Intelligent triage method and system for chest pain

Abstract

The invention discloses a chest pain intelligent triage method and system, and relates to the field of intelligent medical triage, and the method comprises the steps: collecting patient condition information and chest vibration signals, carrying out the standardization processing of the collected data, and extracting a multi-modal feature set; inputting the multi-modal feature set into a chest pain knowledge graph, mapping semantic feature vectors through a BERT semantic embedding model, and screening out graph sub-graphs; on the basis of the atlas subgraphs, disease cause possibility scores are calculated through nonlinear feature correlation analysis, correlation degree optimization is carried out by using a Bayesian network, and a disease cause sorting list and a reasoning path are obtained; a Q-learning reinforcement learning algorithm is used, the chest vibration signals are combined, weight adjustment is conducted on the disease cause sorting list, and a reasoning path is updated; and generating triage priorities and triage suggestions based on the adjusted cause sorting list and the reasoning path. According to the triage recommendation method and the triage recommendation system, the multi-modal data and the Q-learning reinforcement learning algorithm are combined, so that accurate triage priority ranking and personalized triage recommendation generation are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical triage, and particularly to a method and system for intelligent chest pain triage. Background Art

[0002] Chest pain is one of the common emergency symptoms in clinical practice, and its potential causes are numerous, including but not limited to heart diseases, lung diseases, as well as various causes in the digestive system and musculoskeletal system. Traditional chest pain triage methods mainly rely on doctors' clinical experience and routine physical examinations. However, due to the complex and somewhat similar causes of chest pain, traditional methods may not be able to quickly and accurately diagnose the cause in some cases. In recent years, with the rapid development of artificial intelligence technology, intelligent chest pain triage methods have emerged, especially intelligent triage systems that combine technologies such as machine learning, big data analysis, and knowledge graphs, gradually becoming important auxiliary tools in chest pain diagnosis and treatment.

[0003] However, the existing intelligent chest pain triage technologies still have certain deficiencies in many aspects. Firstly, existing intelligent triage systems mostly rely on a single data source, such as clinical examinations, medical history records, or basic physiological data, etc. However, these data often cannot comprehensively and accurately reflect the patient's symptoms and signs, resulting in limitations in the accuracy of diagnosis. Secondly, traditional chest pain triage systems based on rules or classical machine learning models often lack a deep understanding of the complex relationships between causes, leading to limitations in cause ranking and reasoning paths. Especially in complex situations where multiple causes coexist or alternate, the system may not be able to make reasonable judgments. In addition, although the acquisition and analysis of vibration signals have been applied in some fields, how to effectively integrate it into the chest pain triage process and optimize the reasoning process through intelligent algorithms is still an urgent problem to be solved. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for intelligent chest pain triage to solve the problems of insufficient data diversity and limitations in cause ranking of existing chest pain triage methods.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an intelligent triage method for chest pain, which includes collecting patient's condition information and chest vibration signals, performing standardized processing on the collected data, and extracting a multi-modal feature set; inputting the multi-modal feature set into a chest pain knowledge graph, mapping semantic feature vectors through a BERT semantic embedding model, and screening out a sub-graph of the graph; based on the sub-graph of the graph, calculating the probability score of the cause through non-linear feature correlation analysis, and using a Bayesian network for correlation optimization to obtain a list of ranked causes and an inference path; using a Q-learning reinforcement learning algorithm, combined with the chest vibration signal, adjusting the weights of the list of ranked causes and updating the inference path; based on the adjusted list of ranked causes and the inference path, generating a triage priority and triage recommendations.

[0008] As a preferred embodiment of the intelligent triage method for chest pain according to the present invention, wherein: the patient's condition information includes the patient's personal information, description of the main complaint symptoms, and core physical sign data.

[0009] As a preferred embodiment of the intelligent triage method for chest pain according to the present invention, wherein: the steps of collecting the patient's condition information and chest vibration signals, performing standardized processing on the collected data, and extracting a multi-modal feature set are specifically as follows:

[0010] Collect the patient's personal information through a self-service terminal and the patient's mobile device;

[0011] Obtain the patient's subjective feelings about the current chest pain through text input and doctor's inquiry, and summarize the description of the main complaint symptoms;

[0012] Collect the patient's core physical sign data in real time through medical devices;

[0013] Use a chest vibration detection device to non-invasively collect the patient's chest vibration signal;

[0014] Combine the patient's personal information, description of the main complaint symptoms, and core physical sign data into the patient's condition information, and send the patient's condition information and chest vibration signal to a data processing terminal through wireless transmission;

[0015] Perform preprocessing and data standardization on the patient's condition information and chest vibration signal;

[0016] Extract multi-modal features from the processed patient's condition information and chest vibration signal to generate a multi-modal feature set.

[0017] As a preferred embodiment of the intelligent triage method for chest pain according to the present invention, wherein: the steps of inputting the multi-modal feature set into a chest pain knowledge graph, mapping semantic feature vectors through a BERT semantic embedding model, and screening out a sub-graph of the graph are specifically as follows:

[0018] Construct a chest pain knowledge graph based on medical literature, historical cases, and clinical data;

[0019] Input the multi-modal feature set into the chest pain knowledge graph, and map it to a high-dimensional vector space through the BERT semantic embedding model to generate semantic feature vectors;

[0020] Calculate the cosine similarity between the semantic feature vectors and the nodes in the chest pain knowledge graph. The expression is:

[0021]

[0022] Among them, S il is the cosine similarity between the semantic feature vector V i and the node N l in the chest pain knowledge graph. V i is the i-th semantic feature vector, N l is the l-th node in the chest pain knowledge graph, ∥V i ∥ is the norm of the semantic feature vector, ∥N l ∥ is the norm of the node in the chest pain knowledge graph, i is the index coefficient of the number of semantic feature vectors, l is the index coefficient of the number of nodes in the chest pain knowledge graph, and N is the node in the chest pain knowledge graph;

[0023] Filter the subgraph nodes according to the cosine similarity, and retain the association paths between the subgraph nodes to generate a subgraph of the knowledge graph.

[0024] As a preferred solution of the chest pain intelligent triage method described in the present invention, wherein: based on the subgraph of the knowledge graph, calculate the cause possibility score through non-linear feature association analysis, and use the Bayesian network to optimize the association degree to obtain the cause ranking list and the inference path. The specific steps are as follows:

[0025] In the generated subgraph of the knowledge graph, analyze the multi-modal interaction relationship between the semantic feature vectors and the cause nodes, and calculate the cause possibility score. The expression is:

[0026]

[0027] Among them, P k is the cause possibility score, k is the index of the number of cause nodes, the integral interval [0,1] is the standardized time range, n is the total number of features in the multi-modal feature set, α i is the single feature weight coefficient, exp is the exponential function with the natural constant e as the base, ∥V i -N k ∥ is the Euclidean distance, N k is the k-th cause node in the subgraph of the knowledge graph, β is the multi-feature interaction weight coefficient, m is the number of features participating in the multi-feature interaction calculation, j is the index coefficient of m, V j is the j-th semantic feature vector participating in the multi-feature interaction calculation, and dt is the integral identifier;

[0028] Adjust the feature weights based on the Bayesian network model to optimize the association degree between the semantic feature vector and the cause node. The expression is:

[0029]

[0030] Among them, is the updated single feature weight coefficient, is the prior single feature weight coefficient, L(V i |N k ) is the likelihood value of the semantic feature vector V i under the condition of the cause node N k ;

[0031] Arrange the cause nodes in descending order according to the cause possibility score to generate a cause ranking list;

[0032] In the generated sub-graph of the atlas, trace the association path of the cause node with the highest score in the cause ranking list to generate an inference path.

[0033] As a preferred solution of the intelligent chest pain triage method described in the present invention, wherein: the Q-learning reinforcement learning algorithm is used to adjust the weights of the cause ranking list and update the inference path in combination with the chest vibration signal. The specific steps are as follows:

[0034] Extract the time-frequency features of the chest vibration signal through fast Fourier transform and wavelet transform to form a vibration signal feature vector;

[0035] Use the Q-learning reinforcement learning algorithm to adjust the weights of the cause nodes based on the vibration signal feature vector. The expression is:

[0036]

[0037] Among them, Q(s t ,a t ) is the Q value of the current state s t and the action a t . The state s t is the vibration signal feature vector, the action a t is the adjustment of the cause node weight, η is the learning rate, the reward r t is the accuracy evaluation of the inference result of the cause ranking list, γ is the discount factor, is the maximum Q value of all actions in the next state;

[0038] Based on the adjusted weights of the cause nodes, update the inference path. The expression is:

[0039]

[0040] Among them, P is the updated inference path, p is the number of cause nodes in the cause sorting list, and is the weight of the cause node updated through reinforcement learning.

[0041] As a preferred solution of the chest pain intelligent triage method described in the present invention, wherein: based on the adjusted cause sorting list and the inference path, generate triage priorities and triage suggestions, and the specific steps are as follows,

[0042] According to the adjusted cause sorting list, combined with the chest pain knowledge graph, calculate the priority score of each cause;

[0043] Generate a personalized triage suggestion matrix based on the type of cause node and the depth of the inference path;

[0044] Integrate the priority score and the personalized triage suggestion matrix to generate a triage report.

[0045] In a second aspect, the present invention provides a chest pain intelligent triage system, including a condition data collection module, a knowledge graph screening module, a cause sorting and reasoning module, a vibration signal adjustment module, and a triage suggestion generation module; the condition data collection module is used to collect patient condition information and chest vibration signals, perform standardization processing on the collected data, and extract a multi-modal feature set; the knowledge graph screening module is used to input the multi-modal feature set into the chest pain knowledge graph, map semantic feature vectors through the BERT semantic embedding model, and screen out a sub-graph of the graph; the cause sorting and reasoning module is used to calculate the cause possibility score based on the sub-graph of the graph through non-linear feature correlation analysis, and use a Bayesian network to optimize the correlation degree to obtain a cause sorting list and an inference path; the vibration signal adjustment module is used to use the Q-learning reinforcement learning algorithm, combined with the chest vibration signal, adjust the weight of the cause sorting list and update the inference path; the triage suggestion generation module is used to generate triage priorities and triage suggestions based on the adjusted cause sorting list and the inference path.

[0046] In a third aspect, the present invention provides a computer device, including a memory and a processor, where: when the computer program is executed by the processor, it implements any step of the chest pain intelligent triage method described in the first aspect of the present invention.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, where: when the computer program is executed by the processor, it implements any step of the chest pain intelligent triage method described in the first aspect of the present invention.

[0048] The beneficial effects of the present invention are as follows: By collecting the patient's condition information and chest vibration signals, and performing standardized processing and multi-modal feature extraction on the data, the patient's condition information can be comprehensively obtained, and a data basis with strong consistency and high comparability can be provided for subsequent analysis. Inputting these multi-modal features into the chest pain knowledge graph and mapping them to a high-dimensional semantic space through the BERT semantic embedding model can accurately screen out the sub-graphs of the graph most relevant to the patient's condition, thereby realizing personalized etiology reasoning. Through non-linear feature correlation analysis and Bayesian network optimization, the accuracy of the etiology possibility score is further improved, and the reasoning path is optimized. Combining with the Q-learning reinforcement learning algorithm can adjust the weights of the etiology ranking list in real time, continuously optimize the reasoning path, and enhance the adaptive ability. According to the adjusted etiology ranking and reasoning path, generate priority scores and personalized triage suggestions, providing scientific and efficient triage decision support for clinical practice, thereby improving triage efficiency, accuracy, and the utilization rate of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0050] Figure 1 It is a flowchart of the chest pain intelligent triage method in Embodiment 1.

[0051] Figure 2 It is a module diagram of the chest pain intelligent triage system in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings in the specification.

[0053] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0054] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it a separate or selectively exclusive embodiment from other embodiments.

[0055] Example 1, referring to Figure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides an intelligent triage method for chest pain, including the following steps:

[0056] S1: Collect the patient's condition information and chest vibration signals, perform standardization processing on the collected data, and extract a multi-modal feature set.

[0057] Specifically, it includes the following steps:

[0058] S1.1: Collect the patient's personal information through self-service terminals and the patient's mobile devices.

[0059] Specifically, the patient's personal information includes:

[0060] Demographic data such as name, gender, age, weight, height, etc.;

[0061] Previous medical history: such as hypertension, diabetes, coronary heart disease, chronic lung disease, etc.;

[0062] Family history: such as family history of aortic dissection, coronary artery disease, hyperlipidemia;

[0063] Living habits: smoking, drinking, exercise conditions, etc.

[0064] All of the above information is obtained with the user's consent and used for legal purposes.

[0065] S1.2: Obtain the patient's subjective feelings about the current chest pain through text input and doctor's inquiry, and summarize the description of the chief complaint symptoms.

[0066] Specifically, the description of the chief complaint symptoms includes:

[0067] Nature of chest pain: such as compressive sensation, stabbing pain, dull pain;

[0068] Location of chest pain: such as behind the sternum, left chest, right chest;

[0069] Duration of chest pain: such as intermittent or persistent;

[0070] Exacerbating or relieving factors: such as aggravated after activity, relieved after rest, etc.;

[0071] Whether accompanied by other symptoms: such as dyspnea, sweating, nausea, dizziness, left arm radiation pain, etc.

[0072] S1.3: Real-time collect the patient's core vital sign data through medical devices.

[0073] Specifically, the heart rate is recorded by an electrocardiogram monitoring device, the blood pressure is measured by an electronic sphygmomanometer, the blood oxygen saturation is collected by a fingertip pulse oximeter, the body temperature is measured by an infrared thermometer or a contact thermometer, and the respiratory rate of the patient is collected by a chest sensor or manual counting.

[0074] S1.4: Use a chest vibration detection device to non-invasively collect the chest vibration signal of the patient.

[0075] Specifically, it is collected through a chest heart sound sensor (such as an accelerometer or a microphone) to obtain the acoustic vibrations during heart valve closure and myocardial contraction; the airflow characteristics in the chest vibration are collected through a lung sound sensor to detect abnormalities such as dry rales, wet rales, and wheezing; the chest wall tremors are collected through a high-sensitivity vibration sensor (such as a skin surface accelerometer) to capture low-frequency tremor signals for diagnosing etiologies such as aortic dissection.

[0076] S1.5: Combine the patient's personal information, description of the chief complaint symptoms, and core vital sign data into the patient's condition information, and send the patient's condition information and chest vibration signal to the data processing terminal through wireless transmission.

[0077] It should be understood that the chest vibration signal is sent to the data processing terminal through wireless transmission (such as Bluetooth or Wi-Fi), and the data is stored in real time in the database of the hospital information platform or the remote medical platform for subsequent analysis.

[0078] S1.5.1: The patient's condition information includes the patient's personal information, description of the chief complaint symptoms, and core vital sign data.

[0079] S1.6: Preprocess and standardize the patient's condition information and chest vibration signal.

[0080] Specifically, preprocess the collected patient information and chest vibration signal data to remove invalid or noise data. Check the integrity (such as whether key fields are missing) and correctness (such as whether the age range is reasonable) of the patient information data; use a filtering algorithm to remove background noise and extract effective heart sound, lung sound, and tremor data.

[0081] Perform unified standardization processing on data from different sources to ensure that the feature values are within the same scale range. Convert the text description of the patient into structured data. For example, "chest oppression" corresponds to a classification number of 1, and "radiating pain in the left arm" corresponds to a classification number of 2; normalize numerical data such as heart rate and blood pressure to the 0-1 range for subsequent algorithm processing; perform normalization processing on amplitude and frequency, and at the same time extract the main characteristic peaks of the heart sound signal, the spectral distribution characteristics of the lung sound signal, and the low-frequency components of the chest wall tremor signal.

[0082] S1.7: Extract multimodal features from the processed patient condition information and chest vibration signals to generate a multimodal feature set.

[0083] Specifically, the multimodal feature set includes the following:

[0084] Symptom features: including the nature of chest pain, duration, accompanying symptoms, etc.;

[0085] Sign features: including physiological parameters such as heart rate, blood pressure, blood oxygen saturation, etc.;

[0086] Chest vibration signal features: including the amplitude, frequency, and power distribution features of heart sounds, lung sounds, and chest wall tremors;

[0087] Feature format: stored in the form of a structured table or vector as the input data for the next knowledge graph analysis.

[0088] Preferably, by integrating the patient's personal information, chief complaint symptoms, core sign data, and chest vibration signals, through standardized processing and multimodal feature extraction, a comprehensive, accurate, and efficient intelligent triage method for chest pain is provided. Through automated data collection and non-invasive chest vibration signal detection, it overcomes the limitation of traditional triage relying on single physiological indicators and can effectively improve the diagnostic accuracy, especially for the identification of complex diseases such as aortic dissection. In addition, the standardized processing of data ensures the high quality and consistency of information, enabling more accurate correlation analysis of different types of data, thus providing a solid foundation for the intelligent triage of chest pain and improving the efficiency and reliability in the triage process.

[0089] S2: Input the multimodal feature set into the chest pain knowledge graph, map semantic feature vectors through the BERT semantic embedding model, and filter out the sub-graphs of the graph.

[0090] Specifically, it includes the following steps:

[0091] S2.1: Construct a chest pain knowledge graph based on medical literature, historical cases, and clinical data.

[0092] Specifically, the chest pain knowledge graph is a multimodal and multi-level structured data system used to represent the association relationships between chest pain-related symptoms, signs, examination results, medical knowledge, causes, and treatment paths. Its nodes (entities) and edges (relationships) form a network that can support semantic reasoning, path analysis, and dynamic updates.

[0093] S2.1.1: Collect chest pain knowledge graph data.

[0094] Specifically, obtain papers, case analyses, and guidelines related to chest pain from authoritative medical databases (such as PubMed, UpToDate, Medline); obtain standardized medical term data from SNOMED-CT (Medical Terminology Set) for constructing a unified entity representation; obtain disease codes and classifications related to chest pain from ICD-10 / ICD-11 (International Classification of Diseases).

[0095] Extract the mechanisms and diagnosis and treatment paths of chest pain causes from clinical guidelines and medical literature.

[0096] Extract data related to symptoms, examinations, diagnoses, and treatments of chest pain patients from historical cases and clinical data; including examination data such as electrocardiogram, chest CT, echocardiogram, blood test results, etc., as well as the patient's past medical history.

[0097] Through communication with clinicians, cardiovascular specialists, or emergency department specialists, supplement medical experience and diagnosis and treatment logic that are difficult to directly extract from the data in the knowledge graph.

[0098] S2.1.2: Define and extract chest pain knowledge graph entities (nodes) from chest pain knowledge graph data.

[0099] Specifically, chest pain knowledge graph entities include:

[0100] Symptom entity: nature of chest pain (sense of pressure, stabbing pain, burning sensation), duration, location (behind the sternum, left chest), radiation site (left shoulder, jaw).

[0101] Sign entity: abnormal heart rate (tachycardia / bradycardia), abnormal blood pressure, respiratory rate, chest wall tremor.

[0102] Examination entity: electrocardiogram (ST-segment elevation, T-wave inversion), blood examination (troponin, D-dimer), imaging examination (chest CT, coronary angiography).

[0103] Etiology entity: acute coronary syndrome (ACS), aortic dissection, pulmonary embolism, pericarditis, gastroesophageal reflux disease (GERD).

[0104] Treatment entity: drugs (aspirin, clopidogrel), surgeries (PCI, coronary artery bypass grafting), supportive treatments (oxygen therapy, pain relief).

[0105] S2.1.3: Define and extract chest pain knowledge graph relationships (edges) from chest pain knowledge graph data.

[0106] Specifically, define the semantic relationship types between entities and extract specific relationships from the data:

[0107] Symptom - Etiology relationship: nature of chest pain → etiology (such as sense of pressure → acute coronary syndrome).

[0108] Symptom - Examination Relationship: Nature of chest pain → Recommended examination (e.g., chest pain accompanied by dyspnea → Lung CT).

[0109] Examination - Etiology Relationship: Examination result → Etiology (e.g., ST - segment elevation → Acute myocardial infarction).

[0110] Etiology - Treatment Relationship: Etiology → Treatment plan (e.g., Aortic dissection → Emergency surgery).

[0111] Etiology - Complications Relationship: Etiology → Complications (e.g., Acute coronary syndrome → Cardiogenic shock).

[0112] S2.1.3: Compose the extracted entities and relationships of the chest pain knowledge graph into a graph structure.

[0113] Specifically, use triples to represent the knowledge graph. For example:

[0114] <Chest pain accompanied by dyspnea, Recommended examination, Chest CT>

[0115] <ST - segment elevation, Etiology, Acute myocardial infarction>

[0116] <Acute coronary syndrome, Treatment plan, PCI>

[0117] S2.1.4: Define the inference rules of the knowledge graph.

[0118] Specifically, combine medical guidelines and expert knowledge to define inference rules. For example: If the symptom is chest pain accompanied by dyspnea and the examination result shows pulmonary embolism, then infer that the etiology is pulmonary embolism; If the blood test shows elevated troponin and ST - segment elevation on electrocardiogram, then infer that the etiology is acute myocardial infarction.

[0119] S2.2: Input the multi - modal feature set into the chest pain knowledge graph and map it to a high - dimensional vector space through the BERT semantic embedding model to generate semantic feature vectors.

[0120] It should be noted that the BERT model learns the context relationship between features through the self - attention mechanism, maps the multi - modal feature set to a high - dimensional vector space, and forms semantic feature vectors. The bidirectional encoding ability of BERT can capture the complex associations between features, ensuring that the context information of each input feature is fully considered, thus laying a foundation for subsequent semantic analysis.

[0121] S2.3: Calculate the cosine similarity between the semantic feature vectors and the nodes in the chest pain knowledge graph. The expression is:

[0122]

[0123] Among them, S ilFor the semantic feature vector V i and the cosine similarity between the chest pain knowledge graph node N l where V i is the i-th semantic feature vector, and N l is the l-th chest pain knowledge graph node, ∥V i ∥ is the modulus length of the semantic feature vector, ∥N l ∥ is the modulus length of the chest pain knowledge graph node, i is the index coefficient of the number of semantic feature vectors, l is the index coefficient of the number of chest pain knowledge graph nodes, and N is the chest pain knowledge graph node.

[0124] Specifically, the cosine similarity measures the angular difference between two vectors. The closer the value is to 1, the more similar the two vectors are, and the closer the value is to 0, the lower the similarity between them. Through this process, the similarity between each multimodal feature and the nodes of different etiologies, symptoms, pathological mechanisms, etc. in the knowledge graph can be quantified, providing a basis for subsequent screening of the most relevant nodes (i.e., subgraphs).

[0125] S2.4: Screen the subgraph nodes according to the cosine similarity, and retain the association paths between the subgraph nodes to generate a graph subgraph.

[0126] Specifically, sort the cosine similarities between all chest pain knowledge graph nodes and the semantic feature vectors, and select the nodes with higher similarities. By setting a similarity threshold, filter out the nodes with lower similarities, thereby reducing the interference of irrelevant information on subsequent analysis. Further retain the association paths between the screened nodes. These association paths describe the connections between different etiologies, symptoms, or physiological mechanisms. The retention of the association paths can provide important structured information for reasoning and etiological association analysis. Finally, the generated graph subgraph contains the nodes most relevant to the current patient's condition and the association relationships between them, and can provide an accurate semantic framework for subsequent non-linear feature association analysis and reasoning processes.

[0127] Preferably, by constructing a chest pain knowledge graph and combining it with a BERT model for semantic mapping, the patient's condition can be accurately analyzed and rapid diagnosis can be assisted. By extracting multi-level chest pain-related knowledge from medical literature, clinical data, and expert experience, the knowledge graph provides a complete information architecture for the relationships between symptoms, signs, etiologies, examinations, and treatments. The BERT model maps the multimodal feature data to a high-dimensional space through context semantic understanding, calculates the similarity with the knowledge graph nodes, and thus accurately screens out the nodes highly relevant to the patient's condition, optimizing the diagnostic path. The generated graph subgraph can provide accurate and personalized support for clinical decision-making, improve the efficiency and accuracy of diagnosis and treatment, and reduce misdiagnosis.

[0128] S3: Based on the sub-graph of the atlas, calculate the likelihood score of the cause through non-linear feature correlation analysis, and use the Bayesian network to optimize the correlation degree, so as to obtain the list of ranked causes and the inference path.

[0129] Specifically, it includes the following steps:

[0130] S3.1: In the generated sub-graph of the atlas, analyze the multi-modal interaction relationship between the semantic feature vector and the cause node, and calculate the likelihood score of the cause. The expression is:

[0131]

[0132] Among them, P k is the likelihood score of the cause, k is the index of the cause node number, the integration interval [0,1] is the standardized time range, n is the total number of features in the multi-modal feature set, α i is the weight coefficient of a single feature, exp is the exponential function with the natural constant e as the base, ∥V i -N k ∥ is the Euclidean distance, N k is the k-th cause node in the sub-graph of the atlas, β is the multi-feature interaction weight coefficient, m is the number of features participating in the multi-feature interaction calculation, j is the index coefficient of m, V j is the j-th semantic feature vector participating in the multi-feature interaction calculation, and dt is the integration identifier.

[0133] Preferably, for each cause node N k , first calculate the similarity between the multi-modal feature V i -N k ∥ and the cause node through the Euclidean distance, and obtain the contribution of each feature to the cause. At the same time, use the combined function of sine and cosine, sin(πV i ) and cos(πN j ) to simulate the interaction effect between different features. For example, the nature and location of chest pain may have a stronger non-linear relationship with certain causes (such as acute coronary syndrome), and these relationships are calculated through k weighted calculation. weighted calculation.

[0134] S3.2: Adjust the feature weights based on the Bayesian network model to optimize the correlation degree between the semantic feature vector and the cause node. The expression is:

[0135]

[0136] Among them, is the updated weight coefficient of a single feature, is the prior weight coefficient of a single feature, L(V i |N k) is the semantic feature vector V i at the cause node N k The likelihood value under the condition.

[0137] Preferably, a Bayesian network model is used to optimize the weights of each feature. The Bayesian network can combine prior knowledge and observed evidence to infer a more accurate relationship between the feature and the cause node. The updated weights reflect the change in the relative importance of the feature during the diagnosis process. Through this update, the association degree between the cause node and the feature is more accurate and reliable.

[0138] S3.3: Sort the cause nodes in descending order according to the cause possibility scores to generate a cause node sorted list.

[0139] Specifically, according to the cause possibility score P of each cause node k , all cause nodes are sorted from high to low according to the scores. A cause node with a higher score means it has a higher degree of match with the current patient's condition, so it is ranked at the top of the list. The sorted cause nodes generate a cause node sorted list according to the possibility scores, which serves as the basis for subsequent diagnosis and reasoning.

[0140] S3.4: In the generated sub-graph of the knowledge graph, trace the association path of the cause node with the highest score in the cause node sorted list to generate an inference path.

[0141] Specifically, according to the cause node sorted list, select the cause node with the highest score and trace its association path from the knowledge graph. These association paths represent the logical connections between the cause node and other related nodes (such as symptoms, signs, examination results, etc.). By tracing the association path of the cause node, potential causes and their possible treatment plans are gradually deduced. This path provides structured support for subsequent diagnostic decisions and guides clinicians to develop personalized treatment plans.

[0142] Preferably, by combining non-linear feature association analysis and Bayesian network optimization, the accuracy and personalization of chest pain cause diagnosis are effectively improved. By calculating the similarity between multi-modal features and cause nodes and combining non-linear interaction effects, complex cause association relationships are accurately captured; the Bayesian network is used to dynamically adjust feature weights and optimize the association degree between features and causes, thereby improving the reliability of diagnosis. The calculation of cause possibility scores and the generation of the cause node sorted list help identify the most relevant causes, and by tracing the inference path, decision support for personalized treatment plans is provided for clinicians.

[0143] S4: Use the Q-learning reinforcement learning algorithm, combined with chest vibration signals, to adjust the weights of the cause node sorted list and update the inference path.

[0144] Specifically, it includes the following steps:

[0145] S4.1: Extract the time-frequency features of the chest vibration signal through fast Fourier transform and wavelet transform to form a vibration signal feature vector.

[0146] Specifically, apply the fast Fourier transform to the chest vibration signal to convert the chest vibration signal from the time domain to the frequency domain, and obtain the spectral features of the chest vibration signal, including key features such as peak frequency, amplitude, and power spectral density. Use wavelet transform to analyze the time-frequency local characteristics of the chest vibration signal, especially the instantaneous changes in different frequency ranges. Extract the time-frequency features related to heart sounds, lung sounds, and chest wall tremors, such as instantaneous frequency and vibration amplitude within the time window.

[0147] Integrate the time-frequency features extracted by FFT and wavelet transform into a high-dimensional feature vector as the representation of the chest vibration signal. This high-dimensional feature vector contains information such as the spectral distribution, instantaneous frequency, and amplitude change of the chest vibration signal, and can reflect the key characteristics of the chest vibration signal.

[0148] S4.2: Use the Q-learning reinforcement learning algorithm to adjust the weights of the cause nodes based on the vibration signal feature vector. The expression is as follows:

[0149]

[0150] Among them, Q(s t ,a t ) is the Q value of the current state s t and action a t . The state s t is the vibration signal feature vector, and the action a t is the adjustment of the cause node weight. η is the learning rate, and the reward r t is the accuracy evaluation of the inference result of the cause sorting list. γ is the discount factor, is the maximum Q value of all actions in the next state.

[0151] Preferably, the reward is calculated by comparing the difference between the result of the current inference path and the actual cause. If the result of the inference path highly matches the actual cause, the reward is large; if the matching degree is low, the reward is small or negative. In each iteration, the Q value is adjusted according to the feedback reward information to gradually optimize the weights of the cause nodes.

[0152] S4.3: Update the inference path based on the adjusted weights of the cause nodes. The expression is as follows:

[0153]

[0154] Among them, P is the updated inference path, and p is the number of cause nodes in the cause ranking list. is the weight of the cause node updated through reinforcement learning.

[0155] It should be understood that the inference path describes the logical path for deriving potential causes from the current symptoms and signs through the relationships between cause nodes. The updated inference path reflects the relative importance of each cause node in the overall inference process, making the most relevant causes easier to be derived.

[0156] Preferably, by combining the time-frequency feature extraction of chest vibration signals and the Q-learning reinforcement learning algorithm, the weights of cause nodes and the inference path can be continuously optimized, improving the accuracy of cause ranking and the diagnostic efficiency. The time-frequency features extracted through fast Fourier transform and wavelet transform can finely capture the key pathological information in chest vibration signals. Q-learning adjusts the weights of cause nodes through a feedback mechanism and optimizes the inference path, thus providing more personalized and accurate cause analysis.

[0157] S5: Generate triage priorities and triage suggestions based on the adjusted cause ranking list and inference path.

[0158] Specifically, it includes the following steps:

[0159] S5.1: According to the adjusted cause ranking list, combined with the chest pain knowledge graph, calculate the priority score of each cause.

[0160] Specifically, according to the adjusted cause ranking list and the association weights of cause nodes in the inference path, calculate the priority score of each cause. The expression is:

[0161]

[0162] Among them, R k is the cause priority score, P k is the cause possibility score, E k is the medical severity of the cause, and the value range is [1, 5]. For example, aortic dissection is 5 and GERD is 1; (provided by the severity label of the cause in the knowledge graph, such as aortic dissection > acute coronary syndrome > gastroesophageal reflux disease), D k,z is the cause support intensity of the z-th signal feature in the chest vibration signal, and the value range is [0, 1]. Z is the total number of chest vibration signal features, C k is the influence weight of medical resource constraints on the cause, and the value range is [0, 1]. For example, when emergency resources are tense, the priority of non-critical causes will be reduced.

[0163] R kThe higher the value, the higher the urgency of the cause of the disease, and triage should be prioritized. If R k is negative, it means that this cause of the disease does not require immediate treatment or may be a misdiagnosis.

[0164] S5.2: Generate a personalized triage recommendation matrix based on the type of the cause node and the depth of the reasoning path.

[0165] Specifically, generate specific triage recommendations based on the type of the cause node and the depth of the reasoning path.

[0166] The type of the cause node refers to the category or nature of each cause node. According to different medical fields and pathological characteristics, the cause nodes can belong to different types, such as:

[0167] Acute causes (such as myocardial infarction, aortic dissection, etc.): These causes usually require emergency treatment and high-priority triage.

[0168] Chronic causes (such as chronic bronchitis, gastroesophageal reflux disease, etc.): These causes are not urgent, but still require timely diagnosis and treatment.

[0169] Functional diseases (such as anxiety disorder, mild depression, etc.): This type of cause does not directly threaten life, but still affects the patient's health and quality of life.

[0170] Non-critical causes: Such as some minor pathological symptoms, which may be related to other factors, but do not require emergency treatment.

[0171] The depth of the reasoning path refers to the length of the path from the current symptoms and signs to the potential cause of the disease through the reasoning process. A deeper path indicates that more cause nodes or more complex relationships are involved in the reasoning process, and more examinations or professional judgments may be required. The depth of the reasoning path can be understood in the following ways:

[0172] Shallow reasoning path: When the reasoning process directly points to one or a few causes of the disease, the reasoning path is shallow. This situation is related to some common causes of the disease and can usually be diagnosed through basic examinations.

[0173] Deep reasoning path: When the reasoning process involves multiple cause nodes or a complex network of relationships, the reasoning path is deep. At this time, more examinations and expert diagnoses are required to confirm the cause of the disease.

[0174] The depth of the reasoning path helps to determine the complexity of the patient's triage. For a deeper path, multiple examinations or referral to a specialist are required to ensure the accuracy of the diagnosis.

[0175] Triage recommendations include the following:

[0176] Recommended examinations: Based on the reasoning path of the cause nodes, extract the relevant examination nodes and generate specific examination suggestions (such as blood tests, electrocardiograms, chest CTs, etc.).

[0177] Referral departments: Triaging the patient to the appropriate department (such as the Department of Cardiology, Emergency Department, Department of Respiratory Medicine) based on the type of cause nodes.

[0178] Processing time limit: Give clear suggestions on the processing time according to the cause priority (such as immediate examination, follow-up visit within 24 hours).

[0179] Integrate the cause priority score and the reasoning path to generate a personalized triage recommendation matrix. Each row of the personalized triage recommendation matrix corresponds to a cause, and the columns represent examination, department, and time limit suggestions. By sorting each row of the personalized triage recommendation matrix in descending order, select the highest-scoring examination, department, and treatment suggestions.

[0180] S5.3: Integrate the priority score and the personalized triage recommendation matrix to generate a triage report.

[0181] Specifically, integrate the triage priority and triage suggestions to generate a triage report, including the following content:

[0182] Priority diagnosis cause: Display the urgent cause and its possibility score;

[0183] Recommended examinations: List the examination items arranged in priority (such as chest CT, electrocardiogram, D-dimer test);

[0184] Referral departments: Clearly recommend the referral departments (such as the Department of Cardiology, Emergency Department);

[0185] Processing time limit: Suggest the processing time (such as immediate treatment or follow-up visit within 48 hours).

[0186] Preferably, based on the adjusted list of cause rankings and the chest pain knowledge graph, quantify the urgency of the cause through the priority score formula, dynamically generate a personalized triage recommendation matrix in combination with the reasoning path, and integrate it into a triage report, effectively improving the accuracy and efficiency of triage. It not only ensures the priority treatment of high-risk causes and optimizes the allocation of medical resources, but also improves doctor-patient communication and patient experience through clear examination suggestions, referral departments, and processing time limits. At the same time, it provides support for clinical decision-making and training, and has certain clinical application value.

[0187] This embodiment also provides a chest pain intelligent triage system, including:

[0188] A disease condition data acquisition module, which is used to collect the patient's disease condition information and chest vibration signals, perform standardized processing on the collected data, and extract a multi-modal feature set;

[0189] A knowledge graph screening module, which is used to input a multi-modal feature set into a chest pain knowledge graph, map semantic feature vectors through a BERT semantic embedding model, and screen out a sub-graph of the graph;

[0190] An etiology ranking and reasoning module, which is used to calculate the etiology possibility score based on the sub-graph of the graph through non-linear feature correlation analysis, and use a Bayesian network for association degree optimization to obtain an etiology ranking list and an inference path;

[0191] A vibration signal adjustment module, which is used to use a Q-learning reinforcement learning algorithm, combined with chest vibration signals, to adjust the weights of the etiology ranking list and update the inference path;

[0192] A triage recommendation generation module, which is used to generate triage priorities and triage recommendations based on the adjusted etiology ranking list and inference path.

[0193] This embodiment also provides a computer device, which is applicable to the case of a chest pain intelligent triage method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the chest pain intelligent triage method proposed in the above embodiment.

[0194] The computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0195] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the chest pain intelligent triage method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0196] In summary, the present invention: collects the patient's condition information and chest vibration signals, and performs standardized processing and multi-modal feature extraction on the data, can comprehensively obtain the patient's condition information, and provides a data basis with strong consistency and high comparability for subsequent analysis. Inputting these multi-modal features into the chest pain knowledge graph and mapping them to a high-dimensional semantic space through the BERT semantic embedding model can accurately screen out the graph subgraphs most relevant to the patient's condition, thus realizing personalized etiology reasoning. Through non-linear feature correlation analysis and Bayesian network optimization, the accuracy of the etiology possibility score is further improved, and the reasoning path is optimized. Combining with the Q-learning reinforcement learning algorithm can adjust the weights of the etiology ranking list in real time, continuously optimize the reasoning path, and enhance the adaptive ability. According to the adjusted etiology ranking and reasoning path, generate a priority score and personalized triage suggestions, providing scientific and efficient triage decision support for clinical practice, thereby improving triage efficiency, accuracy and the utilization rate of medical resources.

[0197] Example 2, referring to Table 1, is the second embodiment of the present invention. To further verify the technical solution of the present invention, experimental simulation data of the chest pain intelligent triage method are given.

[0198] Thirty chest pain patients were selected as the research objects. Among them, 15 patients were diagnosed using the existing chest pain triage technology (control group), and the other 15 patients were diagnosed using the intelligent chest pain triage method of the present invention (experimental group). The test process includes collecting patient information, chest vibration signals, data standardization processing, multi-modal feature extraction, knowledge graph reasoning and etiology ranking, and generating triage suggestions according to the priority.

[0199] First, demographic data (such as name, gender, age, weight, etc.), past medical history, chief complaint symptoms (such as nature of chest pain, duration, accompanying symptoms, etc.), and core physiological signs (such as heart rate, blood pressure, blood oxygen saturation, respiratory rate, etc.) were collected from the patients in the experimental group and the control group respectively. In the experimental group, a chest vibration sensor was additionally used to collect chest vibration signals, including heart sound signals (amplitude, frequency), lung sound signals (spectrum distribution), and chest wall tremor signals (low-frequency tremor characteristics).

[0200] Next, the multimodal data of the experimental group were normalized. A filtering algorithm was used to denoise and extract signal features (such as peak frequency and power density of the signal). The text descriptions were converted into structured data, and the numerical data were normalized to the range of 0 - 1 to ensure the unity and efficient processing of the features.

[0201] Secondly, the multimodal features of the experimental group were input into the chest pain knowledge graph. Semantic feature vectors were generated through the BERT model, the similarity with the cause nodes was calculated, and subgraphs were screened out. Combining non-linear feature correlation analysis and Bayesian network optimization, a cause ranking list was generated, and the cause weights were adjusted through the Q-learning reinforcement learning algorithm to optimize the reasoning path.

[0202] Finally, triage suggestions were generated for the patients in the experimental group according to the cause priority scores, including recommended examinations (such as blood tests, chest CT, electrocardiogram, etc.), referral departments (such as cardiology department, emergency department), and processing time limits (such as immediate examination or follow-up visit within 24 hours). In the control group, traditional doctor experience combined with a single physiological indicator was used for triage.

[0203] Specifically, it is shown in Table 1 below:

[0204] Table 1 Comparison Record Table of Experimental Data

[0205]

[0206] By analyzing the three groups of experimental data, the technical advantages of the present invention can be clearly seen:

[0207] The average diagnostic accuracy rate of the experimental group was 96.9%, while that of the control group was 85.2%. The highest diagnostic accuracy rate in the experimental group was 97.2% (Experimental Group 3), while the highest in the control group was only 91.4% (Control Group 3). Through multimodal feature extraction and knowledge graph reasoning, the experimental group accurately identified the causes from multi-dimensional data, improving the diagnostic accuracy.

[0208] The average diagnostic time of the experimental group was 13 minutes, while that of the control group was 28.6 minutes. The fastest diagnostic time in the experimental group was 10 minutes (Experimental Group 1), while the fastest in the control group was 25 minutes (Control Group 3). The automated diagnostic process of the present invention effectively reduces manual intervention, improves the diagnostic efficiency, and wins precious treatment time for patients.

[0209] The average identification rate of high-risk etiologies in the experimental group was 96.6%, higher than 65.6% in the control group. The experimental group could accurately identify critical etiologies such as aortic dissection, with the lowest identification rate reaching 96.0% (experimental group 2), while the highest in the control group was only 61.0% (control group 2). This indicates that the present invention performs excellently in the diagnosis of critical etiologies.

[0210] Meanwhile, the experimental group also had advantages in patient satisfaction scores and missed diagnosis rates compared to the control group. The lowest patient satisfaction score in the experimental group reached 9.2 (experimental group 2), while in the control group it was only 7 (control group 1). The highest missed diagnosis rate in the experimental group was only 4% (experimental group 2), while in the control group it reached 20% (control group 1), reflecting the comprehensive advantages of the present invention in diagnostic efficiency, accuracy, and patient experience.

[0211] Through the comparative analysis of the three groups of data, the present invention shows advantages in key indicators such as diagnostic accuracy, diagnostic efficiency, high-risk etiology identification rate, patient satisfaction, and missed diagnosis rate, achieving more efficient, accurate, and safe intelligent triage for chest pain, and demonstrating certain clinical application value.

[0212] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent chest pain triage, characterized by: include, Collect patient condition information and chest vibration signals, standardize the collected data, and extract multimodal feature sets; The multimodal feature set is input into the chest pain knowledge graph, and the semantic feature vector is mapped through the BERT semantic embedding model to filter out the graph subgraph; Based on the graph subgraph, the cause possibility score is calculated through nonlinear feature association analysis, and the Bayesian network is used to optimize the association degree to obtain the cause ranking list and reasoning path; Using the Q-learning reinforcement learning algorithm, combined with chest vibration signals, the cause ranking list is weighted and the reasoning path is updated; Based on the adjusted etiology ranking list and reasoning pathway, triage priorities and triage recommendations were generated.

2. The intelligent chest pain triage method according to claim 1, characterized in that: The patient's condition information includes the patient's personal information, description of the main symptoms and core vital signs data.

3. The intelligent chest pain triage method according to claim 2, characterized in that: The specific steps of collecting patient condition information and chest vibration signals, standardizing the collected data, and extracting a multimodal feature set are as follows: Collect patient personal information through self-service terminals and patient-side mobile devices; Through text input and doctor's inquiry, the patient's subjective feeling of current chest pain is obtained and the description of the main complaint symptoms is summarized; Collect patients’ core vital signs data in real time through medical equipment; Use chest vibration detection equipment to non-invasively collect the patient's chest vibration signal; The patient's personal information, main complaint symptom description and core physical sign data are combined into the patient's condition information, and the patient's condition information and chest vibration signal are sent to the data processing terminal through wireless transmission; Preprocess and normalize the patient's condition information and chest vibration signals in the data processing terminal; Multimodal features are extracted from the processed patient condition information and chest vibration signals to generate a multimodal feature set.

4. The intelligent chest pain triage method according to claim 3, characterized in that: The multimodal feature set is input into the chest pain knowledge graph, the semantic feature vector is mapped through the BERT semantic embedding model, and the graph subgraph is screened out. The specific steps are: Build a chest pain knowledge graph based on medical literature, historical cases, and clinical data; The multimodal feature set is input into the chest pain knowledge graph and mapped to a high-dimensional vector space through the BERT semantic embedding model to generate a semantic feature vector. Calculate the cosine similarity between the semantic feature vector and the node in the chest pain knowledge graph. The expression is: Among them, S il is the semantic feature vector V i and chest pain knowledge graph node N l The cosine similarity between i is the i-th semantic feature vector, N l is the lth chest pain knowledge graph node, ||V i || is the modulus length of the semantic feature vector, ||N l || is the modulus length of the chest pain knowledge graph node, i is the index coefficient of the number of semantic feature vectors, l is the index coefficient of the number of chest pain knowledge graph nodes, and N is the chest pain knowledge graph node; Subgraph nodes are filtered according to cosine similarity, and the associated paths between subgraph nodes are retained to generate a graph subgraph.

5. The intelligent chest pain triage method according to claim 4, characterized in that: Based on the graph subgraph, the cause possibility score is calculated through nonlinear feature association analysis, and the Bayesian network is used to optimize the association degree to obtain the cause ranking list and reasoning path. The specific steps are: In the generated graph subgraph, the multimodal interaction relationship between the semantic feature vector and the cause node is analyzed, and the cause possibility score is calculated, which is expressed as: Among them, P k is the probability score of the cause, k is the index of the number of cause nodes, the integral interval [0,1] is the standardized time range, n is the total number of features in the multimodal feature set, α i is a single feature weight coefficient, exp is an exponential function with the natural constant e as the base, ||V i -N k || is the Euclidean distance, N k is the kth causal node in the graph subgraph, β is the multi-feature interaction weight coefficient, m is the number of features involved in the multi-feature interaction calculation, j is the index coefficient of m, V j is the jth semantic feature vector participating in the multi-feature interaction calculation, dt is the integral identifier; Based on the Bayesian network model, the feature weights are adjusted to optimize the correlation between the semantic feature vector and the cause node. The expression is: in, is the updated single feature weight coefficient, is the prior single feature weight coefficient, L(V i |N k ) is the semantic feature vector V i At the cause node N k Likelihood value under condition; Arrange the cause nodes in descending order according to the cause possibility scores to generate a cause ranking list; In the generated graph subgraph, the associated path of the highest-scoring causal node in the ranked list of causal factors is traced to generate an inference path.

6. The intelligent chest pain triage method according to claim 5, characterized in that: The Q-learning reinforcement learning algorithm is used to adjust the weight of the cause ranking list and update the reasoning path in combination with the chest vibration signal. The specific steps are as follows: Through fast Fourier transform and wavelet transform, the time-frequency characteristics of the chest vibration signal are extracted to form a vibration signal feature vector; Using the Q-learning reinforcement learning algorithm, the weight of the cause node is adjusted based on the vibration signal feature vector. The expression is: Among them, Q(s t ,a t ) is the current state s t and action a t Q value, state s t is the vibration signal feature vector, action a t is the weight adjustment of the cause node, η is the learning rate, and the reward r t is the accuracy evaluation of the cause ranking list inference result, γ is the discount factor, is the maximum Q value of all actions in the next state; Based on the adjusted causal node weights, the reasoning path is updated, and the expression is: Among them, P is the updated reasoning path, p is the number of cause nodes in the cause sorting list, is the weight of the causal node after being updated through reinforcement learning.

7. The intelligent chest pain triage method according to claim 6, characterized in that: The steps of generating triage priorities and triage suggestions based on the adjusted cause ranking list and reasoning path are as follows: Based on the adjusted ranked list of causes and combined with the chest pain knowledge graph, the priority score of each cause is calculated; Generate a personalized triage recommendation matrix based on the type of etiology node and the depth of the reasoning path; The priority score and personalized triage recommendation matrix are integrated to generate triage reports.

8. An intelligent chest pain triage system, based on the intelligent chest pain triage method according to any one of claims 1 to 7, characterized in that: It includes a disease data collection module, a knowledge graph screening module, a cause ranking reasoning module, a vibration signal adjustment module, and a triage suggestion generation module; The condition data collection module is used to collect patient condition information and chest vibration signals, perform standardization on the collected data, and extract a multimodal feature set; The knowledge graph screening module is used to input the multimodal feature set into the chest pain knowledge graph, map the semantic feature vector through the BERT semantic embedding model, and screen out the graph subgraph; The etiology ranking reasoning module is used to calculate the etiology possibility score based on the graph subgraph through nonlinear feature association analysis, and use the Bayesian network to optimize the association degree to obtain the etiology ranking list and reasoning path; The vibration signal adjustment module is used to use the Q-learning reinforcement learning algorithm, combined with the chest vibration signal, to adjust the weight of the cause ranking list and update the reasoning path; The triage suggestion generating module is used to generate triage priorities and triage suggestions based on the adjusted cause ranking list and reasoning path.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the chest pain intelligent triage method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the chest pain intelligent triage method according to any one of claims 1 to 7 are implemented.

Images