Medical data mining and analysis system based on cardiac ultrasound image stream

Abstract

The invention relates to the technical field of medical data processing and computer-aided surgical planning, in particular to a medical data mining and analyzing system based on cardiac ultrasound image streams. The system comprises a data fusion module, a map construction module, a risk prediction module and a decision module. The system obtains a left auricle ultrasonic sequence and mechanical parameters, and constructs a space-time aligned dynamic biomechanical characteristic spectrum; the core of the method is to simulate the dynamic adherence performance of an occluder based on a map, calculate stability, deformation and residual shunt risk indexes, and output risk grades through a model; according to the method, the problem of spatio-temporal dislocation of morphological and mechanical data is effectively solved, digital twinning with authenticity and mechanical accuracy is realized, and accurate decision support is provided for left atrial appendage occlusion.

Description

Technical Field

[0001] This invention relates to the field of medical data processing and computer-aided surgical planning technology, specifically a medical data mining and analysis system based on cardiac ultrasound image streams. Background Technology

[0002] In the preoperative planning of left atrial appendage occlusion, doctors need to select the occluder and make risk predictions based on medical data such as echocardiogram images. Current methods generally adopt a static assessment model, which measures the two-dimensional or three-dimensional geometric dimensions of the left atrial appendage at a specific moment and matches device specifications with clinical experience. Although this method can meet the basic initial size screening, the left atrial appendage is a non-linear soft tissue that undergoes complex dynamic deformation during the cardiac cycle. Furthermore, the imaging data alone lacks spatiotemporal alignment with the tissue's mechanical characteristics, making it difficult to reflect the true tissue compliance. This static method, which ignores dynamic interaction mechanisms, cannot capture the micro-displacement changes and instantaneous loss of contact between rigid devices and flexible tissues during the heartbeat. This makes it difficult to quantify and assess the risk of delayed residual shunt and device instability after surgery, and the selection results often rely on subjective judgment and lack physical basis. Therefore, how to eliminate the spatiotemporal bias of multi-source heterogeneous data and achieve accurate risk quantification and classification based on dynamic interactive simulation has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a medical data mining and analysis system based on cardiac ultrasound image streams. Specifically, the technical solution of this invention includes: A medical data processing platform is used to perform the following steps to support device selection and risk assessment decisions for left atrial appendage occlusion: S1. Acquire multimodal cardiac medical data of the target object, wherein the data includes at least an ultrasound image sequence reflecting the dynamic morphology of the left atrial appendage and a parameter set reflecting its tissue mechanical characteristics; S2. The multimodal cardiac medical data is fused and feature-mined to generate a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage; S3. Based on the left atrial appendage dynamic biomechanical feature map, call the preset occluder compliance risk prediction model to simulate and calculate the dynamic wall adhesion performance dataset under the candidate occluder parameters; S4. Calculate multidimensional risk assessment indicators from the dynamic wall-adhering performance dataset, wherein the indicators include at least the wall-adhering stability index, the deformation compliance index, and the residual diversion risk index. S5. Input the multidimensional risk assessment indicators into the pre-trained clinical risk grading decision model, and output the wall adhesion stability level and dynamic residual shunt risk level for the current candidate occluder parameters as clinical decision support information.

[0004] Preferably, in step S2, generating a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage includes: S21. Extract the temporal information of the volume change of the left atrial appendage during the cardiac cycle from the ultrasound image sequence; S22. Extract the strain and strain rate distribution information of the left atrial wall from the parameter set; S23. Using the standard cardiac cycle as the time axis, the volume change time sequence information and the strain and strain rate distribution information are spatiotemporally aligned and fused to form the dynamic biomechanical feature map of the left atrial appendage.

[0005] Preferably, step S3 involves calling the occluder compliance risk prediction model, including: S31. The dynamic biomechanical feature map of the left atrial appendage, together with the input candidate occluder model and size parameters, are used as model input; S32. The model calculates the dynamic interaction between tissue and instrument during a simulated cardiac cycle and outputs a dynamic adhesion performance dataset including contact area, tissue deformation, and gap distribution.

[0006] Preferably, the wall adhesion stability index calculated in step S4 is obtained through the following method: S41. Extract the instantaneous contact area sequence between the occluder and the inner wall of the left atrial appendage during the simulated cardiac cycle from the dynamic wall adhesion performance dataset; S42. Calculate the integral mean of the instantaneous contact area sequence over a complete cardiac cycle, and compare it with the normalized surface area of ​​the occluder to generate the wall adhesion stability index.

[0007] Preferably, the deformation compliance index calculated in step S4 is obtained in the following way: S51. Obtain simulated key point deformation data of the left atrial appendage after loading the occluder from the dynamic adhesion performance dataset; S52. Retrieve the intraoperative measured deformation reference data corresponding to the simulated key points; S53. Calculate the deviation measure between the simulated key point deformation data and the intraoperative measured deformation reference data, and generate the deformation conformity index after normalization.

[0008] Preferably, the residual diversion risk index calculated in step S4 is obtained through the following method: S61. Based on the dynamic adhesion performance dataset, analyze the distribution of instantaneous micro-gaps between the occluder edge and tissue at each phase of the cardiac cycle; S62. Calculate the probability of blood shunting caused by each instantaneous micro-gap based on the principle of fluid dynamics, and construct a shunting risk probability distribution accordingly; S63. Calculate the information entropy value of the diversion risk probability distribution and define it as the residual diversion risk index.

[0009] Preferably, the output rules of the clinical risk stratification decision model in step S5 include: S71. Preset the safety threshold for each risk assessment indicator; S72. If the wall-attachment stability index is lower than its safety threshold, or the deformation compliance index is lower than its safety threshold, or the residual diversion risk index is higher than its safety threshold, then the decision model outputs a high-risk level judgment. S73. If all risk assessment indicators meet their corresponding safety thresholds, the decision model will output a low-risk level judgment.

[0010] Preferably, the system further includes: The decision feedback optimization module is used to generate occluder parameter adjustment suggestions based on the deviation of each indicator from the threshold when the clinical risk grading decision model outputs a high-risk level judgment, and restart the process of steps S3 to S5 for iterative evaluation until a low-risk level judgment is output.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This system constructs a dynamic biomechanical feature atlas of the left atrial appendage by strictly aligning and fusing ultrasound image sequences reflecting morphological changes with parameter sets reflecting tissue mechanical characteristics using the standard cardiac cycle as the time axis. This mechanism effectively solves the problem of two separate skins caused by the difference in temporal resolution between morphological and mechanical data, as pointed out in the background technology. It can generate a digital twin with both morphological realism and mechanical accuracy, ensuring that the subsequent risk prediction model can be calculated based on the real tissue physical properties, thereby significantly improving the ability to restore the dynamic characteristics of the anatomical structure of a specific patient. 2. This system overcomes the limitations of traditional static dimensional measurement by simulating the dynamic wall-attachment performance of candidate occluders during the cardiac cycle and generating a wall-attachment stability index based on the integral mean over the entire cycle. This technical solution can keenly capture micro-displacement changes and instantaneous contact loss caused by intense movements such as atrial contraction that cannot be observed in static images, thereby quantifying the stability of the device's ability to maintain a seal throughout the cardiac cycle and providing clinicians with intuitive and quantitative physical evidence regarding the risk of device displacement after surgery. 3. This system utilizes fluid dynamics principles to analyze the instantaneous micro-gap distribution between the occluder edge and tissue, and introduces the concept of information entropy to construct a residual shunt risk index. This innovation abstracts complex fluid dynamics phenomena into a single entropy value index, which can detect stubborn shunt channels that, although geometrically small, persist with the heartbeat. Especially when the shunt probability distribution is uniform, i.e., entropy increases, this index can effectively indicate the possibility of unpredictable blood leakage, thereby avoiding the risk of delayed postoperative residual shunt caused by ignoring dynamic blood flow impact. 4. This system achieves a shift from experience-based trial and error to computer-aided automated optimization through a decision feedback optimization module. When risk assessment indicators deviate from the safety threshold, the system can automatically perform attribution analysis and generate parameter adjustment suggestions, such as adjusting the disc diameter or implantation depth, and iteratively calculate until the optimal parameter combination that balances stability and sealing is found. This automated process based on multi-indicator joint constraints not only avoids subjective errors caused by relying solely on doctors' experience, but also greatly shortens the preoperative selection and planning time for complex anatomical structures. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. Example 1:

[0014] Please see Figure 1 A medical data mining and analysis system based on cardiac ultrasound image stream, comprising: A medical data processing platform is used to perform the following steps to support device selection and risk assessment decisions for left atrial appendage occlusion: S1. Acquire multimodal cardiac medical data of the target object, the data including at least ultrasound image sequences reflecting the dynamic morphology of the left atrial appendage and a set of parameters reflecting its tissue mechanical characteristics; S2. Fuse and feature-mining multimodal cardiac medical data to generate a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage; S3. Based on the dynamic biomechanical feature map of the left atrial appendage, the preset occluder compliance risk prediction model is called to simulate and calculate the dynamic wall adhesion performance dataset under the parameters of the candidate occluders. S4. Calculate multidimensional risk assessment indicators from the dynamic wall-adhering performance dataset. The indicators should include at least the wall-adhering stability index, the deformation compliance index, and the residual diversion risk index. S5. Input the multidimensional risk assessment indicators into the pre-trained clinical risk grading decision model, and output the wall adhesion stability level and dynamic residual shunt risk level for the current candidate occluder parameters as clinical decision support information.

[0015] This embodiment details the specific execution logic of the medical data processing platform, aiming to solve the selection error problem caused by neglecting the nonlinear deformation of the left atrial appendage in existing technologies. The system executes the data acquisition steps, connecting a real-time three-dimensional transesophageal ultrasound device and a three-dimensional speckle tracking imaging workstation through a high-speed interface to acquire multimodal cardiac medical data. Among them, the ultrasound image sequence serves as a morphological data source, providing a high frame rate geometric boundary change flow of the left atrial appendage during systole and diastole. The parameter set serves as a functional data source, containing the tissue mechanical characteristics reflecting the longitudinal and circumferential strain of each segment of the left atrial appendage wall during the cardiac cycle. The platform utilizes a spatiotemporal registration algorithm to process the aforementioned heterogeneous data, constructing a dynamic biomechanical feature map of the left atrial appendage. This map, acting as a digital twin, maps tissue stiffness and elastic modulus onto a dynamic geometric mesh. The system invokes an occluder compliance risk prediction model, employing finite element analysis and fluid-structure interaction technology to simulate the dynamic interaction between a rigid occluder and flexible left atrial appendage tissue in a virtual environment, generating a dynamic adhesion performance dataset. Based on this, the system extracts three key quantitative indicators from massive simulation data: adhesion stability index, deformation compliance index, and residual shunt risk index. The clinical risk grading decision model, based on a pre-trained expert knowledge base, responds to the input indicator values ​​and outputs a risk level judgment for the current disc diameter or umbrella depth parameters. This embodiment elevates occluder selection from static size matching to dynamic compliance simulation by constructing a dynamic atlas that includes tissue biomechanical characteristics. This analysis mechanism based on multimodal data fusion enables doctors to predict the micro-displacement changes of the occluder during cardiac beats before surgery, effectively avoiding the risk of delayed residual shunts after surgery due to differences in tissue compliance. Example 2:

[0016] Step S2 generates a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage, including: S21. Extract the temporal information of volume changes of the left atrial appendage during the cardiac cycle from ultrasound imaging sequences; S22. Extract strain and strain rate distribution information of the left atrial wall from the parameter set; S23. Using the standard cardiac cycle as the time axis, the temporal and spatial alignment and data fusion of volume change time sequence information and strain and strain rate distribution information are performed to form a dynamic biomechanical feature map of the left atrial appendage.

[0017] This embodiment further specifies the atlas generation steps in Embodiment 1, aiming to solve the problem of temporal resolution differences between morphological and mechanical data; the system automatically segments real-time three-dimensional transesophageal ultrasound image sequences, extracts the left atrial appendage endothelial boundary, and constructs a dynamic mesh model that changes over time. This allows the system to obtain temporal information on the volume change of the left atrial appendage as the cardiac cycle phase changes. Simultaneously, the system extracts strain and strain rate distribution information of the region of interest from the 3D speckle tracking imaging data stream. This information directly characterizes the tensile capacity of myocardial fibers and the stiffness of local tissues. Strict spatiotemporal alignment and data fusion operations are performed. Specifically, the spatiotemporal registration algorithm employs a cubic spline interpolation function. Discrete timestamps of ultrasound images Mapping to normalized cardiac cycle phase The interpolation formula is ,in, These are the interpolation coefficients. For local time variables; where, coefficients The system constructs and solves for key phase points of the cardiac cycle, such as the peak of the R wave. With the end of the T wave The volume and strain values, along with their first and second derivative continuity conditions, are obtained from a system of linear equations; the specific mathematical expression of the system of linear equations is: ; in, and These represent the time at which the left atrial appendage begins the cardiac cycle. and termination time The volume value, in mL, represents the filling and emptying state of the auricle; and These represent the first derivatives of the volume with respect to time at the start and end times, respectively, i.e., the instantaneous rate of change of volume, in mL / s, used to describe the contraction and relaxation rates of the atrial appendage wall; The function values ​​are the start and end times. The first-order derivative values ​​at the start and end times are given. Solving this system of equations ensures the smoothness of the interpolation curve. Experiments verify that this spatiotemporal alignment algorithm controls the average phase alignment error of multimodal data within a certain range. Within this range, compared to traditional linear interpolation, the overlap of characteristic peaks is improved. In the spatial dimension, the extracted strain tensor is projected using a projection algorithm. Mapping to dynamic mesh model corresponding node The above calculation uses the inverse distance weighted interpolation method, and the specific formula is as follows: ; in, Indicates the first Equivalent scalar strain values ​​for a known data point; With the point to be interpolated as the center and radius... The total number of known data points contained within the spherical neighborhood; Indicates the point to be interpolated and the first... The geometric distance between neighboring data points in three-dimensional Euclidean space, in mm; To prevent regularization parameters from having a denominator of zero, the value is set to [value to be filled in]. This value ensures the stability of numerical calculations while avoiding excessive smoothing of distance weights. This embodiment successfully eliminates phase deviation of multi-source heterogeneous data through a spatiotemporal alignment mechanism centered on the standard cardiac cycle; the generated feature map has both morphological realism and mechanical accuracy, ensuring that subsequent simulation steps can be calculated based on real tissue physical properties, thereby significantly improving the biofidelity of the risk prediction model to the anatomical structure of specific patients. Example 3:

[0018] Step S3 involves calling the bottleneck compliance risk prediction model, including: S31. Use the dynamic biomechanical feature map of the left atrial appendage and the input candidate occluder model and size parameters as model input; S32. The model calculates the dynamic interaction between tissue and instrument during a simulated cardiac cycle and outputs a dynamic adhesion performance dataset that includes contact area, tissue deformation, and gap distribution.

[0019] This embodiment further specifies the risk prediction model invocation steps in Embodiment 1, with the core being the simulation of nonlinear contact processes. The system loads the left atrial appendage dynamic biomechanical feature map generated in step S2 as boundary conditions and material properties, while simultaneously importing a candidate occluder CAD model containing the mechanical parameters of the nickel-titanium alloy skeleton and the characteristics of the flow-blocking membrane. Specifically, the mapping between tissue hardness and elastic modulus adopts the Neo-Hookean hyperelastic constitutive model. ; in, It is the first invariant of the right Cauchy-Green deformation tensor, characterizing the degree of tensile deformation of the material; is the Jacobian determinant of the deformation gradient, representing the rate of volume change; The initial shear modulus of the material, in kPa; Bulk modulus, unit: kPa, used to control the incompressible properties of soft tissues; The strain energy density; in this embodiment, Values Specifically, shear modulus Derived from the local strain rate, its specific functional expression is as follows: ; in, ε is the viscoelastic coefficient, in units of s, used to characterize the damping properties of a tissue under dynamic deformation; For local strain rate; local strain rate The equivalent scalar value is obtained from the strain rate tensor extracted in step S2. The second invariant is calculated, and its specific expression is: ; This calculation ensures that the three-dimensional tensor data can be correctly mapped to the scalar input required by the constitutive model; in this embodiment, Values , Values This parameter setting accurately reflects the dynamic hardening characteristics of the left atrial appendage tissue during the cardiac cycle; the model performs nonlinear contact analysis and uses the penalty function method to calculate the contact force. The formula is ,in, For the penalty stiffness coefficient, Wherein, The value of is determined through convergence testing, and the system in Perform step-by-step calculations within the range, selecting a value that ensures the maximum penetration depth of the node is less than [the specified value]. And the increase rate of computation time is lower than The value, in this embodiment, is calibrated to be [value]. By comparing the measured contact pressure data with that of an in vitro silicone biomimetic model, the average relative error of the model's calculation results is [missing value]. The biomechanical accuracy of the simulation logic was verified; the dynamic balance between the radial support force of the occluder and the recoil force of the left atrial appendage was calculated during the simulated cardiac cycle. The output includes tensor data with a time dimension, namely the dynamic adhesion performance dataset; this dataset records the contact area between the occluder surface nodes and the left atrial appendage inner wall nodes at each time step, used to quantify the size of the sealing area; records tissue deformation, used to quantify the geometric changes after compression; and records the gap distribution, used to quantify the topology of the uncontacted area. This embodiment utilizes a dynamic interaction computing mechanism to capture instantaneous detachment risks or hidden diversion channels that cannot be observed in static images. This simulation output based on physical mechanisms provides a quantitative physical basis for accurate selection, avoiding subjective errors caused by relying solely on experience. Example 4:

[0020] The wall-attachment stability index calculated in step S4 is obtained as follows: S41. Extract the instantaneous contact area sequence between the occluder and the inner wall of the left atrial appendage during the simulated cardiac cycle from the dynamic wall adhesion performance dataset; S42. Calculate the integral mean of the instantaneous contact area sequence over a complete cardiac cycle, and compare it with the normalized surface area of ​​the occluder to generate the wall adhesion stability index.

[0021] This embodiment is a further specification of the calculation logic for the wall adhesion stability index in Embodiment 1; this index The calculation formula is as follows: ; in, The mean RR intervals were extracted from electrocardiogram signals. Sourced from the medical device manufacturer's specification database. Derived from the dynamic wall-attaching performance dataset; The system extracts the time-varying instantaneous contact area sequence from the simulation output. ; for this sequence in a complete cardiac cycle Integral calculations are performed to obtain the time average of the contact area; this average is then compared with the theoretical surface area. Comparison and normalization are performed to generate the adhesion stability index; This embodiment eliminates the bias caused by observation at a single moment by calculating the integral mean over the entire cycle; this index can keenly reflect the contact loss of the occluder during intense motion phases such as atrial contraction, thereby quantifying the stability of the device's ability to maintain a seal throughout the cardiac cycle and providing a direct warning for clinicians regarding the risk of device displacement. Example 5:

[0022] The deformation compliance index calculated in step S4 is obtained in the following way: S51. Obtain simulated key point deformation data of the left atrial appendage after loading the occluder from the dynamic wall adhesion performance dataset, as predicted by the model. S52. Retrieve the intraoperative measured deformation reference data corresponding to the simulated key points; S53. Calculate the deviation measure between the simulated key point deformation data and the intraoperative measured deformation reference data, and generate the deformation conformity index after normalization.

[0023] This embodiment further specifies the calculation logic of the deformation compliance index in Embodiment 1, aiming to verify the credibility of the simulation model; this index The calculation formula is as follows: ; in, For the number of key points, For the first Several key morphological parameters in the simulation, such as the diameter of the anchoring zone, For the corresponding target reference shape data; where, The simulation output is derived from step S3; The reference values ​​are derived from the predicted reference values ​​generated by searching a high-similarity historical case database, or from the diameter and depth of the left atrial appendage anchoring zone measured in an in vitro 3D printed bionic model experiment, and are used as reference data. The system locates key anatomical positions such as the diameter and depth of the anchoring zone from the dataset and obtains deformation data of simulated key points; retrieves the corresponding intraoperative measured data; calculates the average relative error of the simulated value relative to the measured value; and converts the error value into a conformity index to generate a deformation conformity index. This embodiment introduces the deformation conformity index as a confidence index, which effectively assesses the difference between simulation results and actual anatomical conditions. When the index is low, it indicates that the tissue characteristics of the current individual may have abnormalities such as extreme fibrosis, thereby warning doctors to downgrade the reference value of the simulation results and ensuring the rigor of the risk assessment system. Example 6:

[0024] The residual diversion risk index calculated in step S4 is obtained in the following way: S61. Based on the dynamic adhesion performance dataset, analyze the distribution of instantaneous micro-gaps between the occluder edge and tissue at each phase of the cardiac cycle; S62. Calculate the probability of blood shunting caused by each instantaneous micro-gap based on the principle of fluid dynamics, and construct a shunting risk probability distribution accordingly; S63. Calculate the information entropy value of the probability distribution of diversion risk and define it as the residual diversion risk index.

[0025] This embodiment further specifies the calculation logic of the residual diversion risk index in Embodiment 1, introducing the concept of information entropy to assess the risk of complex fluid dynamics; this index The calculation formula is as follows: Initial blood shunt probability sequence for each time phase within the cardiac cycle Normalization is performed to obtain the normalized probability distribution. The calculation formula is: ; Calculate the time distribution entropy based on the normalized probability distribution: ; in, It is a very small positive number, and its value is [value]. This is used to prevent zero overflow errors in logarithmic operations; As a preset constant, For the first The probability of blood shunting at each time phase; The calculation is based on the orifice flow model and the Logistic function: calculating the split flow rate. In this formula, the parameter For flow coefficient, The instantaneous micro-gap area, For translobar pressure difference, For blood density; in this formula, based on experimental data from the sharp-edge orifice, the parameters are... Set as ; Set as Standard blood density; The peak jet velocity vp (in m / s) was extracted from color Doppler flow imaging, and a simplified Bernoulli equation was applied. The coefficients are calculated. Used to complete from arrive Unit conversion, and then to obtain ; in, Sensitivity factor The critical flow threshold; where, the critical flow threshold Set as This corresponds to the minimum jet signal intensity detectable by clinical color Doppler ultrasound; sensitivity factor. Set as This section explicitly states that it possesses inverse flow dimension to ensure the dimensionless nature of the exponential function parameters. This value is derived from a logistic regression analysis of historical cases' shunt flow velocities and postoperative thrombosis risk. Validated with 30 retrospective cases, this index achieves a high accuracy rate in predicting postoperative residual shunt. This effectively quantifies the hydrodynamic risks of tiny gaps; The system analyzes the instantaneous micro-gap distribution between the occluder edge and the tissue, and calculates the shunting probability of each phase using fluid dynamics principles. Construct a full-cycle diversion risk probability distribution; calculate the information entropy value of this distribution using the Shannon entropy principle to generate a residual diversion risk index; This embodiment abstracts complex fluid dynamics phenomena into a single entropy index, breaking through the limitations of traditional geometric gap measurement. This index can keenly capture the persistent risk of shunting that exists with heartbeats even though the geometric gap is small. Especially when the shunting probability distribution is uniform, i.e., entropy increases, it effectively indicates the possibility of unpredictable blood leakage. Example 7:

[0026] The output rules of the clinical risk grading decision model in step S5 include: In this embodiment, the clinical risk grading decision model is essentially a logical judgment module based on a preset threshold. S71. Preset the safety threshold for each risk assessment indicator; S72. If the wall-attachment stability index is lower than its safety threshold, or the deformation compliance index is lower than its safety threshold, or the residual diversion risk index is higher than its safety threshold, the decision model will output a high-risk level judgment. S73. If all risk assessment indicators meet their corresponding safety thresholds, the decision model will output a low-risk level judgment.

[0027] This embodiment further specifies the output rules of the decision model in Embodiment 1; the system loads a set of safety thresholds determined based on large-sample retrospective studies, including wall-attaching stability thresholds. Deformation compliance threshold and residual diversion risk threshold Specifically, each threshold system is determined by... For example, in the analysis of receiver operating characteristic curves of historical surgical data, the point with the maximum Youden index is selected as the determining factor. In this embodiment, the following is set... , , The verification results show that the model applying this set of thresholds has a high sensitivity in identifying high-risk blocking schemes. Specificity is ; The decision model performs Boolean logic operations; responding to the wall-attached stability index. Below Or deformation compliance index Below or residual diversion risk index Higher than In any of these cases, the model outputs a high-risk level judgment; conversely, the model outputs a low-risk level judgment only when all indicators simultaneously meet their corresponding safety threshold requirements. This embodiment constructs a dual insurance mechanism for both blocking effectiveness and prediction reliability through logical rules that combine multiple indicators. This mechanism not only focuses on the physical performance of the device, but also monitors the accuracy of the simulation itself through deformation compliance indicators, effectively avoiding erroneous decisions caused by model distortion, and providing timely and reliable traffic light-style auxiliary information for clinical practice. Example 8:

[0028] The system also includes: The decision feedback optimization module is used to generate occluder parameter adjustment suggestions based on the deviation of each indicator from the threshold when the clinical risk grading decision model outputs a high-risk level judgment, and restarts the process of steps S3 to S5 for iterative evaluation until a low-risk level judgment is output.

[0029] Specifically, the following feedback iteration strategy is implemented: The adjustment step size of the occluder parameters is set. Determined by the following formula: ; in, The preset adjustment coefficient has a value of [value missing]. , The target threshold for the wall adhesion stability index is... This is the current calculated value; if the adjusted parameters lead to a residual diversion risk index... If the step size increases, a backtracking mechanism is triggered, reducing the step size to... And recalculate; at the same time, set the maximum number of iterations. When the number of iterations If no low-risk combination is found, the system outputs the parameter scheme with the highest comprehensive score and terminates the loop as the final recommendation result, without automatically restarting the iteration process. This embodiment extends the functionality of the system in Embodiment 1 by adding closed-loop optimization capabilities. When the decision model outputs a high-risk level, the decision feedback optimization module automatically initiates attribution analysis to identify the specific indicators leading to the high risk. The system generates parameter adjustment suggestions based on the deviation of the indicators. For example, in response to a low wall adhesion stability index, it suggests increasing the occluder disc diameter, or in response to a high residual shunt risk index, it suggests adjusting the implantation depth. The system automatically loads new occluder parameters and re-triggers the simulation and evaluation process in steps S3 to S5. This iterative process continues until the optimal parameter combination for determining a low-risk level is obtained. This embodiment transforms the doctor's experiential trial and error into a computer-aided automated optimization process. In complex anatomical scenarios, this module can quickly traverse multiple parameter combinations, significantly reducing the time cost of preoperative planning and helping doctors find the theoretically optimal implantation solution that balances stability and sealing.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A medical data mining and analysis system based on cardiac ultrasound video stream, characterized in that, include: A medical data processing platform is used to perform the following steps to support device selection and risk assessment decisions for left atrial appendage occlusion: S1. Acquire multimodal cardiac medical data of the target object, wherein the data includes at least an ultrasound image sequence reflecting the dynamic morphology of the left atrial appendage and a parameter set reflecting its tissue mechanical characteristics; S2. The multimodal cardiac medical data is fused and feature-mined to generate a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage; S3. Based on the left atrial appendage dynamic biomechanical feature map, call the preset occluder compliance risk prediction model to simulate and calculate the dynamic wall adhesion performance dataset under the candidate occluder parameters; S4. Calculate multidimensional risk assessment indicators from the dynamic wall-adhering performance dataset, wherein the indicators include at least the wall-adhering stability index, the deformation compliance index, and the residual diversion risk index. S5. Input the multidimensional risk assessment indicators into the pre-trained clinical risk grading decision model, and output the wall adhesion stability level and dynamic residual shunt risk level for the current candidate occluder parameters as clinical decision support information.

2. The medical data mining and analysis system based on cardiac ultrasound image stream according to claim 1, characterized in that, Step S2 generates a spatiotemporally aligned dynamic biomechanical feature map of the left atrial appendage, including: S21. Extract the temporal information of the volume change of the left atrial appendage during the cardiac cycle from the ultrasound image sequence; S22. Extract the strain and strain rate distribution information of the left atrial wall from the parameter set; S23. Using the standard cardiac cycle as the time axis, the volume change time sequence information and the strain and strain rate distribution information are spatiotemporally aligned and fused to form the dynamic biomechanical feature map of the left atrial appendage.

3. The medical data mining and analysis system based on cardiac ultrasound image stream according to claim 1, characterized in that, Step S3 involves calling the bottleneck compliance risk prediction model, including: S31. The dynamic biomechanical feature map of the left atrial appendage, together with the input candidate occluder model and size parameters, are used as model input; S32. The model calculates the dynamic interaction between tissue and instrument during a simulated cardiac cycle and outputs a dynamic adhesion performance dataset including contact area, tissue deformation, and gap distribution.

4. A medical data mining and analysis system based on cardiac ultrasound image stream according to claim 3, characterized in that, The wall-attachment stability index calculated in step S4 is obtained as follows: S41. Extract the instantaneous contact area sequence between the occluder and the inner wall of the left atrial appendage during the simulated cardiac cycle from the dynamic wall adhesion performance dataset; S42. Calculate the integral mean of the instantaneous contact area sequence over a complete cardiac cycle, and compare it with the normalized surface area of ​​the occluder to generate the wall adhesion stability index.

5. A medical data mining and analysis system based on cardiac ultrasound image stream according to claim 4, characterized in that, The deformation compliance index calculated in step S4 is obtained in the following way: S51. Obtain simulated key point deformation data of the left atrial appendage after loading the occluder from the dynamic adhesion performance dataset; S52. Retrieve the intraoperative measured deformation reference data corresponding to the simulated key points; S53. Calculate the deviation measure between the simulated key point deformation data and the intraoperative measured deformation reference data, and generate the deformation conformity index after normalization.

6. A medical data mining and analysis system based on cardiac ultrasound image stream according to claim 5, characterized in that, The residual diversion risk index calculated in step S4 is obtained in the following way: S61. Based on the dynamic adhesion performance dataset, analyze the distribution of instantaneous micro-gaps between the occluder edge and tissue at each phase of the cardiac cycle; S62. Calculate the probability of blood shunting caused by each instantaneous micro-gap based on the principle of fluid dynamics, and construct a shunting risk probability distribution accordingly; S63. Calculate the information entropy value of the diversion risk probability distribution and define it as the residual diversion risk index.

7. A medical data mining and analysis system based on cardiac ultrasound image stream according to claim 6, characterized in that, The output rules of the clinical risk stratification decision model in step S5 include: S71. Preset the safety threshold for each risk assessment indicator; S72. If the wall-attachment stability index is lower than its safety threshold, or the deformation compliance index is lower than its safety threshold, or the residual diversion risk index is higher than its safety threshold, then the decision model outputs a high-risk level judgment. S73. If all risk assessment indicators meet their corresponding safety thresholds, the decision model will output a low-risk level judgment.

8. A medical data mining and analysis system based on cardiac ultrasound image stream according to claim 7, characterized in that, The system also includes: The decision feedback optimization module is used to generate occluder parameter adjustment suggestions based on the deviation of each indicator from the threshold when the clinical risk grading decision model outputs a high-risk level judgment, and restart the process of steps S3 to S5 for iterative evaluation until a low-risk level judgment is output.

Images