Left ventricle quality correction ejection fraction prediction risk system

Abstract

The invention relates to a left ventricular mass correction ejection fraction prediction risk system, which comprises a data acquisition module used for acquiring the measurement data of the left ventricular ejection fraction and the left ventricular mass of a patient; the correction calculation module is in communication connection with the data acquisition module and is used for calculating a correction parameter based on the left ventricle quality according to the measurement data; the risk assessment module is connected with the correction calculation module and used for constructing a risk assessment model based on the correction parameters and the clinical covariables and outputting an individualized risk layering result, and the correction calculation module can analyze and establish a relation model of the correction parameters and the cardiac event risks through a restrictive cubic spline and output the individualized risk layering result. Comprising the steps that data standardization processing is carried out on correction parameters, segmentation fitting is carried out based on preset spline nodes, fitting curve coefficients are solved through a least square method to form a nonlinear relation curve, the model goodness of fit is evaluated through statistical test, and the correction parameters comprise one or more of LVEF / LVM, LVEF / LVM / BSA and LVEF / LVM / BMI.

Description

Technical Field

[0001] This invention relates to the field of cardiovascular disease prognostic assessment technology, and in particular to a left ventricular quality-corrected ejection fraction prediction risk system. Background Technology

[0002] Currently, left ventricular ejection fraction (LVEF) is the most widely used indicator of left ventricular systolic function in clinical practice. Echocardiography methods include M-mode and two-dimensional methods. 1. M-mode echocardiography: Simple and easy to perform, with high temporal resolution, it records the motion of the heart structure in a one-dimensional manner. Placing the sampling line at the level of the mitral valve tip in the long-axis view beside the sternum allows for linear measurement of the left ventricular diameter. Subsequently, LVEF is calculated based on geometric assumptions about the left ventricular morphology, using the formula: (Left ventricular end-diastolic volume - Left ventricular end-systolic volume) / Left ventricular end-diastolic volume × 100%. This is currently the most commonly used measurement method in echocardiogram reports.

[0003] 2. Two-dimensional method: The visual method evaluates systolic function by observing endocardial boundary motion and ventricular wall thickness. For quantitative assessment, the two-dimensional biplane method (modified Simpson's method) can be used to measure left ventricular volume by recording the left ventricular endocardium in the apical four-chamber and two-chamber views at end-diastole and end-systole. The principle is to divide the left ventricle along its long axis into a series of disks of equal height. The volume of each disk is height × disk area (assuming the disk is a perfect circle). The left ventricular volume is the sum of the volumes of all disks, and LVEF is calculated using the formula: (Left ventricular end-diastolic volume - Left ventricular end-systolic volume) / Left ventricular end-diastolic volume × 100%. Currently, the ASE guidelines recommend the two-dimensional biplane method (modified Simpson's method) for measuring ejection fraction.

[0004] CN115620183A discloses a left ventricular ejection fraction (LVEF) prediction device, comprising: a data input unit, an ultrasound image feature representation unit, a contrastive learning unit based on multi-plane two-dimensional echocardiography, and a LVEF result prediction unit; the contrastive learning unit includes: an image representation module, a contrastive image representation module, a negative sample queue storage module, and a feature comparison module; the LVEF result prediction unit is used to process the patient representation h output by the image representation module. k The predicted results are obtained by combining and regression fitting. This invention's left ventricular ejection fraction prediction device, combined with clinical practice of echocardiography, can extract left ventricular-related features from multiple different sections of a patient's two-dimensional echocardiogram, thereby improving the performance of ejection fraction prediction and application in a clinical setting, reducing the workload of professional data annotation, facilitating better clinical research, and helping patients receive better diagnosis, treatment, and prognosis.

[0005] CN118053588A discloses an automatic left ventricular ejection fraction (LVEF) detection method based on deep learning. Addressing the lack of an automatic and accurate method for predicting LVEF values ​​using electrocardiograms (ECGs) in primary healthcare, this invention trains a deep neural network based on ECGs to automatically predict LVEF, solving the problem of insufficient LVEF prediction methods in primary healthcare and providing important reference for clinical diagnosis and treatment. This invention uses a deep learning model to achieve automatic prediction of left ventricular ejection fraction. Tests on multiple ECGs show that the method can accurately estimate LVEF values ​​on most ECGs, and the computation speed meets real-time requirements.

[0006] In the clinical management of patients with chronic heart failure, left ventricular ejection fraction (LVEF) has significant limitations as a primary assessment indicator. First, traditional measurement methods use M-mode echocardiography or the two-dimensional biplane method (modified Simpson's method). The former measures linearly at the mitral valve tip level in the parasternal long-axis section, while the latter calculates volume by recording the endocardial boundary in the apical four-chamber and two-chamber sections. Both methods are based on geometric assumptions and have inherent errors. Second, LVEF is sensitive to stress and has difficulty identifying subtle abnormalities in systolic function. This limits its ability to predict future cardiac events in individuals with mildly reduced or preserved ejection fraction, especially in those with mildly reduced (41-49%) or preserved (≥50%) ejection fraction, where predictive efficacy is significantly reduced.

[0007] Furthermore, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the applicant studied a large number of documents and patents when making this invention, but due to space limitations, not all details and contents were listed in detail. However, this does not mean that the present invention does not possess the features of these prior art. On the contrary, the present invention already possesses all the features of the prior art, and the applicant reserves the right to add relevant prior art to the background art. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a left ventricular mass-corrected ejection fraction (LVEF) prediction risk system to solve at least some of the aforementioned technical problems. Existing technologies using only LVEF to predict cardiac events have poor results. This invention uses LVM-corrected LVEF for prediction, achieving better results, particularly when LVEF / LVM exceeds 4.72~5.53 / kg and LVEF / LVM / BSA exceeds 19.7~24.3m. 2 / kg 2 When the range is within a certain range, it indicates an increased risk of cardiac events.

[0009] This invention discloses a left ventricular quality-corrected ejection fraction (LVEF) risk prediction system, comprising: a data acquisition module for acquiring measurement data of a patient's LVEF and LVM; a correction calculation module, communicatively connected to the data acquisition module, for calculating correction parameters based on left ventricular quality according to the measurement data; and a risk assessment module, connected to the correction calculation module, for constructing a risk assessment model based on the correction parameters and clinical covariates, and outputting individualized risk stratification results. The correction calculation module can establish a relationship model between the correction parameters and cardiac event risk through restricted cubic spline analysis, including data standardization of the correction parameters, piecewise fitting based on preset spline nodes, solving the fitting curve coefficients using the least squares method to form a nonlinear relationship curve, and evaluating the model's goodness of fit through statistical tests. The correction parameters calculated by the correction calculation module include one or more of LVEF / LVM, LVEF / LVM / BSA, and LVEF / LVM / BMI.

[0010] This system overcomes the limitations of traditional single-indicator LVEF in risk prediction by integrating data acquisition, calibration calculation, and risk assessment modules. The calibration calculation module employs restricted cubic spline analysis, eliminating systematic biases between different measurement devices or operators through data standardization. Subsequently, it fits the nonlinear relationship between calibration parameters and cardiac event risk based on preset nodes, capturing complex correlations between calibration parameters and risk (e.g., U-shaped curves). Least squares method is used to solve for the fitting coefficients, ensuring the mathematical rigor of the model parameters, while statistical tests (such as nonlinear hypothesis testing) further verify the model's reliability. By introducing multiple calibration parameters such as LVEF / LVM, LVEF / LVM / BSA, and LVEF / LVM / BMI, the system can select the most appropriate parameter based on the patient's specific physiological characteristics (e.g., body size differences or metabolic load), avoiding predictive bias caused by individual differences in a single parameter. This design enables the risk assessment model to more accurately reflect the true relationship between calibration parameters and cardiac event risk, thereby improving the clinical applicability and individualization of predictions.

[0011] According to a preferred embodiment, the calibration calculation module includes a data processing unit and a storage unit. The data processing unit is used to process echocardiographic data and clinical parameters uploaded by the data acquisition module. The storage unit is communicatively connected to the data acquisition module to call up the LVEF and LVM measurement data acquired by the data acquisition module in real time.

[0012] The real-time communication design between the data processing unit and the storage unit solves the calculation error problem caused by data delay or poor transmission in traditional echocardiographic data processing. The data processing unit rapidly parses the raw LVEF and LVM data from ultrasound images through multi-core parallel computing, while the storage unit ensures that the calibration calculation module can immediately obtain the latest measurement results through a real-time calling mechanism, avoiding inconsistencies in model input caused by data lag. This efficient data processing architecture reduces system response time and reduces redundant processing in intermediate steps through hardware-level data synchronization, thereby improving the real-time performance and stability of calibration calculations. Furthermore, the real-time calling function of the storage unit also supports rapid backtracking of historical data, facilitating subsequent model optimization or source analysis of abnormal data, further ensuring the continuity and repeatability of calibration parameter calculations.

[0013] According to a preferred embodiment, the risk prediction system further includes an input module communicatively connected to the storage unit of the correction calculation module; the input module is used to acquire the subject's basic clinical information and / or physiological parameters, wherein the basic clinical information includes age, gender, smoking history, drinking history, past medical history and / or medication history, and the physiological parameters include height, weight and / or blood pressure; the input module allows operators to manually input data or automatically import data from the hospital's electronic medical record system through an information input interface.

[0014] The data entry module addresses the issues of missing or incorrectly entered clinical covariates in traditional risk assessments through a dual mechanism of manual input and automatic import from the hospital's electronic medical record (EMR) system. Standardized data collection procedures for basic clinical information (such as age, gender, and medical history) and physiological parameters (such as blood pressure and BMI) ensure the completeness and standardization of input data, avoiding model bias caused by inconsistent data formats or human error. For example, standardized blood pressure measurement (averaging three readings after 5 minutes of rest), combined with the high precision of digital blood pressure monitors, effectively reduces the impact of measurement errors on the model. The EMR automatic import function seamlessly integrates with the hospital information system via an interface protocol, directly extracting structured data (such as medication history and laboratory test results), reducing the subjectivity of manual transcription, and thus improving the reliability and timeliness of covariate data.

[0015] According to a preferred embodiment, the data processing unit is configured with a parameter calculation submodule, a nonlinear relationship modeling submodule, and a result verification submodule. The parameter calculation submodule has preset calculation logic for various correction parameters, including calculating LVEF / LVM, LVEF / LVM / BSA, and LVEF / LVM / BMI. The nonlinear relationship modeling submodule establishes a correlation model between the above correction parameters and the risk of cardiac events through restricted cubic spline analysis. The result verification submodule ensures the reliability of the calculation results through range verification, logic verification, and repeatability verification.

[0016] The parameter calculation submodule addresses the limitations of traditional correction methods due to the single parameter selection, by pre-setting calculation logic for multiple correction parameters (such as LVEF / LVM, LVEF / LVM / BSA, and LVEF / LVM / BMI). The nonlinear relationship modeling submodule utilizes restricted cubic spline analysis to flexibly capture the nonlinear correlation between correction parameters and risk (e.g., U-shaped curves), avoiding the simplistic assumptions of linear models regarding complex relationships and thus more realistically reflecting the dynamic changes in physiological parameters and cardiac event risk. The result verification submodule ensures the reasonableness of calculation results from multiple dimensions through range verification (based on clinical reference intervals), logical verification (comparing the consistency of LVEF with correction parameter trends), and repeatability verification (coefficient of variation control). For example, when LVEF is normal but LVEF / LVM significantly deviates from expectations, logical verification triggers a manual review process to avoid misjudgments due to algorithmic bias. This series of verification mechanisms collectively constructs a robust correction parameter calculation framework, significantly reducing the risk of false positives or false negatives and improving the clinical credibility of correction results.

[0017] According to a preferred embodiment, the correction parameters calculated by the correction calculation module can be selected based on the patient's clinical characteristics. LVEF / LVM / BSA is more suitable for scenarios that need to consider individual body size differences, while LVEF / LVM / BMI is more suitable for scenarios that focus on the impact of weight-related factors on cardiac load.

[0018] Dynamically selecting correction parameters based on patient clinical characteristics (such as the presence of hypertension or obesity) solves the problem of traditional fixed parameters neglecting individual differences in correction. For example, LVEF / LVM / BSA corrects for the combined effects of height and weight by introducing body surface area (BSA), avoiding LVEF / LVM bias caused by "tall and thin" or "short and fat" body types, making it particularly suitable for populations with extreme body types or significant racial differences. LVEF / LVM / BMI, on the other hand, focuses on weight-related factors (such as the impact of obesity on cardiac load), directly linking metabolic syndrome and cardiac event risk through BMI correction, providing a more accurate correction basis for patients with hypertension and obesity. This parameter selection strategy, by combining anatomical and metabolic factors, makes the correction parameters more closely match the patient's actual pathophysiological state, thereby improving the accuracy of risk stratification and avoiding predictive bias caused by a single correction dimension.

[0019] According to a preferred embodiment, the risk assessment module is configured with a calculation unit and a clinical data integration interface. The calculation unit includes a risk stratification submodule, a prediction model submodule, a sensitivity analysis submodule, and a result verification submodule. The clinical data integration interface is linked with the hospital system to automatically import the patient's data information, including basic clinical information, medical history information, medication history, and laboratory test data.

[0020] The risk assessment module, through the collaboration of a high-performance computing unit and a clinical data integration interface, addresses the limited predictive power of traditional models caused by data silos. The parallel computing capabilities of the computing unit efficiently handle the joint analysis of multi-dimensional correction parameters (such as continuous variables) and large-scale clinical covariates (such as laboratory test data), significantly shortening model training and prediction time. The clinical data integration interface, through linkage with systems such as HIS and LIS, automatically imports structured and unstructured data (such as medication history and creatinine levels), providing comprehensive covariate support for the model. For example, estimating glomerular filtration rate (eGFR) as an independent covariate can correct for the confounding effects of renal function on cardiac event risk. This multi-source data integration mechanism not only improves the model's predictive efficacy but also enhances its adaptability in complex clinical scenarios, providing a more solid evidence base for individualized risk stratification.

[0021] According to a preferred embodiment, the risk stratification submodule sets a risk stratification threshold based on the distribution characteristics and clinical significance of the calibration parameters to achieve risk level classification and supports fine-tuning of the threshold according to population characteristics; the prediction model submodule adopts a multivariate Cox proportional hazards regression model, with the calibration parameters as the main predictor variables and clinical covariates as adjustment factors, and outputs individualized hazard ratios and confidence intervals; the sensitivity analysis submodule introduces a competing risk model, including non-cardiac cause death as a competing event in the analysis, and evaluates the model stability; the results validation submodule evaluates the reliability and generalization ability of the model through a combination of internal and external validation.

[0022] The risk stratification submodule sets thresholds based on the clinical significance of the calibration parameters (e.g., ≥5.53 / kg for the low-risk group), addressing the poor population adaptability issue caused by fixed thresholds in traditional stratification methods. By allowing operators to fine-tune the thresholds according to population characteristics (e.g., age groups), the system can flexibly adapt to the physiological characteristics of different subgroups. For example, the LVEF / LVM threshold may be elevated in the elderly due to myocardial remodeling, while younger patients require lower thresholds to identify high-risk individuals. The prediction model submodule employs multivariate Cox regression, incorporating covariates such as age and diabetes to eliminate the influence of confounding factors on the association between calibration parameters and risk, thereby enhancing the model's independent predictive value. The sensitivity analysis submodule introduces competing risk models (e.g., the Fine-Gray model), including non-cardiac death as a competing event in the analysis, correcting the estimation bias of traditional Cox models in the presence of competing risks, and ensuring the clinical significance of the hazard ratio (HR). The results validation submodule evaluates the model's discriminative power (C-index) and calibration (calibration curve) through internal cross-validation (k-fold splitting) and external cohort validation, ensuring the model's generalization ability on the training set and independent datasets, and providing reliable quality assurance for clinical applications.

[0023] According to a preferred embodiment, the prediction model submodule preprocesses the input data, including using multiple imputation to handle missing values, identifying outliers based on IQR, and performing logarithmic transformation on non-normally distributed variables; it also selects statistically significant covariates using stepwise regression and stores the model parameters in the model library, supporting real-time access and batch calculation; the prediction model submodule also supports incorporating correction parameters into the model as categorical or continuous variables.

[0024] The preprocessing workflow of the predictive model submodule (such as multiple imputation for missing values ​​and IQR for outlier identification) solves the prediction distortion problem caused by data quality issues in traditional models. For example, multiple imputation fills in missing data by generating multiple reasonable substitute values, avoiding bias caused by simple deletion or mean filling; IQR identifies and removes outliers (such as extremely high BMI), reducing the interference of outliers on model parameter estimation. Logarithmic transformation of non-normally distributed variables (such as blood glucose levels) makes them conform to the statistical assumptions of the regression model, improving the accuracy of parameter estimation. Stepwise regression screens statistically significant covariates, removes redundant or collinear variables, simplifies the model structure, and improves computational efficiency. The model library's storage mechanism supports real-time access and batch computation, enabling prediction results to be quickly applied to clinical decision-making. Simultaneously, parameter version control ensures the traceability of the model iteration process, providing data support for subsequent model updates.

[0025] According to a preferred embodiment, the sensitivity analysis submodule shares the covariate dataset with the prediction model submodule, constructs a competing risk model using the same variable screening strategy, outputs the sub-distributed hazard ratio and confidence interval, and compares it with the HR of the Cox model. When the difference exceeds a preset threshold, the model optimization process is automatically triggered. At the same time, it supports stratified analysis of key covariates and outputs the risk prediction results of each subgroup and the heterogeneity test P value.

[0026] The sensitivity analysis submodule, by sharing the covariate dataset with the prediction model submodule and employing the same variable selection strategy to construct a competing risk model, addresses the risk estimation bias caused by traditional models ignoring competing events (such as non-cardiac death). The comparative analysis of the subdistributed hazard ratio (SHR) and the Cox model's HR provides a clear quantification of the impact of competing risks on prediction results. For example, when the difference between SHR and HR is less than 10%, it indicates that the model remains robust in the presence of competing events. The stratified analysis function evaluates the model's predictive efficacy in different subgroups by grouping by key covariates (such as hypertension status), outputting heterogeneity test p-values ​​to identify potential applicability boundaries of the model. For instance, if the model performs well in diabetic patients (P>0.05) but fails in non-diabetic patients (P<0.05), it suggests the need for stratified optimization of model parameters. This dynamic adjustment mechanism significantly improves the model's applicability in heterogeneous populations, providing a more reliable tool for precision medicine.

[0027] According to a preferred embodiment, the risk prediction system further includes an output module for presenting prediction results, wherein the output module is configured with a display device and / or an output component compatible with a medical-grade printing interface, the display device being able to present the risk assessment results, and the printing interface being able to directly connect to a medical printer to achieve paper-based archiving of reports.

[0028] The output module addresses the limitations of traditional risk assessment results, such as limited presentation formats and inconvenient archiving, through high-resolution display devices and medical-grade printing interfaces. High-resolution display ensures clear visualization of complex charts like risk scores and calibration curves, facilitating intuitive interpretation of model results by physicians. The medical-grade printing interface enables paper-based archiving of reports using standardized formats (such as PDF or DICOM), complying with medical document management standards. For example, risk stratification results can be directly embedded into electronic medical record systems for subsequent follow-up or multidisciplinary consultations. This visual output design not only enhances clinicians' trust in the model but also reduces communication costs through structured reports, promoting the efficient application of risk assessment results in actual clinical practice. Attached Figure Description

[0029] Figure 1 This is a hardware connection diagram of the risk prediction system provided by the present invention.

[0030] Figure 2 This is a schematic diagram of the LVM data acquisition method provided by the present invention.

[0031] Figure 3 This is a comparative graph of LVEF and cardiac event risk provided by the present invention, based on a restricted cubic spline analysis.

[0032] Figure 4This is a comparison chart of LVEF / LVM and cardiac event risk provided by the present invention using restricted cubic spline analysis.

[0033] Figure 5 This is a comparison chart of restricted cubic spline analysis of LVEF / LVM / BSA and cardiac event risk provided by the present invention.

[0034] Figure 6 This is a comparative graph of restricted cubic spline analysis of LVEF / LVM / BMI and cardiac event risk provided by the present invention.

[0035] Figure 7 This is a comparison chart of LVEF between hypertensive patients and non-hypertensive patients, diabetic patients and non-diabetic patients, and patients with dyslipidemia and non-dyslipidemia patients provided by the present invention.

[0036] Figure 8 This invention provides a comparison chart of LVEF / LVM between hypertensive patients and non-hypertensive patients, diabetic patients and non-diabetic patients, and patients with dyslipidemia and non-dyslipidemia patients.

[0037] Figure 9 This invention provides a comparison chart of LVEF / LVM / BSA between patients with hypertension and those without hypertension, patients with diabetes and those without diabetes, and patients with dyslipidemia and those without dyslipidemia.

[0038] Figure 10 This invention provides a comparison chart of LVEF / LVM / BMI between patients with hypertension and those without hypertension, patients with diabetes and those without diabetes, and patients with dyslipidemia and those without dyslipidemia.

[0039] Figure 11 This invention provides a graph showing the cardiac event hazard ratio (HR) and 95% confidence interval (CI) corresponding to LVEF based on the Cox model.

[0040] Figure 12 This invention provides a graph showing the hazard ratio (HR) and 95% confidence interval (CI) of cardiac events corresponding to LVEF / LVM / BMI based on the Cox model.

[0041] Figure 13 This invention provides a graph showing the cardiac event hazard ratio (HR) and 95% confidence interval (CI) for LVEF / LVM (with the 1st quartile as the reference group) based on the Cox model.

[0042] Figure 14 This invention provides a graph showing the cardiac event hazard ratio (HR) and 95% confidence interval (CI) for LVEF / LVM (with the 3rd quartile as the reference group) based on the Cox model.

[0043] Figure 15 This invention provides a graph showing the hazard ratio (HR) and 95% confidence interval (CI) of cardiac events corresponding to LVEF / LVM / BSA (with the 1st quartile as the reference group) based on the Cox model.

[0044] Figure 16 This invention provides a graph showing the hazard ratio (HR) and 95% confidence interval (CI) of cardiac events corresponding to LVEF / LVM / BSA (with the 3rd quartile as the reference group) based on the Cox model.

[0045] List of reference numerals 100: Data acquisition module; 110: Two-dimensional image analysis module; 120: Quality control submodule; 200: Calibration calculation module; 210: Data processing unit; 211: Parameter calculation submodule; 212: Nonlinear relationship modeling submodule; 213: Result verification submodule; 220: Storage unit; 300: Risk assessment module; 310: Calculation unit; 311: Risk stratification submodule; 312: Prediction model submodule; 313: Sensitivity analysis submodule; 314: Result verification submodule; 320: Clinical data integration interface; 400: Input module; 500: Output module. Detailed Implementation

[0046] The following is a detailed explanation with reference to the accompanying drawings.

[0047] like Figure 1 As shown, this invention discloses a left ventricular mass-corrected ejection fraction risk prediction system, which includes: a data acquisition module 100 for acquiring measurement data of the patient's left ventricular ejection fraction (LVEF) and left ventricular mass (LVM); a correction calculation module 200 for calculating correction parameters based on left ventricular mass according to the measurement data; and a risk assessment module 300 for constructing a risk assessment model based on the correction parameters and clinical covariates, and outputting individualized risk stratification results.

[0048] Preferably, the data acquisition module 100 may include an echocardiogram device, wherein the echocardiogram device may be a medical ultrasound system with high-resolution imaging capabilities (such as Philips IE33, epiq 7C, GE Vivid E95, Siemens Redwood S, etc.), and equipped with a broadband transducer suitable for cardiac imaging (such as the S5-1 transducer, with an operating frequency range of 1.0~5.0MHz) to ensure clear images of the left ventricular structure in key sections such as the parasternal long axis, apical four-chamber view, and apical two-chamber view, meeting the needs of subsequent measurements. Simultaneously, to achieve synchronous data recording and storage, the data acquisition module 100 may also be configured with a dedicated data storage unit 220, supporting real-time saving of DICOM standard format data, ensuring the integrity and traceability of the original images and measurement data.

[0049] Preferably, the data acquisition module 100 may have a built-in two-dimensional image analysis module 110, which supports automatic identification and manual correction of the boundaries of the left ventricular end-diastolic and end-systolic phases. The automatic identification algorithm is based on a deep learning model and optimizes the identification accuracy by training a large number of clinically labeled images. The manual correction function allows operators to adjust the endocardial boundaries by touch or mouse operation to eliminate the interference of image artifacts or complex anatomical structures on the measurement.

[0050] Preferably, the data acquisition module 100 can use a two-dimensional dual-plane method (modified Simpson's method) to acquire LVEF data. The specific operation procedure is as follows: Under the guidance of the echocardiography equipment in the data acquisition module 100, the operator acquires dynamic images of the apical four-chamber view and the apical two-chamber view, respectively, recording images for three consecutive cardiac cycles for each view; the data acquisition module 100 automatically selects frame images from end-diastole (the moment corresponding to the peak of the R wave on the electrocardiogram) and end-systole (the moment of minimum ventricular volume), and records these images in these two... In the frame image, the boundary of the left ventricular endocardium is depicted by combining the above-mentioned automatic recognition and manual correction. The left ventricle is virtually divided into a series of disks of equal height along the long axis. The volume of each disk (height × disk area, assuming the disk is a perfect circle) is calculated, and the volumes of all disks are summed to obtain the left ventricular end-diastolic volume (LVEDV) and left ventricular end-systolic volume (LVESV). Finally, the LVEF value is calculated using the formula LVEF=[(LVEDV-LVESV) / LVEDV]×100%.

[0051] Preferably, such as Figure 2As shown, the data acquisition module 100 can acquire LVM data by measuring the geometric parameters of the left ventricle. Specific steps include: acquiring a short-axis image of the left ventricle in a parasternal long-axis view (below the level of the mitral valve tip); measuring the end-diastolic interventricular septal thickness (IVSTd), left ventricular end-diastolic diameter (LVEDD), and left ventricular posterior wall thickness (LVPWTd). All these parameters are measured using a two-dimensional image direct measurement method, avoiding areas of myocardial echo heterogeneity or calcification. Each parameter is measured three times consecutively, and the average value is taken as the final result. The parameters are then substituted into the preset LVM calculation formula: LVM = 0.8 × 1.04 × [(IVSTd + LVEDD + LVPWTd)]. 3 -LVEDD 3 The LVM value was automatically calculated by the software using 0.6g. The coefficients and constants in the formula were set based on clinically validated left ventricular mass calculation standards to ensure clinical consistency of the results.

[0052] Furthermore, to ensure data quality, the data acquisition module 100 can also be configured with a quality control submodule 120, whose functions include: image quality assessment, which automatically filters out images that do not meet the measurement requirements (such as images with obvious motion artifacts or unclear endocardial visualization) by analyzing indicators such as signal-to-noise ratio and boundary clarity, and prompts the operator to re-acquire the image; parameter rationality verification, which sets reference ranges for parameters such as IVSTd, LVEDD, and LVPWTd based on the physiological range of normal populations, and marks measurements outside the ranges, requiring operator verification; and repeatability testing, which calculates the coefficient of variation for three consecutive measurements of the same patient, and prompts re-measurement when the coefficient of variation is greater than 5%, in order to reduce the impact of operational errors on the data. In addition to using automatic algorithms to objectively evaluate images, operators can also subjectively score the image quality (e.g., on a 1-5 scale) after each measurement, focusing on the clarity of the endocardial boundary, the accuracy of the section angle, and the degree of interference from motion artifacts, in order to improve the accuracy of image quality assessment. In addition, for certain special cases (such as patients with severe valvular disease or arrhythmia), the data acquisition module 100 can use alternative measurement methods (such as three-dimensional echocardiography or cardiac magnetic resonance imaging) for verification. For cases with image quality defects, the data acquisition module 100 supports completion through historical data interpolation or machine learning models to reduce data loss due to missing images.

[0053] Preferably, the correction calculation module 200 can calculate correction parameters based on left ventricular mass and establish a nonlinear relationship model between the correction parameters and the risk of cardiac events. The correction calculation module 200 may include a high-performance data processing unit 210 and a storage unit 220. The data processing unit 210 uses a processor with multi-core computing capabilities, supporting parallel computing to quickly process large amounts of echocardiographic data and clinical parameters. The storage unit 220 is linked with the data acquisition module 100 and can call up raw data such as LVEF and LVM acquired by the data acquisition module 100 in real time.

[0054] Preferably, the risk prediction system may also be configured with an input module 400 that is communicatively connected to the storage unit 220 of the correction calculation module 200. The input module 400 can acquire the patient's basic clinical information and related physiological parameters. The basic clinical information may include age, gender, smoking history, drinking history, past medical history (such as hypertension, diabetes, stroke, etc.) and / or medication history (such as hypoglycemic drugs, antihypertensive drugs, lipid-lowering drugs, etc.), which can be manually entered by the operator through an integrated information input interface or automatically imported from the hospital's electronic medical record (EMR) system. The physiological parameters may include height, weight and / or blood pressure, where height is measured using a standard right-angle measuring instrument with a fixed tape measure (accurate to 0.1cm), weight is measured using a calibrated scale (accurate to 0.1kg), and blood pressure is measured three times using a digital blood pressure monitor (such as OMRON HBP-1300 or equivalent model) after the patient has rested in a seated position for 5 minutes, and the average value is taken. The acquisition of the above parameters follows standardized operating procedures to ensure the reliability of the data. If the calibration calculation module 200 needs to use the patient's basic clinical information and / or relevant physiological parameters to calculate the calibration parameters, the input module 400 can establish a communication connection with the calibration calculation module 200, and the operator can manually input the parameters or automatically import them from the hospital's electronic medical record (EMR) system.

[0055] Preferably, the data processing unit 210 may be configured with a parameter calculation submodule 211, a nonlinear relationship modeling submodule 212, and a result verification submodule 213. Each submodule achieves data interaction through a standardized interface to ensure the continuity and accuracy of the calculation process.

[0056] Preferably, the parameter calculation submodule 211 has preset calculation logic for multiple correction parameters to adapt to the needs of different clinical scenarios. For the calculation of LVEF / LVM, the core is to divide the LVEF value (unit: %) acquired by the data acquisition module 100 by the LVM value (unit: g), i.e., LVEF / LVM = LVEF ÷ LVM, with the result rounded to two decimal places and the unit being % / g. To eliminate the influence of body size differences on the correction results, the parameter calculation submodule 211 can support composite correction combined with physiological indicators, including but not limited to body surface area (BSA) and body mass index (BMI). The calculation of LVEF / LVM / BSA involves further dividing the LVEF / LVM result by BSA (unit: m²). 2 That is, LVEF / LVM / BSA = (LVEF ÷ LVM) ÷ BSA, with units of % / (g·m). 2 The calculation of LVEF / LVM / BMI is the result of LVEF / LVM divided by BMI (unit: kg / m²). 2 That is, LVEF / LVM / BMI = (LVEF ÷ LVM) ÷ BMI, with units of % / (g·kg / m²). 2 It should be noted that BSA can be calculated using the DuBois formula, i.e., BSA = 0.007184 × height. 0.725 ×weight 0.425 BMI is calculated by dividing weight (kg) by the square of height (m), i.e., BMI = weight ÷ (height × height). The above formula is preset in the parameter calculation submodule 211 and can be automatically calculated based on the height and weight data uploaded by the input module 400 without manual intervention.

[0057] Preferably, the nonlinear relationship modeling submodule 212 can establish a relationship model between the correction parameters and the risk of cardiac events through restricted cubic spline (RCS) analysis. The nonlinear relationship modeling submodule 212 first performs data standardization on the input correction parameters (such as LVEF / LVM, LVEF / LVM / BSA, LVEF / LVM / BMI), converting the parameter values ​​into standardized scores conforming to a normal distribution. Then, based on preset spline nodes (such as 5 nodes), the standardized parameters are piecewise fitted. The node positions are set according to the distribution characteristics of clinical data, typically covering the 5%, 25%, 50%, 75%, and 95th percentiles of the parameter values. Next, the least squares method is used to solve for the fitting curve coefficients of each segment, forming a continuous nonlinear relationship curve. For LVEF / LVM and LVEF / LVM / BSA, the model focuses on capturing their U-shaped association with the risk of cardiac events, i.e., identifying the parameter range with the lowest risk (e.g., LVEF / LVM 4.72~5.53 / kg, LVEF / LVM / BSA 19.7~24.3m). 2 / kg 2 The model's goodness of fit is evaluated through statistical tests (such as nonlinear hypothesis testing) to ensure that the curve accurately reflects the true relationship between the parameters and the risk. Figures 3-6 This clearly demonstrates the U-shaped relationship. For example, when LVEF / LVM is below 4.72 / kg or above 5.53 / kg, the risk of cardiac events increases significantly, providing a direct basis for setting risk stratification thresholds.

[0058] Preferably, the result verification submodule 213 is used to ensure the reliability of the correction calculation results, which is achieved through multiple verification mechanisms: First, range verification, based on the clinically known normal reference range, makes a reasonable judgment on the calculated LVEF / LVM, LVEF / LVM / BSA and / or LVEF / LVM / BMI. If the result exceeds the preset range (e.g., LVEF / LVM < 1.0 / kg or > 10.0 / kg), it is marked as abnormal and prompts to re-examine the original data; Second, logic verification, compares the changing trend of individual LVEF values ​​with the correction parameters. If a contradiction occurs (e.g., LVEF is normal but LVEF / LVM is significantly reduced), a manual review process is triggered to ensure that the calculation logic is without deviation; Third, repeatability verification, performs multiple calculations on the same set of original data. If the coefficient of variation of the result exceeds 3%, it is determined to be unstable and the calculation engine is automatically restarted for reprocessing.

[0059] Preferably, the correction calculation module 200 also has parameter selection and switching functions, allowing operators to select the most suitable correction parameters based on the patient's clinical characteristics (such as whether they have hypertension or diabetes). LVEF / LVM, as a baseline correction indicator, is suitable for most individuals with mildly reduced or preserved ejection fraction (LVEF 41%~60%), especially for routine risk screening scenarios where body shape or weight factors are not particularly considered. Body surface area (BSA) integrates height and weight, reducing LVEF / LVM bias caused by "tall and thin" or "short and fat" body types. Therefore, LVEF / LVM / BSA is preferred when individual body shape differences need to be considered (such as tall or short patients). For example, LVEF / LVM / BSA is more accurate for groups with significant height differences (such as children and adults, different ethnic groups), or in situations where special body shapes (such as athletes, patients with dwarfism) may affect the absolute value of LVM. BMI directly reflects the relationship between weight and height and is closely related to the cardiac burden caused by obesity. Therefore, when focusing on the impact of weight-related factors (such as obesity and metabolic syndrome) on cardiac load, LVEF / LVM / BMI should be used preferentially. For example, for patients with hypertension and obesity (BMI ≥ 28 kg / m²), 2For patients with weight-related metabolic diseases such as diabetes, LVEF / LVM / BMI can be additionally correlated with the load pressure of weight on the left ventricle, helping to determine the impact of metabolic factors on cardiac events. Figures 7-10 The differences in parameters among the aforementioned populations were clearly presented. For example, the LVEF / LVM of hypertensive patients (4.44±1.21 / kg) was significantly lower than that of non-hypertensive patients (4.77±1.20 / kg), providing a clinical basis for parameter selection.

[0060] Preferably, the calibration calculation module 200 also supports storing the calculation process and results in the storage unit 220 in the form of a log. The log content includes the original data, calculation steps, intermediate results, final calibration parameter values ​​and model fitting parameters, which facilitates subsequent traceability and auditing.

[0061] Preferably, the risk assessment module 300 integrates the correction parameters and clinical covariates output by the correction calculation module 200 to construct an accurate risk assessment model, enabling stratified and quantitative prediction of cardiac event risks. Furthermore, the risk assessment module 300 can be configured with a high-performance computing unit 310 and a clinical data integration interface 320. The computing unit 310 uses a processor supporting parallel computing, capable of rapidly processing multi-dimensional correction parameters and large-scale clinical sample data, ensuring real-time computation of the risk model. The clinical data integration interface 320 can be linked with hospital information systems (HIS), laboratory information systems (LIS), and other hospital systems, automatically importing patients' basic clinical information (such as age, gender, smoking history, and drinking history), medical history (such as hypertension, diabetes, and stroke), medication history (such as hypoglycemic drugs, antihypertensive drugs, and lipid-lowering drugs), and laboratory test data (such as blood glucose, blood lipids, creatinine, and estimated glomerular filtration rate), providing comprehensive covariate support for the risk assessment model.

[0062] Preferably, the calculation unit 310 may integrate a risk stratification submodule 311, a prediction model submodule 312, a sensitivity analysis submodule 313, and a result verification submodule 314. Each submodule works collaboratively through a standardized data bus to ensure the systematic and rigorous nature of the risk assessment process.

[0063] Preferably, the risk stratification submodule 311 can set risk stratification thresholds based on the distribution characteristics and clinical significance of the correction parameters to classify patients' risk levels. For example, for the LVEF / LVM parameter, the risk stratification submodule 311 can divide the patient into a low-risk group (≥5.53 / kg), an intermediate reference group (4.72–5.53 / kg), and a high-risk group (<4.72 / kg) based on the lowest risk interval determined by restricted cubic spline analysis; for the LVEF / LVM / BSA parameter, the risk stratification submodule 311 can correspondingly divide the patient into a low-risk group (≥24.3m). 2 / kg2 Intermediate reference group (19.7–24.3m) 2 / kg 2 ) and high-risk group (<19.7m) 2 / kg 2 The aforementioned thresholds can be set based on the non-linear correlation between correction parameters and cardiac event risk, where the intermediate reference group represents the lowest risk range, and the high-risk and low-risk groups correspond to the two ends of the increased risk, respectively. The risk stratification submodule 311 can have a built-in threshold management function, allowing operators to fine-tune the thresholds according to different population characteristics (such as age groups and disease subtypes), and automatically record adjustment logs for traceability.

[0064] Preferably, the prediction model submodule 312 adopts a multivariate Cox proportional hazards regression model as its core algorithm. This model uses the correction parameters (LVEF / LVM, LVEF / LVM / BSA, or LVEF / LVM / BMI) as the main predictor variables, while incorporating the aforementioned clinical covariates as adjustment factors. The model calculates the hazard coefficients of each variable using maximum likelihood estimation, ultimately outputting individualized hazard ratios (HR) and 95% confidence intervals (CI). During model computation, the input data is first preprocessed, including missing value handling (using multiple imputation), outlier identification (based on IQR), and variable transformation (logarithmic transformation of non-normally distributed variables). Then, stepwise regression is used to screen statistically significant covariates to ensure model simplicity and predictive power. The final generated model parameters (such as regression coefficients and standard errors) are stored in a model library, supporting real-time retrieval and batch calculation. Furthermore, the prediction model submodule 312 supports incorporating correction parameters into the model as categorical variables (based on risk stratification thresholds) or continuous variables. Operators can choose the appropriate variable form according to clinical needs. Continuous variables retain fine-grained information about parameter changes, while categorical variables are easier for clinical interpretation. Table 1 presents the predictive power of different correction parameters in unadjusted and multi-adjusted models. For example, LVEF / LVM in the fully adjusted model has an HR of 0.79 (95% CI 0.66–0.94, P = 0.008), validating its independent predictive value. Figures 11-16 Further visualizations show the risk differences stratified by the quartiles of the adjusted parameters. For example, the HR of the lowest quartile group of LVEF / LVM is 2.01 (95% CI 1.13~3.56), which intuitively presents the strength of the association between the parameter and the risk.

[0065] Table 1

[0066] Note: Model 1: Adjusted for gender and age. Model 2: Adjusted for variables in Model 1 plus smoking status, alcohol consumption, diabetes, dyslipidemia, hypertension, and stroke. Model 3: Adjusted for variables in Model 2 plus hypoglycemic treatment, antihypertensive treatment, lipid-lowering treatment, and eGFR.

[0067] Preferably, the sensitivity analysis submodule 313 is used to evaluate the stability and robustness of the risk model. Its core is the introduction of a competing risk model (such as the Fine-Gray subdistributed risk model), incorporating non-cardiac cause death as a competing event into the analysis to correct the estimation bias of the traditional Cox model in the presence of competing risks. This submodule shares the covariate dataset with the prediction model submodule 312, uses the same variable selection strategy to construct the competing risk model, outputs the subdistributed hazard ratio (SHR) and 95% CI, and compares it with the HR of the Cox model. If the difference is less than a preset threshold (e.g., 10%), the model result is considered stable; if the difference exceeds the threshold, the model optimization process is automatically triggered to readjust the variable weights to reduce bias. Simultaneously, the sensitivity analysis submodule 313 supports stratified analysis of key covariates (such as age and hypertension status) to evaluate the model's applicability in different subgroups, outputting the risk prediction results and heterogeneity test p-values ​​for each subgroup, providing a detailed basis for the model's clinical application. Table 2 shows the comparison of results with and without competition risk adjustment. For example, the low LVEF / LVM group had an HR of 2.00 (95% CI 1.08~3.71) in the Cox model and an SHR of 1.91 (95% CI 1.34~3.52) in the Fine-Gray model. The difference was small, which verified the stability of the model.

[0068] Table 2

[0069] Note: Low, medium, and high groups are defined as follows: left ventricular ejection fraction <64%, 64-67%, and ≥67%; for LVEF / LVM, <4.72 g / kg, 4.72-5.53 g / kg, and ≥5.53 g / kg; for LVEF / LVM / BSA, <19.7 m² / kg. 2 ), 19.7~24.3 (m) 2 / kg 2 ), ≥24.3 (m) 2 / kg 2 ); LVEF / LVM / BMI (n=2616) are all less than 19.7 (m 2 / kg 2 ), 19.7~24.3 (m) 2 / kg2 ) and ≥24.3 (m 2 / kg 2 ). It is expressed as HR (95% CI), P-value, and SHR (95% CI).

[0070] Abbreviations: BMI, Body Mass Index; BSA, Body Mass Index; CI, Confidence Interval; FGR, Fine Gray; HR, Hazard Ratio; LVEF, Left Ventricular Ejection Fraction; LVM, Left Ventricular Mass; SHR, Subdistribution Hazard Ratio.

[0071] Preferably, the outcome validation submodule 314 combines internal and external validation to ensure the reliability of the risk assessment results. Internal validation employs k-fold cross-validation (k=5 or 10), randomly dividing historical sample data into training and validation sets. A model is built on the training set, and predictive efficacy is evaluated on the validation set. The model's discriminative power and calibration are quantified by calculating the consistency index (C-index), calibration curve, and area under the patient operating characteristic curve (ROC) (AUC). External validation supports importing clinical data from independent cohorts, repeating the model computation process, and comparing the consistency between predicted and actual outcomes. If the difference between the AUC of external validation and internal validation is less than 0.05, the model is considered to have good generalization ability. The outcome validation submodule 314 also features a model update trigger mechanism. When the amount of new follow-up data reaches 20% of the historical sample size, it automatically prompts the operator to retrain the model to incorporate the latest clinical outcome information, ensuring dynamic matching between model parameters and population risk characteristics.

[0072] Preferably, the risk prediction system may also be configured with an output module 500 that presents the prediction results in the form of a visual report or risk score to assist clinical decision-making. Furthermore, the output module 500 may be configured with a display device and an output component compatible with a medical-grade printing interface. The display device has high resolution to ensure clear presentation of the risk assessment results; the printing interface conforms to medical document output standards and can be directly connected to a medical printer for paper-based archiving of reports.

[0073] It should be noted that the specific embodiments described above are exemplary. Those skilled in the art can devise various solutions inspired by the disclosure of this invention, and these solutions all fall within the scope of this invention and its protection. Those skilled in the art should understand that this specification and its accompanying drawings are illustrative and do not constitute a limitation on the claims. The scope of protection of this invention is defined by the claims and their equivalents. This specification contains multiple inventive concepts; phrases such as "preferred" or "according to a preferred embodiment" indicate that the corresponding paragraph discloses an independent concept. The applicant reserves the right to file divisional applications based on each inventive concept. Throughout the text, the feature introduced by "preferred" is only an optional mode and should not be construed as mandatory. Therefore, the applicant reserves the right to abandon or delete relevant preferred features at any time.

Claims

1. A left ventricular quality-corrected ejection fraction prediction risk system, characterized in that, It includes: The data acquisition module (100) is used to acquire the patient's LVEF and LVM measurement data; The calibration calculation module (200) is communicatively connected to the data acquisition module (100) and is used to calculate calibration parameters based on left ventricular mass according to the measurement data; The risk assessment module (300), connected to the correction calculation module (200), is used to construct a risk assessment model based on correction parameters and clinical covariates, and output individualized risk stratification results. The correction calculation module (200) can establish a relationship model between correction parameters and cardiac event risk through restricted cubic spline analysis. This includes standardizing the correction parameters, performing piecewise fitting based on preset spline nodes, solving the fitting curve coefficients using the least squares method to form a nonlinear relationship curve, and evaluating the model's goodness of fit through statistical testing. The correction parameters calculated by the correction calculation module (200) include one or more of LVEF / LVM, LVEF / LVM / BSA and LVEF / LVM / BMI.

2. The system according to claim 1, characterized in that, The calibration calculation module (200) includes a data processing unit (210) and a storage unit (220). The data processing unit (210) is used to process the echocardiogram data and clinical parameters uploaded by the data acquisition module (100). The storage unit (220) is communicatively connected to the data acquisition module (100) to call the LVEF and LVM measurement data acquired by the data acquisition module (100) in real time.

3. The system according to claim 1 or 2, characterized in that, It also includes an input module (400) that is communicatively connected to the storage unit (220) of the correction calculation module (200); the input module (400) is used to acquire the subject's basic clinical information and / or physiological parameters, wherein the basic clinical information includes age, gender, smoking history, drinking history, past medical history and / or medication history, and the physiological parameters include height, weight and / or blood pressure; the input module (400) allows the operator to manually input data through an information input interface or automatically import data from the hospital's electronic medical record system.

4. The system according to any one of claims 1 to 3, characterized in that, The data processing unit (210) is configured with a parameter calculation submodule (211), a nonlinear relationship modeling submodule (212), and a result verification submodule (213), wherein, The parameter calculation submodule (211) has preset calculation logic for a variety of correction parameters, including the calculation of LVEF / LVM, LVEF / LVM / BSA and LVEF / LVM / BMI; The nonlinear relationship modeling submodule (212) establishes a correlation model between the above-mentioned correction parameters and the risk of cardiac events through restricted cubic spline analysis; The result verification submodule (213) ensures the reliability of the calculation results through range verification, logic verification and repeatability verification.

5. The system according to any one of claims 1 to 4, characterized in that, The correction parameters calculated by the correction calculation module (200) can be selected according to the patient's clinical characteristics. Among them, LVEF / LVM / BSA is more suitable for scenarios that need to consider individual body size differences, while LVEF / LVM / BMI is more suitable for scenarios that focus on the impact of weight-related factors on cardiac load.

6. The system according to any one of claims 1 to 5, characterized in that, The risk assessment module (300) is configured with a calculation unit (310) and a clinical data integration interface (320). The calculation unit (310) includes a risk stratification submodule (311), a prediction model submodule (312), a sensitivity analysis submodule (313), and a result verification submodule (314). The clinical data integration interface (320) is linked with the hospital system to automatically import the patient's data information, including basic clinical information, medical history information, medication history, and laboratory test data.

7. The system according to any one of claims 1 to 6, characterized in that, The risk stratification submodule (311) sets the risk stratification threshold based on the distribution characteristics and clinical significance of the correction parameter, realizes the risk level classification, and supports fine-tuning the threshold according to population characteristics; the prediction model submodule (312) adopts a multivariate Cox proportional hazards regression model, with the correction parameter as the main predictor variable, and incorporates clinical covariates as adjustment factors, outputting individualized hazard ratios and confidence intervals; the sensitivity analysis submodule (313) introduces a competing risk model, including non-cardiac cause death as a competing event in the analysis, and evaluates the stability of the model; the result validation submodule (314) evaluates the reliability and generalization ability of the model through a combination of internal and external validation.

8. The system according to any one of claims 1 to 7, characterized in that, The prediction model submodule (312) preprocesses the input data, including using multiple interpolation to handle missing values, identifying outliers based on IQR, and performing logarithmic transformation on non-normally distributed variables; The statistically significant covariates are selected by stepwise regression and the model parameters are stored in the model library, supporting real-time access and batch calculation; the prediction model submodule (312) also supports the inclusion of correction parameters in the model in the form of categorical or continuous variables.

9. The system according to any one of claims 1 to 8, characterized in that, The sensitivity analysis submodule (313) shares the covariate dataset with the prediction model submodule (312), constructs a competing risk model using the same variable screening strategy, outputs the sub-distributed risk ratio and confidence interval, and compares it with the HR of the Cox model. When the difference exceeds the preset threshold, the model optimization process is automatically triggered. At the same time, it supports hierarchical analysis of key covariates and outputs the risk prediction results and heterogeneity test P value of each subgroup.

10. The system according to any one of claims 1 to 9, characterized in that, It also includes an output module (500) for presenting prediction results, wherein the output module (500) is configured with a display device and / or an output component with a medical-grade printing interface, the display device being able to present risk assessment results, and the printing interface being able to be directly connected to a medical printer to achieve paper archiving of reports.

Images