Auxiliary triage method based on multi-dimensional symptom correlation analysis and reception feedback correction

Abstract

This invention relates to the field of auxiliary triage technology and discloses an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction. This invention aims to solve the problems of inaccurate assessment of multi-dimensional concurrent symptom joint diagnosis mapping in existing auxiliary triage methods and the lack of technical solutions for dynamic weight updates based on real patient feedback. This invention preprocesses and standardizes multi-source symptoms and signs to construct a disease association network based on conditional probability and joint conditional probability; it extracts local relationship networks and compares extreme value weights to generate triage slips; then, it obtains real confirmed diagnosis records after patient admission to construct standard samples, and locally recalculates the weights of corresponding directed connections. This invention achieves objective quantification of the diagnostic orientation of concurrent features and establishes a dynamic update mechanism for the disease association network.

Description

Technical Field

[0001] This invention relates to the field of auxiliary triage technology, specifically, to an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction. Background Technology

[0002] Assisted triage refers to the crucial medical service step in which, at the initial stage of a patient's visit, medical information technology is used to conduct a preliminary assessment of the patient's subjective symptoms and objective signs, thereby guiding them to the appropriate department. Currently, most existing technologies in this field rely on pre-set static rule dictionaries or simple keyword matching mechanisms. Conventional assisted triage methods generally collect the patient's single chief complaint or basic physiological indicators and perform point-to-point mapping and matching with the corresponding routine diseases in each department to achieve preliminary automatic or semi-automatic triage.

[0003] However, in real-world clinical settings, patients often present with multiple complex concurrent symptoms, and the course of their illness is frequently influenced by individual physical signs and the spatial and temporal environment of the visit. Existing triage technologies, when dealing with such multi-dimensional concurrent features, often rely on fixed conditional combination rules or simple linear superposition, making it difficult to accurately quantify the overall correlation depth and true diagnostic direction when complex symptom clusters coexist. Furthermore, these triage schemes generally lack a closed-loop feedback mechanism based on objective diagnostic data, failing to dynamically update static triage rules according to the actual final diagnosis after admission. This results in existing technologies consistently exhibiting rigid triage decisions and limited accuracy in matching departments for multi-dimensional concurrent symptoms when facing complex and ever-changing real-world cases. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, this invention provides an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction, aiming to solve the technical problems of existing auxiliary triage methods being inaccurate in assessing the weight of multi-dimensional complication symptoms and lacking the technical ability to dynamically update weights based on real patient feedback.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction, characterized in that it includes: Step S1, Data Collection: Obtain the patient's chief symptoms and physiological values, as well as the time and space information of the visit; Step S2, Data Preprocessing: Clean the chief complaint symptoms to obtain chief complaint symptom text, and convert the physiological values ​​into physiological symptom text; combine the chief complaint symptom text, the physiological symptom text, and the spatiotemporal information of the visit to generate a preprocessed dataset; Step S3, Network Construction: Based on standardized historical medical data, establish symptom nodes, spatiotemporal nodes, and disease nodes, and configure corresponding department labels for the disease nodes; establish directed connections from the symptom nodes, the spatiotemporal nodes to the disease nodes, calculate the initial weights of the directed connections according to the mapping ratio between nodes, and calculate the co-occurrence gain weight when multiple nodes point to the same disease node, so as to generate a disease association network. Step S4, Symptom Feature Standardization: Replace the non-standard symptom descriptions contained in the preprocessed dataset with standard symptom terms to generate standardized symptom feature information.

[0007] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the method further includes the following steps: Step S5, Local Atlas Construction: In the disease association network, lock the symptom nodes and spatiotemporal nodes of the standardized symptom feature information, and spread outward along the directed connection with the locked nodes as the center to generate a local relationship network. Step S6, Generate a candidate department set: Traverse the initial weights and co-occurrence gain weights pointing to the same disease node in the local relation network, select the highest weight as the disease's adaptation score, match the corresponding department, and set them into a candidate department set; Step S7, Triage Reference Output: Output the candidate department set from high to low according to the adaptation score to the triage terminal, and generate a triage slip according to the final triage instruction returned by the triage terminal; Step S8, Patient Feedback and Weight Adjustment: Obtain actual confirmed records, extract the real confirmed diseases from the actual confirmed records, and associate them with the corresponding standardized symptom feature information and spatiotemporal information of the visit to transform them into standard sample data, and add them to the standardized historical medical data to adjust the corresponding initial weights and co-occurrence gain weights in the disease association network.

[0008] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of obtaining the patient's chief complaint symptoms, physiological values, and spatiotemporal information of the visit includes: receiving a natural language speech stream or text stream input from the front-end device of the triage desk, and extracting the chief complaint symptoms; reading the patient's body temperature, blood pressure, and heart rate data as the physiological values ​​by calling the application programming interface of the in-hospital vital sign measurement device; extracting the current network time protocol timestamp to obtain the current visit time, and receiving the patient's recent travel history actively entered by the patient on the front-end device of the triage desk to record the patient's recent main activity areas, and combining the current visit time and the recent main activity areas into the spatiotemporal information of the visit.

[0009] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of generating the preprocessing dataset includes: segmenting the chief complaint into independent lexical units and removing irrelevant words, extracting keywords to form the chief complaint text; then comparing the physiological values ​​with a preset medical health threshold range, and mapping the physiological values ​​to physiological symptom text representing abnormal states according to the degree of deviation from the medical health threshold range; and structurally combining the chief complaint text, the physiological symptom text, and the spatiotemporal information of the patient visit as features of different dimensions to generate a multi-dimensional feature vector as the preprocessing dataset.

[0010] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of calculating weights to generate a disease association network includes: statistically analyzing the frequency of each symptom node and spatiotemporal node appearing independently and ultimately diagnosed as a certain disease node in the standardized historical medical data; calculating the conditional probability of a single node pointing to the corresponding disease node; using the conditional probability as the initial weight of the directed connection; mining concurrent multi-node combinations in the historical medical data; statistically analyzing the co-occurrence frequency of the multi-node combinations pointing to the same disease node; calculating the joint conditional probability of the multi-node combinations pointing to the same disease node based on the co-occurrence frequency; and using the joint conditional probability as the co-occurrence gain weight.

[0011] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of generating standardized symptom feature information includes: acquiring standard medical terms as standard items, and collecting colloquial expressions with semantically consistent meanings with the standard items as non-standard items; establishing a many-to-one unidirectional mapping relationship between the non-standard items and the standard items to form a thesaurus; statistically analyzing the co-occurrence frequency and contextual distribution characteristics of words in the medical corpus, and constructing a feature mapping matrix of the correspondence between words and multi-dimensional spatial coordinates; extracting each of the standard items and non-standard items in the thesaurus into a multi-dimensional numerical array containing semantic information using the feature mapping matrix, as a reference text vector; using the feature mapping matrix to extract semantic features and perform multi-dimensional spatial mapping on the non-standard symptom expressions in the preprocessed dataset, converting them into text vectors; calculating the cosine value of the text vector and each of the reference text vectors as a semantic correlation score; selecting the standard medical terms corresponding to the terms with the highest semantic correlation scores that exceed a set matching threshold, replacing the non-standard symptom expressions, and generating the standardized symptom feature information.

[0012] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of extracting and generating a local relationship network includes: matching symptom nodes and spatiotemporal nodes consistent with the standardized symptom feature information in the disease association network, and locking the matching symptom nodes and spatiotemporal nodes; starting from the locked symptom nodes and spatiotemporal nodes, expanding outward along the directed connection to the connected disease nodes, and obtaining the initial weight corresponding to the directed connection and the co-occurrence gain weight when multiple nodes are combined, to form a local relationship network; if the initial weight and co-occurrence gain weight of the locked node pointing to a certain disease node are both zero, then the current branch path is determined to be invalid, and the corresponding directed connection and the disease node are removed from the local relationship network.

[0013] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of generating the candidate department set includes: for any disease node in the local relationship network, obtaining all initial weights and co-occurrence gain weights pointing to the disease node; comparing the numerical values ​​of the obtained initial weights of a single node and the co-occurrence gain weights of multiple nodes, and extracting the one with the highest value as the fit score of the disease node; extracting the department tags configured for the disease node to match the corresponding department category, and sorting them in descending order according to the fit score from high to low, with the disease node as the main axis, to generate the candidate department set.

[0014] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of generating a triage slip includes: outputting the candidate department set to the display interface of the triage terminal in the form of a visual list, and intuitively presenting the disease node, the corresponding matching score, and the department in the visual list; receiving the department confirmation instruction returned by the triage terminal, parsing the department confirmation instruction to obtain the finally selected target department, and using the confirmation result of the target department as the final triage instruction; extracting the target department information from the final triage instruction, and combining it with the obtained patient queue sequence number for data encapsulation to generate a triage slip.

[0015] As a preferred embodiment of the auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction described in this invention, the specific process of adjusting the weights includes: receiving the actual diagnosis record returned by the hospital information system after the patient's visit, and extracting the real diagnosed disease from the actual diagnosis record; structurally binding the real diagnosed disease with the standardized symptom feature information and the spatiotemporal information of the patient at the beginning of the visit, and converting it into standard sample data; appending the standard sample data to the historical medical data; and recalculating the conditional probability and joint conditional probability between corresponding nodes based on the updated historical medical data, so as to dynamically update the initial weight and co-occurrence gain weight of the corresponding directed connection.

[0016] The beneficial effects of this invention are as follows: It preprocesses and standardizes multi-source symptoms and signs, constructs a disease association network based on conditional probability and joint conditional probability; it extracts local relationship networks and compares extreme value weights to generate triage slips; then, it obtains actual confirmed diagnosis records after admission to construct standard samples, and locally recalculates the weights of corresponding directed connections. This invention achieves objective quantification of the diagnostic orientation of concurrent features and establishes a dynamic update mechanism for the disease association network. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the overall process for an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction.

[0019] Figure 2 A flowchart for data preprocessing and standardization.

[0020] Figure 3Flowchart for constructing and calculating weights for disease association networks.

[0021] Figure 4 This is a flowchart for patient feedback and partial network correction. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0025] Example 1 Reference Figures 1-4 This is one embodiment of the present invention, which provides an auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction, specifically including the following steps: Step S1, Data Collection: Obtain the patient's chief symptoms and physiological values, as well as the time and space information of the visit.

[0026] The specific process of obtaining the patient's chief complaint symptoms, physiological values, and spatiotemporal information of the consultation includes: receiving the natural language speech or text stream input from the front-end device of the triage desk and extracting it to form the chief complaint symptoms; reading the patient's body temperature, blood pressure, and heart rate data as physiological values ​​by calling the application programming interface of the in-hospital vital sign measurement device; extracting the current network time protocol timestamp to obtain the current consultation time; and receiving the patient's recent travel history actively entered by the patient at the front-end device of the triage desk to record the patient's recent main activity areas, and combining the current consultation time with the recent main activity areas to form the spatiotemporal information of the consultation.

[0027] Step S2, data preprocessing: clean the chief complaint symptoms to obtain chief complaint symptom text, and convert the physiological values ​​into physiological symptom text; combine the chief complaint symptom text, the physiological symptom text, and the spatiotemporal information of the visit to generate a preprocessed dataset.

[0028] The specific process of generating the preprocessing dataset includes: segmenting the chief complaint into independent lexical units and removing irrelevant words, extracting keywords to form the chief complaint text; then comparing the physiological values ​​with a preset medical health threshold range, and mapping the physiological values ​​to physiological symptom text representing the abnormal state according to the degree of deviation from the medical health threshold range; and then structurally combining the chief complaint text, physiological symptom text, and spatiotemporal information of the consultation as features of different dimensions to generate a multidimensional feature vector as the preprocessing dataset.

[0029] In the front-end interaction stage, the system first receives the patient's spoken natural language audio stream or input text stream in real time through the triage desk. For the audio stream, a speech recognition engine is used to convert it into text. Since the patient's original expression usually contains a lot of colloquial redundancy, a comprehensive basic dictionary containing medical terminology and everyday language is needed. After obtaining the original text, the obtained text string is scanned and truncated from left to right according to the set character window length. The truncated string fragments are compared one by one with the entries in the comprehensive basic dictionary. If they match, they are broken and extracted into independent word units. This process is repeated until the continuous whole sentence is completely divided into multiple independent word units. In order to remove interference information that has no actual medical diagnostic significance, a stop word list is compiled in advance based on historical consultation records and a general corpus. After word segmentation, each independent word unit is compared with the stop word list to effectively filter out subjective modifiers and modifiers such as "I", "I feel", "probably", and "a little", thereby accurately extracting core keywords with practical diagnostic value such as "headache", "nausea", and "chest tightness", and finally forming a clean and structured text of the chief complaint symptoms.

[0030] In terms of objective physiological data collection, the system directly connects to and calls the application programming interfaces (APIs) of in-hospital self-service vital sign measurement devices (such as electronic blood pressure monitors, infrared thermometers, etc.) through standard medical data transmission protocols (such as the Health Layer 7 protocol or rapid medical interoperability resources) to read the patient's current body temperature, systolic and diastolic blood pressure, heart rate and other physiological values ​​in real time.

[0031] To enable subsequent atlas calculations for numerical and textual data within a unified feature space, a text-based mapping of physiological values ​​is required. This mapping process first establishes upper and lower limits for the normal values ​​of various physiological indicators based on commonly used clinical diagnostic standards and a vital sign reference database, thus constructing a preset medical health threshold range. After reading the physiological values, the values ​​are compared with the corresponding medical health threshold ranges. When a physiological value exceeds the medical health threshold range, the deviation from the normal threshold is calculated, and according to a preset step-matrix mapping rule, the physiological value is mapped to a physiological symptom representing an abnormal state. For example, if the medical health threshold range for adult axillary temperature is set to 36.0℃ to 37.2℃, and the measured temperature is 39.5℃, indicating a significant deviation from this range, it is mapped to the physiological symptom text "high fever" according to the step-matrix mapping rule. Similarly, blood pressure values ​​exceeding the upper threshold are mapped to physiological symptom texts such as "high blood pressure."

[0032] Furthermore, by extracting the Network Time Protocol (NTP) timestamp of the current node, the time period of the patient's visit (e.g., springtime, late at night) is identified. Combined with the patient's recent travel history information entered on front-end devices such as registration machines, the patient's recent main activity areas are recorded. The identified time period and recorded activity areas are then combined to form the patient's spatiotemporal information. Finally, the cleaned text of the chief complaint, the mapped text of physiological symptoms, and the extracted spatiotemporal information are concatenated and combined according to a fixed structural dimension to generate a multidimensional feature vector, which is then used as a preprocessed dataset input to the downstream module.

[0033] Step S3, Network Construction: Based on standardized historical medical data, establish symptom nodes, spatiotemporal nodes, and disease nodes, and configure corresponding department labels for the disease nodes; establish directed connections from the symptom nodes, the spatiotemporal nodes to the disease nodes, calculate the initial weights of the directed connections according to the mapping ratio between nodes, and calculate the co-occurrence gain weight when multiple nodes point to the same disease node, so as to generate a disease association network.

[0034] The specific process of calculating weights to generate a disease association network includes: statistically analyzing the frequency of each symptom node and spatiotemporal node appearing independently and ultimately diagnosed as a certain disease node in standardized historical medical data; calculating the conditional probability of a single node pointing to the corresponding disease node; using the conditional probability as the initial weight of the directed connection; mining concurrent multi-node combinations in historical medical data; statistically analyzing the co-occurrence frequency of multi-node combinations pointing to the same disease node; and calculating the joint conditional probability of multi-node combinations pointing to the same disease node based on the co-occurrence frequency, using the joint conditional probability as the co-occurrence gain weight.

[0035] Step S4, Symptom Feature Standardization: Replace the non-standard symptom descriptions contained in the preprocessed dataset with standard symptom terms to generate standardized symptom feature information.

[0036] The specific process of generating standardized symptom feature information includes: acquiring standard medical terms as standard items and collecting colloquial expressions with semantically consistent meanings as non-standard items; establishing a many-to-one unidirectional mapping relationship between non-standard items and standard items to construct a thesaurus; statistically analyzing the co-occurrence frequency and contextual distribution features of words in the medical corpus to construct a feature mapping matrix corresponding to the correspondence between words and multidimensional spatial coordinates; extracting each standard and non-standard item in the thesaurus into a multidimensional numerical array containing semantic information using the feature mapping matrix, as a baseline text vector; using the feature mapping matrix to extract semantic features and perform multidimensional spatial mapping on the non-standard symptom expressions in the preprocessed dataset, transforming them into text vectors; calculating the cosine value of the text vector and each baseline text vector as the semantic association score; selecting the standard medical terms corresponding to the terms with the highest semantic association scores that exceed a set matching threshold, replacing the non-standard symptom expressions, and generating standardized symptom feature information.

[0037] To construct the disease association network, it is necessary to read and parse standardized historical medical data (such as electronic medical records); then, extract independent symptoms from the historical medical data as symptom nodes, extract the time and geographical region of the visit as spatiotemporal nodes, and extract the final clinical diagnosis as disease nodes. Simultaneously, obtain the standard list of medical specialties published by the national health administration department or the hospital's current departmental structure specifications, using this as a departmental classification directory, and assign a departmental label to each disease node based on this directory (e.g., assign the label "general surgery" to the "acute appendicitis" node).

[0038] After extracting the nodes, directed connections are established from the aforementioned symptom nodes and spatiotemporal nodes to the corresponding disease nodes. To quantify the indicative strength of a single feature for a disease, the frequency of a single symptom node (e.g., "right lower abdominal pain") or a single spatiotemporal node appearing independently in standardized historical medical data and ultimately diagnosed as a corresponding disease node is statistically analyzed. Based on the ratio of this frequency to the total frequency of the single node in the entire dataset, the conditional probability of a single node pointing to a corresponding disease node is calculated, and this conditional probability is used as the initial weight of the corresponding directed connection. The initial weight calculation formula is as follows. for:

[0039] in, Represents a single node Pointing to a specific disease node The initial weights assigned to the directed connections; Indicates a single node When the disease occurs independently, it is ultimately diagnosed. The conditional probability; This indicates that within the complete set of standardized historical medical data, a single node... The total frequency of independent occurrences is used as the basic denominator for calculation; This indicates that within the same historical medical data, a single node With specific disease nodes The frequency of co-occurrence is used as the numerator in the calculation.

[0040] The calculated conditional probabilities objectively quantify the unidirectional indicative significance of a single symptom or spatiotemporal background for a specific disease.

[0041] Considering that patients often exhibit multidimensional symptoms during actual medical visits, it is necessary to further mine for co-occurring multi-node combinations in historical medical data (e.g., a combination of "fever," "right lower abdominal pain," and "elevated white blood cell count"). Subsequently, the co-occurrence frequency of such multi-node combinations pointing to the same specific disease node is statistically analyzed. Based on this co-occurrence frequency and the total frequency of the combined feature, the joint conditional probability of the multi-node combination ultimately being diagnosed as the aforementioned specific disease is calculated. The obtained joint conditional probability is directly used as the co-occurrence gain weight when multiple nodes jointly point to the same disease node. Its co-occurrence gain weight... Calculation formula:

[0042] in, Represents a feature set composed of multiple nodes. They all point to a specific disease node. The co-occurrence gain weight generated at that time; This represents a combined feature set consisting of multiple concurrent feature nodes, i.e. ; Indicates the combination of feature sets Under the premise of complete concurrency, the final diagnosis was the disease. The joint conditional probability; This represents the combined feature set within the entire set of standardized historical medical data. The total frequency of concurrent occurrences as a whole is used as the basic denominator in the calculation. This represents the combined feature set. With specific disease nodes The total frequency of co-diagnosed cases is used as the numerator in the calculation.

[0043] By calculating and quantifying the additional departmental mapping correlation generated when a specific symptom cluster is accurately captured, which far exceeds the simple addition of single nodes, a higher-order weighting basis is provided for subsequent calculation of the optimal fit score.

[0044] To eliminate semantic ambiguity caused by patients' colloquial expressions and the physiological symptom text generated by mapping in the pre-processing stage, standard medical terms from medical literature and the International Classification of Diseases (ICD) are obtained as standard items, and everyday colloquial expressions that are semantically consistent with the standard items are collected as non-standard items. A many-to-one unidirectional mapping association is established from multiple non-standard items to a single standard item, thereby constructing a thesaurus.

[0045] To achieve deep semantic understanding, it is necessary to statistically analyze the co-occurrence frequency of each word in the medical corpus and examine the contextual distribution characteristics of the target word within its surrounding words of a fixed span. Based on these distribution characteristics, a feature mapping matrix is ​​constructed to correspond to the words in a multidimensional spatial coordinate system. Using this feature mapping matrix, the standard and non-standard items in the thesaurus are projected into a multidimensional vector space, extracting them as a multidimensional numerical array containing continuous semantic information, which serves as the baseline text vector.

[0046] Finally, the preprocessed dataset from the data preprocessing module is received. Using the same feature mapping matrix, feature calculation and multidimensional spatial mapping are performed on the non-standard symptom descriptions in the preprocessed dataset, transforming them into corresponding text vectors. Subsequently, using spatial vector arithmetic rules, the cosine value of the angle between this text vector and each baseline text vector in the thesaurus is calculated in multidimensional space. The calculated cosine value is used as the semantic association score, representing semantic similarity. After comparing all semantic association scores, the term with the highest semantic association score, exceeding a set matching threshold, is selected. The standard medical term corresponding to this term is extracted and used to replace the non-standard symptom descriptions in the preprocessed dataset, ultimately generating standardized symptom feature information that can accurately match nodes in the disease association network.

[0047] Step S5, Local Atlas Construction: In the disease association network, lock the symptom nodes and spatiotemporal nodes of the standardized symptom feature information, and spread outward along the directed connection with the locked nodes as the center to generate a local relationship network.

[0048] The specific process of extracting and generating a local relational network includes: matching symptom nodes and spatiotemporal nodes that are consistent with standardized symptom feature information in the disease association network, and locking the matching symptom nodes and spatiotemporal nodes; starting from the locked symptom nodes and spatiotemporal nodes, expanding outward along directed lines to connected disease nodes, and obtaining the initial weights corresponding to the directed lines and the co-occurrence gain weights when multiple nodes are combined, so as to form a local relational network; if the initial weights and co-occurrence gain weights of the locked node pointing to a certain disease node are both zero, then the current branch path is determined to be invalid, and the corresponding directed lines and disease nodes are removed from the local relational network.

[0049] Step S6: Generate a candidate department set: Traverse the initial weights and co-occurrence gain weights pointing to the same disease node in the local relation network, select the highest weight as the disease's adaptation score, and match the corresponding department to form a candidate department set.

[0050] The specific process of generating a candidate department set includes: for any disease node in the local relation network, obtaining all initial weights and co-occurrence gain weights pointing to the disease node; comparing the numerical values ​​of the obtained initial weights of a single node and the co-occurrence gain weights of multiple nodes, and extracting the one with the highest value as the fit score of the disease node; extracting the department tags configured for the disease node to match the corresponding department category, and sorting them in descending order of fit score from high to low with the disease node as the main axis to generate a candidate department set.

[0051] Step S7, Triage Reference Output: Output the candidate department set from high to low according to the adaptation score to the triage terminal, and generate a triage slip based on the final triage instruction returned by the triage terminal.

[0052] The specific process of generating a triage slip includes: outputting the candidate department set as a visual list to the display interface of the triage terminal, and visually presenting the disease node, corresponding matching score, and department in the visual list; receiving the department confirmation instruction returned by the triage terminal, parsing the department confirmation instruction to obtain the final selected target department, and using the confirmation result of the target department as the final triage instruction; extracting the target department information from the final triage instruction, and combining it with the obtained patient queue sequence number for data encapsulation to generate a triage slip.

[0053] Based on standardized symptom feature information, precise string matching is performed in the disease association network to identify symptom nodes and spatiotemporal nodes related to the current medical visit. Then, starting from these identified nodes, the network expands outward layer by layer along directed connections to find all potential disease nodes directly connected to them. During this expansion, the initial weights on the corresponding directed connections and the co-occurrence gain weights related to the current node combination are extracted simultaneously, thus pruning the disease association network into a local relationship network only related to the current patient's characteristics. If the initial weight and co-occurrence gain weight of a identified node pointing to a disease node are both null or zero, it indicates that such concurrent cases do not exist in the historical data. Therefore, the graph branch path is deemed invalid, and the corresponding connection and disease node are directly removed from the local relationship network.

[0054] For any disease node in the local relational network, obtain all initial weights and co-occurrence gain weights pointing to that disease node. Compare the numerical values ​​of the initial weights of a single node with the co-occurrence gain weights of multiple nodes, and extract the highest value as the fitness score for that disease node. Extract the department tags configured for that disease node to match the corresponding department categories. Using that disease node as the main axis, sort the departments in descending order of fitness score to generate a candidate department set.

[0055] The candidate department set is output to the triage terminal's display interface in the form of a visual list, presenting disease nodes, corresponding adaptation scores, and department categories. The system receives department confirmation instructions triggered by the triage terminal on the display interface for the target patient destination, parses these instructions to obtain the final selected target department, and uses the confirmation result of the target department as the final triage instruction. The target department information is extracted from the final triage instruction and combined with the obtained patient queue number for data encapsulation to generate a triage slip. This triage slip guides patients to the waiting area of ​​the corresponding target department and serves as proof of access to the backend reception system for retrieving the patient's previously standardized symptom information.

[0056] Step S8, Patient Feedback and Weight Adjustment: Obtain actual confirmed records, extract the real confirmed diseases from the actual confirmed records, and associate them with the corresponding standardized symptom feature information and spatiotemporal information of the visit to transform them into standard sample data, and add them to the standardized historical medical data to adjust the corresponding initial weights and co-occurrence gain weights in the disease association network.

[0057] The specific process of adjusting the weights includes receiving the actual diagnosis records returned by the hospital information system after the patient's visit, extracting the real diagnosed diseases from the actual diagnosis records; structurally binding the real diagnosed diseases with the standardized symptom characteristics and spatiotemporal information of the patient at the beginning of the visit, and converting them into standard sample data; appending the standard sample data to the historical medical data; and recalculating the conditional probability and joint conditional probability between corresponding nodes based on the updated historical medical data, so as to dynamically update the initial weights and co-occurrence gain weights of the corresponding directed connections.

[0058] This process involves receiving the actual confirmed diagnosis record returned by the patient after treatment and settlement of medical records in the hospital information system. The record is then parsed to extract the standardized diagnostic name or disease code, which is then identified as the actual confirmed disease. Using the patient's visit serial number as the primary key, a reverse search is performed in the database to extract standardized symptom features and visit time-space information generated by the front-end during the initial stage of the visit. The retrieved initial feature data is then combined with the extracted actual confirmed disease data at the field level to generate standard sample data with initial symptoms and time-space information as feature conditions and the actual confirmed disease as the target result.

[0059] The completed standard sample data is appended and stored in historical medical data to accumulate data for the basic medical sample database. To avoid wasting computational resources by blindly recalculating the entire disease association network, the newly added standard sample data is analyzed to extract specific symptom nodes, specific spatiotemporal nodes, and actual diagnosed diseases, which are then mapped to corresponding nodes in the graph. Subsequently, the corresponding directed connections that cross these nodes are accurately located in the disease association network.

[0060] For the identified directed connections, a weight recalculation logic is triggered. Based on the updated full historical medical data, the latest cumulative frequency of the occurrence of specific symptom nodes or spatiotemporal nodes independently and diagnosed as disease nodes is recalculated. The latest conditional probability is calculated using the ratio of this cumulative frequency to the total occurrence frequency of a single node. Simultaneously, the latest co-occurrence frequency of multiple nodes occurring concurrently and diagnosed as disease nodes is calculated, and the latest joint conditional probability is calculated. The newly calculated latest conditional probabilities and joint conditional probabilities are written into the attribute fields of the corresponding directed connections, replacing the original initial weights and co-occurrence gain weights, thus completing the local targeted update and structural correction of the disease association network weights.

[0061] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction, characterized in that, include: Step S1, Data Collection: Obtain the patient's chief symptoms and physiological values, as well as the time and space information of the visit; Step S2, Data preprocessing: Clean the chief complaint symptoms to obtain chief complaint symptom text, and convert the physiological values ​​into physiological symptom text; The chief complaint text, the physiological symptom text, and the spatiotemporal information of the medical visit are combined to generate a preprocessed dataset. Step S3, Network Construction: Based on standardized historical medical data, establish symptom nodes, spatiotemporal nodes, and disease nodes, and configure corresponding department labels for the disease nodes; establish directed connections from the symptom nodes, the spatiotemporal nodes to the disease nodes, calculate the initial weights of the directed connections according to the mapping ratio between nodes, and calculate the co-occurrence gain weight when multiple nodes point to the same disease node, so as to generate a disease association network. Step S4, Symptom Feature Standardization: Replace the non-standard symptom descriptions contained in the preprocessed dataset with standard symptom terms to generate standardized symptom feature information.

2. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 1, characterized in that, The method further includes the following steps: Step S5, Local Atlas Construction: In the disease association network, lock the symptom nodes and spatiotemporal nodes of the standardized symptom feature information, and spread outward along the directed connection with the locked nodes as the center to generate a local relationship network. Step S6, Generate a candidate department set: Traverse the initial weights and co-occurrence gain weights pointing to the same disease node in the local relation network, select the highest weight as the disease's adaptation score, match the corresponding department, and set them into a candidate department set; Step S7, Triage Reference Output: Output the candidate department set from high to low according to the adaptation score to the triage terminal, and generate a triage slip according to the final triage instruction returned by the triage terminal; Step S8, Patient Feedback and Weight Adjustment: Obtain actual confirmed records, extract the real confirmed diseases from the actual confirmed records, and associate them with the corresponding standardized symptom feature information and spatiotemporal information of the visit to transform them into standard sample data, and add them to the standardized historical medical data to adjust the corresponding initial weights and co-occurrence gain weights in the disease association network.

3. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 1, characterized in that, The specific process of obtaining the patient's chief complaint symptoms, physiological values, and spatiotemporal information of the consultation includes: receiving a natural language speech stream or text stream input from the front-end device of the triage desk, and extracting the chief complaint symptoms; reading the patient's body temperature, blood pressure, and heart rate data as the physiological values ​​by calling the application interface of the in-hospital vital sign measurement device; extracting the current network time protocol timestamp to obtain the current consultation time, and receiving the patient's recent travel history actively entered by the patient on the front-end device of the triage desk to record the patient's recent main activity areas, and combining the current consultation time and the recent main activity areas to form the consultation spatiotemporal information.

4. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 1, characterized in that, The specific process of generating the preprocessed dataset includes: segmenting the chief complaint into independent lexical units and removing irrelevant words, extracting keywords to form the chief complaint text; then comparing the physiological values ​​with a preset medical health threshold range, and mapping the physiological values ​​to physiological symptom text representing abnormal states according to the degree of deviation from the medical health threshold range; and structurally combining the chief complaint text, the physiological symptom text, and the spatiotemporal information of the medical visit as features of different dimensions to generate a multidimensional feature vector as the preprocessed dataset.

5. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 1, characterized in that, The specific process of calculating weights to generate a disease association network includes: statistically analyzing the frequency of each symptom node and spatiotemporal node appearing independently and ultimately diagnosed as a certain disease node in the standardized historical medical data; calculating the conditional probability of a single node pointing to the corresponding disease node; using the conditional probability as the initial weight of the directed connection; mining concurrent multi-node combinations in the historical medical data; statistically analyzing the co-occurrence frequency of the multi-node combinations pointing to the same disease node; calculating the joint conditional probability of the multi-node combinations pointing to the same disease node based on the co-occurrence frequency; and using the joint conditional probability as the co-occurrence gain weight.

6. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 1, characterized in that, The specific process for generating standardized symptom feature information includes: acquiring standard medical terms as standard items and collecting colloquial expressions with semantically consistent meanings as non-standard items; establishing a many-to-one unidirectional mapping relationship between the non-standard items and the standard items to form a thesaurus; statistically analyzing the co-occurrence frequency and contextual distribution features of words in the medical corpus to construct a feature mapping matrix relating words to multidimensional spatial coordinates; extracting each standard and non-standard item in the thesaurus into a multidimensional numerical array containing semantic information using the feature mapping matrix, as a baseline text vector; using the feature mapping matrix to extract semantic features and perform multidimensional spatial mapping on the non-standard symptom expressions in the preprocessed dataset, converting them into text vectors; calculating the cosine value of the text vector and each baseline text vector as a semantic association score; selecting the standard medical term corresponding to the term with the highest semantic association score that exceeds a set matching threshold, replacing the non-standard symptom expression, and generating the standardized symptom feature information.

7. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 2, characterized in that, The specific process of extracting and generating a local relational network includes: matching symptom nodes and spatiotemporal nodes that are consistent with the standardized symptom feature information in the disease association network, and locking the matching symptom nodes and spatiotemporal nodes; starting from the locked symptom nodes and spatiotemporal nodes, expanding outward along the directed connection to the connected disease nodes, and obtaining the initial weights corresponding to the directed connection and the co-occurrence gain weights when multiple nodes are combined, to form a local relational network; if the initial weights and co-occurrence gain weights of the locked node pointing to a certain disease node are both zero, then the current branch path is determined to be invalid, and the corresponding directed connection and the disease node are removed from the local relational network.

8. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 2, characterized in that, The specific process of generating the candidate department set includes: for any disease node in the local relationship network, obtaining all the initial weights and co-occurrence gain weights pointing to the disease node; comparing the values ​​of the initial weights of a single node and the co-occurrence gain weights of multiple nodes, and extracting the one with the highest value as the fit score of the disease node; extracting the department tags configured for the disease node to match the corresponding department category, and sorting them in descending order according to the fit score, with the disease node as the main axis, to generate the candidate department set.

9. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 2, characterized in that, The specific process of generating a triage slip includes: outputting the candidate department set to the display interface of the triage terminal in the form of a visual list, and visually presenting the disease node, the corresponding matching score, and the department in the visual list; receiving the department confirmation instruction returned by the triage terminal, parsing the department confirmation instruction to obtain the finally selected target department, and using the confirmation result of the target department as the final triage instruction; extracting the target department information from the final triage instruction, and combining it with the obtained patient queue sequence number for data encapsulation to generate a triage slip.

10. The auxiliary triage method based on multi-dimensional symptom correlation analysis and patient feedback correction according to claim 2, characterized in that, The specific process of adjusting the weights includes: receiving the actual diagnosis record returned by the hospital information system after the patient's visit, and extracting the real diagnosed disease from the actual diagnosis record; structurally binding the real diagnosed disease with the standardized symptom feature information and the spatiotemporal information of the patient at the beginning of the visit, and converting it into standard sample data; appending the standard sample data to the historical medical data; and recalculating the conditional probability and joint conditional probability between corresponding nodes based on the updated historical medical data, so as to dynamically update the initial weight and co-occurrence gain weight of the corresponding directed connection.

Images