Coronary artery risk prediction method and system for coronary heart disease patient

Abstract

The invention relates to a coronary artery risk prediction method and system for a coronary heart disease patient, and aims to solve the limitation that a traditional static evaluation model is difficult to integrate multi-source dynamic data. Collecting multi-source heterogeneous data of a patient from a hospital information system, a gene database and wearable equipment; the method comprises the following steps: respectively extracting key features through a special processing module: analyzing unstructured text and image plaque features by adopting a deep learning model, calculating a multi-gene risk score, capturing a physiological parameter attenuation trend by utilizing time sequence analysis, constructing a three-stage cascade integration model, screening high-contribution features at the first stage, and constructing a three-stage cascade integration model; the secondary stage integrates space-time dynamic characteristics through an attention mechanism network, the final stage integrates static and dynamic factors to generate a dynamic risk score, a visual risk trend chart and a high-risk factor thermodynamic diagram are output, and real-time early warning and linkage are performed through a medical terminal to generate a personalized intervention scheme. According to the method, collaborative analysis and dynamic risk tracking of multi-dimensional data are realized, and the early warning capability and the clinical decision-making efficiency are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of medical big data technology, and in particular to a method and system for predicting coronary artery risk in patients with coronary heart disease. Background Art

[0002] The field of coronary artery disease risk prediction currently faces significant limitations. Traditional clinical assessments primarily rely on static indicators (such as blood lipids and blood pressure) and single imaging studies, making it difficult to integrate multidimensional dynamic risk factors. Existing prediction models are typically constructed based on a single data source and lack comprehensive analysis of genetic susceptibility, long-term physiological changes, and the evolution of plaque characteristics. This results in insufficient sensitivity in identifying high-risk patients and is particularly prone to overlooking early warning signs of asymptomatic progressive lesions.

[0003] Existing technologies face bottlenecks in processing multi-source, heterogeneous data. Unstructured text in electronic medical records, quantitative features from medical images, genomic data, and time-series physiological parameters generated by wearable devices all differ significantly in data structure and spatiotemporal scale, making an effective collaborative analysis framework yet to be established. Conventional prediction methods have limited feature extraction depth for unstructured text, lack accuracy in automated identification of imaging plaque characteristics, and few models can simultaneously process the interactive effects of static clinical data and dynamic physiological monitoring signals.

[0004] Clinical practice urgently needs dynamic, personalized risk warning tools. Current risk scoring systems generally lack real-time update mechanisms, failing to reflect the impact of short-term patient status changes on prognosis. Furthermore, the model's decision-making process lacks interpretability, making it difficult for physicians to quickly identify core risk drivers. Furthermore, existing systems are often independent of the diagnosis and treatment process, failing to directly translate risk prediction results into intervention strategies and leverage medical resources, limiting their clinical translational value. Summary of the Invention

[0005] To achieve the above objectives, this application provides the following technical solutions: According to a first aspect of the present invention, the present invention claims protection for a method for predicting coronary risk in patients with coronary heart disease, comprising: S1, determining a target patient, and acquiring multi-source heterogeneous risk basic data of the target patient from a multi-source database; S2, obtaining hierarchical characteristic parameters of each multi-source heterogeneous risk basic data using different processing models based on the multi-source heterogeneous risk basic data; S3, inputting the hierarchical characteristic parameters into a three-level cascade integrated model, and outputting a dynamic risk prediction score for the target patient; S4, generating a dynamic risk trend graph and a high-risk factor contribution heat map based on the dynamic risk prediction score, and pushing an early warning signal in real time through the medical terminal.

[0006] Furthermore, the S1 further includes: Acquire the target patient's clinical data, genomic data, and dynamic physiological monitoring data in real time from hospital information systems, gene databases, and wearable devices, where the clinical data at least includes electronic medical records (EMRs), coronary CT angiography (CCTA) images, and laboratory test reports; Said S2 further includes: The BERT model is used to extract key pathological features from unstructured text in clinical data, and a 3D convolutional neural network is used to automatically quantify coronary artery calcium scores (CACS) and plaque characteristics in CCTA images. The genetic risk value of coronary heart disease was calculated using the polygenic risk score (PRS) model for genomic data; Extract the heart rate variability (HRV) attenuation index from dynamic physiological data using a time series anomaly detection algorithm; Said S3 further includes: The hierarchical feature parameters output by S2 are input into a three-level cascade integration model, wherein the first level uses XGBoost to screen feature importance, the second level uses a bidirectional LSTM network with an integrated attention mechanism to process temporal features, and the third level uses random forest to integrate static and dynamic features to generate a dynamic risk prediction score.

[0007] Furthermore, the extraction of the plaque features in S2 includes: The U-Net++ segmentation model was used to identify non-calcified plaques, mixed plaques, and low-density plaques in CCTA images. The ratio of plaque volume to vascular cross-sectional area (PV / VA) was calculated as a quantitative indicator of high-risk plaques; The calcification area identification of S2 includes: Automatically outline the vascular intima border on the cross section of CT images; Pixel areas with CT values ​​greater than 130HU were marked as calcified plaques; The total volume of calcified plaques was calculated by continuous scanning along the vascular axis.

[0008] Furthermore, the construction of the polygenic risk score PRS model of S2 also includes: SNP sites associated with coronary heart disease were extracted from the UK Biobank database, and LASSO regression was used to screen significant sites; Generate patient-specific PRS values ​​based on the target population allele frequency-corrected weights; Said S2 also includes the generation of a genetic risk index, prioritizing the screening of 20 gene loci that are strongly associated with coronary heart disease in the target population; The weighting coefficient is set according to the multiple of each site that increases the risk of coronary heart disease; The weighted values ​​of all high-risk sites carried by the patient are accumulated as the final index.

[0009] Furthermore, a feature interaction module is introduced into the three-level cascade integration model of S3, and the interaction terms of clinical features and genomic features are generated through an explicit feature cross network, and the interaction terms and original features are input into a bidirectional LSTM network. The second level of processing in S3 includes arranging the heart rate variability data for 7 consecutive days in chronological order, automatically identifying the attenuation trend of variability that has continued to decrease for more than 5 days, and inputting the attenuation slope as an independent risk factor into the third level model; The third level of processing in S3 includes multiplying the genetic risk index by the calcification volume percentage to generate a composite index. When the composite index is greater than a threshold, the final risk level is automatically increased.

[0010] Furthermore, the attention mechanism of S3 is specifically implemented as follows: A temporal attention layer is superimposed on the bidirectional LSTM output layer to dynamically weight the physiological monitoring data at different time points; The correlation between attention weights and risk scores is optimized via backpropagation; Before executing S3, the following steps are also included: Smoothing and filtering of abnormal peaks in ECG data caused by motion interference; Remove blurred frames caused by respiratory motion artifacts in CT images.

[0011] Furthermore, the method further comprises: In the continuous learning step of the model, when the amount of new patient data reaches a threshold, the incremental learning process is started, and the elastic weight curing (EWC) algorithm is used to prevent catastrophic forgetting of the model; The online model update step receives data on newly diagnosed coronary heart disease patients from the hospital every month, retrains only the third-level decision tree model and retains the original feature screening rules, and triggers a full model update when the accuracy of the validation set changes by more than 5%.

[0012] Furthermore, the generation of the high-risk factor contribution heat map of S4 includes: The SHAP value is used to interpret the model output, calculate the marginal contribution of each feature to the risk score, and generate a feature importance map by sorting the absolute value of the contribution; The top 10% of calcified plaques by volume are highlighted in red, and a numerical dashboard showing genetic risk index and dynamic physiological indicators is displayed next to the vascular model.

[0013] Furthermore, the dynamic physiological monitoring data of S1 includes: The resting heart rate, QT interval dispersion, and standard deviation of blood oxygen saturation fluctuations at night for 7 consecutive days were used to suppress sensor noise through Kalman filtering; The processing of the dynamic physiological monitoring data also includes counting the number of abnormal blood oxygen levels during sleep at night, and activating an emergency data retransmission mechanism when the number of abnormalities in a single night exceeds 10 times.

[0014] Furthermore, the dynamic risk prediction score is used to: Generate personalized intervention strategy recommendations, including medication adjustment suggestions and review cycle planning, and link with the hospital's resource scheduling system to automatically schedule follow-up visits with cardiologists; An intervention priority list pops up on the physician's workstation, sorted in descending order based on risk score, and a draft report is automatically generated containing the recommended date for a follow-up CT scan and a statin dose adjustment plan.

[0015] According to a second aspect of the present invention, the present invention claims protection for a coronary risk prediction system for patients with coronary heart disease, comprising: one or more processors; A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the coronary risk prediction method for patients with coronary heart disease.

[0016] The present invention relates to a method and system for predicting coronary risk in patients with coronary heart disease, aiming to address the limitation of traditional static assessment models that are difficult to integrate multi-source dynamic data. Multi-source heterogeneous data of patients are collected from hospital information systems, gene databases and wearable devices; key features are extracted respectively through dedicated processing modules: a deep learning model is used to parse unstructured text and image plaque features, a polygenic risk score is calculated, and time series analysis is used to capture the attenuation trend of physiological parameters, and a three-level cascade integration model is constructed. The first level screens high-contribution features, the second level fuses spatiotemporal dynamic features through an attention mechanism network, and the final level integrates static and dynamic factors to generate a dynamic risk score, outputs a visual risk trend graph and a high-risk factor heat map, and generates personalized intervention plans through real-time warnings and linkage through medical terminals. The present invention realizes the collaborative analysis of multi-dimensional data and dynamic risk tracking, significantly improving early warning capabilities and clinical decision-making efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a workflow diagram of a method for predicting coronary artery risk in patients with coronary heart disease as claimed in an embodiment of the present application; Figure 2 A second workflow diagram of a method for predicting coronary artery risk in patients with coronary heart disease as claimed in an embodiment of the present application; Figure 3 This is a structural module diagram of a coronary risk prediction system for patients with coronary heart disease as claimed in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0019] The terms "first", "second", "third" in the present application are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0020] Reference to "embodiments" herein means that the specific features, structures or properties described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily all refer to the same embodiments, nor is it necessarily independent or alternative embodiments to other embodiments. It is explicitly and implicitly understood by a person skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] According to the first embodiment of the present application, the present application claims to protect a coronary risk prediction method for coronary heart disease patients, with reference to Figure 1 , comprising: S1, determining a target patient, collecting and acquiring multi-source heterogeneous risk basic data of the target patient from a multi-source database; S2, obtaining layered feature parameters of each multi-source heterogeneous risk basic data based on the multi-source heterogeneous risk basic data by using different processing models; S3, inputting the layered feature parameters into a three-level cascaded integrated model to output a dynamic risk prediction score of the target patient; S4, generating a dynamic risk trend graph and a high-risk factor contribution heat map based on the dynamic risk prediction score, and pushing an early warning signal in real time through a medical terminal.

[0022] In this embodiment, when collecting multi-source heterogeneous data, the patient's electronic medical records, including medical history records, medication records, and surgical records, are extracted from the structured database of the hospital information system; the original DICOM file of the coronary CT angiography is obtained from the medical imaging archiving system; the continuous heart rate and blood oxygen data collected by the medical-grade electrocardiogram monitor worn by the patient are received in real time via the Bluetooth transmission protocol; and the genetic test report containing single nucleotide polymorphism sites is obtained from the genetic sequencing agency; For clinical data processing, the unstructured text in electronic medical records is fed into a pre-trained medical language model, which outputs key pathological feature vectors containing symptoms of myocardial ischemia and angina pectoris. Three-dimensional reconstruction of coronary artery CT images automatically identifies calcified areas on the vessel wall and calculates the percentage of calcified volume to the total vessel volume. For genomic data processing, the patient's gene loci are compared with the risk loci in the coronary heart disease genetic database, and the frequency of high-risk loci is counted; the genetic risk index is generated based on the clinical weight coefficients of different loci; For dynamic physiological data processing, ECG data is aggregated by hourly segments, the standard deviation of adjacent heartbeat intervals is calculated as an indicator of heart rate variability, and abnormal periods and durations when blood oxygen saturation is below 92% are marked; When constructing the dynamic risk prediction model, the hierarchical feature parameters generated by S2 are arranged in chronological order and input into the cascade prediction system: The first level uses the gradient boosting decision tree to filter out the top 30% important features; The second level uses a recurrent neural network with temporal memory function to analyze the changing trends of dynamic physiological characteristics over time; The third level integrates static features and trend analysis results, and generates a risk score of 0-100 through voting by multiple decision trees; When outputting risk visualization, the risk score is converted into three-color warning levels: red (high risk), yellow (medium risk), and green (low risk); a visual map is generated with the three-dimensional model of the coronary artery and the location of the plaque superimposed, and the area and location coordinates of the calcified area are marked.

[0023] Furthermore, the S1 further includes: Acquire the target patient's clinical data, genomic data, and dynamic physiological monitoring data in real time from hospital information systems, gene databases, and wearable devices, where the clinical data at least includes electronic medical records (EMRs), coronary CT angiography (CCTA) images, and laboratory test reports; Reference Figure 2 , said S2 further includes: The BERT model is used to extract key pathological features from unstructured text in clinical data, and a 3D convolutional neural network is used to automatically quantify coronary artery calcium scores (CACS) and plaque characteristics in CCTA images. The genetic risk value of coronary heart disease was calculated using the polygenic risk score (PRS) model for genomic data; Extract the heart rate variability (HRV) attenuation index from dynamic physiological data using a time series anomaly detection algorithm; Said S3 further includes: The hierarchical feature parameters output by S2 are input into a three-level cascade integration model, in which the first level uses XGBoost to screen feature importance, the second level uses a bidirectional LSTM network with an integrated attention mechanism to process time series features, and the third level uses random forest to integrate static and dynamic features to generate a dynamic risk prediction score.

[0024] Furthermore, the extraction of the plaque features in S2 includes: The U-Net++ segmentation model was used to identify non-calcified plaques, mixed plaques, and low-density plaques in CCTA images. The ratio of plaque volume to vascular cross-sectional area (PV / VA) was calculated as a quantitative indicator of high-risk plaques; The calcification area identification of S2 includes: Automatically outline the vascular intima border on the cross section of CT images; Pixel areas with CT values ​​greater than 130HU were marked as calcified plaques; The total volume of calcified plaques was calculated by continuous scanning along the vascular axis.

[0025] Furthermore, the construction of the polygenic risk score PRS model of S2 also includes: SNP sites associated with coronary heart disease were extracted from the UK Biobank database, and significant sites were screened using LASSO regression. Generate patient-specific PRS values ​​based on the target population allele frequency-corrected weights; Said S2 also includes the generation of a genetic risk index, prioritizing the screening of 20 gene loci that are strongly associated with coronary heart disease in the target population; The weighting coefficient is set according to the multiple of each site that increases the risk of coronary heart disease; The weighted values ​​of all high-risk sites carried by the patient are accumulated as the final index.

[0026] Furthermore, a feature interaction module is introduced into the three-level cascade integration model of S3, and the interaction terms of clinical features and genomic features are generated through an explicit feature cross network, and the interaction terms and original features are input into a bidirectional LSTM network. The second level of processing in S3 includes arranging the heart rate variability data for 7 consecutive days in chronological order, automatically identifying the attenuation trend of variability that has continued to decrease for more than 5 days, and inputting the attenuation slope as an independent risk factor into the third level model; The third level of processing in S3 includes multiplying the genetic risk index by the calcification volume percentage to generate a composite index. When the composite index is greater than a threshold, the final risk level is automatically increased.

[0027] Furthermore, the attention mechanism of S3 is specifically implemented as follows: A temporal attention layer is superimposed on the bidirectional LSTM output layer to dynamically weight the physiological monitoring data at different time points; The correlation between attention weights and risk scores is optimized via backpropagation; Before executing S3, the following steps are also included: Smoothing and filtering of abnormal peaks in ECG data caused by motion interference; Remove blurred frames caused by respiratory motion artifacts in CT images.

[0028] Furthermore, the method further comprises: In the continuous learning step of the model, when the amount of new patient data reaches a threshold, the incremental learning process is started, and the elastic weight curing (EWC) algorithm is used to prevent catastrophic forgetting of the model; The online model update step receives data on newly diagnosed coronary heart disease patients from the hospital every month, retrains only the third-level decision tree model and retains the original feature screening rules, and triggers a full model update when the accuracy of the validation set changes by more than 5%.

[0029] Furthermore, the generation of the high-risk factor contribution heat map of S4 includes: The SHAP value is used to interpret the model output, calculate the marginal contribution of each feature to the risk score, and generate a feature importance map by sorting the absolute value of the contribution; The top 10% of calcified plaques by volume are highlighted in red, and a numerical dashboard showing genetic risk index and dynamic physiological indicators is displayed next to the vascular model.

[0030] Furthermore, the dynamic physiological monitoring data of S1 includes: The resting heart rate, QT interval dispersion, and standard deviation of blood oxygen saturation fluctuations at night for 7 consecutive days were used to suppress sensor noise through Kalman filtering; The processing of the dynamic physiological monitoring data also includes counting the number of abnormal blood oxygen levels during sleep at night, and activating an emergency data retransmission mechanism when the number of abnormalities in a single night exceeds 10 times.

[0031] Furthermore, the dynamic risk prediction score is used to: Generate personalized intervention strategy recommendations, including medication adjustment suggestions and review cycle planning, and link with the hospital's resource scheduling system to automatically schedule follow-up visits with cardiologists; An intervention priority list pops up on the physician's workstation, sorted in descending order based on risk score, and a draft report is automatically generated containing the recommended date for a follow-up CT scan and a statin dose adjustment plan.

[0032] According to the second embodiment of the present invention, the present invention claims a coronary risk prediction system for patients with coronary heart disease, referring to Figure 3 ,include: one or more processors; A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the coronary risk prediction method for patients with coronary heart disease.

[0033] In this embodiment, the target patient multi-source heterogeneous data collection is aimed at obtaining clinical data, and extracting the patient's electronic health record from the hospital information system structured database, including: Disease diagnosis entries and disease course description text in medical history records; Drug categories and medication cycles in the long-term medication list; Summary of interventional treatment modalities and postoperative follow-up in surgical records.

[0034] Retrieve original coronary CT angiography image files from the medical imaging archiving system, including the complete cardiac scan sequence, and integrate biomarker test results from laboratory test reports; Genomic data access: Obtain patient gene sequencing reports from genetic testing institutions through encrypted transmission protocols, and analyze the single nucleotide polymorphism site information and gene annotations in the reports; During dynamic physiological monitoring, a Bluetooth low-power communication link is established with a medical-grade ECG monitor to receive and cache continuously collected ECG waveforms and blood oxygen saturation original signals in real time, and an automatic retransmission mechanism is activated for interrupted data streams.

[0035] For hierarchical feature parameter extraction, including clinical data processing and text feature extraction, the unstructured text of electronic medical records is input into the pre-trained medical semantic understanding model: Identify and encode myocardial ischemia-related symptom description entities, extract semantic features of angina attack frequency and duration, and output pathological feature vector representation Coronary image analysis includes vascular reconstruction and calcification identification, isotropic spatial reconstruction of CT image sequences, automatic segmentation of the coronary artery tree in three-dimensional space, detection of continuous areas where the CT values ​​of the vessel wall exceed the calcification threshold, and calculation of the ratio of the volume of the calcified area to the total volume of the vessel segment.

[0036] When quantifying plaque features, a multi-layer segmentation network is used to distinguish plaque subtypes, including non-calcified / mixed / low-density. The relative area ratio of the plaque to the vascular cross-section is calculated based on the three-dimensional morphology of the plaque, and the morphological characteristics of vulnerable plaques, such as surface irregularity, are marked.

[0037] The genomic data were processed for genetic risk modeling, and the patient's gene loci were matched with the coronary heart disease genetic database. Based on the weight coefficient corrected for population frequency, the multi-locus risk effect value was aggregated to output a standardized genetic risk index. Perform ECG signal analysis for dynamic physiological data processing, process continuous ECG data in segments with fixed durations, calculate the degree of discrete fluctuations between adjacent heartbeat intervals as an indicator of autonomic nervous function, identify motion artifacts in the signal, and perform data repair; Monitor blood oxygen abnormalities, detect the duration of time when blood oxygen saturation is below the clinical warning line, count the frequency of abnormal events per unit time, and calculate the coefficient of variation of blood oxygen fluctuation amplitude.

[0038] For the three-level cascade dynamic risk assessment, the first level is feature importance screening, which receives all hierarchical feature parameters and classifies them by source. The association strength between the features and coronary events is evaluated through a tree structure model, and the feature subset with the highest association strength is retained; The second level is temporal trend modeling. It constructs a time series matrix for dynamic physiological characteristics and uses a recurrent network structure with memory units to capture the continuous change pattern of indicators such as heart rate variability. It automatically identifies the decay trajectory of continuously decreasing variability, outputs a mathematical representation of the decay trend, and superimposes an attention mechanism to dynamically weight key time nodes. The third level is multimodal feature fusion for feature interaction processing, generating cross-combination terms of clinical features and genetic features, and calculating the multiplicative composite factor of calcification index and genetic risk; An integrated decision-making mechanism is adopted to input static features, time series trends and interaction items into multiple decision trees. A comprehensive risk score of 0-100 is generated based on voting based on the tree model. When the composite factor exceeds the preset critical value, the risk level escalation rule is activated.

[0039] A dynamic risk map is generated for risk visualization and early warning. The risk score is mapped to a three-color warning level (red / yellow / green). By drawing a trend curve of the risk score over time, the time nodes corresponding to clinical intervention events are marked, and the critical time period of risk level transition is highlighted. The construction of a 3D coronary artery model integrates CT image reconstruction results with plaque location information, uses color gradients to render the spatial distribution of different plaque types, and superimposes the contours of calcified areas on the surface of the vascular model to generate an interactively rotatable atlas of the coronary anatomy. The contribution of high-risk factors is analyzed using model interpretability technology to calculate feature contribution values, which are then sorted by contribution to generate a heat map. Plaque features and genetic factors are highlighted for contribution, and displayed alongside the 3D model: a real-time data dashboard for key physiological indicators, a genetic risk index level indicator, an early warning push mechanism, and a pop-up alert to the attending physician's terminal when the risk level enters the high-risk range. A summary report containing core risk factors is automatically generated, and personalized intervention recommendations are pushed. Recommendations for drug adjustment and dosage optimization include imaging review window planning and specialist clinic appointment recommendations.

[0040] The incremental learning process of the system continuous optimization mechanism of this embodiment includes regularly receiving new patient data and initiating model updates, using an anti-forgetting learning strategy to retain historical knowledge, and triggering full model retraining when model performance fluctuations exceed a tolerance threshold; The quality control module is used to perform integrity verification on the raw data, discard physiological signal segments with severe motion artifacts, exclude blurred CT image frames caused by respiratory motion, and establish a failure retry mechanism for the feature extraction process.

[0041] This program ensures clinical practicality through the above design: Deeply integrate the static risk of multi-source genomes and dynamic physiological changes to capture the multi-dimensional characterization of coronary artery lesions; Using time-series sensitivity to identify progressive attenuation patterns of indicators such as heart rate variability, providing early warning of potential functional deterioration; Execute interpretable output to visualize high-risk plaques in three dimensions, and use heat maps to quantify the contribution of each factor to assist clinical decision-making; A closed-loop intervention plan is formed, and the risk score directly links the generation of treatment recommendations and the scheduling of medical resources to form a closed-loop prevention and control system.

[0042] This example demonstrates that by implementing each technology step by step, a closed loop has been achieved: from data acquisition to feature processing to model decision-making to clinical output. While excluding diagnostic methods, a complete, patent-eligible technical solution has been constructed using quantitative features (such as the 130HU calcification threshold), dynamic mechanisms (such as blood oxygen retransmission), and clinical integration (automatic medical order generation).

[0043] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0044] In addition, the functional units in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

[0045] The above detailed description of the specific embodiments of the invention is intended only as an example, and the present application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions of the invention are also within the scope of the present application. Therefore, equivalent changes, modifications, and improvements made without departing from the spirit and scope of the present application should be included within the scope of the present application.

Claims

1. A method for predicting coronary risk in patients with coronary heart disease, characterized in that: include: S1, determining a target patient, and acquiring multi-source heterogeneous risk basic data of the target patient from a multi-source database; S2, obtaining hierarchical characteristic parameters of each multi-source heterogeneous risk basic data using different processing models based on the multi-source heterogeneous risk basic data; S3, inputting the hierarchical characteristic parameters into a three-level cascade integrated model, and outputting a dynamic risk prediction score for the target patient; S4, generating a dynamic risk trend graph and a high-risk factor contribution heat map based on the dynamic risk prediction score, and pushing an early warning signal in real time through a medical terminal.

2. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 1, characterized in that: Said S1 further comprises: Acquire the target patient's clinical data, genomic data, and dynamic physiological monitoring data in real time from hospital information systems, gene databases, and wearable devices, where the clinical data at least includes electronic medical records (EMRs), coronary CT angiography (CCTA) images, and laboratory test reports; Said S2 further includes: The BERT model is used to extract key pathological features from unstructured text in clinical data, and a 3D convolutional neural network is used to automatically quantify coronary artery calcium scores (CACS) and plaque characteristics in CCTA images. The genetic risk value of coronary heart disease was calculated using the polygenic risk score (PRS) model for genomic data; Extract the heart rate variability (HRV) attenuation index from dynamic physiological data using a time series anomaly detection algorithm; Said S3 further includes: The hierarchical feature parameters output by S2 are input into a three-level cascade integration model, in which the first level uses XGBoost to screen feature importance, the second level uses a bidirectional LSTM network with an integrated attention mechanism to process time series features, and the third level uses random forest to integrate static and dynamic features to generate a dynamic risk prediction score.

3. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 2, characterized in that: The extraction of the plaque features in S2 includes: The U-Net++ segmentation model was used to identify non-calcified plaques, mixed plaques, and low-density plaques in CCTA images. The ratio of plaque volume to vascular cross-sectional area (PV / VA) was calculated as a quantitative indicator of high-risk plaques; The calcification area identification of S2 includes: Automatically outline the vascular intima border on the cross section of CT images; Pixel areas with CT values ​​greater than 130HU were marked as calcified plaques; The total volume of calcified plaques was calculated by continuous scanning along the vascular axis.

4. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 2, wherein: The construction of the polygenic risk score PRS model of S2 also includes: SNP sites associated with coronary heart disease were extracted from the UK Biobank database, and significant sites were screened using LASSO regression. Generate patient-specific PRS values ​​based on the target population allele frequency-corrected weights; Said S2 also includes the generation of a genetic risk index, prioritizing the screening of 20 gene loci that are strongly associated with coronary heart disease in the target population; The weighting coefficient is set according to the multiple of each site that increases the risk of coronary heart disease; The weighted values ​​of all high-risk sites carried by the patient are accumulated as the final index.

5. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 2, characterized in that: A feature interaction module is introduced into the three-level cascade integration model of S3, which generates interaction terms between clinical features and genomic features through an explicit feature cross network, and inputs the interaction terms and original features into a bidirectional LSTM network. The second level of processing in S3 includes arranging the heart rate variability data for 7 consecutive days in chronological order, automatically identifying the attenuation trend of variability that has continued to decrease for more than 5 days, and inputting the attenuation slope as an independent risk factor into the third level model; The third level of processing in S3 includes multiplying the genetic risk index by the calcification volume percentage to generate a composite index. When the composite index is greater than a threshold, the final risk level is automatically increased.

6. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 2, characterized in that: The S3 attention mechanism is specifically implemented as follows: A temporal attention layer is superimposed on the bidirectional LSTM output layer to dynamically weight the physiological monitoring data at different time points; The correlation between attention weights and risk scores is optimized via backpropagation; Before executing S3, the following steps are also included: Smoothing and filtering of abnormal peaks in ECG data caused by motion interference; Remove blurred frames caused by respiratory motion artifacts in CT images.

7. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 2, characterized in that: The method further includes: In the continuous learning step of the model, when the amount of new patient data reaches a threshold, the incremental learning process is started, and the elastic weight curing (EWC) algorithm is used to prevent catastrophic forgetting of the model; The online model update step receives data on newly diagnosed coronary heart disease patients from the hospital every month, retrains only the third-level decision tree model and retains the original feature screening rules, and triggers a full model update when the accuracy of the validation set changes by more than 5%.

8. The method for predicting coronary artery risk in patients with coronary heart disease according to claim 1, characterized in that: The generation of the S4 high-risk factor contribution heat map includes: The SHAP value is used to interpret the model output, calculate the marginal contribution of each feature to the risk score, and generate a feature importance map by sorting the absolute value of the contribution; The top 10% of calcified plaques by volume are highlighted in red, and a numerical dashboard showing genetic risk index and dynamic physiological indicators is displayed next to the vascular model.

9. The method for predicting coronary risk in patients with coronary heart disease according to claim 1, characterized in that: The dynamic physiological monitoring data of S1 include: The resting heart rate, QT interval dispersion, and standard deviation of blood oxygen saturation fluctuations at night for 7 consecutive days were used to suppress sensor noise through Kalman filtering; The processing of the dynamic physiological monitoring data also includes counting the number of abnormal blood oxygen levels during sleep at night, and activating an emergency data retransmission mechanism when the number of abnormalities in a single night exceeds 10; The dynamic risk prediction score is used to: Generate personalized intervention strategy recommendations, including medication adjustment suggestions and review cycle planning, and link with the hospital's resource scheduling system to automatically schedule follow-up visits with cardiologists; An intervention priority list pops up on the physician's workstation, sorted in descending order based on risk score, and a draft report is automatically generated containing the recommended date for a follow-up CT scan and a statin dose adjustment plan.

10. A coronary risk prediction system for patients with coronary heart disease, characterized in that: include: one or more processors; A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement a method for predicting coronary risk in patients with coronary heart disease according to any one of claims 1 to 9.

Images