and the pooling or the last hidden state at the end is taken as the branch output vector . The training and hyperparameter determination method: The LSTM hidden state dimension is determined through cross-validation or Bayesian optimization on the labeled data, and the loss function is minimized with the task-related metric (such as regression error or classification accuracy) as the target. The spatial variation coefficient is input into the second feature processing branch, the spatial anisotropic elastic ratio is first mapped into a graph structure node feature, and the topological relationship between positions is constructed by using a graph convolution network to capture the local and global spatial dependence, and then the variation features of key nodes are aggregated by using an attention pooling layer, and a spatial heterogeneity feature vector is output. Let the graph node feature matrix be , and the adjacency matrix be . The single-layer graph convolution update is: ; the parameters are as follows: : initial node feature matrix (composed of local anisotropic elastic ratio) : adjacency matrix of graph, representing the connection strength between nodes : obtained by : degree matrix : learnable weight matrix of graph convolution layer : nonlinear activation function : node feature matrix after graph convolution. Based on the node representation, attention weights are calculated and weighted summation is performed to obtain the graph representation . Hyperparameters and weight determination method: adjacency matrix construction (based on spatial distance or correlation threshold) and pooling window size are determined by optimizing task indicators on the validation set. The morphological parameters are input into the third feature processing branch (geometric morphological branch). This branch inputs the geometric representation of the region boundary (curvature sequence or boundary mapping graph) into the two-dimensional convolution layer to extract local texture and geometric patterns, and then captures multi-scale shape irregularity features through spatial pyramid pooling, outputting a geometric morphological feature vector. Pooling is performed on the convolution output Z in a set of multi-scale pooling windows , and is obtained by concatenation. Parameter description: : feature map after two-dimensional convolution : qth pooling scale window : multi-scale morphological feature vector after concatenation. Network training: the number of kernels in each convolution layer, the pooling scale, etc. are selected by hyperparameter search and training set performance driving. The above three branches output: fiber continuity feature vector ; spatial heterogeneity feature vector ; geometric morphological feature vector .

[0040] Specifically, the specific process of outputting the meniscus tear quantitative evaluation value through the cross-branch attention mechanism to fuse the feature representations output by each branch is as follows: the fiber continuity feature, spatial heterogeneity feature and geometric morphological feature representations output by the three feature processing branches are concatenated to construct a multi-dimensional feature tensor, and the importance weights of each feature channel are calculated through the cross-branch attention mechanism, wherein the query vector comes from the fiber continuity feature, and the key-value pair comes from the spatial heterogeneity feature and the geometric morphological feature; the attention weights are normalized by the softmax function to enhance important features and suppress unimportant features; the weighted feature representation is input into the fully connected layer for dimension reduction fusion, and the sigmoid activation function is used to output a meniscus tear quantitative evaluation value in the range of 0 to 1, which comprehensively reflects the fiber structure integrity, tissue heterogeneity and lesion morphological features.

[0041] In this implementation, the feature concatenation and query / key / value construction, first concatenate the three-branch outputs to form a multi-dimensional feature tensor, and then construct the query, key and value vectors for attention to calculate the cross-modal attention weight. The query comes from the fiber continuity branch (emphasizing the prior of mechanical continuity), and the key and value come from the spatial heterogeneity branch and the geometric morphology branch (providing position and morphology evidence). Representation: Let the concatenated vector be . Construct the query , key matrix , value matrix : ; Parameter description: ; fiber continuity branch output vector; ; spatial heterogeneity branch output vector; ; geometric morphology branch output vector; ; query projection matrix; ; key projection matrix; ; value projection matrix; ; query vector; , ; key and value matrix. Attention weight calculation and normalization, using dot product attention mechanism to calculate attention energy and normalization: ; where the weighted representation is: ; Parameter description: ; the column vector of the key matrix ; ; the column vector of the value matrix ; ; normalized attention weight; ; scaling factor (usually a scalar of the dimension of the key vector or determined by a trainable parameter) to stabilize the dot product value; ; cross-branch feature representation obtained by attention weighted fusion. Weight determination method: scaling factor and projection matrix are learned through data backpropagation; if manually set, take the square root of the dimension of the key vector as the initial value, and fine-tune on the validation set. Feature enhancement and fully connected fusion, concatenate and the original fiber continuity feature or add residual to input several fully connected layers for dimension reduction and fusion, finally output a single value score. Dimension reduction and activation: let the fusion vector be , the fully connected output is: ; ; Parameter description: ; fusion input vector (contains and possible residual information); ; fully connected layer weight matrix; ; fully connected bias; ; internal activation function (ReLU) of the fully connected layer; ; non-normalized output of the fully connected layer; ; final meniscus tear quantification score after sigmoid mapping, ranging from 0 to 1, closer to 1 indicates higher tear severity or probability. Training strategy and hyperparameter determination method, projection matrix , , and fully connected weights learned on labeled data by standard backpropagation; scaling factor in attention normalization can be initialized as the dimension of the key vector and fine-tuned as a trainable parameter; if needed, set attention threshold or sparsify weights (e.g. clip min to improve interpretability), threshold can be selected by optimizing the F1 or AUC behavior of the model on the validation set; the loss function can use weighted binary cross-entropy or joint loss with interpretability constraints (e.g. training both regional localization and overall score), and the weights are determined by grid search.

[0042] Please refer to Figure 2 , a meniscus tear quantification system based on ultrasound shear wave elastography, comprising the following modules: a data acquisition module for fixing the knee joint at a predetermined flexion angle, exciting shear waves along the circumferential fiber direction and perpendicular direction of the meniscus by a high-frequency linear array ultrasound probe, and collecting ultrasound radio frequency signals of the affected and healthy menisci under unloaded and standardized axial load conditions; an elastic atlas construction module for calculating the shear wave propagation velocity field in the meniscal tissue based on the ultrasound radio frequency signals through time domain analysis, establishing elastic modulus distribution maps along the circumferential fiber direction and perpendicular direction, respectively, calculating the anisotropic elastic ratio by comparing the elastic modulus values in the two directions, and constructing a three-dimensional anisotropic elastic atlas combining the elastic modulus changes under load conditions; a feature parameter extraction module for obtaining the elastic continuity index along the fiber direction by calculating the elastic modulus gradient change of adjacent regions along the fiber direction based on the three-dimensional anisotropic elastic atlas, obtaining the spatial variation coefficient of the anisotropic elastic ratio by statistically analyzing the dispersion degree of the anisotropic elastic ratio in the entire meniscal region, and obtaining morphological parameters including boundary curvature and shape irregularity by analyzing the geometric characteristics of abnormal elastic regions; a quantification module for inputting the elastic continuity index, spatial variation coefficient, and morphological parameters into three feature processing branches of a multi-branch neural network for feature processing, weighting and fusing the feature representations output by each branch through cross-branch attention mechanism, and outputting the meniscus tear quantification score.

[0043] In this embodiment, the data acquisition module functions to provide high-quality, standardized raw ultrasound signal data for the entire evaluation system. By setting the flexion angle of the knee joint and controlling the excitation direction of the shear wave, this module can collect ultrasound radio frequency signals of the meniscus under different directions and stress states under strictly consistent spatial and load conditions, providing a basic data source for subsequent elasticity analysis. The design of this module ensures the repeatability of the data and the controllability of the tissue state, making the evaluation results more reliable. The elasticity atlas construction module functions to convert the ultrasound radio frequency signals into visualized data of the elastic distribution of the tissue. This module first establishes the shear wave propagation velocity field through time domain analysis and wave speed calculation, and calculates the spatial distribution of the elastic modulus accordingly; then it obtains the anisotropy characteristics through the elastic comparison between directions, and further reconstructs the three-dimensional anisotropic elasticity atlas in combination with the load response information. The core significance of this module lies in that it not only reflects the mechanical differences of the meniscal tissue in different directions, but also reveals the dynamic elastic response of the lesion area under stress, providing a physical basis for quantifying the tear degree. The feature parameter extraction module functions to extract key quantitative indicators reflecting the tissue structure and lesion characteristics from the elasticity atlas. This module first calculates the elastic gradient change along the fiber direction, thereby obtaining the elastic continuity index representing the fiber continuity and structural integrity; secondly, it analyzes the spatial dispersion of the anisotropic elastic ratio in the entire meniscal region to obtain the spatial variation coefficient, which is used to reflect the unevenness of the anisotropic distribution of the tissue; finally, it identifies the abnormal elastic region and extracts geometric morphological features such as boundary curvature and shape irregularity to reflect the structural abnormality of the lesion area. This module plays a role in extracting high-value features from complex atlas in the system, and is the core link of quantitative analysis. The quantitative evaluation module functions to fuse multi-dimensional features and output the quantitative evaluation results of meniscus tear. This module processes and learns the feature expression of different types of features through a multi-branch neural network, with each branch focusing on different biological aspects of information: structural continuity, spatial heterogeneity, and geometric morphological abnormalities; then it realizes feature weighted fusion through cross-branch attention mechanism to improve the recognition ability of the model for complex lesion patterns; finally, it outputs a continuous quantitative value reflecting the comprehensive evaluation results of the meniscus tear degree. This module realizes the transformation from physical elastic features to intelligent quantitative discrimination, enabling the system to have automated and objective diagnostic capabilities.

[0044] In summary, the present application has at least the following effects:

[0045] The application discloses a method and system for quantitatively evaluating meniscus internal tear based on ultrasonic shear wave elastography, and realizes fine visual representation of meniscus tissue mechanical characteristics through three-dimensional anisotropic elastic atlas construction based on ultrasonic shear wave elastography; the multi-dimensional feature quantitative analysis of the elastic continuity index, the spatial variation coefficient and the morphological parameter can comprehensively reflect the internal structure continuity damage and local tear degree of the meniscus; the multi-branch neural network fuses the multi-dimensional elastic features, realizes objective quantitative evaluation of the meniscus tear degree, improves the sensitivity and quantitative level of the tear detection, and provides a reliable basis for clinical minimally invasive surgery planning and rehabilitation evaluation.

[0046] Those skilled in the art will appreciate that embodiments of the application can be supplied as methods, systems or computer program products. Accordingly, the application can be embodied in the form of complete hardware embodiments, complete software embodiments or embodiments combining software and hardware aspects. Furthermore, the application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage etc.) having computer-usable program code embodied therein.

[0047] The application is described in reference to the flowcharts and / or block diagrams of the systems, devices (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions, which are executed via the processor of the computer or other programmable data processing device, generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in the flowchart

[0048] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in the flowchart

[0049] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the steps of the functions specified in the one or more blocks.

[0050] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the preferred embodiments by those of skill in the art once they have the benefit of the present disclosure. Therefore, the appended claims are intended to encompass within their scope all possible variations and modifications of the preferred embodiments. The preferred embodiments of the application described herein are not meant to be limiting, but rather are meant to be illustrative only. It is therefore desired to be protected in the broadest scope of the appended claims to encompass all changes and modifications of the preferred embodiments of the application.

[0051] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended to include all such changes and modifications in the scope of the application recited in the claims and their equivalents.

Claims

1. A method for quantitatively evaluating an amount of an inner meniscus tear based on ultrasonic shear wave elastography, characterized by, The method comprises the following steps: S1. Fix the knee joint at a preset flexion angle, and excite shear waves along the direction of the meniscal annular fiber and the perpendicular direction by a high-frequency linear array ultrasonic probe, respectively, to collect ultrasonic radio frequency signals of the affected and healthy menisci under a no-load state and a standardized axial load state; S2. According to the ultrasonic radio frequency signals, the propagation velocity field of the shear wave in the meniscal tissue is calculated through time domain analysis, and the elastic modulus distribution maps along the annular fiber direction and the perpendicular direction are established, respectively, the anisotropic elastic ratio is calculated by comparing the elastic modulus values in the two directions, and the three-dimensional anisotropic elastic atlas is constructed in combination with the elastic modulus change under the load state; S3. Based on the three-dimensional anisotropic elastic atlas, the elastic continuity index in the annular fiber direction is obtained by calculating the elastic modulus gradient change of adjacent regions along the fiber direction, the spatial variation coefficient of the anisotropic elastic ratio is obtained by statistically analyzing the dispersion degree of the anisotropic elastic ratio in the whole meniscal region, and the morphological parameters including the boundary curvature and the shape irregularity are obtained by analyzing the geometric characteristics of the abnormal elastic region; S4. The elastic continuity index, the spatial variation coefficient and the morphological parameters are respectively input into three feature processing branches of a multi-branch neural network for feature processing, the feature representations output by each branch are weighted and fused through a cross-branch attention mechanism, and a meniscal tear quantitative evaluation value is output.

2. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: Step S1 comprises the following steps: The knee joint is fixed at a preset flexion angle, the meniscal body is exposed to the effective detection range of the ultrasonic probe, the probe direction is positioned along the annular fiber direction and the perpendicular direction of the meniscus by a high-frequency linear array ultrasonic probe, and the shear wave excitation direction is ensured to have a specific spatial relationship with the fiber structure; The baseline ultrasonic radio frequency signal is collected under a no-load state, the initial elastic properties of the meniscus are recorded, a preset standardized axial load is applied to simulate the weight-bearing state of the knee joint, and the ultrasonic radio frequency signal under the load state is collected; The measurement process is repeated for the affected and healthy menisci to ensure the consistency of the measurement conditions.

3. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: According to the ultrasonic radio frequency signals, the propagation velocity field of the shear wave in the meniscal tissue is calculated through time domain analysis, and the elastic modulus distribution maps along the annular fiber direction and the perpendicular direction are established. The specific process is as follows: The collected ultrasonic radio frequency signals are subjected to beam synthesis and signal enhancement processing, the time domain characteristics of the shear wave propagation are extracted, the propagation process of the shear wave in the meniscal tissue is tracked through a motion estimation algorithm, and the propagation velocity of each point is calculated; According to the constitutive relationship between the shear wave propagation velocity and the elastic modulus, the elastic modulus values along the annular fiber direction and the perpendicular direction are calculated, and the elastic modulus distribution maps along the annular fiber direction and the perpendicular direction are simultaneously generated through spatial interpolation and gridding processing.

4. The method of quantitatively evaluating meniscal tear according to claim 3, wherein: The specific process of calculating the anisotropic elastic ratio by comparing the elastic modulus values in the two directions and constructing the three-dimensional anisotropic elastic atlas in combination with the elastic modulus change under the load state is as follows: The elastic modulus ratio of the annular fiber direction to the perpendicular direction is calculated point by point, an anisotropic elastic ratio matrix is established, the change characteristics of the elastic modulus under the load state are analyzed, and a load response coefficient is calculated. The anisotropic elasticity ratio matrix is spatially registered and feature-fused with the load response coefficient, and a three-dimensional anisotropic elasticity atlas containing spatial position information is constructed through a three-dimensional reconstruction algorithm; The elasticity features under different load states are integrated to form a three-dimensional anisotropic elasticity atlas of dynamic load response.

5. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: The specific process of obtaining the elasticity continuity index of the annular fiber direction by calculating the elasticity modulus gradient change along the adjacent regions of the fiber direction is as follows: Based on the three-dimensional anisotropic elasticity atlas, a reference path of the fiber direction is determined, a continuous elasticity modulus sampling point sequence is extracted along the path, the elasticity modulus gradient change between adjacent sampling points is calculated, and a distribution curve of the elasticity modulus along the fiber direction is established; The gradient fluctuation characteristics of each section in the distribution curve are statistically analyzed through a sliding window, the standard deviation and coefficient of variation of the gradient in the window are calculated, the gradient statistical characteristics of each window are integrated, and an elasticity continuity index reflecting the integrity of the fiber structure is constructed.

6. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: The specific process of obtaining the spatial variation coefficient of the anisotropic elasticity ratio by statistically analyzing the dispersion degree of the anisotropic elasticity ratio in the entire meniscus region is as follows: Based on the three-dimensional anisotropic elasticity atlas, the anisotropic elasticity ratio of each spatial position in the meniscus region is extracted, a spatial distribution matrix of the anisotropic elasticity ratio is constructed, the statistical characteristics of the spatial distribution matrix are analyzed, and the local range and spatial autocorrelation index of the anisotropic elasticity ratio are calculated; The dispersion degree of the anisotropic elasticity ratio in the entire meniscus region is quantified through variogram analysis, and the local statistical characteristics and overall dispersion degree are combined to construct the spatial variation coefficient of the anisotropic elasticity ratio.

7. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: The specific process of obtaining the morphological parameters including boundary curvature and shape irregularity through analyzing the geometric characteristics of the abnormal elasticity region is as follows: Based on the three-dimensional anisotropic elasticity atlas, the abnormal elasticity region is identified through a region growing algorithm, the complete boundary profile of the region is extracted, the local curvature of each point on the boundary profile is calculated, the irregular features of the boundary are analyzed, and a curvature distribution atlas is established; The complexity of the region shape is quantified through polygon approximation, the shape irregularity index is calculated, and the complete set of morphological parameters is constructed by integrating the boundary curvature distribution characteristics and the shape irregularity index.

8. The method of quantitatively evaluating meniscal tear according to claim 1, wherein: The specific process of inputting the elasticity continuity index, spatial variation coefficient, and morphological parameters into three feature processing branches of a multi-branch neural network for feature processing is as follows: The elasticity continuity index is input into the first feature processing branch, the local sequence pattern features of the fiber structure are extracted through a one-dimensional convolution layer, and the elasticity continuity change law along the fiber direction is captured through a long short-term memory network; The spatial variation coefficient is input into the second feature processing branch, the spatial topological relationship of the anisotropic elasticity ratio is constructed through a graph convolution network, and the variation features of the key region are aggregated using an attention pooling layer; The morphological parameters are input into the third feature processing branch, the geometric features of the boundary profile are extracted through a two-dimensional convolution layer, and the multi-scale shape irregularity features are captured using a spatial pyramid pooling; The fiber continuity features, spatial heterogeneity features, and geometric and morphological features extracted through the deep network are output by each feature processing branch.

9. The method of quantitatively evaluating meniscal tear according to claim 8, wherein: The specific process of outputting the meniscus tear quantitative evaluation value through the cross-branch attention mechanism to weight and fuse the feature representations of each branch output is as follows: The fiber continuity feature, the spatial heterogeneity feature and the geometric morphological feature representation output by the three feature processing branches are spliced to construct a multi-dimensional feature tensor, and the importance weight of each feature channel is calculated through the cross-branch attention mechanism, wherein the query vector comes from the fiber continuity feature, and the key-value pair comes from the spatial heterogeneity feature and the geometric morphological feature; The attention weight is normalized through the softmax function to realize the enhancement of important features and the suppression of non-important features; The weighted feature representation is input into the full connection layer for dimension reduction fusion, and the sigmoid activation function is used to output the meniscus tear quantitative evaluation value in the range of 0 to 1, which comprehensively reflects the fiber structure integrity, tissue heterogeneity and lesion morphological features.

10. A system for quantitatively evaluating meniscal intrasubstance tear based on ultrasonic shear wave elastography, applied to the method for quantitatively evaluating meniscal intrasubstance tear based on ultrasonic shear wave elastography according to any one of claims 1-9, characterized in that, It includes the following modules: The data acquisition module is used to fix the knee joint at a predetermined flexion angle, excite shear waves along the circular fiber direction and the vertical direction of the meniscus respectively through a high-frequency linear array ultrasonic probe, and collect ultrasonic radio frequency signals of the affected and healthy menisci under no load and standardized axial load conditions; The elastic atlas construction module is used to calculate the shear wave propagation velocity field in the meniscus tissue through time domain analysis according to the ultrasonic radio frequency signal, establish the elastic modulus distribution map along the circular fiber direction and the vertical direction respectively, calculate the anisotropic elastic ratio by comparing the elastic modulus values in the two directions, and construct a three-dimensional anisotropic elastic atlas combined with the elastic modulus change under the load state; The feature parameter extraction module is used to calculate the elastic continuity index along the fiber direction by calculating the elastic modulus gradient change of adjacent regions along the fiber direction based on the three-dimensional anisotropic elastic atlas, calculate the spatial variation coefficient of the anisotropic elastic ratio by calculating the discrete degree of the anisotropic elastic ratio in the entire meniscus region, and calculate the morphological parameters including boundary curvature and shape irregularity by analyzing the geometric characteristics of the abnormal elastic region; The quantitative evaluation module is used to input the elastic continuity index, the spatial variation coefficient and the morphological parameters into the three feature processing branches of the multi-branch neural network for feature processing, weight and fuse the feature representations output by each branch through the cross-branch attention mechanism, and output the meniscus tear quantitative evaluation value.

Images