Medical inquiry system and method based on service consultation management

Abstract

The invention relates to the technical field of medical services, in particular to a medical inquiry system and method based on service consultation management, and the method comprises the following steps: S1, building a health file of a patient; s2, storing key information extracted from each round of consultation in a patient health file; s3, dynamic tracking of multi-round consultation contexts: ensuring that each round of consultation automatically obtains key information in previous consultation by using a context tracking technology, and associating the key information with current consultation content to obtain latest consultation content; s4, according to the latest consultation content, analyzing the change trend of the symptoms of the patient through a time sequence prediction model; and S5, storing the current multi-round inquiry consultation path and the key information in a patient health file. According to the method, the problems of information isolated island and resource waste in the multi-department inquiry process are effectively solved, and compared with a traditional fixed inquiry path, the method improves the diagnosis and treatment flexibility and the medical experience of the patient.

Description

Technical Field

[0001] The present invention relates to the field of medical service technology, and in particular to a medical consultation system and method based on service consultation management. Background Art

[0002] With the continuous development of medical informatization, patients' medical data are gradually digitized in the form of electronic medical records and health records. However, in the existing medical consultation system, patients' multiple rounds of consultation data are often stored in isolation, lacking effective association and dynamic management mechanisms, leading to the following problems:

[0003] In the existing technology, the patient's consultation data mostly exists in the form of independent records, lacking the function of automatically associating with the historical consultation content. Doctors need to manually review a large amount of historical records, which is time-consuming and prone to missing important information, affecting the comprehensiveness and efficiency of diagnosis and treatment.

[0004] The traditional consultation process is usually a pre-designed fixed path, which is difficult to adjust dynamically based on the patient's real-time feedback and changes in the patient's condition. Especially when the patient's symptoms worsen or the condition is complicated and requires collaboration from multiple departments, the consultation path lacks flexibility, which can easily lead to waste of diagnosis and treatment resources and a bad experience for patients.

[0005] Although some medical systems have initially introduced artificial intelligence technologies, such as voice recording or keyword extraction, there are still deficiencies in high-level intelligent technologies such as semantic analysis of patient feedback and dynamic prediction of disease trends, and they are unable to provide accurate diagnosis and treatment support for individual patients. In particular, in complex consultation scenarios, existing systems have difficulty understanding the contextual relationships in natural language, resulting in insufficient accuracy in diagnosis and treatment recommendations. Summary of the invention

[0006] The present invention provides a medical consultation system and method based on service consultation management.

[0007] The medical consultation method based on service consultation management includes the following steps:

[0008] S1, establishment of patient health records: collect basic information and medical history data of patients through self-reported medical history and electronic medical record management platform, generate and store patient health records;

[0009] S2, semantic analysis based on natural language processing technology: during the process of multiple rounds of consultation, the consultation content of each round is recorded and stored in real time, including patient feedback and doctor's diagnosis and treatment opinions. Natural language processing technology is used to analyze the patient's feedback and doctor's diagnosis and treatment opinions in each round of consultation, extract key information, and perform semantic understanding. After the analysis is completed, the key information extracted from each round of consultation is stored in the patient's health record;

[0010] S3, dynamic tracking of multi-round consultation context: Use context tracking technology to ensure that each round of consultation automatically obtains key information from previous consultations and associates it with the current consultation content to obtain the latest consultation content. The latest consultation content incorporates historical consultation content after association analysis, so that doctors can quickly understand patients' health changes and needs;

[0011] S4, context-based consultation path adjustment: According to the latest consultation content, analyze the trend of disease changes, analyze the trend of patient symptoms through time series prediction model, and adjust the patient's subsequent consultation path based on the results of the disease trend analysis, including giving priority to recommending departments related to high-risk symptoms;

[0012] S5, stores the consultation paths and key information of these multiple rounds of medical consultation in the patient's health record.

[0013] Optionally, the establishment of the patient health record in S1 specifically includes: collecting the patient's basic information and medical history data through the patient's self-reported medical history and electronic medical record management platform, the medical history data including family history, allergy history, and previous treatment records; integrating the collected data through data cleaning and standardization processing to generate a patient health record, and storing the health record in a unified cloud platform database.

[0014] Optionally, the key information in S2 includes:

[0015] Symptom description: "Headache", "Fever", "Cough";

[0016] Test results: "blood sugar 12.5mmol / L", "abnormal electrocardiogram";

[0017] Symptom severity and duration: "condition based on test results and doctor's interview", "how long symptoms last";

[0018] Drug name: "Amoxicillin", "Ibuprofen";

[0019] Treatment recommendations: "Take it three times a day for two weeks";

[0020] Patient feedback: "Symptoms have been relieved, but there is still slight pain";

[0021] The input data of the natural language processing technology is patient feedback and doctor's diagnosis and treatment opinions. The natural language processing technology specifically includes:

[0022] Word embedding: converting patient feedback and doctor's diagnosis and treatment opinions into vector representations that can be understood by computers;

[0023] BiLSTM: Capture contextual information through bidirectional BiLSTM to understand the relationship between words and semantic information in the context;

[0024] Conditional Random Field (CRF): Use the CRF layer to assign appropriate labels to each word to ensure entity labeling consistency;

[0025] Key information extraction: Extract key information through labeled entity information to form structured data.

[0026] Optionally, the BiLSTM processing is based on a recurrent neural network network architecture, and simultaneously captures the contextual information before and after the word, specifically including:

[0027] The forward LSTM processes the input sequence T = {w1,w2,...,w n}:

[0028] The backward LSTM processes the input sequence in reverse order:

[0029] Concatenate the forward and backward LSTM outputs to get the final representation h for each time step t t :

[0030]

[0031] Finally, BiLSTM generates a contextual representation (i.e., word vector) for each word, such as fully capturing the contextual information of headache, thereby helping the model understand its semantics in this context.

[0032] Among them, x t is the input vector at time step t, i.e., the word embedding vector x∈R d , is the hidden state of the forward LSTM at time step t, capturing the contextual information from time 1 to t, is the hidden state of the forward LSTM at time step t-1, is the hidden state of the backward LSTM at time step t, capturing the contextual information from time n to t, is the hidden state of the backward LSTM at time step t+1, h t is the final hidden state at time step t, is the forward hidden state and the backward hidden state The splicing result is Representation vector concatenation operation, concatenating the forward and backward hidden states into a complete representation.

[0033] Optionally, the conditional random field is used for sequence labeling, specifically including:

[0034] Input context representation h from BiLSTM t, and perform sequence labeling through the CRF layer;

[0035] Label score: The CRF layer determines the most appropriate label sequence by calculating the score of each label sequence. The scoring formula is: Among them, s(H,Y) is the score of the labeled sequence Y, which indicates the degree of match between sequence Y and input H. is the label y t The weight vector of is the label y t The bias, is the transfer score between labels, indicating that t to y t+1 The transition probability, n represents the length of the sequence, that is, the number of words in the input text;

[0036] Normalized score: Normalize the labeled sequence through the softmax function to obtain the probability distribution of each labeled sequence: Where P(Y|H) is the conditional probability of sequence Y, which means the probability of occurrence of sequence Y given input H, exp(s(H,Y)) is the unnormalized score of labeled sequence Y, ∑ Y′ exp(s(H,Y′)) is the sum of the unnormalized scores of all possible labeled sequences and is used for normalization;

[0037] Predict the label sequence: Use the Viterbi algorithm to select the label sequence Y with the highest score * as output.

[0038] Optionally, the S3 specifically includes:

[0039] S31, using context tracking algorithms, automatically extracts and analyzes historical consultation records related to the current consultation in the patient's health file at the beginning of each consultation, including historical symptom descriptions, symptom severity and duration, examination results, drug names, treatment recommendations, and patient feedback;

[0040] S32, by matching keywords between historical consultation data and current consultation content, analyzing semantic similarity and mining association rules, identifying key information in the current consultation that is consistent with the historical consultation content, and associating it with the current consultation content;

[0041] S32, during the current consultation process, based on the key information of each round of consultation, the status of historical key information is updated, the associated historical consultation content is merged with the current consultation content, the latest consultation content is generated, and stored in the patient's health record.

[0042] Optionally, the keyword matching is used to identify the same or similar keywords between the current consultation and the historical consultation, and to determine the relevance by calculating the matching degree;

[0043] The semantic similarity analysis is based on cosine similarity, which is used to evaluate the similarity between the current consultation and the historical consultation content. Each consultation record (including historical consultation and current consultation) is converted into a vector representation, and then their semantic similarity is measured by cosine similarity.

[0044] The association rule mining generates rules based on the frequent patterns in the consulting data to predict future trends or behaviors. Suppose there is a set of frequent item sets I = {i1, i2, ..., i k}, then generate association rules That is, if condition A holds, then condition B also holds. The strength of a rule is measured by support, confidence, and lift:

[0045] Support: Where |A∩B| is the number of data items that satisfy both A and B, and |D| is the size of the total data set;

[0046] Confidence: Where |A| is the number of data items that satisfy A;

[0047] Lift: Lift measures the effect of condition A on condition B. If Lift>1, it means there is a strong correlation between A and B.

[0048] Optionally, analyzing the changing trend of the patient's symptoms by the time series prediction model in S4 specifically includes:

[0049] The time series data of the symptom severity and duration of the patient's consultation content are represented as {u1,u2,..,u t},u t For the latest consultation content, predict the future condition status through the time series prediction model t+e :u t+e =f(u t ,u t-1 ,...,u t-n ;X t ), where u t is the severity and duration of symptoms in the current round of consultation, u t-1 ,…,u t-n The degree and duration of symptoms in the historical consultation, u t+e : The predicted symptom severity and duration at time e in the future, X tis the relevant data in the current consultation, including drug use and treatment recommendations. f is the time series prediction model function, which uses the long short-term memory network LSTM model to output the trend of disease changes: according to the predicted value u t+e , determine whether symptoms are improving, worsening, or unchanged.

[0050] Optionally, based on the trend of the disease condition change, the patient's subsequent consultation path is adjusted, and the adjustment rules of the consultation path are as follows:

[0051] According to the predicted trend of disease changes, high-risk departments related to disease changes are recommended first:

[0052] Among them, P(Department b |u t+k ) represents the probability of department b being recommended, f 科室b (u t+e ) represents the prediction of symptom u by department b t+e The correlation function of .

[0053] The medical consultation system based on service consultation management is used to implement the above-mentioned medical consultation method based on service consultation management, and includes the following modules:

[0054] Health record management module: collects the patient's basic information and medical history data through the patient's self-reported medical history and electronic medical record management platform, and generates and stores the patient's health record;

[0055] Semantic analysis module: used to record and store patient feedback and doctor's treatment opinions in real time during the consultation process, analyze the consultation content through natural language processing technology, extract key information, and store the extracted key information in the patient's health record;

[0056] Context tracking module: Utilizes context tracking technology to automatically obtain key information from previous consultations and associate it with current consultation content to generate the latest consultation content that incorporates historical consultation content;

[0057] Path adjustment module: used to analyze the changing trend of patient symptoms through the time series prediction model according to the latest consultation content, and adjust the patient's subsequent consultation path based on the changing trend of the condition;

[0058] Storage update module: used to dynamically store the multi-round consultation path and extracted key information in the patient's health record for use and analysis in subsequent consultation sessions.

[0059] Beneficial effects of the present invention:

[0060] The present invention, through context tracking technology, can dynamically acquire and analyze the patient's historical consultation data, and deeply associate it with the current consultation content. With the help of keyword matching, semantic similarity analysis and association rule mining technology, it can not only accurately identify the potential association between historical consultations and current consultations, but also automatically update and store key information to ensure the data consistency and integrity of the diagnosis and treatment process. In particular, the introduction of semantic analysis technology makes it possible to understand complex expressions in natural language, thereby improving the accuracy and intelligence level of consultation data analysis, reducing information omissions and redundant collection, and greatly improving the diagnosis and treatment efficiency of doctors.

[0061] The present invention analyzes the changing trend of the patient's condition through a time series prediction model, combines the latest consultation content and historical consultation records, and realizes dynamic adjustment of the patient's subsequent consultation path. It can give priority to recommending relevant departments according to the changing trend of the condition and ensure that the patient receives personalized diagnosis and treatment services updated in real time, effectively solving the problems of information islands and resource waste in the multi-department consultation process. Compared with the traditional fixed consultation path, this method improves the flexibility of diagnosis and treatment and the patient's medical experience.

[0062] The present invention combines the latest content of each round of patient consultation with historical data and stores them in health records, builds a dynamically updated patient health data chain, extracts frequently occurring symptom combinations and disease risks from patient historical data, thereby assisting doctors in discovering potential conditions at an early stage, realizing data-driven intelligent decision support, and improving the scientificity and reliability of medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0065] Figure 2 Schematic diagram of system modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0067] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0068] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0069] like Figure 1 As shown, the medical consultation method based on service consultation management includes the following steps:

[0070] S1, establishment of patient health records: collect basic information and medical history data of patients through self-reported medical history and electronic medical record management platform, generate and store patient health records;

[0071] S2, semantic analysis based on natural language processing technology: during the process of multiple rounds of consultation, the consultation content of each round is recorded and stored in real time, including patient feedback and doctor's diagnosis and treatment opinions. Natural language processing technology is used to analyze the patient's feedback and doctor's diagnosis and treatment opinions in each round of consultation, extract key information, and perform semantic understanding. After the analysis is completed, the key information extracted from each round of consultation is stored in the patient's health record;

[0072] S3, dynamic tracking of multi-round consultation context: Use context tracking technology to ensure that each round of consultation automatically obtains key information from previous consultations and associates it with the current consultation content to obtain the latest consultation content. The latest consultation content incorporates historical consultation content after association analysis, so that doctors can quickly understand patients' health changes and needs;

[0073] S4, context-based consultation path adjustment: According to the latest consultation content, analyze the trend of disease changes, analyze the trend of patient symptoms through time series prediction model, and adjust the patient's subsequent consultation path based on the results of the disease trend analysis, including giving priority to recommending departments related to high-risk symptoms;

[0074] S5, stores the consultation paths and key information of these multiple rounds of medical consultation in the patient's health record.

[0075] The establishment of patient health records in S1 specifically includes: collecting patients' basic information and medical history data through self-reported medical history and electronic medical record management platform, and the medical history data includes family history, allergy history, and previous treatment records; integrating the collected data through data cleaning and standardization processing to generate patient health records, and storing the health records in a unified cloud platform database so that they can be called and updated at any time to ensure the integrity, real-time and privacy protection of patient information; health records can be shared among multiple medical institutions, and support cross-platform data synchronization to ensure the continuity and accuracy of medical services.

[0076] Duplicate data removal: There may be duplicate records in patient health records (due to different data sources, some patients' basic information and medical history data may be recorded repeatedly). The deduplication operation is performed through hash comparison, setting a unique identifier ID to compare the data records. If ID1 = ID2, the duplicate records are removed.

[0077] Numerical standardization: Convert all numerical data into a unified standardized format to ensure consistent units and data scales, including Zscore standardization: Among them, X is the original data, μ is the mean, σ is the standard deviation, X norm is the standardized value.

[0078] Minimum and maximum normalization: Among them, X norm is the normalized value, X min and X max are the minimum and maximum values ​​in the data column.

[0079] Standardization of categorical data: For categorical data (symptoms, diagnosis, etc.), convert categorical variables into binary values ​​through one-hot encoding. For example, convert "fever" into "[1,0,0]" and convert "cough" into "[0,1,0]".

[0080] The key information in S2 includes:

[0081] Symptom description: "Headache", "Fever", "Cough";

[0082] Test results: "blood sugar 12.5mmol / L", "abnormal electrocardiogram";

[0083] Symptom severity and duration: "condition based on test results and doctor's interview", "how long symptoms last";

[0084] Drug name: "Amoxicillin", "Ibuprofen";

[0085] Treatment recommendations: "Take it three times a day for two weeks";

[0086] Patient feedback: "Symptoms have been relieved, but there is still slight pain";

[0087] The input data of natural language processing technology is patient feedback and doctor's diagnosis and treatment opinions. Natural language processing technology specifically includes:

[0088] Word embedding: converting patient feedback and doctor's diagnosis and treatment opinions into vector representations that can be understood by computers;

[0089] BiLSTM: Capture contextual information through bidirectional BiLSTM to understand the relationship between words and semantic information in the context;

[0090] Conditional Random Field (CRF): Use the CRF layer to assign appropriate labels to each word to ensure entity labeling consistency;

[0091] Key information extraction: Extract key information through labeled entity information to form structured data.

[0092] Word embedding calculation process: For each word w i (e.g. headache, ibuprofen), which is converted into a fixed-dimensional vector x through word embedding i , x∈R d , where d is the dimension of the vector: x i =Embed(w i ), Embed represents the embedding operation. The embedded word vector can capture the semantic relationship between words through the model.

[0093] BiLSTM processes a network architecture based on a recurrent neural network, capturing the contextual information before and after a word. When processing natural language, BiLSTM can learn and understand the long-distance dependencies between words, especially in medical texts. Correctly understanding the contextual information is crucial to the diagnosis results, including:

[0094] The forward LSTM processes the input sequence T = {w1,w2,...,w n}:

[0095] The backward LSTM processes the input sequence in reverse order:

[0096] Concatenate the forward and backward LSTM outputs to get the final representation h for each time step t t :

[0097]

[0098] Finally, BiLSTM generates a contextual representation (i.e., word vector) for each word, such as fully capturing the contextual information of headache, thereby helping the model understand its semantics in this context.

[0099] Among them, x t is the input vector at time step t, i.e., the word embedding vector x∈R d , is the hidden state of the forward LSTM at time step t, capturing the contextual information from time 1 to t, is the hidden state of the forward LSTM at time step t-1, is the hidden state of the backward LSTM at time step t, capturing the contextual information from time n to t, is the hidden state of the backward LSTM at time step t+1, h t is the final hidden state at time step t, is the forward hidden state and the backward hidden state The splicing result is Representation vector concatenation operation, concatenating the forward and backward hidden states into a complete representation.

[0100] Conditional random fields are used for sequence labeling. They assign appropriate labels to each word (e.g., symptoms, medications, medical history, etc.) based on the contextual information. They can effectively model the dependencies between labels to ensure the consistency and correctness of the final labeled sequence. Specifically, they include:

[0101] Input context representation h from BiLSTM t , and perform sequence labeling through the CRF layer;

[0102] Label score: The CRF layer determines the most appropriate label sequence by calculating the score of each label sequence. The scoring formula is: Among them, s(H,Y) is the score of the labeled sequence Y, which indicates the degree of match between sequence Y and input H. is the label y t The weight vector of is the label y t The bias, is the transfer score between labels, indicating that t to y t+1 The transition probability, H = {h1,h2,...,h n} is the context representation sequence from BiLSTM, containing the hidden states of all time steps, Y = {y1, y2..., y n} is the label sequence, y t represents the label of time step t, and n represents the length of the sequence, that is, the number of words in the input text;

[0103] Normalized score: Normalize the labeled sequence through the softmax function to obtain the probability distribution of each labeled sequence: Where P(Y|H) is the conditional probability of sequence Y, which means the probability of occurrence of sequence Y given input H, exp(s(H,Y)) is the unnormalized score of labeled sequence Y, ∑ Y′ exp(s(H,Y′)) is the sum of the unnormalized scores of all possible labeled sequences and is used for normalization;

[0104] Predict the label sequence: Use the Viterbi algorithm to select the label sequence Y with the highest score * As output, Y * = argmax Y s(H,Y), where Y * is the label sequence with the highest score, that is, the predicted optimal label sequence, argmax Y It means taking the labeling sequence Y that maximizes the score s(H,Y).

[0105] According to the label of each word obtained through the BiLSTM and CRF networks, various types of key information can be extracted from the text, including symptoms, drugs, treatment recommendations, patient feedback, etc.

[0106] The labeled sequence obtained by the CRF layer (for example, "headache" is labeled as the "symptom" category, and "ibuprofen" is labeled as the "drug" category) is output as structured information:

[0107] Symptom description: such as "headache" → symptom

[0108] Drug name: such as "ibuprofen" → drug

[0109] Treatment suggestions: such as "continue to take the medicine for two days" → treatment plan

[0110] Patient feedback: such as "slightly relieved" → feedback

[0111] Finally, a structured data is obtained that reflects the patient's health status and the doctor's advice.

[0112] S3 specifically includes:

[0113] S31, using context tracking algorithms, automatically extracts and analyzes historical consultation records related to the current consultation in the patient's health file at the beginning of each consultation, including historical symptom descriptions, symptom severity and duration, examination results, drug names, treatment recommendations, and patient feedback;

[0114] S32, by matching keywords between historical consultation data and current consultation content, analyzing semantic similarity and mining association rules, identifying key information in the current consultation that is consistent with the historical consultation content, and associating it with the current consultation content;

[0115] S32, during the current consultation process, based on the key information of each round of consultation, the status of historical key information is updated, the associated historical consultation content is merged with the current consultation content, the latest consultation content is generated, and stored in the patient's health record.

[0116] Keyword matching is used to identify the same or similar keywords between the current consultation and the historical consultation, and to determine the relevance by calculating the matching degree;

[0117] Suppose the keyword set in the current consultation is Q current ={q1,q2,...,q n}, the keyword set in historical consultation is Q history ={h1,h2,...,h m}, the keyword matching degree is calculated by the Jaccard similarity coefficient:

[0118] in:

[0119] |Q current ∩Q history | represents the number of intersections between the current query and the historical query, that is, the number of keywords that match the current query and the historical query, |Q current ∪Q history | represents the number of keywords in the current consultation and the historical consultation, that is, the total number of all keywords after deduplication. The returned value is between 0 and 1. The closer the value is to 1, the higher the degree of keyword matching between the current consultation and the historical consultation.

[0120] Semantic similarity analysis is based on cosine similarity, which is used to evaluate the similarity between the current consultation and the historical consultation content. Each consultation record (including historical consultation and current consultation) is converted into a vector representation, and then their semantic similarity is measured by cosine similarity;

[0121] Suppose the current consultation and historical consultation are transformed into vectors V current and V history , then the cosine similarity is calculated by the following formula:

[0122] Among them, V current is the vector representation of the current consultation content, usually obtained using word embedding, V history is the vector representation of the historical consultation content, V current ·V historyis the dot product of the current consultation vector and the historical consultation vector (i.e., the weighted sum of two vectors in the same dimension), V current and V history are the Euclidean norms (i.e., the modulus length of each vector) of the current consultation vector and the historical consultation vector, respectively. The calculated value range is from -1 to 1, where the closer the value is to 1, the more similar the semantics are, and the value of -1 indicates complete dissimilarity;

[0123] Association rule mining generates rules based on frequent patterns in consulting data to predict future trends or behaviors. Suppose there is a set of frequent item sets I = {i1, i2, ..., i k}, then generate association rules That is, if condition A holds, then condition B also holds. The strength of a rule is measured by support, confidence, and lift:

[0124] Support: Where |A∩B| is the number of data items that satisfy both A and B, and |D| is the size of the total data set;

[0125] Confidence: Where |A| is the number of data items that satisfy A;

[0126] Lift: Lift measures the effect of condition A on condition B. If Lift>1, it means there is a strong correlation between A and B.

[0127] The three technologies are used in combination and complement each other to analyze data from different perspectives to improve the accuracy and reliability of the association.

[0128] Keyword matching: used to quickly filter out data that is obviously related in the current consultation and historical consultations, and narrow the scope of analysis (for example, matching the keywords "headache", "fever").

[0129] Semantic similarity analysis: Based on the results of keyword matching, further analyze the semantic similarity of current consultation and historical data to capture potential associations. For example, "chest tightness" may be semantically related to "heart discomfort", and even if the keywords are different, they can be identified through semantic analysis.

[0130] Association rule mining: Extract frequent symptom combinations, medication regimens and other rules based on historical data to supplement the results of semantic analysis. For example, identify the association of "headache + hypertension" and automatically suggest possible diagnostic directions in the current consultation.

[0131] Joint use process:

[0132] Keyword matching quickly screens the potentially relevant parts of the current consultation and historical consultation records, and calculates the keyword matching score; semantic similarity analysis further calculates the semantic similarity of the records screened out from the keyword matching results to identify deep-level associations; association rule mining extracts frequent item sets and association rules from historical data, applies these rules to the current consultation, and prompts potential related information or predictions.

[0133] The analysis of the changing trend of patient symptoms through the time series prediction model in S4 specifically includes:

[0134] The time series data of the symptom severity and duration of the patient's consultation content are represented as {u1,u2,..,u t},u t For the latest consultation content, predict the future condition status through the time series prediction model t+e :u t+e =f(u t ,u t-1 ,...,u t-n ;X t ), where u t is the severity and duration of symptoms in the current round of consultation, u t-1 ,…,u t-n The degree and duration of symptoms in the historical consultation, u t+e : The predicted symptom severity and duration at time e in the future, X t is the relevant data in the current consultation, including drug use and treatment recommendations. f is the time series prediction model function, which uses the long short-term memory network LSTM model to output the trend of disease changes: according to the predicted value u t+e , determine whether symptoms are improving, worsening, or unchanged.

[0135] The LSTM model solves the long sequence dependency problem by introducing a gating mechanism (input gate, forget gate, output gate), describing the calculation process of the LSTM model at time step t, and is used to predict the next moment u t+1 .

[0136] Based on the trend of disease changes, adjust the patient's subsequent consultation path. The adjustment rules of the consultation path are as follows:

[0137] According to the predicted trend of disease changes, high-risk departments related to disease changes are recommended first:

[0138] Among them, P(Department b |u t+k ) represents the probability of department b being recommended, f 科室b (u t+e ) represents the prediction of symptom u by department bt+e The correlation function of .

[0139] like Figure 2 As shown, the medical consultation system based on service consultation management is used to implement the above-mentioned medical consultation method based on service consultation management, and includes the following modules:

[0140] Health record management module: collects the patient's basic information and medical history data through the patient's self-reported medical history and electronic medical record management platform, and generates and stores the patient's health record;

[0141] Semantic analysis module: used to record and store patient feedback and doctor's treatment opinions in real time during the consultation process, analyze the consultation content through natural language processing technology, extract key information, and store the extracted key information in the patient's health record;

[0142] Context tracking module: Utilizes context tracking technology to automatically obtain key information from previous consultations and associate it with current consultation content to generate the latest consultation content that incorporates historical consultation content;

[0143] Path adjustment module: used to analyze the changing trend of patient symptoms through the time series prediction model according to the latest consultation content, and adjust the patient's subsequent consultation path based on the changing trend of the condition;

[0144] Storage update module: used to dynamically store the multi-round consultation path and extracted key information in the patient's health record for use and analysis in subsequent consultation sessions.

[0145] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0146] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A medical consultation method based on service consultation management, characterized in that: The following steps are involved: S1, establishment of patient health records: collect basic information and medical history data of patients through self-reported medical history and electronic medical record management platform, generate and store patient health records; S2, semantic analysis based on natural language processing technology: during the process of multiple rounds of consultation, the consultation content of each round is recorded and stored in real time, including patient feedback and doctor's diagnosis and treatment opinions. Natural language processing technology is used to analyze the patient's feedback and doctor's diagnosis and treatment opinions in each round of consultation, extract key information, and perform semantic understanding. After the analysis is completed, the key information extracted from each round of consultation is stored in the patient's health record; S3, dynamic tracking of multi-round consultation context: using context tracking technology to ensure that each round of consultation automatically obtains key information from previous consultations and associates it with the current consultation content to obtain the latest consultation content, which incorporates the historical consultation content after association analysis; S4, context-based consultation path adjustment: According to the latest consultation content, analyze the trend of disease changes, analyze the trend of patient symptoms through time series prediction model, and adjust the patient's subsequent consultation path based on the results of the disease trend analysis, including giving priority to recommending departments related to high-risk symptoms; S5, stores the consultation paths and key information of these multiple rounds of medical consultation in the patient's health record.

2. The medical consultation method based on service consultation management according to claim 1, characterized in that: The establishment of the patient health record in S1 specifically includes: collecting the patient's basic information and medical history data through the patient's self-reported medical history and electronic medical record management platform, the medical history data includes family history, allergy history, and previous treatment records; integrating the collected data through data cleaning and standardization processing to generate patient health records, and storing the health records in a unified cloud platform database.

3. The medical consultation method based on service consultation management according to claim 1, characterized in that: The key information in S2 includes: symptom description, examination results, symptom severity and duration, medication name, treatment recommendations, and patient feedback; The input data of the natural language processing technology is patient feedback and doctor's diagnosis and treatment opinions. The natural language processing technology specifically includes: Word embedding: convert patient feedback and doctor's diagnosis and treatment opinions into vector representation; BiLSTM: Capture contextual information through bidirectional BiLSTM to understand the relationship between words and semantic information in the context; Conditional Random Field (CRF): Use the CRF layer to assign labels to each word to ensure entity labeling consistency; Key information extraction: Extract key information through labeled entity information to form structured data.

4. The medical consultation method based on service consultation management according to claim 3 is characterized in that: The BiLSTM process is based on a recurrent neural network architecture and captures the contextual information of the word, including: The forward LSTM processes the input sequence T = {w1,w2,...,w n }: The backward LSTM processes the input sequence in reverse order: Concatenate the forward and backward LSTM outputs to get the final representation h for each time step t t : Finally, BiLSTM generates a contextual representation (i.e., word vector) for each word, such as completely capturing the contextual information of headache, thereby helping the model understand its semantics in this context. Among them, x t is the input vector at time step t, i.e., the word embedding vector x∈R d , is the hidden state of the forward LSTM at time step t, capturing the contextual information from time 1 to t, is the hidden state of the forward LSTM at time step t-1, is the hidden state of the backward LSTM at time step t, capturing the contextual information from time n to t, is the hidden state of the backward LSTM at time step t+1, h t is the final hidden state at time step t, is the forward hidden state and the backward hidden state The splicing result is Representation vector concatenation operation, concatenating the forward and backward hidden states into a complete representation.

5. The medical consultation method based on service consultation management according to claim 3 is characterized in that: The conditional random field is used for sequence labeling, specifically including: Input context representation h from BiLSTM t , and perform sequence labeling through the CRF layer; Label score: The CRF layer determines the label sequence by calculating the score of each label sequence. The scoring formula is: Among them, s(H,Y) is the score of the labeled sequence Y, which indicates the degree of match between sequence Y and input H. is the label y t The weight vector of is the label y t The bias of is the transfer score between labels, indicating that t to y t+1 The transition probability, n represents the length of the sequence, that is, the number of words in the input text; Normalized score: Normalize the labeled sequence through the softmax function to obtain the probability distribution of each labeled sequence: Where P(Y|H) is the conditional probability of sequence Y, which means the probability of occurrence of sequence Y given input H, exp(s(H,Y)) is the unnormalized score of labeled sequence Y, ∑ Y′ exp(s(H,Y′)) is the sum of the unnormalized scores of all possible labeled sequences and is used for normalization; Predict the label sequence: Use the Viterbi algorithm to select the label sequence Y with the highest score * as output.

6. The medical consultation method based on service consultation management according to claim 1, characterized in that: The S3 specifically includes: S31, using context tracking algorithms, automatically extracts and analyzes historical consultation records related to the current consultation in the patient's health file at the beginning of each consultation, including historical symptom descriptions, symptom severity and duration, examination results, drug names, treatment recommendations, and patient feedback; S32, by matching keywords between historical consultation data and current consultation content, analyzing semantic similarity and mining association rules, identifying key information in the current consultation that is consistent with the historical consultation content, and associating it with the current consultation content; S32, during the current consultation process, based on the key information of each round of consultation, the status of historical key information is updated, the associated historical consultation content is merged with the current consultation content, the latest consultation content is generated, and stored in the patient's health record.

7. The medical consultation method based on service consultation management according to claim 1, characterized in that: The keyword matching is used to identify the same or similar keywords between the current consultation and the historical consultation, and to determine the relevance by calculating the matching degree; The semantic similarity analysis is based on cosine similarity, which is used to evaluate the similarity between the current consultation and the historical consultation content. Each consultation record is converted into a vector representation, and then their semantic similarity is measured by cosine similarity. The association rule mining generates rules based on the frequent patterns in the consulting data to predict future trends or behaviors. Suppose there is a set of frequent item sets I = {i1, i2, ..., i k }, then generate association rules That is, if condition A holds, then condition B also holds. The strength of a rule is measured by support, confidence, and lift: Support: Where |A∩B| is the number of data items that satisfy both A and B, and |D| is the size of the total data set; Confidence: Where |A| is the number of data items that satisfy A; Lift: Lift measures the effect of condition A on condition B. If Lift>1, it means there is a strong correlation between A and B.

8. The medical consultation method based on service consultation management according to claim 1, characterized in that: Analyzing the changing trend of the patient's symptoms by the time series prediction model in S4 specifically includes: The time series data of the symptom severity and duration of the patient's consultation content are represented as {u1,u2,..,u t },u t For the latest consultation content, predict the future condition status through the time series prediction model t+e :u t+e =f(u t ,u t-1 ,...,u t-n ;X t ), where u t is the severity and duration of symptoms in the current round of consultation, u t-1 ,…,u t-n The degree and duration of symptoms in the historical consultation, u t+e : The predicted symptom severity and duration at time e in the future, X t is the relevant data in the current consultation, including drug use and treatment recommendations. f is the time series prediction model function, which outputs the trend of disease changes: according to the predicted value u t+e , determine whether symptoms are improving, worsening, or unchanged.

9. The medical consultation method based on service consultation management according to claim 8, characterized in that: Based on the trend of the disease, the patient's subsequent consultation path is adjusted. The adjustment rules of the consultation path are as follows: According to the predicted trend of disease changes, high-risk departments related to disease changes are recommended first: Among them, P(Department b |u t+k ) represents the probability of department b being recommended, f 科室b (u t+e ) represents the prediction of symptom u by department b t+e The correlation function of .

10. A medical consultation system based on service consultation management, used to implement the medical consultation method based on service consultation management as claimed in any one of claims 1 to 9, characterized in that: Includes the following modules: Health record management module: collects the patient's basic information and medical history data through the patient's self-reported medical history and electronic medical record management platform, and generates and stores the patient's health record; Semantic analysis module: used to record and store patient feedback and doctor's treatment opinions in real time during the consultation process, analyze the consultation content through natural language processing technology, extract key information, and store the extracted key information in the patient's health record; Context tracking module: Utilizes context tracking technology to automatically obtain key information from previous consultations and associate it with current consultation content to generate the latest consultation content that incorporates historical consultation content; Path adjustment module: used to analyze the changing trend of patient symptoms through the time series prediction model according to the latest consultation content, and adjust the patient's subsequent consultation path based on the changing trend of the condition; Storage update module: used to dynamically store the multi-round consultation path and extracted key information into the patient's health record.

Images