Medical assistance method and system based on voice recognition and medical information knowledge base

Abstract

The invention discloses a medical assistance method and system based on voice recognition and a medical information knowledge base, and relates to the technical field of medical information. According to the invention, the medical information of the patient is recorded by using the intelligent voice recognition system, and matching correction is carried out, so that the accuracy and efficiency of voice operation in a medical scene can be improved, and character errors caused by manual input or data recording are reduced; by utilizing the prediction model and according to the historical patient medical information, the habits of the medical personnel are determined, and the next operation of the medical personnel is predicted, so that the time for manually searching and calling data can be shortened, the waste of time resources is greatly reduced, and the working efficiency of the medical personnel is effectively improved; detecting whether the medical information formulated by the medical staff is wrong or not by using the anomaly detection model, and reminding the medical staff to prevent medical accidents; a comprehensive patient medical information knowledge base is established, intelligent operation prediction and prompt functions are provided, user experience is improved, and intelligentization of various medical operations is supported.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a medical assistance method and system based on speech recognition and medical information knowledge base. Background Art

[0002] In the modern medical environment, doctors and nurses need to handle a large amount of medical information and operations, such as prescribing medical orders, executing medical orders, recording medical records, and filling out temperature sheets. These operations are not only cumbersome, but also prone to errors due to human negligence, which in turn affects the patient's treatment effect. The current application of speech recognition technology in the medical field is not yet widespread enough, especially the intelligent prediction and prompt functions in medical scenarios still need to be improved. Therefore, the development of a medical assistance method and system based on speech recognition and medical information knowledge base can effectively improve the efficiency and accuracy of medical operations, which is of great practical significance. Summary of the invention

[0003] The purpose of the present invention is to provide a method for visually and dynamically constructing an intelligence analysis system to improve the above technical problems.

[0004] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0005] A medical assistance method based on speech recognition and medical information knowledge base, comprising:

[0006] S1. Obtain the medical information of patients in the inpatient department and store it in the patient medical information knowledge base;

[0007] S2, inputting the patient's medical information into an intelligent speech recognition system for recognition, and outputting a speech recognition result; the intelligent speech recognition system is a PDA model;

[0008] S3, matching and correcting the speech recognition result to obtain corrected text data;

[0009] S4, inputting the corrected text data into a prediction model, and outputting a prediction result; the prediction model adopts a sequence-to-sequence model;

[0010] S5, giving a voice prompt for the prediction result, obtaining feedback from medical staff on the prediction result, updating the prediction model or executing the prediction result and updating the corrected text data;

[0011] S6. Input the updated text data into an anomaly detection model, and output an anomaly detection result; the anomaly detection model includes an NLP module, a knowledge graph module and an isolation forest model;

[0012] S7. Based on the abnormality detection result, adjust the updated text data, and store the adjusted text data in a patient medical information knowledge base.

[0013] Furthermore, the patient medical information includes the patient's basic information, diagnosis information, information on issued medical orders, historical information on all issued medical orders in the current department, medication status, examination and test results, treatment and surgery records, nursing records, health assessment, lifestyle, and psychological assessment.

[0014] Furthermore, the patient medical information knowledge base includes a parallel basic information module, a diagnosis information module, an issued medical order information module, a historical information module of all issued medical orders in the current department, a medication status module, an examination and test result module, a treatment and surgery record module, a nursing record module, a health assessment module, a lifestyle module, and a psychological assessment module; each module stores the corresponding patient medical information in chronological order.

[0015] Furthermore, matching and correcting the speech recognition result to obtain corrected text data includes:

[0016] S3-1, segmenting the speech recognition result to obtain corresponding segmented text data;

[0017] S3-2, calculating the edit distance between each of the text data and the patient medical information in the patient medical information knowledge base, and selecting the minimum edit distance corresponding to each of the text data;

[0018] S3-3, based on the minimum edit distance, matching each of the text data with the patient medical information in the patient medical information knowledge base to obtain a corresponding matching result;

[0019] S3-4. Based on the matching result, proofread each of the text data to obtain the corrected text data.

[0020] Furthermore, the sequence-to-sequence model is a Seq2Seq model; the Seq2Seq model includes an encoder and a decoder; and both the encoder and the decoder use LSTM units.

[0021] Furthermore, the prediction result is given a voice prompt, and feedback from medical staff on the prediction result is obtained, and the prediction model is updated or the prediction result is executed and the corrected text data is updated, including:

[0022] S5-1, voice broadcasting the prediction result, and collecting feedback from the PDA model medical staff through the intelligent voice recognition system; the prediction result is the prediction of the next operation of the medical staff; the feedback is the answer of the medical staff, that is, "yes" or "no";

[0023] S5-2, judging whether the feedback is "yes"; if so, executing the corresponding prediction result, updating the corrected text data, and entering S6; otherwise, training the sequence-to-sequence model, and entering S5-3;

[0024] S5-3, recording the feedback of "No" and the corresponding prediction result and corrected text data as the historical data for the sequence-to-sequence model training;

[0025] S5-4, inputting the historical data into the sequence-to-sequence model, and constructing a corresponding objective function; the objective function adopts a cross entropy loss function;

[0026] S5-6. Based on the objective function, the Adam optimizer is used to adjust the model parameters of the sequence-to-sequence model to obtain an updated sequence-to-sequence model; the updated sequence-to-sequence model is used for the next round of prediction.

[0027] Furthermore, the updated text data is input into the anomaly detection model, and the anomaly detection result is output, including:

[0028] S6-1, taking the patient's basic information, diagnosis information, prescribed medical information, examination and test results in the patient's medical information as knowledge nodes, and taking the relationships between the patient's basic information, diagnosis information, prescribed medical information, examination and test results as edges, to construct a corresponding knowledge graph;

[0029] S6-2, based on the knowledge graph, inputting the updated text data into the NLP module, and outputting relevant medical information data; the relevant medical information data is the keywords of the prescribed medical advice information in the updated text data;

[0030] S6-3, inputting the relevant medical information data into the isolation forest model to obtain an abnormality detection result; the abnormality detection result includes abnormal information of the prescribed medical order and normal information of the prescribed medical order;

[0031] S6-4, broadcast the issued medical order information corresponding to the abnormal detection result that the issued medical order information is abnormal, wait for the voice reply of the medical staff, and enter S7.

[0032] Furthermore, the training process of the isolation forest model is:

[0033] S6-3-1. Obtain historical relevant medical information data and perform random sampling to obtain sampled historical relevant medical information data as a subsample set;

[0034] S6-6-2. Build an isolated decision tree corresponding to each sub-sample set, and randomly select features and split points in the sub-sample set to divide the isolated decision tree into a left subtree and a right subtree;

[0035] S6-6-3, repeat S6-6-2 until all isolated decision trees corresponding to sub-samples are isolated, completing the construction of the isolation forest model;

[0036] S6-6-4, inputting the historical relevant medical information data into the isolation forest model, and calculating the path length of the historical relevant medical information data on each decision tree;

[0037] S6-6-5. Based on the path length, calculate the average path length corresponding to the historical related medical information data;

[0038] S6-6-6. Calculate an anomaly score based on the path length and the average path length;

[0039] S6-6-7. Determine a training anomaly detection result based on the anomaly score;

[0040] S6-6-8. Based on the training anomaly detection result, the parameters of the isolation forest model are adjusted to complete the training of the isolation forest model.

[0041] A medical assistance system based on speech recognition and medical information knowledge base, comprising a medical information acquisition module, a speech recognition module, a text matching correction module, an operation prediction module, a data update module, an abnormality detection module, a data adjustment module and a medical data storage module; wherein;

[0042] Medical information acquisition module, used to obtain medical information of patients in the inpatient department;

[0043] A speech recognition module is used to recognize the patient's medical information through an intelligent speech recognition system to obtain a speech recognition result;

[0044] A text matching and correction module, used to match and correct the speech recognition result to obtain corrected text data;

[0045] An operation prediction module is used to predict the corrected text data through a prediction model to obtain a prediction result;

[0046] A data updating module, used to give a voice prompt of the prediction result, obtain feedback from medical staff on the prediction result, update the prediction model or execute the prediction result and update the corrected text data;

[0047] An anomaly detection module is used to detect the updated text data using the anomaly detection model and output an anomaly detection result;

[0048] A data adjustment module, used for adjusting the updated text data based on the abnormality detection result;

[0049] The medical data storage module is used to store the medical information of patients in the inpatient department and the adjusted text data storage.

[0050] The beneficial effects of the present invention are:

[0051] 1. The present invention uses an intelligent voice recognition system to input patient medical information and perform matching correction, which can improve the accuracy and efficiency of voice operations in medical scenarios and reduce text errors caused by manual input or data entry;

[0052] 2. The present invention uses a prediction model and historical patient medical information to determine the habits of medical staff, predict the next operation of the medical staff, and execute the predicted operation, which can reduce the time for manual search and call of data, greatly reduce the waste of time resources, and effectively improve the work efficiency of medical staff;

[0053] 3. The present invention uses an anomaly detection model to detect whether the medical information formulated by medical staff is incorrect, and reminds medical staff to prevent medical accidents;

[0054] 4. The present invention provides an intelligent medical assistance system, establishes a comprehensive patient medical information knowledge base, provides intelligent operation prediction and prompt functions, improves user experience, and supports the intelligence of multiple medical operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 is a flow chart of a method in an embodiment of the present invention;

[0057] Figure 2 is an edit distance graph in an embodiment of the present invention; wherein (a) is a schematic diagram of an initialization matrix, (b) is a schematic diagram of a filling matrix, (c) is a schematic diagram of a filling matrix, (d) is a schematic diagram of a filling matrix, and (e) is a schematic diagram of a filling matrix;

[0058] Figure 3 4 is a system structure diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0060] See also Figure 1 , this embodiment provides a medical assistance method based on speech recognition and medical information knowledge base, which includes:

[0061] S1. Obtain the medical information of patients in the inpatient department and store it in the patient medical information knowledge base;

[0062] The patient medical information includes the patient's basic information, diagnosis information, information on issued medical orders, historical information on all issued medical orders in the current department, medication status, examination and test results, treatment and surgery records, nursing records, health assessment, lifestyle, and psychological assessment.

[0063] The patient medical information is integrated and stored in the patient medical information knowledge base, that is, the format of the patient medical information is unified using a unified coding system (ICD-10, etc.), and the patient medical information in the unified format is added with fields and stored in the patient medical information knowledge base. The added field is used to store the pinyin of the patient medical information. For example, if "amoxicillin" exists in the medication situation, "amoxicillin" needs to be stored in the form of "amxl".

[0064] The patient medical information knowledge base includes a parallel basic information module, a diagnosis information module, an issued medical order information module, a historical information module of all issued medical orders in the current department, a medication status module, an examination and test result module, a treatment and surgery record module, a nursing record module, a health assessment module, a lifestyle module, and a psychological assessment module; the patient medical information knowledge base has a hierarchical organization, and the corresponding patient medical information is stored in each module in chronological order, that is, it is stored in chronological order.

[0065] The data structure library of the patient medical information knowledge base supports fast access and update. Each patient's medical information is stored according to the corresponding table structure. As shown in Table 1, Table 2, Table 3, Table 4, Table 5, Table 6, Table 7, Table 8, Table 9, Table 10, and Table 11, the table structures of all data types of patient medical information are displayed.

[0066] Table 1 Basic information

[0067]

[0068] Table 2 Diagnostic information

[0069]

[0070] Table 3 Information on prescribed medical advice

[0071]

[0072] Table 4 Historical information of all medical orders issued by the current department

[0073]

[0074] Table 5 Medication status

[0075]

[0076]

[0077] Table 6 Inspection and test results

[0078]

[0079] Table 7 Treatment and surgical records

[0080]

[0081] Table 8 Nursing Record

[0082]

[0083]

[0084] Table 9 Health Assessment

[0085]

[0086] Table 10 Lifestyle

[0087]

[0088] Table 11 Psychological Assessment

[0089]

[0090]

[0091] All table structures contain a primary key; the primary key is used to uniquely identify each record; PatientID is used as a foreign key to associate records in different tables and is associated with the patient's basic information, which facilitates fast access and update; Pinyin code fields (such as AssessmentType Pinyin, etc.) are used to improve data query efficiency and are suitable for scenarios where Pinyin is input or searched.

[0092] To support fast access to the database, the patient medical information knowledge base adopts indexing and batch update operations to improve access efficiency. To ensure data security and privacy, the patient medical information knowledge base uses encryption technology to protect patient information. The patient medical information knowledge base sets access rights for different user roles to ensure that only authorized personnel can access sensitive information.

[0093] Create indexes on NamePinyin, DiagnosisPinyin, and PatientID, that is:

[0094] CREATE INDEX idx_name_pinyin ON PatientInfo(NamePinYIN);

[0095] CREATE INDEX idx_diagnosis_pinyin ON Diagnosis(DiagnosisPinyin);

[0096] CREATE INDEX idx_patient_id ON Diagnosis(PatientID);

[0097] Create a composite index on PatientID and Date to optimize time range queries, namely:

[0098] CREATE INDEX idx_patient_date ON Diagnosis(PatientID,Date);

[0099] Partitioning the stored data by date or other dimensions helps improve query speed. For example, the inspection result table is partitioned by year, that is:

[0100] PARTITION BY RANGE(YEAR(Date))(

[0101] PARTITION p2023 VALUES LESS THAN(2024)

[0102] PARTITION p2024 VALUES LESS THAN(2025) );

[0104] Introduce a cache layer (such as Redis) to store frequently accessed data in the patient medical information knowledge base and reduce the database load of the patient medical information knowledge base.

[0105] Batch update is to update multiple records at one time, which can reduce the overhead of connecting to the patient medical information knowledge base database, use more stored procedures and transactions, and operate cache (Redis) in the program code. For example:

[0106] BEGIN;

[0107] UPDATE PatientInfo;

[0108] SET Age=CASE

[0109] WHEN PatientID=1 THEN 30

[0110] WHEN PatientID=2THEN 40

[0111] ELSE Age

[0112] END

[0113] WHERE PatientID IN(1,2);

[0114] UPDATE Diagnosis

[0115] SET Diagnosis=CASE

[0116] WHEN PatientID=1THEN'Updated Diagnosis 1'

[0117] WHEN PatientID=2THEN'Updated Diagnosis 2'

[0118] ELSE Dignosis

[0119] END

[0120] WHERE PatientID IN(1,2);

[0121] COMMIT;

[0122] S2. Input the patient's medical information into an intelligent speech recognition system for recognition, and output a speech recognition result; the intelligent speech recognition system is a PDA model; the PDA model is a speech model integrated in the PDA by iFLYTEK; in addition, the intelligent speech recognition system can select a speech model with a higher recognition rate on the market.

[0123] Doctors and nurses can shout the keyword "Xiaohui Xiaohui" to the PDA model to realize the voice wake-up function and let the voice model answer. After receiving the voice of the operator (medical staff), the voice model converts the voice into text. Medical staff use the voice function to input the patient's medical information into the intelligent voice recognition system for recognition, which greatly reduces the time of manual input, improves work efficiency, reduces human input errors, and facilitates subsequent search and review.

[0124] S3, matching and correcting the speech recognition result to obtain corrected text data;

[0125] In the process of speech recognition, problems that do not conform to medical standards or character errors may occur easily. For example, myocardial infarction is recognized as heart attack, which requires correction. That is, matching and correcting the speech recognition results to obtain corrected text data includes:

[0126] S3-1, segmenting the speech recognition result to obtain corresponding segmented text data;

[0127] S3-2, calculating the edit distance between each of the text data and the patient medical data in the patient medical information knowledge base, and selecting the minimum edit distance corresponding to each of the text data;

[0128] Define a two-dimensional array dp[i][j], which represents the edit distance between the first i characters of string s1 and the first j characters of string s2. Use dynamic programming to calculate the edit distance. The corresponding process is: fill in two strings, that is, dp[i][0]=i, dp[0][j]=j; dp[i][0]=i means that one of the strings is inserted i times to become the first i characters in string s1; dp[0][j]=j means that one of the strings is inserted j times to become the first j characters in string s2.

[0129] For any i and j, if s1[i-1] == s2[j-1], then dp[i][j] == dp[i-1][j-1]; if s1[i-1] ! == s2[j-1], then dp[i][j] == min(dp[i-1][j-1]+1); where dp[i-1][j]+1 represents a delete operation, dp[i][j-1]+1 represents an insert operation, dp[i-1][j-1]+1 represents a replace operation, s1[i-1] represents the first i-1 characters in string s1, and s2[j-1] represents the first j-1 characters in string s2.

[0130] S3-3, based on the minimum edit distance, matching each of the text data with the patient medical data in the patient medical information knowledge base to obtain a corresponding matching result;

[0131] S3-4. Based on the matching result, proofread each of the text data to obtain the corrected text data.

[0132] Taking the medical information of patient Zhang as an example, the corresponding patient medical data is:

[0133] Patient name: Zhang; Gender: Male; Age: 45; Diagnosis information: "Myocardial infarction": ["Myocardial infarction", "Heart attack", "Heart attack"], "Hypertension": ["Hypertension", "Gao Xueya", "High blood pressure"], "Diabetes": ["Diabetes", "Glycosuria", "Diabetes"], "Pneumonia": ["Pneumonia", "Lung infection", "Lung infection"]; Patient's medical prescription information: "Aspirin": ["Aspirin", "Aspirin", "Aspirin"], "Lovastatin": ["Lovastatin", "Lova", "Statin"]; All medical prescription history information of the current department: "Physical examination": "Physical examination", "Physical examination", "Comprehensive examination"], "Electrocardiogram": ["Electrocardiogram", "ECG", "Heart examination"]; Medication status: "Amoxicillin ”: ["Amoxicillin", "Amoxicillin", "Antibiotics"]; Examination and test results: "Normal ECG": ["Normal ECG", "No abnormal ECG"], "High blood sugar": ["High blood sugar", "High sugar"]; Treatment and surgery records: "Heart bypass": ["Heart bypass", "Heart surgery"]; Nursing records: "Daily blood pressure monitoring": ["Daily blood pressure monitoring", "Daily blood pressure check"]; Health assessment: "Overweight": ["Overweight", "Overweight"], "Normal": ["Normal", "No abnormal"]; Lifestyle: "Smoking": ["Smoking", "Smoking"], "Drinking": ["Drinking", "Drinking"]; Psychological assessment: "Depression": ["Depression", "Low mood"], "Anxiety": ["Anxiety", "Tense"]

[0134] like Figure 2 As shown in (a) in

[0135] For example Figure 2 As shown in (b) in

[0136] dp[1][1] (comparing "心" in the recognition result with "心" in the patient's medical information knowledge base): s1[0] == s2[0], no operation is needed, dp[1][1] = dp[0][0] = 0;

[0137] For example Figure 2 As shown in (c) in

[0138] dp[1][2] (comparing "心" in the recognition result with "心梗" in the patient's medical information knowledge base): s1[0] != s2[1], replacement is needed, dp[1][2] = min(dp[0][2] + 1, dp[1][1] + 1, dp[0][1] + 1) = min(3, 1, 2) = 1;

[0139] For example Figure 2 As shown in (d) in

[0140] dp[2][1] (comparing "心更" in the recognition result with "心" in the patient's medical information knowledge base): s1[1] != s2[0], replacement is needed, dp[2][1] = min(dp[1][1] + 1, dp[2][0] + 1, dp[1][0] + 1) = min(1, 3, 2) = 1;

[0141] For example Figure 2 As shown in (e) in

[0142] dp[2][2] (comparing "心" in the recognition result with "心" in the patient's medical information knowledge base): s1[1] != s2[1], replacement is needed, dp[2][2] = min(dp[1][2] + 1, dp[2][1] + 1, dp[1][1] + 1) = min(2, 2, 1) = 1.

[0143] The edit distance is obtained according to the formula dp[len(s1)][len(s2)], that is, dp[2][2]=1, and the edit distance between "myocardial infarction" and "heart infarction" is 1.

[0144] Calculate the edit distance EditDistance between "heart infarction" and the terms in the diagnostic information, that is, EditDistance("myocardial infarction", "heart infarction") = 1, EditDistance("heart attack", "heart infarction") = 4, EditDistance("heart infarction", "hypertension") = 3, EditDistance("diabetes", "heart infarction") = 4, EditDistance("pneumonia", "heart infarction") = 3; It can be seen that the minimum edit distance is 1, and the corresponding term is "myocardial infarction", so the matching result is: BestMatch("heart infarction") = "myocardial infarction".

[0145] Proofread the matching result, and replace the wrong words or phrases with the correct medical terms, that is, CorrectedWords = ["patient", "has", "myocardial infarction"]; Recombine the proofread words into text, that is, CorrectedText = "The patient has myocardial infarction".

[0146] Call the preset API interface of the intelligent speech recognition system to complete the storage and update of the recognition results, and provide training data for subsequent behavior prediction and intelligent reminder.

[0147] S4. Input the corrected text data into the prediction model, and output the prediction result; The prediction model uses a sequence-to-sequence model;

[0148] The sequence-to-sequence model is a Seq2Seq model; The Seq2Seq model includes an encoder and a decoder; Both the encoder and the decoder use LSTM units, that is, an LSTM network model.

[0149] The input of the corrected text data into the prediction model and the output of the prediction result include:

[0150] S4-1. Input the corrected text data into the encoder, and output the corresponding context vector. The corresponding formula is:

[0151] ht = LSTM(xt, ht-1);

[0152] Among them, ht and ht-1 respectively represent the current context vector and the context vector obtained by the previous round of encoding, xt represents the corrected text data, and LSTM(·) represents the LSTM unit.

[0153] S4-2, input the context vector to the decoder, and output the prediction result. The corresponding formula is:

[0154] st=LSTM(yt-1,st-1,c);

[0155]

[0156] Among them, st and st-1 represent the current initial prediction result and the initial prediction result obtained by the previous round of decoding, respectively, c represents the current context vector, W represents the weight parameter of the decoder, and yt-1 represents the prediction result obtained by the previous round of decoding. represents the current prediction result, and softmax(·) represents the activation function.

[0157] S5, giving a voice prompt for the prediction result, obtaining feedback from medical staff on the prediction result, updating the prediction model or executing the prediction result and updating the corrected text data;

[0158] The step of providing a voice prompt for the prediction result, obtaining feedback from medical staff on the prediction result, updating the prediction model, or executing the prediction result and updating the corrected text data includes:

[0159] S5-1, voice broadcasting the prediction result, and collecting feedback from the PDA model medical staff through the intelligent voice recognition system; the prediction result is the prediction of the next operation of the medical staff; the feedback is the answer of the medical staff, that is, "yes" or "no";

[0160] S5-2, judging whether the feedback is "yes"; if so, executing the corresponding prediction result, updating the corrected text data, and entering S6; otherwise, training the sequence-to-sequence model, and entering S5-3;

[0161] S5-3, recording the feedback of "No" and the corresponding prediction result and corrected text data as the historical data for the sequence-to-sequence model training;

[0162] S5-4, inputting the historical data into the sequence-to-sequence model, and constructing a corresponding objective function; the objective function adopts a cross entropy loss function;

[0163] S5-6. Based on the objective function, the Adam optimizer is used to adjust the model parameters of the sequence-to-sequence model to obtain an updated sequence-to-sequence model; the updated sequence-to-sequence model is used for the next round of prediction.

[0164] S6. Input the updated text data into an anomaly detection model, and output an anomaly detection result; the anomaly detection model includes an NLP module, a knowledge graph module and an isolation forest model;

[0165] The updated text data is input into the anomaly detection model, and the anomaly detection result is output, including:

[0166] S6-1, taking the patient's basic information, diagnosis information, prescribed medical information, examination and test results in the patient's medical information as knowledge nodes, and taking the relationships between the patient's basic information, diagnosis information, prescribed medical information, examination and test results as edges, to construct a corresponding knowledge graph;

[0167] S6-2, based on the knowledge graph, inputting the updated text data into the NLP module, and outputting relevant medical information data; the relevant medical information data is the keywords of the prescribed medical advice information in the updated text data;

[0168] S6-3, inputting the relevant medical information data into the isolation forest model to obtain an abnormality detection result; the abnormality detection result includes abnormal information of the prescribed medical order and normal information of the prescribed medical order;

[0169] The training process of the isolation forest model:

[0170] S6-3-1. Obtain historical relevant medical information data and perform random sampling to obtain sampled historical relevant medical information data as a subsample set;

[0171] S6-6-2. Build an isolated decision tree corresponding to each sub-sample set, and randomly select features and split points in the sub-sample set to divide the isolated decision tree into a left subtree and a right subtree;

[0172] S6-6-3, repeat S6-6-2 until all isolated decision trees corresponding to sub-samples are isolated, completing the construction of the isolation forest model;

[0173] S6-6-4, inputting the historical relevant medical information data into the isolation forest model, and calculating the path length h(x) of the historical relevant medical information data on each decision tree;

[0174] S6-6-5. Based on the path length, calculate the average path length corresponding to the historical related medical information data;

[0175] S6-6-6. Based on the path length and the average path length, calculate the anomaly score s(X,n), and the corresponding formula is:

[0176]

[0177] Where n represents the number of historical relevant medical information data, H(n-1) represents the n-1th harmonic number, X represents the subsample, h(X) represents the path length of subsample X in a decision tree, E(h(X)) represents the average path length of the path length h(X) of subsample X in all decision trees, and c(n) represents the expected path length;

[0178] S6-6-7. Based on the anomaly score, determine the training anomaly detection result; if the anomaly score is greater than a threshold, the medical advice in the historical related medical information data is used as the training anomaly detection result; wherein the threshold is 0.5.

[0179] S6-6-8. Based on the training anomaly detection result, the parameters of the isolation forest model are adjusted to complete the training of the isolation forest model.

[0180] S6-4, voice broadcast the issued medical order information corresponding to the abnormal detection result of the issued medical order information abnormality, wait for the voice reply of the medical staff, and enter S7. After the medical staff of the inpatient department completes the daily patient inspection, after the doctor and the nurse have voice communication, a voice prompt is given after a logical error is detected.

[0181] Suppose a patient with chest pain is initially diagnosed with unstable angina, and the doctor issues medical orders including bed rest, sublingual nitroglycerin, ECG monitoring and other conventional treatments for angina. In subsequent electrocardiogram and myocardial enzyme tests, it was found that the patient's myocardial enzymes (such as troponin) increased sharply, and the electrocardiogram showed ST segment elevation with an arched back, which is typical of acute myocardial infarction. This indicates a conflict with the doctor's diagnosis and medical orders, and the doctor needs to be urgently notified to adjust the medical orders and change them to the treatment plan for acute myocardial infarction. Therefore, when the doctor issues treatment measures, the model detects problems based on the latest patient information medical knowledge base data and voice reminds the reasons.

[0182] The doctor's diagnosis and medical advice are input as updated text data into the NLP module to extract relevant medical information, that is, the voice input is bed rest, nitroglycerin, and the output is patient demand: bed rest, medicine: nitroglycerin. Among them, since there are many types of medical advice given by doctors, there are about 30,000 classified according to the name of the medical advice, so it is temporarily set to 30,000 data, and each medical advice name is encoded into a 30,000-dimensional vector using the bag of words model, where each dimension corresponds to a medical advice name. Assuming that a patient's medical advice is: {bed rest, nitroglycerin}, it is converted into a vector X = [1, 0, 0, ..., 1, 0] (30,000 dimensions, where the corresponding medical advice position is 1).

[0183] The abnormal score of the doctor's order is calculated. If the abnormal score is 0.7, which is greater than the threshold of 0.5, it means that the order is abnormal. The doctor's order is reported by voice, and the voice prompt is: "Since the patient's electrocardiogram shows ST segment elevation and elevated myocardial enzyme levels, it is recommended to adjust the treatment plan to acute myocardial infarction." Wait for the doctor's reply.

[0184] S7. Based on the abnormality detection result, adjust the updated text data, and store the adjusted text data in a patient medical information knowledge base.

[0185] like Figure 3 As shown, a medical assistance system based on speech recognition and medical information knowledge base includes a medical information acquisition module, a speech recognition module, a text matching correction module, an operation prediction module, a data update module, an abnormality detection module, a data adjustment module and a medical data storage module; wherein;

[0186] Medical information acquisition module, used to obtain medical information of patients in the inpatient department;

[0187] A speech recognition module is used to recognize the patient's medical information through an intelligent speech recognition system to obtain a speech recognition result;

[0188] A text matching and correction module, used to match and correct the speech recognition result to obtain corrected text data;

[0189] An operation prediction module is used to predict the corrected text data through a prediction model to obtain a prediction result;

[0190] A data updating module, used to give a voice prompt of the prediction result, obtain feedback from medical staff on the prediction result, update the prediction model or execute the prediction result and update the corrected text data;

[0191] An anomaly detection module is used to detect the updated text data using the anomaly detection model and output an anomaly detection result;

[0192] A data adjustment module, used for adjusting the updated text data based on the abnormality detection result;

[0193] The medical data storage module is used to store the medical information of patients in the inpatient department and the adjusted text data storage.

[0194] The medical assistance system is suitable for the inpatient department of the hospital, especially after the medical staff completes the daily inspection, the medical data corresponding to each patient needs to be updated and stored.

[0195] In summary, the present invention utilizes an intelligent voice recognition system to input patient medical information and performs matching correction, which can improve the accuracy and efficiency of voice operations in medical scenarios and reduce text errors caused by manual input or data entry; utilizes a predictive model and historical patient medical information to determine the habits of medical staff, predict the next operation of the medical staff, and execute the predicted operation, which can reduce the time for manual search and call of data, greatly reduce the waste of time resources, and effectively improve the work efficiency of medical staff; utilizes an anomaly detection model to detect whether the medical information formulated by medical staff is incorrect, and reminds medical staff to prevent medical accidents; provides an intelligent medical assistance system, establishes a comprehensive patient medical information knowledge base, provides intelligent operation prediction and prompt functions, improves user experience, and supports the intelligence of multiple medical operations.

[0196] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A medical assistance method based on speech recognition and medical information knowledge base, characterized in that: include: S1. Obtain the medical information of patients in the inpatient department and store it in the patient medical information knowledge base; S2, inputting the patient's medical information into an intelligent speech recognition system for recognition, and outputting a speech recognition result; the intelligent speech recognition system is a PDA model; S3, matching and correcting the speech recognition result to obtain corrected text data; S4, inputting the corrected text data into a prediction model, and outputting a prediction result; the prediction model adopts a sequence-to-sequence model; S5, giving a voice prompt for the prediction result, obtaining feedback from medical staff on the prediction result, updating the prediction model or executing the prediction result and updating the corrected text data; S6. Input the updated text data into an anomaly detection model, and output an anomaly detection result; the anomaly detection model includes an NLP module, a knowledge graph module and an isolation forest model; S7. Based on the abnormality detection result, adjust the updated text data, and store the adjusted text data in a patient medical information knowledge base.

2. A medical assistance method based on speech recognition and medical information knowledge base according to claim 1, characterized in that: The patient medical information includes the patient's basic information, diagnosis information, information on issued medical orders, historical information on all issued medical orders in the current department, medication status, examination and test results, treatment and surgery records, nursing records, health assessment, lifestyle, and psychological assessment.

3. A medical assistance method based on speech recognition and medical information knowledge base according to claim 2, characterized in that: The patient medical information knowledge base includes a parallel basic information module, a diagnosis information module, an issued medical order information module, a historical information module of all issued medical orders in the current department, a medication status module, an examination and test result module, a treatment and surgery record module, a nursing record module, a health assessment module, a lifestyle module, and a psychological assessment module; each module stores the corresponding patient medical information in chronological order.

4. A medical assistance method based on speech recognition and medical information knowledge base according to claim 1, characterized in that: The matching and correction of the speech recognition result to obtain corrected text data includes: S3-1, segmenting the speech recognition result to obtain corresponding segmented text data; S3-2, calculating the edit distance between each of the text data and the patient medical information in the patient medical information knowledge base, and selecting the minimum edit distance corresponding to each of the text data; S3-3, based on the minimum edit distance, matching each of the text data with the patient medical information in the patient medical information knowledge base to obtain a corresponding matching result; S3-4. Based on the matching result, proofread each of the text data to obtain the corrected text data.

5. A medical assistance method based on speech recognition and medical information knowledge base according to claim 1, characterized in that: The sequence-to-sequence model is a Seq2Seq model; the Seq2Seq model includes an encoder and a decoder; the encoder and the decoder both use LSTM units.

6. A medical assistance method based on speech recognition and medical information knowledge base according to claim 1, characterized in that: The step of providing a voice prompt for the prediction result, obtaining feedback from medical staff on the prediction result, updating the prediction model, or executing the prediction result and updating the corrected text data includes: S5-1, voice broadcasting the prediction result, and collecting the feedback of the PDA model medical staff through the intelligent voice recognition system; the prediction result is the prediction of the next operation of the medical staff; the feedback is the answer of the medical staff, that is, "yes" or "no"; S5-2, judging whether the feedback is "yes"; if so, executing the corresponding prediction result, updating the corrected text data, and entering S6; otherwise, training the sequence-to-sequence model, and entering S5-3; S5-3, recording the feedback of "No" and the corresponding prediction results and corrected text data as the historical data for the sequence-to-sequence model training; S5-4, inputting the historical data into the sequence-to-sequence model, and constructing a corresponding objective function; the objective function adopts a cross entropy loss function; S5-6. Based on the objective function, the Adam optimizer is used to adjust the model parameters of the sequence-to-sequence model to obtain an updated sequence-to-sequence model; the updated sequence-to-sequence model is used for the next round of prediction.

7. A medical assistance method based on speech recognition and medical information knowledge base according to claim 2, characterized in that: The updated text data is input into the anomaly detection model, and the anomaly detection result is output, including: S6-1, taking the patient's basic information, diagnosis information, prescribed medical information, examination and test results in the patient's medical information as knowledge nodes, and taking the relationships between the patient's basic information, diagnosis information, prescribed medical information, examination and test results as edges, to construct a corresponding knowledge graph; S6-2, based on the knowledge graph, inputting the updated text data into the NLP module, and outputting relevant medical information data; the relevant medical information data is the keywords of the prescribed medical advice information in the updated text data; S6-3, inputting the relevant medical information data into the isolation forest model to obtain an abnormality detection result; the abnormality detection result includes abnormal information of the prescribed medical order and normal information of the prescribed medical order; S6-4, broadcast the issued medical order information corresponding to the abnormal detection result that the issued medical order information is abnormal, wait for the voice reply of the medical staff, and enter S7.

8. A medical assistance method based on speech recognition and medical information knowledge base according to claim 7, characterized in that: The training process of the isolation forest model: S6-3-1. Obtain historical relevant medical information data and perform random sampling to obtain sampled historical relevant medical information data as a subsample set; S6-6-2. Build an isolated decision tree corresponding to each sub-sample set, and randomly select features and split points in the sub-sample set to divide the isolated decision tree into a left subtree and a right subtree; S6-6-3, repeat S6-6-2 until all isolated decision trees corresponding to sub-samples are isolated, completing the construction of the isolation forest model; S6-6-4, inputting the historical relevant medical information data into the isolation forest model, and calculating the path length of the historical relevant medical information data on each decision tree; S6-6-5. Based on the path length, calculate the average path length corresponding to the historical related medical information data; S6-6-6. Calculate an anomaly score based on the path length and the average path length; S6-6-7. Determine a training anomaly detection result based on the anomaly score; S6-6-8. Based on the training anomaly detection result, the parameters of the isolation forest model are adjusted to complete the training of the isolation forest model.

9. A medical assistance system based on speech recognition and medical information knowledge base, used to implement a medical assistance method based on speech recognition and medical information knowledge base as described in any one of claims 1 to 9, characterized in that: It includes a medical information acquisition module, a speech recognition module, a text matching and correction module, an operation prediction module, a data update module, an anomaly detection module, a data adjustment module and a medical data storage module; in; Medical information acquisition module, used to obtain medical information of patients in the inpatient department; A speech recognition module is used to recognize the patient's medical information through an intelligent speech recognition system to obtain a speech recognition result; A text matching and correction module, used to match and correct the speech recognition result to obtain corrected text data; An operation prediction module is used to predict the corrected text data through a prediction model to obtain a prediction result; A data updating module, used to give a voice prompt of the prediction result, obtain feedback from medical staff on the prediction result, update the prediction model or execute the prediction result and update the corrected text data; An anomaly detection module is used to detect the updated text data using the anomaly detection model and output an anomaly detection result; A data adjustment module, used for adjusting the updated text data based on the abnormality detection result; The medical data storage module is used to store the medical information of patients in the inpatient department and the adjusted text data storage.

Images