ECM-cartilage repair image analysis method and system based on deep learning

Abstract

The invention relates to the technical field of medical image analysis, and discloses an ECM-cartilage repair image analysis method and system based on deep learning. The method comprises the following steps: acquiring multi-modal medical image data in a cartilage repair process; respectively extracting a structure contour line and a component distribution contour line by adopting a deep learning segmentation model; calculating geometric features based on the contour line, constructing a stress analysis model, and calculating a stress concentration coefficient as a mechanical risk feature; calculating a spatial distribution consistency index by combining the structure contour line and the component distribution contour line; and finally, a comprehensive repair quality index is calculated by fusing the geometric features, the mechanical risk features and the distribution consistency index, risk cause tracing is carried out, and a graphical comprehensive evaluation report is generated. According to the method, automatic, multi-dimensional and quantitative evaluation of the cartilage repair quality can be realized, and the objectivity and clinical applicability of evaluation are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image analysis, in particular to an ECM-cartilage repair image analysis method and system based on deep learning. BACKGROUND

[0002] Cartilage injury repair is a common problem in orthopedics and sports medicine, and its repair effect evaluation usually relies on manual observation and subjective scoring of histological sections, which has limitations such as non-uniform evaluation standard, dependence on expert experience, and difficulty in quantification. In the prior art, although individual studies have attempted to use image analysis methods to assist in evaluation, most methods only focus on morphological features, lacking comprehensive consideration of functional characteristics of cartilage tissue (such as mechanical properties, composition distribution).

[0003] In addition, traditional evaluation methods cannot effectively integrate multi-modal image information (such as structural images and composition images), making it difficult to reveal the internal functional state and potential risks of the repaired tissue. Although deep learning technology has made progress in medical image segmentation, its application in cartilage repair evaluation is still limited to single tasks, and a systematic evaluation system from structure to function and from morphology to composition has not yet been formed.

[0004] Therefore, there is an urgent need in the art for a cartilage repair evaluation method that can deeply integrate multi-modal image information, comprehensively consider structural and functional features, and have automatic and quantitative capabilities, to improve the comprehensiveness, accuracy and clinical practicability of the evaluation Therefore, the present application provides an ECM-cartilage repair image analysis method and system based on deep learning. SUMMARY

[0005] The embodiments of the present specification provide the following technical solutions: Step S1, collecting multi-modal medical image data in the cartilage repair process, the multi-modal medical image data comprising a first modality image and a second modality image; Step S2, processing the first modality image using a first deep learning segmentation model and extracting the structural contour line of the new ECM region, and processing the second modality image using a second deep learning segmentation model and extracting the composition distribution contour line of the specific ECM composition; Step S3, calculating geometric features based on the structural contour line, constructing a stress analysis model based on the structural contour line, analyzing based on the stress analysis model, calculating a stress concentration coefficient as a mechanical risk feature based on the analysis result, and calculating a spatial distribution consistency index based on the structural contour line and the composition distribution contour line; Step S4, calculating a comprehensive repair quality index based on the geometric features, the mechanical risk features and the distribution consistency index, and performing risk cause tracing to generate a comprehensive evaluation report.

[0006] Compared with the prior art, the application has at least the following beneficial effects: In the technical scheme provided in the application, by fusing multiple features such as geometric shapes, mechanical risks and component distributions, comprehensive quantitative evaluation of cartilage repair quality is realized, and the limitation of a single perspective of a traditional method is overcome; automatic extraction of structure contours and component contours is realized based on a deep learning model, analysis efficiency is greatly improved, and human intervention and subjective errors are reduced; through a stress concentration coefficient and a distribution consistency index, risk early warning of potential mechanical failure and component distribution abnormality of the repaired tissue is realized, and clinical foresight is possessed; and a comprehensive evaluation report is generated to display the repair quality and risk areas in a graphical manner, thereby facilitating intuitive understanding of doctors and clinical decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0008] Figure 1 An embodiment schematic diagram of an ECM-cartilage repair image analysis method based on deep learning in the embodiments of the application; Figure 2 An embodiment schematic diagram of an ECM-cartilage repair image analysis system based on deep learning in the embodiments of the application. DETAILED DESCRIPTION

[0009] The embodiments of the application provide an ECM-cartilage repair image analysis method and system based on deep learning. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the application and the above drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0010] For the convenience of understanding, the specific process of the embodiments of the application will be described below. Please refer to Figure 1 An embodiment of an ECM-cartilage repair image analysis method based on deep learning in the embodiments of the application includes: Step S1, collect multi-modal medical image data in the process of cartilage repair, the multi-modal medical image data comprising a first modality image and a second modality image; Specifically, multi-modal medical image data in the process of cartilage repair is collected, the multi-modal medical image data comprising a first modality image for representing tissue macrostructure information and a second modality image for representing spatial distribution information of specific extracellular matrix components, the first modality image being a hematoxylin-eosin staining histological image, the hematoxylin-eosin staining histological image being able to clearly show the morphology, distribution and basic microscopic structure of cells, thus being very suitable for defining the overall boundary of the new cartilage region and not being interfered by specific ECM components, and the second modality image being a type II collagen immunohistochemical staining image, type II collagen being the most characteristic ECM component of mature hyaline cartilage, and its distribution quantity, position and continuity directly reflecting the quality of the repaired cartilage, and the component distribution being able to be displayed with high contrast by immunohistochemistry.

[0011] Step S2, a first deep learning segmentation model is used to process the first modality image and extract the structural contour line of the new ECM region, and a second deep learning segmentation model is used to process the second modality image and extract the component distribution contour line of the specific ECM component; Specifically, the first modality image is processed by using a pre-trained deep learning segmentation model, and the new cartilage extracellular matrix region is recognized and segmented, and the boundary of the new cartilage extracellular matrix region is extracted to generate a structural contour line, and at the same time, the second modality image is processed by using a pre-trained second deep learning segmentation model, and the target extracellular matrix component enrichment region specifically expressed in the image is recognized and segmented, the specific ECM component refers to the target extracellular matrix component, and the boundary of the target extracellular matrix component enrichment region is extracted to generate a component distribution contour line, the first deep learning segmentation model and the second deep learning segmentation model are both pre-trained by a large amount of labeled data, and can recognize the fine texture and boundary features of different tissues, so as to realize accurate classification, and the two models are used to process the two images respectively, so as to ensure that the structure and component information are independently and accurately extracted.

[0012] Step S3, calculate the geometric feature based on the structural contour line, construct a stress analysis model based on the structural contour line, analyze based on the stress analysis model, calculate the stress concentration coefficient as the mechanical risk feature based on the analysis result, and calculate the spatial distribution consistency index based on the structural contour line and the component distribution contour line; Specifically, geometric features based on structural contour lines are used to quantify the regularity of the extracellular matrix structure of newly formed chondrocytes. These geometric features include the rate of change of contour line radius, contour line concavity, and contour line surface roughness. The specific calculation methods will be explained in detail later. Geometric features can quantitatively describe the regularity of the extracellular matrix structure. Generally, the more regular the structure and the closer it is to natural cartilage, the higher the repair quality. To predict dynamic function and potential failure risks from static structure, a two-dimensional plane stress analysis model is also constructed based on the structural contour lines. Analysis based on the stress analysis model involves assigning the model preset linear elastic material properties that simulate newly formed cartilage tissue, applying uniformly distributed micro-amplitude static pressure loads to the model boundaries, and performing micro-finite element analysis to obtain the analysis results. The specific analysis process will be explained in detail later. In detail, the analysis results include stress tensors from multiple nodes. Based on the analysis results, the stress concentration factor is calculated as a mechanical risk characteristic. The specific calculation method will be explained in detail later. According to mechanical principles, irregularities such as sharp corners and holes in a structure will cause stress to be significantly higher than the average level. A higher stress concentration factor indicates that the corresponding area is more prone to damage under physiological load. The stress concentration factor is a bridging parameter from morphology to function. In order to assess the quality and spatial rationality of the sublimation composition of the extracellular matrix, the component distribution consistency index is also calculated. If the proteoglycans responsible for compression resistance and the collagen fibers responsible for tensile resistance are not spatially distributed, for example, only one is present and the other is missing, or they are distributed in different areas, then the overall mechanical properties of the tissue will be reduced. The component distribution consistency index quantifies the effectiveness of this component synergy.

[0013] Step S4: Calculate the comprehensive repair quality index based on geometric features, mechanical risk features and distribution consistency index, trace the causes of risks, and generate a comprehensive assessment report.

[0014] Specifically, to integrate information from multiple dimensions into an easily understandable comprehensive score and diagnose major risks, a weighted fusion method is used to unify features with different dimensions and meanings into a comprehensive repair quality index. To trace the root causes of risks, a comparative method is used to identify which feature parameters deviate abnormally from the normal range. For example, if the stress concentration coefficient is abnormally high while other parameters are basically normal, it will be judged as a predominantly mechanically weak risk, which can directly guide doctors to focus on the structural stability of that area. To convert complex quantitative data into intuitive and clinically interpretable graphical reports, a comprehensive assessment report is also generated. The comprehensive assessment report graphically displays the structural outline and uses a continuous gradient color mapping scheme to color the outline. Different colors correspond to local repair quality scores from high to low. At the same time, classification markers of major risk causes are overlaid at key positions on the structural outline. The comprehensive assessment report also lists the comprehensive repair quality index in numerical form and displays the calculated values ​​of each feature and their comparison with the preset reference range in tabular form.

[0015] By coordinating the above steps, an automated, multi-dimensional, and quantitative assessment of cartilage repair quality can be achieved. This method is applicable to various cartilage injury repair scenarios, especially for articular cartilage defects caused by sports injuries, aging degeneration, and disease erosion. Whether it is cartilage tearing caused by acute trauma or cartilage wear caused by chronic strain, this application can accurately analyze images of the repair process, assist doctors in judging the repair effect, adjusting treatment plans, and improving the success rate and quality of cartilage repair.

[0016] In one specific embodiment, calculating geometric features based on structural contour lines includes the following steps: Using the center of the smallest circumcircle of the structural contour as the reference center point, calculate the radial distance between the reference center point and multiple measurement points distributed at predetermined angular intervals on the structural contour. Then calculate the rate of change of radial distance between adjacent measurement points and the variance of all distance rates of change. Use the variance as the rate of change of contour radius. Calculate the shortest distance from each measurement point on the structural contour line to the edge of the smallest circumcircle, and take the maximum value among all the shortest distances as the concavity of the contour line.

[0017] Specifically, the higher the quality of cartilage repair, the more regular and smooth the extracellular matrix contour becomes. This is manifested in a low rate of radius change, small local curvature, shallow concavity, and low surface roughness. To calculate the rate of radius change of the structural contour, first, the smallest circumcircle of the structural contour is found, and its center is defined as the reference center point of the entire contour. Then, starting from a predetermined starting point on the structural contour, a ray is drawn clockwise or counterclockwise around the center at fixed angular intervals (e.g., 1 degree). At each angle, a ray originating from the center is drawn, intersecting the structural contour at an intersection point. This intersection point is used as a measurement point, and the distance from each measurement point to the center is taken as the radial distance. The radial distance is then calculated. The radius is the radius at the corresponding angle. By mapping the angle to its corresponding radius, a discrete function can be obtained to describe the change of the radius with the angle. The rate of change of the radius between adjacent measurement points is calculated. Since there are multiple radii, multiple rates of change of the radius can be calculated. In order to calculate a common rate of change of the radius, the variance of all the rates of change of the radius is calculated. This variance is used as the rate of change of the radius of the structural contour. The rate of change of the radius can quantify the steepness of the undulation of the structural contour. The higher the quality of cartilage repair, the gentler the change of the radius. Conversely, a large rate of change of the radius means that there are steep peaks and valleys in the contour. It means that the internal structure of the cartilage repair tissue has formed large depressions or protrusions due to structural collapse, indicating that the quality of cartilage repair is low.

[0018] The smallest circumcircle of the structural contour represents the most compact and ideal smooth contour that can enclose the structural contour. For each measurement point on the structural contour, a line is formed connecting the center of the circle and the measurement point. This line intersects the smallest circumcircle at a point. The distance between this point and the measurement point is taken as the shortest distance from each measurement point on the structural contour to the edge of the smallest circumcircle. Since there are multiple measurement points, multiple shortest distances can be obtained. The maximum value among the shortest distances is taken as the concavity of the structural contour. In cartilage repair, concavity corresponds to tissue defects, erosion, or hypoplasia. Therefore, the largest concavity is used as the concavity of the structural contour.

[0019] In one specific embodiment, calculating geometric features based on structural contour lines further includes the following steps: The radial distance of each measurement point on the structural contour is obtained to generate a radial distance sequence. Gaussian filtering is applied to the radial distance sequence to smooth it and obtain a new radial distance sequence. For each measurement point, the deviation between the original radial distance and the smoothed radial distance is calculated. The root mean square error of all deviations is calculated and the root mean square error is used as the surface roughness of the contour.

[0020] Specifically, to calculate the surface roughness of the contour line, the radial distance corresponding to each measurement point of the original structural contour line is obtained to generate a radial distance sequence. The radial distance sequence is smoothed using Gaussian filtering. Gaussian filtering can effectively smooth out high-frequency fluctuations (fine, rapid undulations) in the radial distance sequence while preserving its low-frequency trend, resulting in a new radial distance sequence. Then, for each measurement point, the deviation value between the original radial distance and the smoothed radial distance is calculated, and the root mean square error of all deviation values ​​is calculated. The root mean square error is used as the surface roughness of the contour line. High-quality cartilage ECM should have a relatively smooth surface, while a rough surface indicates that the cartilage structure is loose or that there are minor ulcers or peelings. It also reflects poor microstructural order at the cellular and fibrous levels. By calculating multiple geometric features, the cartilage recovery is evaluated from different perspectives, so that the subsequent analysis of cartilage recovery based on geometric features can obtain more accurate analysis results.

[0021] In one specific embodiment, the analysis is performed based on a stress analysis model, specifically including the following steps: The structural outline is transformed into a two-dimensional planar geometric model as a stress analysis model. The stress analysis model is discretized into mesh elements, and each mesh element is assigned mechanical property parameters representing newly formed cartilage tissue. Boundary conditions are defined and loads are applied. Based on the mechanical property parameters, boundary conditions, and loads of each mesh element, a system of linear equations is generated. The system of linear equations is solved to obtain the stress tensor of each mesh element. Based on preset calculation rules, the stress tensor is converted into an equivalent equivalent stress. The analysis result is the equivalent stress of each mesh element.

[0022] Specifically, to transform the biological image problem into an engineering mechanics problem, finite element analysis is performed based on a stress analysis model. First, the structural outline is transformed into a two-dimensional planar geometric model. The region enclosed by the structural outline is defined as the planar domain to be analyzed. The geometric model is discretized into a large number of small, simple mesh elements (quadrilateral elements). Each mesh element is assigned mechanical property parameters (material properties) representing newly formed cartilage tissue. These mechanical property parameters include Young's modulus and Poisson's ratio. Young's modulus represents the stiffness of the material and can be non-uniformly replicated based on image readings, or assigned a typical value representing newly formed cartilage based on relevant material literature (e.g., (Select a value from 1-10 MPa) To distinguish it from mature cartilage, Poisson's ratio represents the tendency of a material to deform laterally under pressure. For cartilage tissue, this value is usually set between 0.3 and 0.45. Boundary conditions are also defined for each mesh element to constrain the displacement of the bottom boundary nodes of the model, simulating their connection with the underlying bone or rigid tissue. The displacement of the bottom nodes of the model is usually set to zero to simulate the firm bond between the repair tissue and the rigid subchondral bone. Applying a load refers to applying a uniformly distributed micro-amplitude static pressure inside the model and on the upper surface to simulate synovial fluid pressure or slight physiological contact stress. The magnitude is usually 0.1 MPa.

[0023] Each mesh element is analyzed independently: Based on elasticity theory, a stiffness characteristic description, i.e., an element stiffness matrix, is established for each mesh element. This matrix quantifies the relationship between the force and displacement at each vertex when the mesh element is subjected to force. The value depends on the shape, size, and material properties (also mechanical performance parameters) of the mesh element. The element stiffness matrices of all mesh elements are assembled according to their connection relationships to generate a global stiffness matrix. At the same time, external loads are integrated into a global load vector. Based on the principle of mechanical equilibrium, it is ensured that the forces of all connected mesh elements at each vertex of the mesh element are balanced with the external loads. The mechanical control equations for the entire repair area are established. Then, boundary conditions are introduced to generate a system of linear equations with nodal displacements as unknowns. Solving the linear equations yields the displacement solution, i.e., the stress tensor. The preset calculation rule refers to the stress-strain relationship. Then, based on the stress-strain relationship, the von Mises stress distribution of each mesh element, i.e., the equivalent stress mentioned above, is calculated. Subsequently, the maximum stress value and the average stress value of the entire repair area can be extracted based on the equivalent stress, providing quantitative data for subsequent mechanical risk assessment.

[0024] It should be noted that the repair area mentioned above refers to the area enclosed by the structural outline.

[0025] In one specific embodiment, the stress concentration factor is calculated based on the analysis results, specifically including the following steps: The analysis results include the equivalent stress at each vertex in each mesh element. The average value of all equivalent stresses is calculated as the average equivalent stress. The maximum equivalent stress among all equivalent stresses is obtained, and the result obtained by dividing the maximum equivalent stress by the average equivalent stress is used as the stress concentration factor.

[0026] Specifically, the average equivalent stress reflects the stress level after the external load applied to the repaired tissue is evenly distributed throughout the entire repaired tissue. It serves as a baseline value, representing the overall stress level of the tissue when there is no stress concentration. The average equivalent stress is used as a benchmark reference to assess the severity of stress concentration. The maximum equivalent stress identifies the mechanically weakest and most dangerous point in the entire cartilage repair area. The material at this point is closest to the critical state of damage, acting as a risk locator. The maximum equivalent stress indicates that certain local areas of the cartilage repair area are under abnormally high mechanical loads. These areas are potential starting points for microcrack initiation and tissue tearing. Dividing the maximum equivalent stress by the average equivalent stress yields the stress concentration factor. The stress concentration factor is used to quantify the factor by which local stress is amplified due to structural irregularities (such as depressions, protrusions, etc.). A stress concentration factor of 1 indicates perfectly uniform stress distribution and an ideal state in the cartilage repair area. A factor greater than 1 indicates the presence of stress concentration. The larger the stress concentration factor, the more severe the stress concentration phenomenon and the greater the negative impact of structural irregularities on the mechanical environment.

[0027] Stress concentration factor can isolate the effects of load and purely assess cartilage structural defects. During cartilage recovery, the geometry of the repair area remains constant regardless of whether the patient walks or runs (corresponding to different external loads, which affect the mean stress). The stress concentration factor eliminates the influence of load magnitude and directly reflects the quality of the structure itself. A poor-quality repair tissue will exhibit a high stress concentration factor even under low loads. In biology, stress concentration factor predicts a variety of different situations: fatigue damage, accumulation of micro-damage, or repair failure. This is because high stress concentration points accelerate the fatigue and breakage of collagen fibers, which may cause the repair tissue to gradually expand from micro-damage to larger tears, or even lead to the rupture of the repair tissue, causing pain and inflammation.

[0028] In one specific embodiment, the spatial distribution consistency index is calculated based on the structural contour line and the component distribution contour line, specifically including the following steps: A structural binary mask is generated based on the structural contour line. In the structural binary mask, the pixel values ​​inside the contour line are set to 1, and the pixel values ​​outside the contour line are set to 0. A component binary mask is generated based on the component distribution contour line. In the component binary mask, the pixel values ​​inside the component contour line are set to 1, and the pixel values ​​outside the component contour line are set to 0. A first region composed of pixels with a value of 1 in both masks is obtained. A second region composed of pixels with a value of 1 in any mask is also obtained. The result of dividing the area of ​​the first region by the area of ​​the second region is used as the spatial distribution consistency index.

[0029] Specifically, the structural contour line defines the overall region of newly formed cartilage, and the component distribution contour line defines the region enriched by a specific ECM component (i.e., type II collagen). A structural binary mask is generated based on the structural contour line. In this mask, pixels inside the contour line are set to 1 (represented as white), and pixels outside the contour line are set to 0 (represented as black), marking the region of newly formed cartilage. A component binary mask is generated based on the component distribution contour line. In this mask, pixels inside the component contour line are set to 1, and pixels outside the component contour line are set to 0, marking the region enriched by type II collagen. A first region is obtained, consisting of pixels with a value of 1 in both masks. This first region represents the area containing both structural and component binary masks. The ideal region for a specific ECM component is defined as follows: the second region refers to the area composed of all pixels with a pixel value of 1 in either the structural binary mask or the component binary mask. This represents all regions with structure but no component, with component but no structure, and with both. The area of ​​the first region divided by the area of ​​the second region is used as the spatial distribution consistency index. Specifically, the total number of pixels in the first region and the total number of pixels in the second region can be counted and divided. If the spatial distribution consistency index is equal to 1, it means that the component region and the structural region completely overlap, indicating that the structural framework of the newly formed cartilage is perfectly filled by the target ECM component, indicating excellent repair quality. If the spatial distribution consistency index is equal to 0, it means that the component region and the structural region have almost no overlap, which means that although a structure has been formed, the necessary functional components (such as type II collagen) are lacking, indicating poor repair quality.

[0030] Spatial distribution consistency index complements the shortcomings of pure morphological assessment from a biochemical perspective, and can reveal the functional maturity of cartilage repair: even if the geometric morphology is relatively regular, if the effective components are not present or misplaced, that is, if the spatial distribution consistency index is low, its mechanical function and long-term stability still cannot be guaranteed.

[0031] In one specific embodiment, the comprehensive repair quality index is calculated based on geometric features, mechanical risk features, and distribution consistency index, specifically including the following steps: The feature parameters are divided into two categories. The first category of feature parameters is subjected to a first standardization process to obtain the corresponding standardized value, and the second category of feature parameters is subjected to a second standardization process to obtain the corresponding standardized value. Different weights are set for different feature parameters. The weights of the feature parameters and their corresponding standardized values ​​are weighted and summed. The result of the weighted sum is used as the final comprehensive repair quality index.

[0032] Specifically, to integrate multiple dimensions of feature parameters into a single comprehensive score, the feature parameters are first divided into two categories. Feature parameters include geometric features, mechanical risk features, and spatial distribution consistency index. Geometric features include the rate of change of contour radius, contour concavity, and contour surface roughness. Mechanical risk features refer to the stress concentration factor. Therefore, the multiple dimensions of feature parameters include the rate of change of contour radius, contour concavity, contour surface roughness, stress concentration factor, and spatial distribution consistency index. These feature parameters are divided into two categories: the first category of feature parameters refers to those inversely proportional to cartilage repair quality; that is, the smaller these feature parameters are, the better the cartilage repair quality. The second category of feature parameters refers to those directly proportional to cartilage repair quality; that is, the larger these feature parameters are, the better the cartilage repair quality. The first category of feature parameters includes the stress concentration factor, the rate of change of contour radius, contour concavity, and contour surface roughness. The second category of feature parameters includes the spatial distribution consistency index. The first standardization process for the first category of feature parameters is: Standardized value = 1 - (Current value of feature parameter / Maximum historical value of feature parameter). The second category of feature parameters is then... The second standardization process refers to: Standardized value = Current value of characteristic parameter / Reference ideal value of characteristic parameter. The reference ideal value of characteristic parameter is calculated by statistically analyzing historical cases of good cartilage repair quality, taking the most frequently occurring value of the corresponding characteristic parameter as the reference ideal value. If the characteristic parameter is greater than the reference ideal value, the corresponding standardized value is set to 1. After obtaining the standardized value, different weights are assigned to different characteristic parameters. The weight allocation principle can be: assign the highest weight to stress concentration factor and spatial distribution consistency index, because stress concentration factor is directly related to mechanical failure risk, and spatial distribution consistency index is directly related to biological functional maturity, which is the core issue of greatest clinical concern; assign a moderate weight to contour concavity, for example, , because contour concavity reflects macroscopic defects; assign the lowest weight to contour radius change rate and contour surface roughness, because these two describe more local details and surface characteristics of risk. Other more reasonable methods can also be used to set the weights, such as weighted summation based on the weights of the characteristic parameters and the corresponding standardized values, and the weighted summation result is used as the final comprehensive repair quality index.

[0033] In one specific embodiment, risk causal tracing includes the following steps: Based on a sample database of known repair results, a standard reference value is set for each feature parameter. Based on the standard reference value and the weight value of the feature parameter, the feature contribution of each feature parameter is calculated. Based on the feature contribution and the physical and biological meaning of the corresponding feature, the risk causes are identified.

[0034] Specifically, firstly, based on a large sample database of known repair results, a standardized reference score is set for each feature parameter. For example, the average value of historically well-repaired feature parameters can be used as the corresponding standardized reference score. Then, the current feature parameter is subtracted from the standardized reference score to obtain the difference. The absolute value of the difference is taken, and multiplied by the weight set for the corresponding feature parameter to obtain the feature contribution. The larger the feature contribution, the greater the impact of the defect represented by the corresponding feature on the repair quality. The features with the largest feature contributions (one or two) are identified, and based on the physical characteristics of these features... The biological meaning is mapped to the preset main risk cause category. For example, if the characteristic contribution of the stress concentration coefficient is the highest, it is judged as the main risk of mechanical weakness, indicating that there are sharp stress concentration points in the structure that are very easy to damage. If the characteristic contribution of the spatial distribution consistency index is the highest, it is judged as the main risk of abnormal component distribution, indicating that the repair tissue lacks necessary components such as type II collagen. If the characteristic contribution of the contour concavity is the highest, it is judged as the main risk of structural defect, indicating that there is serious tissue loss. If the characteristic contribution of the contour radius change rate or contour roughness is the highest, it is judged as structural irregularity, indicating that there are morphological problems such as poor integration and surface ulceration.

[0035] In one specific embodiment, different weights are set for different feature parameters, specifically including the following steps: A dataset of cartilage repair samples was collected. For each sample, the corresponding feature parameters and clinical standard scores were obtained. The standardized feature parameters were used as independent variables and the clinical standard scores were used as dependent variables for multiple regression analysis. After the regression analysis, a corresponding regression coefficient was calculated for each independent variable. These regression coefficients were standardized and used as the weights of the corresponding feature parameters.

[0036] Specifically, to more rationally set weights for the feature parameters, corresponding feature parameters and corresponding clinical standard scores are obtained based on a dataset of historical cartilage repair samples. The clinical standard scores are determined by multiple senior pathology experts based on experience and feature parameter data to determine the corresponding repair level for each feature parameter. Then, the standardized feature parameters (i.e., the standardized values ​​of the feature parameters) are used as independent variables. The standardization method can use the two standardization methods in the step of calculating the comprehensive repair quality index. The clinical standard scores are used as dependent variables for multiple regression analysis. After the regression analysis, a corresponding regression coefficient is calculated for each independent variable, i.e., each feature parameter. These regression coefficients are standardized (e.g., normalized so that their sum is 1). The standardized coefficients are used as the weights of the corresponding feature parameters. The above method learns from a large number of real samples through a mathematical model to determine the weights, directly reflecting the actual predictive ability and contribution of each feature parameter to the final clinical outcome, making the weight setting more scientific and reasonable.

[0037] The above describes a deep learning-based ECM-cartilage repair image analysis method according to embodiments of this application. The following describes a deep learning-based ECM-cartilage repair image analysis system according to embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the deep learning-based ECM-cartilage repair image analysis system in this application includes: The acquisition unit acquires multimodal medical image data during the cartilage repair process. The multimodal medical image data includes a first modality image and a second modality image. The processing unit uses a first deep learning segmentation model to process the first modality image and extract the structural contour lines of the newly formed ECM region, and uses a second deep learning segmentation model to process the second modality image and extract the component distribution contour lines of specific ECM components. The calculation unit calculates geometric features based on the structural contour line, constructs a stress analysis model based on the structural contour line, performs analysis based on the stress analysis model, calculates the stress concentration factor as a mechanical risk feature based on the analysis results, and also calculates the spatial distribution consistency index based on the structural contour line and the component distribution contour line. The analysis unit calculates the comprehensive repair quality index based on geometric features, mechanical risk features, and distribution consistency index, and also traces the causes of risks to generate a comprehensive assessment report.

[0038] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0039] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0040] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A deep learning-based ECM-cartilage repair image analysis method, characterized by, The method comprises: Step S1, collecting multi-modal medical image data in the cartilage repair process, the multi-modal medical image data comprising a first modality image and a second modality image; Step S2, processing the first modality image using a first deep learning segmentation model and extracting a structure contour line of a new ECM region, processing the second modality image using a second deep learning segmentation model and extracting a component distribution contour line of a specific ECM component; Step S3, calculating a geometric feature based on the structure contour line, constructing a stress analysis model based on the structure contour line, analyzing based on the stress analysis model, calculating a stress concentration coefficient as a mechanical risk feature based on the analysis result, and further calculating a spatial distribution consistency index based on the structure contour line and the component distribution contour line; Step S4, calculating a comprehensive repair quality index based on the geometric feature, the mechanical risk feature and the distribution consistency index, and further performing risk cause tracing to generate a comprehensive evaluation report.

2. The method of claim 1, wherein, The geometric feature is calculated based on the structure contour line, comprising: Taking the center of the minimum circumscribed circle of the structure contour line as a reference center point, calculating the radial distance between the reference center point and a plurality of measurement points distributed at a predetermined angle interval on the structure contour line, then calculating the radial distance variation rate between adjacent measurement points, and further calculating the variance of all distance variation rates as the contour line radius variation rate; Calculating the shortest distance from each measurement point on the structure contour line to the edge of the minimum circumscribed circle, and taking the maximum value of all shortest distances as the contour line concave.

3. The method of claim 2, wherein, The geometric feature is calculated based on the structure contour line, further comprising: Generating a radial distance sequence by obtaining the radial distance of each measurement point of the structure contour line, applying Gaussian filtering to the radial distance sequence to smooth the radial distance sequence to obtain a new radial distance sequence, for each measurement point, calculating the deviation value of the original radial distance and the smoothed radial distance, calculating the root mean square error of all deviation values, and taking the root mean square error as the contour line surface roughness.

4. The method of claim 1, wherein, The analysis is performed based on the stress analysis model, comprising: Converting the structure contour line into a two-dimensional plane geometry model as a stress analysis model, discretizing the stress analysis model into grid elements, assigning mechanical property parameters representing the new cartilage tissue to each grid element, defining boundary conditions and applying loads, generating a linear equation system based on the mechanical property parameters of each grid element, boundary conditions and loads, solving the linear equation system to calculate the stress tensor of each grid element, and converting the stress tensor into equivalent equivalent stress based on a pre-designed calculation rule, and the analysis result is the equivalent stress of each grid element.

5. The method of claim 1, wherein, The stress concentration coefficient is calculated based on the analysis result, comprising: The analysis result includes the equivalent stress of each vertex in each grid element, the average value of all equivalent stresses is calculated as the average equivalent stress, the maximum equivalent stress in all equivalent stresses is obtained, and the result value obtained by dividing the maximum equivalent stress by the average equivalent stress is taken as the stress concentration coefficient.

6. The method of claim 1, wherein, The spatial distribution consistency index is calculated based on the structure contour line and the component distribution contour line, comprising: The structural binary mask is generated based on the structural contour line, in which the pixel value inside the contour line is set to 1 and the pixel value outside the contour line is set to 0; the component binary mask is generated based on the component distribution contour line, in which the pixel value inside the component contour line is set to 1 and the pixel value outside the component contour line is set to 0; a first region composed of pixel points with a pixel value of 1 in both masks is obtained, and a second region composed of pixel points with a pixel value of 1 in any mask is also obtained; and a result value obtained by dividing the area of the first region by the area of the second region is taken as the spatial distribution consistency index.

7. The method of claim 1, wherein, The comprehensive repair quality index is calculated based on the geometric feature, the mechanical risk feature and the distribution consistency index, including: The feature parameters are divided into two categories, the first standardization processing is performed on the first category of feature parameters to obtain corresponding standardized values, the second standardization processing is performed on the second category of feature parameters to obtain corresponding standardized values, different weights are set for different feature parameters, and weighted summation is performed based on the weights of the feature parameters and the corresponding standardized values, and the result value of the weighted summation is taken as the final comprehensive repair quality index.

8. The method of claim 7, wherein, The risk cause is traced, including: Based on the sample database of known repair results, a standard reference value is set for each feature parameter, the feature contribution degree of each feature parameter is calculated based on the standard reference value and the weight value of the feature parameter, and the risk cause is identified based on the feature contribution degree and the physical and biological meaning of the corresponding feature.

9. The method of claim 7, wherein, Different weights are set for different feature parameters, including: A data set of cartilage repair samples is collected, for each sample, the corresponding feature parameters are obtained, the clinical standard score is obtained, the standardized feature parameters are taken as independent variables, and the clinical standard score is taken as dependent variable, and multiple regression analysis is performed, after regression analysis, a corresponding regression coefficient is calculated for each independent variable, the regression coefficients are standardized, and the standardized regression coefficients are taken as the weights of the corresponding feature parameters.

10. A deep learning-based ECM-cartilage repair image analysis system for implementing a deep learning-based ECM-cartilage repair image analysis method according to any one of claims 1-9, characterized in that, The system includes: The acquisition unit acquires multi-modal medical image data in the cartilage repair process, and the multi-modal medical image data includes first modality images and second modality images; The processing unit processes the first modality images using a first deep learning segmentation model and extracts structural contour lines of the new ECM region, and processes the second modality images using a second deep learning segmentation model and extracts component distribution contour lines of specific ECM components; The calculation unit calculates the geometric feature based on the structural contour line, constructs a stress analysis model based on the structural contour line, analyzes based on the stress analysis model, calculates the stress concentration coefficient as the mechanical risk feature based on the analysis result, and also calculates the spatial distribution consistency index based on the structural contour line and the component distribution contour line; The analysis unit calculates the comprehensive repair quality index based on the geometric feature, the mechanical risk feature and the distribution consistency index, and also traces the risk cause, and generates a comprehensive evaluation report.

Images