Pre-diagnosis system for coronary heart disease

Abstract

The invention relates to the technical field of cardiovascular diagnosis, and discloses a pre-diagnosis system for coronary heart disease, comprising: a data acquisition module for acquiring coronary artery images, dynamic physiological data and serum metabolite detection data; the data preprocessing module is used for standardizing data; the feature extraction module is used for extracting multi-dimensional features and constructing a cross-dimensional score set; the diagnosis analysis module is used for constructing a multi-feature fusion diagnosis model and pre-diagnosing the coronary heart disease probability and risk level of a patient; and the result output module is used for generating a visual report and establishing a database. According to the method, the image, physiological and metabolic multi-dimensional data are fused, the random forest algorithm is utilized to construct the multi-feature fusion diagnosis model, accurate pre-diagnosis of the coronary heart disease probability and the risk level of the patient is realized, the diagnosis efficiency and accuracy are improved, and reliable reference is provided for clinic.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cardiovascular diagnosis, in particular to a pre-diagnosis system for coronary heart disease. BACKGROUND

[0002] With the increasing incidence of cardiovascular diseases year by year, coronary heart disease as a high mortality cardiovascular disease, the demand for early pre-diagnosis is increasingly urgent. Current coronary artery imaging equipment, patient wearable devices and serum detection equipment are increasingly mature, providing support for synchronous acquisition of vascular structure image, dynamic physiological data of electrocardiogram and blood pressure and serum metabolite detection data, and the development of machine learning technology also lays a foundation for the fusion analysis of multi-source medical data.

[0003] Traditional coronary heart disease diagnosis relies on a single data source, such as evaluating vascular lesions only through imaging or judging the disease condition only according to physiological indicators, which has the defects of simple feature extraction and lack of cross-dimensional data interaction analysis; and no systematic feature credibility verification and weight distribution mechanism is established, which is easy to cause diagnosis deviation due to data redundancy or key features being ignored, and it is difficult to accurately calculate the incidence probability and risk level of coronary heart disease.

[0004] The limitations of single data diagnosis, rough feature processing, insufficient cross-dimensional fusion, and lack of credibility verification and reasonable weight distribution in the prior art result in lack of diagnosis accuracy and reliability, which cannot effectively meet the early pre-diagnosis needs of coronary heart disease.

[0005] Therefore, the present application provides a pre-diagnosis system for coronary heart disease to solve the above problems. SUMMARY

[0006] In view of this, the present application provides a pre-diagnosis system for coronary heart disease to solve the problems of limitations of single data diagnosis, rough feature processing, insufficient cross-dimensional fusion, and lack of credibility verification and reasonable weight distribution in the prior art, which result in lack of diagnosis accuracy and reliability, and cannot effectively meet the early pre-diagnosis needs of coronary heart disease.

[0007] In one aspect, the present application provides a pre-diagnosis system for coronary heart disease, comprising: A data acquisition module is configured to acquire image data of a patient's coronary artery vascular structure through a coronary artery imaging device; is further configured to connect a patient wearable device and acquire dynamic physiological data of electrocardiogram and blood pressure of the patient in real time; and is further configured to interface with a serum detection device to acquire metabolite detection data related to coronary heart disease in the patient's serum; A data preprocessing module is configured to standardize and preprocess the image data, dynamic physiological data and metabolite detection data; The feature extraction module is configured to extract multi-dimensional features of blood vessel structure features, dynamic physiological data trend features, and metabolite concentration features from the preprocessed data. The diagnostic analysis module is configured to use the multi-dimensional features as input variables and a clinically confirmed diagnosis result as a label to construct a multi-feature fusion diagnostic model according to a random forest algorithm, and to use the multi-feature fusion diagnostic model to predict the probability and risk level of coronary heart disease of a patient. The result output module is configured to generate a visual diagnostic report of the diagnostic result of the diagnostic analysis module and establish a patient diagnosis database, and to output the visual diagnostic report to a patient intelligent terminal.

[0008] Further, the feature extraction module comprises a feature refinement unit and a feature interaction unit. The feature refinement unit is configured to finely extract the multi-dimensional features. The feature refinement unit finely extracts the blood vessel structure features into vessel stenosis, plaque features, and vessel branch angles. The feature refinement unit processes the dynamic physiological data trend features into heart rate variability, blood pressure circadian rhythm, and electrocardiogram ST segment trend. The feature refinement unit processes the metabolite concentration features into good and bad cholesterol ratio, triglyceride concentration, and homocysteine concentration.

[0009] Further, the feature interaction unit is configured to eliminate duplicates of the finely extracted multi-dimensional features according to a preset feature correlation threshold, and to construct a cross-dimensional feature score set of images, physiology, and metabolism according to the multi-dimensional features after the elimination of duplicates.

[0010] Further, the preset feature correlation threshold comprises a preset similar feature correlation threshold and a preset diagnostic correlation threshold. The elimination of duplicates comprises that the feature interaction unit determines whether the correlation value of two similar multi-dimensional features exceeds the preset similar feature correlation threshold. If not, the feature interaction unit retains both features. If yes, the feature interaction unit determines the correlation of the two similar multi-dimensional features with the coronary heart disease diagnosis result respectively. If the correlation of a multi-dimensional feature with the coronary heart disease diagnosis result does not exceed the preset diagnostic correlation threshold, the feature interaction unit removes the feature. If the correlation of a multi-dimensional feature with the coronary heart disease diagnosis result exceeds the preset diagnostic correlation threshold, the feature interaction unit retains the feature.

[0011] Further, the deduplication elimination further comprises: when judging the correlation of the two similar multi-dimensional features and the coronary heart disease diagnosis result, if the correlation of the two similar multi-dimensional features and the coronary heart disease diagnosis result both exceeds the preset diagnosis correlation threshold, comparing the correlation numerical values of the two similar multi-dimensional features and the coronary heart disease diagnosis result, retaining the feature with the larger correlation numerical value, and removing the feature with the smaller correlation numerical value.

[0012] Further, the diagnosis analysis module comprises a feature evaluation unit, a weight allocation unit and a hierarchical diagnosis unit. The feature evaluation unit is configured to verify the credibility of the cross-dimensional feature score set. The feature evaluation unit presets a credibility evaluation rule for image, physiological and metabolic features respectively. If a feature meets the rule, the feature evaluation unit marks the feature as credible. If a feature does not meet the rule, the feature evaluation unit performs secondary verification on the feature in combination with whether other features meet the rule.

[0013] Further, the weight allocation unit is configured to adjust the score weight of each dimensional feature in the cross-dimensional feature score set. When the feature evaluation unit performs secondary verification: If a feature does not meet the rule, and the other two features both meet the rule, the feature evaluation unit marks the cross-dimensional feature score set as credible after the weight allocation unit reduces the score weight of the feature. If more than two features do not meet the rule, the feature evaluation unit marks the cross-dimensional feature score set as untrustworthy, and re-verifies a new cross-dimensional feature score set.

[0014] Further, the hierarchical diagnosis unit is configured to input the multi-dimensional features of a patient to be diagnosed into the multi-feature fusion diagnosis model, calculate the coronary heart disease probability of the patient according to the cross-dimensional feature score set marked as credible, and determine the coronary heart disease risk level of the patient according to a preset probability range.

[0015] Further, the process of determining the coronary heart disease risk level of the patient by the hierarchical diagnosis unit comprises: If the coronary heart disease probability of the patient is less than the preset probability range, the hierarchical diagnosis unit determines that the coronary heart disease risk level of the patient is low risk. If the coronary heart disease probability of the patient is in the preset probability range, the hierarchical diagnosis unit determines that the coronary heart disease risk level of the patient is medium risk. If the coronary heart disease probability of the patient is greater than the preset probability range, the hierarchical diagnosis unit determines that the coronary heart disease risk level of the patient is high risk.

[0016] Further, the visual diagnosis report comprises patient basic information, key indicators of multi-dimensional features, coronary heart disease probability, risk level and clinical suggestions.

[0017] Compared with the prior art, the present application has the following beneficial effects: The present application integrates coronary artery image, dynamic physiological and serum metabolite data through a multi-source data acquisition module, realizes comprehensive information collection, ensures data effectiveness and feature comprehensiveness through standardized preprocessing and multi-dimensional feature extraction, constructs a multi-feature fusion model with the help of a random forest algorithm, improves the accuracy and reliability of coronary heart disease pre-diagnosis, provides intuitive reference for clinical through a visual report and database construction, facilitates data management, at the same time, pushes the result to the patient terminal, is beneficial to early intervention, and overall improves the comprehensiveness, accuracy and practicality of coronary heart disease pre-diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present application. Moreover, the same reference numerals in different figures represent the same or similar components. In the drawings: Figure 1 A functional block diagram of a pre-diagnosis system for coronary heart disease provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the present disclosure are shown. It is to be understood that the present disclosure can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art. It is to be noted that the embodiments and features of the present application can be combined with each other, without conflict, in the present application. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0020] The existing coronary heart disease pre-diagnosis system has the problems of single data diagnosis limitation, rough feature processing, insufficient cross-dimension fusion, lack of credibility verification and reasonable weight distribution, resulting in lack of diagnosis accuracy and reliability, and cannot effectively meet the early pre-diagnosis needs of coronary heart disease. Therefore, the present application provides a pre-diagnosis system for coronary heart disease to solve the above problems.

[0021] Reference Figure 1 In some embodiments of the present application, a pre-diagnosis system for coronary heart disease comprises a data acquisition module, a data preprocessing module, a feature extraction module, a diagnosis analysis module and a result output module.

[0022] Specifically, the data acquisition module acquires imaging data of the patient's coronary artery structure through coronary imaging equipment; it is also used to connect to the patient's wearable device and acquire dynamic physiological data of the patient's electrocardiogram and blood pressure in real time; and it is also used to interface with serum detection equipment to acquire detection data of metabolites related to coronary heart disease in the patient's serum.

[0023] Specifically, the data acquisition module uses a multi-source data collaborative acquisition mechanism to connect to coronary artery imaging equipment such as CT angiography machines to acquire vascular structure image data, connects to wearable devices such as smart bracelets to transmit real-time dynamic physiological data such as electrocardiogram and blood pressure, and establishes a data interaction channel with serum testing equipment such as fully automated biochemical analyzers to acquire metabolite detection data.

[0024] Specifically, the data preprocessing module performs standardized preprocessing on image data, dynamic physiological data, and metabolite detection data. Preprocessing includes data cleaning, format unification, and dimensional standardization. Data cleaning includes outlier removal and missing value imputation. Format unification includes converting the DICOM format of CTA images to the system's common JPEG format (preserving key structural information), converting the waveform files of ECG data into numerical sequences, and converting the text reports of metabolites into structured numerical values. Dimensional standardization includes using "Z-score standardization" (converting the data to a normal distribution with mean = 0 and standard deviation = 1) to process metabolite concentration and physiological data, and using "min-max normalization" to process image pixel values ​​to ensure that all data are within the same numerical range.

[0025] Specifically, the feature extraction module extracts multi-dimensional features from the preprocessed data, including vascular structure features, dynamic physiological data trend features, and metabolite concentration features, through a multi-dimensional feature deep analysis mechanism.

[0026] Specifically, vascular structural features include: structural parameters directly related to coronary heart disease extracted from coronary artery images, such as vascular stenosis: the ratio of the diameter at the stenosis site to the diameter at the normal site; plaque characteristics: plaque stability, degree of calcification, size, density, location, and thickness of the fibrous cap; and vascular branching angles. Vascular structural features are the core indicators for judging the degree of vascular lesions.

[0027] Dynamic physiological data trend characteristics include: indicators of "change patterns" extracted from continuous physiological data, such as heart rate variability (HRV): the standard deviation of heart rate fluctuations, reflecting autonomic nervous function; blood pressure diurnal rhythm: the difference between daytime and nighttime blood pressure, abnormalities indicate decreased vascular regulation capacity; and ECG ST segment trend: the duration of ST segment elevation / depression. Dynamic physiological data trend characteristics can indicate the risk of myocardial ischemia.

[0028] Metabolite concentration characteristics include: the "concentration and ratio" of serum metabolites related to coronary heart disease, such as the LDL-C / HDL-C ratio (the ratio of bad cholesterol to good cholesterol, the higher the ratio, the higher the risk), triglyceride concentration, homocysteine ​​concentration, etc. Metabolite concentration characteristics reflect the impact of the blood biochemical environment on blood vessels.

[0029] Specifically, the diagnostic analysis module is used to construct a multi-feature fusion diagnostic model based on the random forest algorithm, using multi-dimensional features as input variables and clinical diagnosis results as labels. The multi-feature fusion diagnostic model is then used to pre-diagnose the probability and risk level of coronary heart disease in patients.

[0030] Specifically, the diagnostic analysis module collects historical data on patients' multi-dimensional features and clinical diagnoses from the feature extraction module, dividing it into a training set and a test set in a 7:3 ratio. The training set is used to train a random forest model: 100 decision trees are randomly generated, and each tree learns the mapping relationship between "feature → diagnostic result" based on random samples and random features. By adjusting parameters such as "number of decision trees" and "tree depth," the model's accuracy is verified using the test set until the accuracy meets clinical requirements (e.g., ≥85%), thus completing the construction of the multi-feature fusion diagnostic model. The diagnostic analysis module inputs the multi-dimensional feature matrix of the patient to be diagnosed into the trained diagnostic analysis module. The model calculates the probability of having coronary heart disease through voting by 100 decision trees. The diagnostic analysis module presets the probability range of having coronary heart disease according to clinical guidelines, such as low risk <20%, medium risk 20%-70%, and high risk >70%, and then corresponds the calculated patient coronary heart disease probability to the risk level.

[0031] Specifically, the results output module generates a visual diagnostic report from the diagnostic analysis module, establishes a patient diagnostic database, and outputs the visual diagnostic report to the patient's smart terminal.

[0032] Specifically, the results output module automatically converts the diagnostic results from the diagnostic analysis module into a visual diagnostic report. Then, it stores the patient's full-link data (raw data, preprocessing results, features, and reports) in a diagnostic database (such as MySQL) indexed by "patient ID + examination time," and supports medical staff to query historical diagnostic records by ID. Finally, the visual report is sent to the patient's smart terminal via APP push, SMS link, etc., and doctors can view and export the report in the system backend.

[0033] Understandably, the raw multidimensional data is too large, and directly inputting it into a multi-feature fusion diagnostic model will cause the model complexity to soar and the prediction accuracy to drop. The pathogenesis of coronary heart disease is the result of the combined effects of "vascular structural lesions + physiological function abnormalities + metabolic disorders". Therefore, it is necessary to use the feature extraction module to specifically extract three types of features that match the pathological mechanism to achieve data dimensionality reduction and information focus.

[0034] As can be seen, the above embodiments integrate "structural + physiological + metabolic" information through multi-source data collection, avoiding the risk of missed diagnoses due to single data (such as relying solely on imaging); standardized preprocessing improves data quality, the ensemble learning characteristics of the random forest model reduce diagnostic errors, and the accuracy is higher than traditional subjective judgment; visual reports facilitate doctors' rapid decision-making, the database supports case traceability, and patient terminal push assists in early intervention; the full automation of the process from data collection to diagnosis to push reduces manual operation and shortens the diagnostic cycle; new cases accumulated in the diagnostic database can be used to optimize the model, continuously improve diagnostic performance, and adapt to changes in clinical needs.

[0035] Reference Figure 1 In some embodiments of this application, the feature extraction module includes a feature refinement unit and a feature interaction unit; Feature refinement units are used to refine the extraction of multi-dimensional features: The feature refinement unit extracts vascular structural features into vascular stenosis, plaque features, and vascular branching angles. The feature refinement unit extracts dynamic physiological features into heart rate variability, blood pressure diurnal rhythm, and ECG ST segment trend. The feature refinement unit extracts metabolite features into the ratio of good to bad cholesterol, triglyceride concentration, and homocysteine ​​concentration.

[0036] Specifically, the feature extraction module's feature refinement unit is equipped with a feature deep deconstruction model, which processes image features into vascular stenosis, plaque features, and vascular branch angles. Dynamic physiological features are processed into heart rate variability, blood pressure diurnal rhythm, and ECG ST segment trends through a time-series physiological data fluctuation analysis model. Metabolite features are processed into the ratio of good to bad cholesterol, triglyceride concentration, and homocysteine ​​concentration through a metabolite network topology analysis model.

[0037] Understandably, the feature extraction module extracts deeper features reflecting the pathological mechanism of coronary heart disease from basic features, such as plaque stability being directly related to the risk of myocardial infarction, making the features more diagnostically specific and reducing interference from irrelevant information.

[0038] Reference Figure 1In some embodiments of this application, the feature interaction unit is used to remove duplicates from the finely extracted multi-dimensional features according to a preset feature relevance threshold, and to construct a cross-dimensional feature scoring set of image, physiology and metabolism based on the deduplicated multi-dimensional features.

[0039] Specifically, the preset feature relevance thresholds include: preset similar feature relevance thresholds and preset diagnostic relevance thresholds; Deduplication and deduplication include: the feature interaction unit determines whether the correlation value of two similar multi-dimensional features exceeds a preset similarity feature correlation threshold. If the number does not exceed a certain limit, the feature interaction unit will retain two features simultaneously. If the number of similar features exceeds a certain threshold, the feature interaction unit will then determine the correlation between the two similar multidimensional features and the diagnosis of coronary heart disease: If the correlation between a certain multidimensional feature and the diagnosis result of coronary heart disease does not exceed the preset diagnostic correlation threshold, the feature interaction unit will remove the feature. If the correlation between a certain multidimensional feature and the diagnosis result of coronary heart disease exceeds the preset diagnostic correlation threshold, the feature interaction unit will retain the feature.

[0040] Specifically, deduplication and elimination also include: when judging the correlation between two similar multidimensional features and the diagnosis result of coronary heart disease: if the correlation between the two similar multidimensional features and the diagnosis result of coronary heart disease exceeds the preset correlation threshold with the diagnosis result of coronary heart disease, then compare the correlation values ​​between the two similar multidimensional features and the diagnosis result of coronary heart disease, retain the feature with the larger correlation value, and remove the feature with the smaller correlation value.

[0041] Specifically, combining clinical data and machine learning experience, a preset feature relevance threshold is established. For the similarity feature relevance threshold, the Pearson correlation coefficient algorithm is used to calculate the relevance of similar features. If the Pearson correlation coefficient is ≥0.8, meaning the linear correlation between two features exceeds 80%, they are considered highly similar. For the preset diagnostic relevance threshold, the mutual information algorithm is used to calculate the relevance of each feature to the "clinical diagnosis result of coronary heart disease" (label: confirmed / not confirmed). If the mutual information value is ≥0.5, meaning the correlation between the feature and the clinical diagnosis result of coronary heart disease exceeds 50%, it is considered diagnostically significant.

[0042] Specifically, the following is a concrete example of pairwise comparison and deduplication of nine detailed features (three imaging features, three physiological features, and three metabolic features): If the correlation between two features is ≥0.8 (highly similar): Example 1: The mutual information value of vascular stenosis in the imaging dimension is 0.68, and the mutual information value of plaque size in the plaque feature is 0.62. The correlation between vascular stenosis and plaque size is 0.85. Since both are strongly correlated with the clinical diagnosis of coronary heart disease, vascular stenosis with a higher mutual information value is retained, and plaque size is discarded. Example 2: The mutual information value of the good cholesterol to bad cholesterol ratio in the metabolic dimension is 0.65, and the mutual information value of triglyceride concentration is 0.48. The correlation between the two is 0.82. Since the mutual information value of triglyceride concentration is <0.5, it is directly eliminated. If the correlation between two features is <0.8 (no redundancy): both are retained. For example, the mutual information value of heart rate variability in the physiological dimension is 0.58, and the mutual information value of blood pressure diurnal rhythm is 0.52. The correlation between the two is 0.65, and both are retained.

[0043] Specifically, the deduplicated multi-dimensional features are used to construct cross-dimensional feature scoring sets for imaging, physiology, and metabolism according to the "dimensional intersection" principle, including four types of scoring set combinations: Combination 1: Imaging + Physiology, such as plaque characteristics + heart rate variability; Combination 2: Imaging + Metabolism, such as plaque characteristics + good and bad cholesterol ratio; Combination 3: Physiological + Metabolic, such as the circadian rhythm index of blood pressure + triglyceride concentration; Combination 4: Imaging + Physiology + Metabolism, such as plaque characteristics + heart rate variability + triglyceride concentration.

[0044] Specifically, the diagnostic analysis module includes a feature evaluation unit, a weight allocation unit, and a hierarchical diagnostic unit. The weight allocation unit first calculates feature importance using logistic regression or random forest algorithms, contributing weights to the feature allocation diagnosis in combination 4. For example, the plaque feature score has a weight of 0.4, the heart rate variability score has a weight of 0.3, and the triglyceride concentration score has a weight of 0.3. Then, the weight allocation unit uses a weighted summation method to calculate the comprehensive score for each scoring base combination: Combination score = (Feature 1 × Weight 1) + (Feature 2 × Weight 2) + ... + (Feature n × Weight n), with a score range of 0-10. Higher scores indicate a higher risk of coronary heart disease.

[0045] It is understandable that the feature score weights in combination 1, combination 2 and combination 3 are all 0.5.

[0046] Specifically, the feature evaluation unit is used to verify the credibility of cross-dimensional feature scoring sets: The feature evaluation unit pre-sets confidence evaluation rules for the imaging, physiological and metabolic features in combination 4; If a feature meets the rules, the feature evaluation unit will mark the feature as trustworthy; If a feature does not conform to the rule, the feature evaluation unit will perform a secondary verification of that feature by combining the verification of other features with the rule.

[0047] Understandably, the preset confidence assessment rules are based on clinical standards and equipment parameter settings.

[0048] Specifically, the weight allocation unit is also used to adjust the score weights of features in each dimension of the cross-dimensional feature scoring set; When the feature evaluation unit performs secondary validation: If a certain feature does not meet the rules, but the other two features do meet the rules, the feature evaluation unit will mark the cross-dimensional feature score set as trustworthy after the weight allocation unit reduces the score weight of that feature. If two or more features do not meet the rules, the feature evaluation unit will mark the cross-dimensional feature score set as untrustworthy and re-verify the new cross-dimensional feature score set.

[0049] Specifically, the following is an example of the feature evaluation unit verifying the feature credibility of combination 4: First, the feature evaluation unit compares each feature of combination 4 against the preset rules one by one: If a feature in a certain dimension fully meets all the rules of the corresponding dimension, it is marked as "credible". For example, if a patient's vascular stenosis feature meets the requirements of "clarity ≥30dB, artifact rate <10%, segmentation accuracy ≥90%", it is marked as credible. If a certain dimension feature violates any rule, it is marked as "preliminarily unreliable". For example, a patient's heart rate variability feature is marked as "preliminarily unreliable" because "the effective data ratio is 85% < 90%".

[0050] Then, the feature evaluation unit performs secondary verification on the initially unreliable features: For features marked as initially unreliable, the reliability status of other features of the same patient is verified: Before the second validation, the weight allocation unit assigns initial weights (totaling 1) to the imaging, physiological, and metabolic features in the cross-dimensional feature scoring set based on historical clinical data and feature importance analysis.

[0051] For example, in the “Imaging + Physiology + Metabolism” scoring set, the initial weights are: imaging characteristics (vascular stenosis) 0.4, physiological characteristics (heart rate variability) 0.3, and metabolic characteristics (good cholesterol to bad cholesterol ratio) 0.3.

[0052] Scenario 1: A single feature does not conform to the rule, while the other two features do. If a certain dimension feature (such as metabolic feature) is judged as unreliable because "detection error 7% > preset threshold 5%", but image features (which meet the clarity and artifact rate rules) and physiological features (which meet the data integrity rules) are all reliable.

[0053] The weighting unit reduces the weight of this untrusted feature by 50%-70% of the initial weight, while simultaneously fine-tuning and increasing the weights of the other two trustworthy features, keeping the total weight of 1.

[0054] Example: The initial weight of metabolic features was 0.3, which was reduced to 0.1 (a decrease of 67%); the weight of imaging features was increased from 0.4 to 0.45, and the weight of physiological features was increased from 0.3 to 0.45. The total weight after adjustment is 1 (0.45+0.45+0.1=1).

[0055] Result: The feature evaluation unit marked the adjusted cross-dimensional feature scoring set as "trustworthy" and proceeded to the subsequent hierarchical diagnostic process.

[0056] Scenario 2: Two or more features do not conform to the rules If both imaging features (artifact rate 20% > 10%) and physiological features (valid data percentage 75% < 90%) are unreliable, only metabolic features are reliable.

[0057] The weight allocation unit does not adjust the weights because multi-dimensional features are unreliable and adjustments cannot compensate for systematic errors.

[0058] Conclusion: The cross-dimensional feature scoring set was determined to be "unreliable overall" due to anomalies in the multi-dimensional data, which may be caused by a problem in the data acquisition process. In this case, the diagnostic analysis module triggered the system to feed back to the data preprocessing module, which reprocessed the patient's data and generated a new cross-dimensional scoring set, which then re-entered the verification process.

[0059] Understandably, by reducing the weight of unreliable features, the interference of random errors on diagnosis can be reduced, while the dominant position of reliable features can be maintained, making the diagnostic results more consistent with the patient's actual condition and improving accuracy. By adjusting the weight of a single unreliable feature rather than eliminating it, potential diagnostic information can be avoided due to local data problems, enabling the system to adapt to the "local noise in data" scenario commonly seen in clinical practice.

[0060] Specifically, the hierarchical diagnostic unit is used to input the multi-dimensional features of the patient to be diagnosed into the multi-feature fusion diagnostic model, calculate the patient's probability of coronary heart disease based on the cross-dimensional feature score set marked as credible, and determine the patient's risk level of coronary heart disease based on the preset probability range.

[0061] Understandably, the cross-dimensional feature score set, after being adjusted and marked as "credible," is output to the hierarchical diagnostic unit as the core input data for calculating the probability and risk level of coronary heart disease.

[0062] Specifically, the process by which a stratified diagnostic unit determines a patient's risk level for coronary artery disease includes: If the patient's probability of coronary heart disease is less than the preset probability range, the stratified diagnostic unit will determine the patient's risk level of coronary heart disease as low risk. If the patient's probability of coronary heart disease is within the preset probability range, the stratified diagnostic unit will determine the patient's coronary heart disease risk level as medium risk. If a patient's probability of having coronary heart disease is greater than the preset probability range, the stratified diagnostic unit will determine the patient's risk level as high.

[0063] Specifically, the hierarchical diagnostic unit receives a weighted "credible cross-dimensional feature score set", for example, combination 4: vascular stenosis: weight 0.45, heart rate variability: weight 0.45, good cholesterol ratio: weight 0.1; The hierarchical diagnostic unit inputs the features of combination 4 into the multi-feature fusion diagnostic model to calculate the probability of coronary heart disease in the patient to be diagnosed.

[0064] If the probability of coronary heart disease is less than 20%, the stratified diagnostic unit will determine it as low risk and output a clinical recommendation of "having a routine physical examination once a year to monitor changes in blood pressure and blood lipids". If the probability of coronary heart disease is between 20% and 70%, the stratified diagnostic unit will determine it as medium risk and provide a clinical recommendation to "complete a coronary CTA and exercise stress test within 3 months to further clarify the diagnosis"; If the probability of coronary heart disease is >70%, the stratified diagnostic unit will determine it as high risk and output a clinical recommendation: "High risk of myocardial infarction, it is recommended to seek medical attention within 24 hours and evaluate the indications for coronary angiography and interventional treatment."

[0065] Understandably, the stratified diagnostic unit outputs the risk level and corresponding clinical recommendations to the results output module.

[0066] Reference Figure 1 In some embodiments of this application, the visualized diagnostic report includes: basic patient information, key indicators of multi-dimensional characteristics, probability of coronary heart disease, risk level, and clinical recommendations.

[0067] Specifically, the key indicators of multidimensional features are core diagnostic parameters extracted from the three dimensions of imaging, physiology, and metabolism, and retained after feature extraction, evaluation, and weight adjustment. These parameters form the basis for calculating the probability of coronary heart disease and include: Imaging dimensions: such as the degree of vascular stenosis, e.g., "40% stenosis of the left anterior descending artery"; plaque stability, e.g., moderately stable. Physiological dimensions: such as heart rate variability, such as "85ms, within the normal range"; blood pressure diurnal rhythm, such as a 12% drop in blood pressure at night, which is a normal rhythm; Metabolic dimensions: such as the ratio of good to bad cholesterol, such as 4.2, which is higher than the normal threshold of 3.5; homocysteine ​​concentration, such as "12 μmol / L, which is normal".

[0068] As can be seen, the pre-diagnostic system for coronary heart disease in the above embodiments, by integrating multi-source data such as coronary artery imaging, wearable dynamic physiology, and serum metabolites, overcomes the limitations of traditional single data. Combined with standardized preprocessing to eliminate format differences, it comprehensively captures disease pathological signals and avoids missed diagnoses caused by incomplete data. It refines general features into core clinical indicators and intelligently removes redundancy while retaining high-diagnostic-value features, providing accurate input for subsequent diagnosis and improving the correlation between features and the disease. A multi-feature fusion model is constructed based on the random forest algorithm, coupled with feature credibility verification and dynamic weight adjustment, reducing the interference of single-feature errors. By using probability stratification to clarify "low / medium / high" risk levels, the diagnosis is transformed from vague numerical values ​​into clinically actionable evidence, reducing misdiagnosis. By generating a visual report containing key indicators, risk levels, and personalized suggestions, it is simultaneously pushed to both medical and patient terminals, balancing doctors' rapid decision-making with patients' health awareness. A diagnostic database is established to support long-term disease tracking, assisting primary healthcare, achieving early warning, and improving overall diagnostic efficiency and health management levels.

[0069] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0070] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0071] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A pre-diagnostic system for coronary heart disease, characterized in that, include: The data acquisition module is used to acquire imaging data of the patient's coronary artery structure through coronary imaging equipment; it is also used to connect to the patient's wearable device to acquire dynamic physiological data of the patient's electrocardiogram and blood pressure in real time; and it is also used to interface with serum detection equipment to acquire detection data of metabolites related to coronary heart disease in the patient's serum. The data preprocessing module is used to perform standardized preprocessing on image data, dynamic physiological data, and metabolite detection data. The feature extraction module is used to extract multi-dimensional features from the preprocessed data, including vascular structure features, dynamic physiological data trend features, and metabolite concentration features. The diagnostic analysis module is used to construct a multi-feature fusion diagnostic model based on the random forest algorithm, using the multi-dimensional features as input variables and the clinical diagnosis results as labels, and to use the multi-feature fusion diagnostic model to prediagnose the probability and risk level of coronary heart disease in patients. The result output module is used to generate a visual diagnostic report from the diagnostic results of the diagnostic analysis module, establish a patient diagnostic database, and output the visual diagnostic report to the patient's smart terminal.

2. The pre-diagnostic system for coronary heart disease according to claim 1, characterized in that, The feature extraction module includes a feature refinement unit and a feature interaction unit; The feature refinement unit is used to refine the extraction of the multi-dimensional features: The feature refinement unit extracts the vascular structural features into vascular stenosis, plaque features, and vascular branching angles. The feature refinement unit processes the dynamic physiological data trend features into heart rate variability, blood pressure diurnal rhythm, and ECG ST segment trend. The feature refinement unit processes the metabolite concentration features into the ratio of good to bad cholesterol, triglyceride concentration, and homocysteine ​​concentration.

3. The pre-diagnostic system for coronary heart disease according to claim 2, characterized in that, The feature interaction unit is used to remove duplicates from the finely extracted multi-dimensional features according to a preset feature correlation threshold, and to construct a cross-dimensional feature scoring set of imaging, physiology and metabolism based on the deduplicated multi-dimensional features.

4. The pre-diagnostic system for coronary heart disease according to claim 3, characterized in that, The preset feature relevance thresholds include: a preset similar feature relevance threshold and a preset diagnostic relevance threshold. The deduplication and elimination process includes: the feature interaction unit determining whether the correlation value of two similar multi-dimensional features exceeds a preset similar feature correlation threshold. If the number does not exceed a certain limit, the feature interaction unit will retain two features simultaneously. If the correlation exceeds the threshold, the feature interaction unit will determine the correlation between the two similar multidimensional features and the diagnosis result of coronary heart disease: If the correlation between a certain multidimensional feature and the diagnosis result of coronary heart disease does not exceed a preset diagnostic correlation threshold, the feature interaction unit removes the feature. If the correlation between a certain multidimensional feature and the diagnosis result of coronary heart disease exceeds a preset diagnostic correlation threshold, the feature interaction unit will retain the feature.

5. A pre-diagnostic system for coronary heart disease according to claim 4, characterized in that, The deduplication and elimination process further includes: when judging the correlation between two similar multidimensional features and the diagnosis result of coronary heart disease: if the correlation between the two similar multidimensional features and the diagnosis result of coronary heart disease exceeds a preset diagnostic correlation threshold, then compare the correlation values ​​between the two similar multidimensional features and the diagnosis result of coronary heart disease, retain the feature with the larger correlation value, and remove the feature with the smaller correlation value.

6. A pre-diagnostic system for coronary heart disease according to claim 5, characterized in that, The diagnostic analysis module includes a feature evaluation unit, a weight allocation unit, and a hierarchical diagnostic unit; The feature evaluation unit is used to verify the credibility of the cross-dimensional feature scoring set: The feature evaluation unit pre-sets confidence evaluation rules for imaging, physiological and metabolic features; If a feature meets the rule, the feature evaluation unit marks the feature as trustworthy; If a feature does not conform to the rule, the feature evaluation unit will perform a secondary verification of the feature by combining the other features to see if they conform to the rule.

7. A pre-diagnostic system for coronary heart disease according to claim 6, characterized in that, The weight allocation unit is used to adjust the score weights of each dimension feature in the cross-dimensional feature score set; When the feature evaluation unit performs secondary verification: If a certain feature does not conform to the rule, but the other two features do conform to the rule, the feature evaluation unit will mark the cross-dimensional feature score set as reliable after the weight allocation unit reduces the score weight of the feature. If two or more features do not meet the rules, the feature evaluation unit will mark the cross-dimensional feature scoring set as unreliable and re-verify the new cross-dimensional feature scoring set.

8. A pre-diagnostic system for coronary heart disease according to claim 7, characterized in that, The hierarchical diagnostic unit is used to input the multi-dimensional features of the patient to be diagnosed into the multi-feature fusion diagnostic model, calculate the patient's coronary heart disease probability based on the cross-dimensional feature score set marked as reliable, and determine the patient's coronary heart disease risk level based on a preset probability range.

9. A pre-diagnostic system for coronary heart disease according to claim 8, characterized in that, The process by which the stratified diagnostic unit determines a patient's risk level for coronary heart disease includes: If the patient's probability of coronary heart disease is less than the preset probability range, the stratified diagnostic unit determines the patient's risk level of coronary heart disease as low risk. If the patient's probability of coronary heart disease is within a preset probability range, the stratified diagnostic unit determines the patient's coronary heart disease risk level as medium risk. If the patient's probability of coronary heart disease is greater than the preset probability range, the stratified diagnostic unit determines the patient's coronary heart disease risk level as high risk.

10. A pre-diagnostic system for coronary heart disease according to claim 9, characterized in that, The visualized diagnostic report includes: basic patient information, key indicators of multi-dimensional characteristics, probability of coronary heart disease, risk level, and clinical recommendations.

Images