Emergency level determination method, device and equipment for patient and storage medium

Abstract

The application discloses a method, device and equipment for determining an emergency level of a patient, and a storage medium. The method comprises the following steps: determining at least one first symptom of the patient; determining a first main symptom and at least one first accompanying symptom from the at least one first symptom; determining a target reference accompanying symptom from at least one reference accompanying symptom in a preset first accompanying symptom library according to a first similarity between each reference accompanying symptom and each reference accompanying symptom in the at least one reference accompanying symptom; the first accompanying symptom library is an accompanying symptom library corresponding to the first main symptom; and determining an emergency level corresponding to the target reference accompanying symptom as the emergency level of the patient.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and to, but is not limited to, a method, apparatus, device, and storage medium for determining the emergency level of a patient. Background Technology

[0002] In related technologies, medical staff typically input numerical vital sign data of emergency patients measured by instruments into the interface of the hospital information system. In the main complaint text box of the interface, they select multiple symptom descriptive words from a preset symptom dictionary. For symptom descriptive words not in the symptom dictionary, medical staff can manually input them. Finally, medical staff manually set the emergency level to a certain level from 1 to 4 and manually input it into the interface of the hospital information system.

[0003] However, due to the limited symptom lexicon, it is difficult to accurately describe the current condition of emergency patients. In addition, the manual grading method requires medical staff to be proficient in emergency grading standards and have rich medical experience; otherwise, it is easy to cause unreasonable grading. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus, device, and storage medium for determining the emergency level of a patient.

[0005] In a first aspect, embodiments of this application provide a method for determining the emergency level of a patient, the method comprising: determining at least one first symptom of the patient; determining a first primary symptom and at least one first accompanying symptom from the at least one first symptom; determining a target reference accompanying symptom from the at least one reference accompanying symptom based on a first similarity between each of the first accompanying symptom and each of the reference accompanying symptoms in a preset first accompanying symptom database; the first accompanying symptom database being an accompanying symptom database corresponding to the first primary symptom; and determining the emergency level corresponding to the target reference accompanying symptom as the emergency level of the patient.

[0006] Secondly, embodiments of this application provide an emergency level determination device for a patient, comprising: a first determination module for determining at least one first symptom of the patient; a classification module for determining a first primary symptom and at least one first accompanying symptom from the at least one first symptom; a screening module for determining a target reference accompanying symptom from the at least one reference accompanying symptom based on a first similarity between each of the first accompanying symptom and each of the reference accompanying symptoms in a preset first accompanying symptom database; the first accompanying symptom database being an accompanying symptom database corresponding to the first primary symptom; and a second determination module for determining the emergency level corresponding to the target reference accompanying symptom as the emergency level of the patient.

[0007] Thirdly, embodiments of this application provide an apparatus including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the steps in the method for determining the emergency level of a patient as described in the first aspect of embodiments of this application.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for determining the emergency level of a patient as described in the first aspect of embodiments of this application.

[0009] In this embodiment, by identifying the first primary symptom and the first accompanying symptom from the patient's first symptom, and determining the target reference accompanying symptom based on the similarity between the first accompanying symptom and the reference accompanying symptom in the first accompanying symptom database, and determining the emergency level corresponding to the target reference accompanying symptom as the patient's emergency level, the patient's emergency level can be accurately, reasonably, and efficiently determined, so that emergency resources can be used rationally. Attached Figure Description

[0010] Figure 1 This is a schematic diagram illustrating an application scenario of a method for determining the emergency level of a patient, as described in an embodiment of this application.

[0011] Figure 2 This is a flowchart illustrating a method for determining the emergency level of a patient according to an embodiment of this application.

[0012] Figure 3 This is a schematic diagram illustrating a process for determining a first symptom using a trained named entity recognition network, according to an embodiment of this application.

[0013] Figure 4 This is a schematic diagram of a process for determining a patient's first symptom according to an embodiment of this application;

[0014] Figure 5 This is a schematic diagram of a reference main symptom node according to an embodiment of this application;

[0015] Figure 6 This is a schematic diagram of a first accompanying symptom database according to an embodiment of this application;

[0016] Figure 7 This is a flowchart illustrating a training method for a feature extraction model according to an embodiment of this application;

[0017] Figure 8 This is a schematic diagram of a feature extraction process based on a BERT network according to an embodiment of this application;

[0018] Figure 9 This is a schematic diagram illustrating an embodiment of the present application for extracting one-dimensional features of a first accompanying symptom;

[0019] Figure 10 This is a flowchart illustrating a method for acquiring patient condition data according to an embodiment of this application;

[0020] Figure 11 This is a flowchart illustrating a loss function calculation method according to an embodiment of this application;

[0021] Figure 12 This is a schematic diagram of the composition of a patient emergency level determination device according to an embodiment of this application;

[0022] Figure 13 This is a schematic diagram of a hardware entity of the device according to an embodiment of this application. Detailed Implementation

[0023] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0024] Figure 1 This is a schematic diagram illustrating an application scenario of a method for determining the emergency level of a patient according to an embodiment of this application. See also... Figure 1 Medical staff can voice-input patient symptoms, i.e., a description of the condition 102, into the front-end interface 101 (also known as the HIS system interface) of the Hospital Information System (HIS). The description of the condition 102 can be a voice description of the condition. The front-end interface 101 sends the voice description of the condition to the back-end triage service 103 (also known as the triage service) of the HIS. The back-end triage service 103 obtains the text description of the condition through speech-to-text technology. The back-end triage service 103 can determine the patient's emergency level (i.e., the triage result) based on the text description of the condition 104, and send the emergency level to the front-end interface 101 for display, so that medical staff can record and view it.

[0025] In another embodiment, medical staff can directly input patient symptoms, i.e., condition description 102, into the front-end interface 101. The condition description can be a text description of the condition. The front-end interface 101 sends the condition description text to the back-end triage service 103 of the HIS. The back-end triage service 103 can determine the patient's emergency level based on the condition description text and send the emergency level to the front-end interface 101 for display, so that medical staff can record and view it.

[0026] Figure 2 This is a flowchart illustrating a method for determining the emergency level of a patient according to an embodiment of this application. Figure 2 As shown, the method includes:

[0027] Step 202: Identify at least one primary symptom in the patient;

[0028] The patient may be an emergency room visitor; the first symptom may be a numerical single performance indicator, such as a body temperature of 36 degrees Celsius, a respiratory rate of 25 breaths per minute, a heart rate of 100 beats per minute, etc.; the first symptom may also be a text-based risk performance indicator, such as "cardiac arrest", "persistent chest pain", "active or severe blood loss", "tinnitus", "headache" and "vomiting", etc.

[0029] Among them, the values ​​of individual performance indicators can be directly obtained from precision testing instruments, and the thresholds of emergency levels and corresponding indicator items are relatively clear, making it easy to classify emergency cases based on individual performance indicators; however, text-based risk performance indicators may come from descriptions by patients or medical staff, and due to the richness of language, it is difficult to completely match the descriptions of standard risk indicator items in the classification standards.

[0030] Step 204: Identify a first primary symptom and at least one first accompanying symptom from the at least one primary symptom;

[0031] The at least one accompanying symptom can be any first symptom other than the first primary symptom.

[0032] Step 206: Based on the first similarity between each of the first accompanying symptoms and at least one of the reference accompanying symptoms in the preset first accompanying symptom database, determine the target reference accompanying symptom from the at least one reference accompanying symptoms; the first accompanying symptom database is the accompanying symptom database corresponding to the first primary symptom.

[0033] Each primary symptom can correspond to a database of accompanying symptoms. Table 1 shows a database of accompanying symptoms provided in this application embodiment. Referring to Table 1, the database of accompanying symptoms includes multiple reference accompanying symptoms belonging to different emergency levels. Level 1 reference accompanying symptoms include cardiac arrest, respiratory arrest, airway obstruction or suffocation, signs of shock, and confirmed myocardial infarction, etc.; Level 2 reference accompanying symptoms include severe dyspnea, circulatory disorders, acute stroke, persistent chest pain, and acute poisoning, etc.; Level 3 reference accompanying symptoms include acute asthma, head injury, intermittent epileptic seizures, persistent vomiting, and moderate non-cardiac chest pain, etc.; Level 4 reference accompanying symptoms include stable condition, mild symptoms, mild pain without dangerous features, and mild mental and behavioral abnormalities, etc. As can be seen from Table 1, Level 1 is the highest level with the most severe symptoms, and Level 4 is the lowest level with the mildest symptoms.

[0034] Table 1

[0035] level Accompanying symptom index database 1 Cardiac arrest / respiratory arrest / airway obstruction or suffocation / signs of shock / confirmed myocardial infarction... 2 Severe respiratory distress / circulatory disturbance / acute stroke / persistent chest pain / acute poisoning... 3 Acute asthma / head trauma / intermittent seizures / persistent vomiting / moderate non-cardiac chest pain... 4 Stable condition / mild symptoms / minor pain without dangerous features / mild mental and behavioral abnormalities...

[0036] Step 208: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0037] If the target reference accompanying symptoms include grade 1 "cardiac arrest" and grade 3 "acute asthma", then the highest grade 1 can be used as the emergency level for the patient.

[0038] It should be noted that the emergency triage method for patients proposed in the embodiments of this application can be used to perform emergency triage of single performance indicators and emergency triage of risk performance indicators, respectively. The triage results of the two indicators are then comprehensively analyzed, and the comprehensive analysis result is used as the emergency level of the patient.

[0039] The emergency triage method for patients proposed in the embodiments of this application can also be used to classify emergency risk performance indicators. The correspondence between emergency level and the value of individual performance indicators can be used to classify individual performance indicators. The classification results of the two indicators are comprehensively analyzed, and the comprehensive analysis result is used as the emergency level of the patient.

[0040] In this embodiment, by identifying the first primary symptom and the first accompanying symptom from the patient's first symptom, and determining the target reference accompanying symptom based on the similarity between the first accompanying symptom and the reference accompanying symptom in the first accompanying symptom database, and determining the emergency level corresponding to the target reference accompanying symptom as the patient's emergency level, the patient's emergency level can be accurately, reasonably, and efficiently determined, so that emergency resources can be used rationally.

[0041] This application also provides a method for determining the emergency level of a patient, the method comprising:

[0042] Step S202: Obtain the patient's condition description information;

[0043] The description of the illness may include the patient's onset time and main symptoms, and may include "intermittent tinnitus, acute headache and vomiting", "cardiac arrest, severe blood loss", etc.

[0044] Step S204: Based on the description of the illness and the trained symptom recognition model, determine at least one first symptom of the patient;

[0045] The description of the illness can be a voice description, text description, or video description; the symptom recognition model can be a Name Entity Recognition (NER) network. A Name Entity Recognition network can be constructed by connecting a Bidirectional Encoder Representation from Transformers (BERT) network followed by a Long Short-Term Memory (LSTM) network and a Conditional Random Field (CRF) network.

[0046] Figure 3 This is a schematic diagram illustrating a process for determining a first symptom using a trained named entity recognition network, as described in an embodiment of this application. (See also...) Figure 3 The description of the illness text 301 is processed by the BERT network 302, LSTM network 303 and CRF network 304 of the named entity recognition network to identify multiple first symptoms 305, including symptom 1, symptom 2, ... symptom n.

[0047] The BERT network is a pre-trained network model primarily used to extract features from the text describing the patient's condition. In BERT, position embedding tells the model the positional information of the input words, meaning that positional information is weakened during computation. However, positional information is crucial in sequence labeling tasks, and even directional information is essential. Therefore, this embodiment uses LSTM to learn the dependencies in the observation sequence, and finally uses CRF to learn the relationships in the state sequence, outputting a label for each input word.

[0048] Figure 4 This is a flowchart illustrating a method for determining a patient's first symptom according to an embodiment of this application. See also... Figure 4 The system can input the description information of the illness 401 "tinnitus, acute headache and vomiting" into the named entity recognition network 402, and the named entity recognition network 402 will output the corresponding first symptoms 403 "tinnitus", "headache" and "vomiting".

[0049] Step S206: Based on the second similarity between each of the first symptoms and each of the reference main symptoms in the preset main symptom database, determine the first main symptom and at least one first accompanying symptom from the at least one first symptom;

[0050] See Figure 4In one embodiment, based on the similarity between the first symptom and the reference primary symptom, "tinnitus" can be identified as the first primary symptom 404 from the first symptoms 403 "tinnitus", "headache" and "vomiting", "headache" and "vomiting" can be identified as the first accompanying symptoms 405, and the clause containing the first accompanying symptom 405, "acute headache and vomiting", can be the accompanying symptom description clause 406.

[0051] In one embodiment, a main symptom database can be built based on the neo4j graph database. The main symptom database can also be called a main symptom knowledge graph. The main symptom database includes multiple reference main symptoms. Each reference main symptom is an independent node, and each node can be associated with its replacement nodes. Figure 5 This is a schematic diagram of a reference main symptom node according to an embodiment of this application. See also... Figure 5 The abdominal pain node 50 may include lower abdominal pain 501, lower abdominal distension 502, lower abdominal cramping 503, upper abdominal pain 504, and right lower quadrant tenderness 505. If the named entity recognition symptom part identifies... Figure 5 Any of the reference main symptoms can be specified as "abdominal pain".

[0052] Step S208: Based on the first similarity between each of the first accompanying symptoms and at least one reference accompanying symptoms in a preset first accompanying symptom database, a target reference accompanying symptom is determined from the at least one reference accompanying symptoms; the first accompanying symptom database is an accompanying symptom database corresponding to the first primary symptom.

[0053] Step S210: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0054] In this embodiment, the patient's first symptom is determined based on the patient's condition description information and symptom recognition model, thereby improving the accuracy and intelligence of symptom determination; the first primary symptom is determined by calculating the similarity between the first symptom and the reference primary symptoms in the preset primary symptom database, thereby improving the accuracy of the determined first primary symptom.

[0055] This application also provides a method for determining the emergency level of a patient, the method comprising:

[0056] Step S302: Obtain the patient's voice description of their condition and convert the voice description of their condition into text description of their condition; or, obtain the patient's text description of their condition.

[0057] Step S304: Based on the description of the illness and the trained symptom recognition model, determine at least one first symptom of the patient;

[0058] Step S306: Determine the second similarity between each of the first symptoms and each of the reference main symptoms according to the preset matching order;

[0059] Step S308: Identify the reference main symptom corresponding to the highest second similarity as the candidate main symptom;

[0060] Step S310: Determine the first primary symptom based on the candidate primary symptoms;

[0061] Step S312: Identify the other primary symptoms besides the primary symptom as primary accompanying symptoms;

[0062] Step S314: Determine the first textual feature for each of the first accompanying symptoms;

[0063] Step S316: Determine the second textual features for each reference accompanying symptom;

[0064] Step S318: Determine the similarity between each of the first text features and each of the second text features;

[0065] Step S320: Identify the reference accompanying symptoms corresponding to the second text features with similarity greater than the first threshold as the target reference accompanying symptoms;

[0066] Step S322: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0067] Specifically, the second textual features of each reference accompanying symptom in the first accompanying symptom database can be extracted. The second textual features can be represented by a one-dimensional vector [y1, y2, ... y]. n [x1, x2, ..., x] represents the first text feature of the clause containing the first accompanying symptom, and the first text feature can be represented by a one-dimensional vector [x1, x2, ..., x]. n This application embodiment indicates that the cosine similarity algorithm can be used to calculate the similarity between two vectors. The similarity value between the vectors is the similarity value between the two statements; a threshold T can be preset based on experience. If the similarity value is greater than the threshold T, it is considered to match this standard risk statement (i.e., the target reference accompanying symptom), and the patient is classified according to the emergency level to which the standard risk statement belongs; if there are multiple reference accompanying symptoms with similarity data greater than T, the reference accompanying symptom with the first matched similarity value greater than the threshold T can be determined as the target reference accompanying symptom according to the matching order. The cosine similarity calculation method of the first text feature and the second text feature is shown in the following formula (1):

[0068]

[0069] Figure 6This is a schematic diagram of a first accompanying symptom database according to an embodiment of this application. See also... Figure 6 The patient's voice description of the condition 61 can be converted into text description of the condition 62. Based on the text description of the condition 62, the first symptom 63 (i.e., symptom entity) is determined. From the first symptom 63, the first main symptom 631 and the first accompanying symptom 632 are determined. The first text feature of each clause containing the first accompanying symptom 632 is extracted using the BERT network.

[0070] The first primary symptom 631 (i.e., primary symptom 1) corresponds to the first accompanying symptom database 6311, which includes multiple first accompanying symptoms belonging to different emergency levels, such as "cardiac arrest" at level 1, "severe dyspnea" at level 2, "acute asthma" at level 3, and "stable condition" at level 4. The second text feature of each of the reference accompanying symptoms can be extracted using the BERT network, and the similarity between each of the first text feature and each of the second text features can be calculated. The reference accompanying symptom with the first matching similarity value greater than the threshold T is determined as the target reference accompanying symptom, and the emergency level (i.e. emergency treatment level) corresponding to the target reference accompanying symptom is determined as the emergency level of the patient.

[0071] In this embodiment, the patient's voice description of their condition can be obtained, converted into text, and then input into the symptom recognition model. Alternatively, the patient's text description of their condition can be directly obtained and input into the symptom recognition model, thus improving the diversity of the obtained condition description information. By determining the similarity between the first text feature and the second text feature, a reference accompanying symptom with a high similarity to the first accompanying symptom can be more accurately identified as the target reference accompanying symptom.

[0072] This application also provides a method for determining the emergency level of a patient, the method comprising:

[0073] Step S402: Obtain the patient's voice description of their condition and convert the voice description of their condition into text description of their condition; or, obtain the patient's text description of their condition.

[0074] Step S404: Based on the description of the illness and the trained symptom recognition model, determine at least one first symptom of the patient;

[0075] Step S406: Determine the second similarity between each of the first symptoms and each of the reference main symptoms according to a preset matching order;

[0076] Step S408: Identify the reference main symptom corresponding to the highest second similarity as the candidate main symptom;

[0077] Wherein, the reference main symptom includes the normative main symptom and the non-normative main symptom associated with the normative main symptom, the candidate main symptom also includes the normative main symptom and the associated non-normative main symptom; see also Figure 5 The standard primary symptom is "abdominal pain," and non-standard primary symptoms associated with "abdominal pain" include "lower abdominal pain," "lower abdominal distension," "lower abdominal colic," "upper abdominal pain," and "right lower abdominal tenderness," etc.

[0078] Step S410: If the candidate primary symptom is a standard primary symptom, the candidate primary symptom is determined as the first primary symptom;

[0079] Step S412: If the candidate primary symptom is an irregular primary symptom, determine the regular primary symptom associated with the candidate primary symptom as the first primary symptom; see [link to relevant documentation]. Figure 5 If the candidate primary symptom is "abdominal pain", then "abdominal pain" can be identified as the primary symptom. If the candidate primary symptom is any one of "lower abdominal pain", "lower abdominal distension", "lower abdominal colic", "upper abdominal pain" and "right lower abdominal tenderness", then the associated "abdominal pain" can be identified as the primary symptom.

[0080] Step S414: Identify the other first symptoms besides the first primary symptom as the first accompanying symptoms;

[0081] Step S416: Determine the first textual feature for each of the first accompanying symptoms;

[0082] Step S418: Determine the second textual features for each reference accompanying symptom;

[0083] Step S420: Determine the similarity between each of the first text features and each of the second text features;

[0084] Step S422: Identify the reference accompanying symptoms corresponding to the second text features with similarity greater than the first threshold as the target reference accompanying symptoms;

[0085] Step S424: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0086] In this embodiment of the application, when the candidate main symptom is not standardized, the standardized main symptom associated with the candidate main symptom can be determined as the first main symptom, thereby improving the standardization of the determination of the first main symptom.

[0087] This application also provides a method for determining the emergency level of a patient, the method comprising:

[0088] Step S502: Obtain at least one historical medical record text;

[0089] Figure 7This is a flowchart illustrating a training method for a feature extraction model according to an embodiment of this application. See also... Figure 7 The historical medical record text 701 is the collected sample data, which can be "Hepatitis B surface antigen and antibody test result positive". The total length of the historical medical record text is 11, that is, it includes 11 characters.

[0090] Step S504: Replace a portion of the text at a preset replacement position in each of the historical medical condition texts to obtain the corresponding historical medical condition sample;

[0091] See Figure 7 A certain percentage (e.g., 15%) of the characters, approximately two characters, such as "liver" and "positive", can be selected from the entire historical medical record text "hepatitis B two-and-a-half test result is positive" to construct a training sample. The replaced sentence "hepatitis B two-and-a-half test result is [MASK]" is used as the historical medical record text 702.

[0092] Step S506: Extract the fourth textual features of each of the historical disease samples using a feature extraction model;

[0093] The feature extraction module is a deep learning model for extracting text features. The feature extraction model can be a BERT network. Words replaced with [MASK] can be considered as covered. The feature extraction model predicts the covered words based on the context.

[0094] Figure 8 This is a schematic diagram of a feature extraction process based on a BERT network according to an embodiment of this application. See [link to relevant documentation]. Figure 8 Training sample sequence 801 is input into BERT network 80. The training sample can be historical disease samples. After passing through BERT network 80, sample features 802 are output. The sample features can be fourth text features. The BERT network includes an embedding layer 803, a multi-head attention layer 804, a feed forward network layer 805, and an add&norm layer 806. Sequence 801 first enters the embedding layer 803, and then reaches the multi-head attention layer 804, the feed forward network layer 805, and the add&norm layer 806. The outputs of the multi-head attention layer 804 and the feed forward network layer 805 are sent to the add&norm layer 806 for processing. The add&norm layer 806 includes residual structures and layer normalization.

[0095] Step S508: Use the feature extraction model to extract the fifth text feature at the preset replacement position in each of the historical medical texts;

[0096] Among them, the fifth text feature 704 corresponding to the two [MASK] positions (i.e., the preset replacement positions) can be extracted from the fourth text feature 703.

[0097] Step S510: Based on the text feature dictionary and each of the fifth text features, determine the word at the preset replacement position in each of the historical disease texts;

[0098] Specifically, based on the text feature dictionary and each of the fifth text features, it is possible to predict the character 705 mapped to the [MASK] position in the dictionary, that is, the position of the character at the [MASK] position in the text feature dictionary is predicted.

[0099] Step S512: Update the feature extraction model based on the character at each preset replacement position and the one-hot encoding of each preset replacement position;

[0100] Step S514: Obtain the patient's voice description of their condition and convert the voice description of the condition into text description of the condition; or, obtain the patient's text description of their condition.

[0101] Step S516: Based on the description of the illness and the trained symptom recognition model, determine at least one first symptom of the patient;

[0102] Step S518: Determine the second similarity between each of the first symptoms and each of the reference main symptoms according to the preset matching order;

[0103] Step S520: When there are multiple reference main symptoms corresponding to the highest second similarity, the reference main symptom with the highest matching order among the reference main symptoms corresponding to the highest second similarity is determined as the candidate main symptom;

[0104] Among them, the first matched reference main symptom can be identified as the candidate main symptom.

[0105] Step S522: If the candidate primary symptom is a standard primary symptom, the candidate primary symptom is determined as the first primary symptom;

[0106] Step S524: If the candidate main symptom is an irregular main symptom, determine the regular main symptom associated with the candidate main symptom as the first main symptom;

[0107] Step S526: Identify the other primary symptoms besides the primary symptom as primary accompanying symptoms;

[0108] Step S528: Extract the third text features for each of the first accompanying symptoms using the trained feature extraction model;

[0109] See Figure 8The training sample sequence 801 is input into the BERT network 80. The training sample can be the clause containing the first accompanying symptom. After passing through the BERT network 80, the sample feature 802 is output. The sample feature can be the third text feature.

[0110] Step S530: Perform average pooling on each of the third text features to obtain the first text feature corresponding to the first accompanying symptom;

[0111] in, Figure 9 This is a schematic diagram illustrating the extraction of one-dimensional features of a first accompanying symptom according to an embodiment of this application. See also... Figure 9 The text features of the clause containing the first accompanying symptom 901 can be extracted based on the pre-trained BERT network model 91, namely the third text feature 902. The dimension of the third text feature 902 can be represented as [sequence_length, embedding_size]. After average pooling (Avg Pool) 903, a one-dimensional vector with the dimension [embedding_size] is output to represent the features of the entire sentence, namely the first text feature 904 of the first accompanying symptom.

[0112] Step S532: Determine the second textual features for each reference accompanying symptom;

[0113] Step S534: Determine the similarity between each of the first text features and each of the second text features;

[0114] Step S536: Identify the reference accompanying symptoms corresponding to the second text features with similarity greater than the first threshold as the target reference accompanying symptoms;

[0115] Step S538: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0116] In this embodiment, by updating and training the feature extraction model using historical medical text, the accuracy of feature extraction can be improved. When there are multiple reference main symptoms corresponding to the highest second similarity, a candidate main symptom can be determined from the reference main symptoms according to the matching order, thus identifying a unique candidate main symptom. By performing average pooling on the extracted text features, one-dimensional text features can be obtained, thereby reducing computation and memory consumption and simplifying network complexity.

[0117] This application also provides a method for determining the emergency level of a patient, the method comprising:

[0118] Step S602: Obtain the patient's historical medical history text from at least one medical service website using a web crawler engine;

[0119] Alternatively, one can search for patients' historical medical records in the electronic medical record database using keywords;

[0120] Figure 10 This is a flowchart illustrating a method for obtaining patient condition data according to an embodiment of this application. See also... Figure 10 The web crawler engine 1001 can crawl the question and answer data of the medical and health website database 1002 to obtain disease description information, such as "positive result of hepatitis B two-and-a-half test", as historical medical condition text. The historical medical condition text is then processed and stored in the database 1003, which can be a MySQL database. The medical and health website database 1002 can also be called a medical service website, which can be a number of pre-selected online medical and health consultation service websites that provide real doctors.

[0121] The crawler engine 1001 can also allocate and manage current crawler tasks. The crawling module 1004 can crawl patient condition data from various websites and pages in parallel through the downloader 1005 as historical condition text and send it to the crawler engine 1001. The crawler engine 1001 processes the crawling results, such as removing some useless words (interjections, auxiliary words, or punctuation marks, etc.). The processed data is then stored in the database 1003.

[0122] In addition, historical medical records can also be obtained from the hospital's electronic medical record database. By extracting symptom description data with keywords such as "chief complaint", "present illness", "past medical history", and "physical examination" from the electronic medical records of multiple patients, historical medical records can be obtained.

[0123] Step S604: Replace a portion of the text at a preset replacement position in each of the historical medical condition texts to obtain the corresponding historical medical condition sample;

[0124] Step S606: Using the embedding layer, each historical disease sample is segmented to obtain an ordered word set of the corresponding historical disease sample;

[0125] Step S608: Vectorize the words in each ordered word set to obtain the word matrix of the corresponding ordered word set;

[0126] The embedding layer can be obtained by summing Token Embedding, Segment Embeddings, and PositionEmbeddings.

[0127] Token Embeddings: These are word vectors (each word corresponds to a Chinese character). The first word is a CLS marker, which can be used for subsequent classification tasks, and the last word's SEP marker indicates the end of the sentence.

[0128] Segment Embeddings: Used to distinguish between two types of sentences, because the input to BERT can be a single sentence or two sentences. When there are two sentences, Segment Embeddings uses two identifiers to mark the two sentences respectively.

[0129] Position Embeddings: The position of each word in a sentence is very important information. Position Embeddings are used to extract the positional features of each word in the sentence.

[0130] Step S610: Using the transformation layer, based on each word matrix and the weight matrix of the feature extraction model, obtain the fourth text feature of each historical disease sample;

[0131] The weight matrices include Q, K, and V matrices, i.e., Query, Key, and Value matrices. The Transformer layer is composed of concatenated outputs from multiple Attention layers. The input to each Attention layer is the Q, K, and V matrices, which are obtained by performing different matrix transformations on the output (word matrix) of the Embedding layer. The Q and K matrices undergo scaling, followed by softmax processing, and then multiplied by the V matrix to obtain the final output. The three weight matrices, Query, Key, and Value, are used: Query is used to find relationships with other inputs, and Key is specifically used to handle inputs that are related to itself.

[0132] Step S612: Use the feature extraction model to extract the fifth text feature at the preset replacement position in each of the historical medical texts;

[0133] Step S614: Based on the text feature dictionary and each of the fifth text features, determine the word at the preset replacement position in each of the historical disease texts;

[0134] Step S616: Determine the cross-entropy loss value based on the word at each preset replacement position and the one-hot code at each preset replacement position;

[0135] Step S618: If the cross-entropy loss value is greater than the second threshold, use the gradient descent algorithm to adjust the weight matrix of the feature extraction model until the cross-entropy loss value is not greater than the second threshold.

[0136] The one-hot encoding, also known as one-hot encoding, uses the one-hot representation (the vector length is the length of the dictionary, where the value at the corresponding character position in the dictionary is 1, and the rest are 0) of the character (part of the text) replaced by [MASK] to represent the training label data. The vector of the predicted character at the preset replacement position is represented as the prediction data. The cross-entropy loss value between the prediction data and the label data is calculated as the final loss function. The process of iteratively optimizing the network weights of the weight matrix through gradient descent to continuously reduce the loss value is the update and training process of the feature extraction model.

[0137] Figure 11 This is a flowchart illustrating a loss function calculation method according to an embodiment of this application. See also... Figure 11 First, training samples 1101 with [MASK] tags can be obtained. Then, the full sentence features of the training samples can be extracted based on the BERT network 1102 to obtain the representation of the [MASK]-tagged words 1103. The one-hot representation of the [MASK]-tagged words 1104 can be obtained. The cross-entropy loss between the two can be calculated 1105 and the cross-entropy loss 1105 can be used as the loss.

[0138] Step S620: Obtain the patient's voice description of their condition and convert the voice description of their condition into text description of their condition; or, obtain the patient's text description of their condition.

[0139] Step S622: Based on the description of the illness and the trained symptom recognition model, determine at least one first symptom of the patient;

[0140] Step S624: Determine the second similarity between each of the first symptoms and each of the reference main symptoms according to the preset matching order;

[0141] Step S626: If there are multiple reference main symptoms corresponding to the highest second similarity, the reference main symptom with the highest matching order among the reference main symptoms corresponding to the highest second similarity is determined as the candidate main symptom;

[0142] Step S628: If the candidate primary symptom is a standard primary symptom, the candidate primary symptom is determined as the first primary symptom;

[0143] Step S630: If the candidate main symptom is an irregular main symptom, the regular main symptom associated with the candidate main symptom is determined as the first main symptom;

[0144] Step S632: Identify the other primary symptoms besides the primary symptom as primary accompanying symptoms;

[0145] Step S634: Extract the third text features for each of the first accompanying symptoms using the trained feature extraction model;

[0146] Step S636: Perform average pooling on each of the third text features to obtain the first text feature corresponding to the first accompanying symptom;

[0147] Step S638: Determine the second textual features for each reference accompanying symptom;

[0148] Step S640: Determine the similarity between each of the first text features and each of the second text features;

[0149] Step S642: Identify the reference accompanying symptoms corresponding to the second text features with similarity greater than the first threshold as the target reference accompanying symptoms;

[0150] Step S644: Determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0151] In this embodiment, the cross-entropy loss value is determined based on the word at each preset replacement position and the one-hot encoding at each preset replacement position. The cross-entropy loss value is used as a loss function to update and train the feature extraction model, thereby improving the accuracy of feature extraction by the feature extraction model.

[0152] This application's embodiments belong to the field of Natural Language Processing (NLP). It involves techniques such as web crawling, unsupervised learning, text feature extraction, and text feature similarity judgment. A large amount of unlabeled medical symptom description text is collected through web crawling, and a deep learning model for medical text feature extraction is trained using unsupervised learning techniques. By judging the similarity between the current patient's symptom description text and the grading standard text, the patient can be triaged to level 1 to 4.

[0153] With the rapid development of emergency medicine in my country and the increasing number of emergency visits year by year, emergency triage plays an increasingly important role as the primary step in emergency patient care. Emergency triage standards are divided into two types of indicators: individual indicators and standard risk indicators. Individual indicators are numerical indicators, also known as individual performance indicators, such as a body temperature of 36 degrees Celsius, a respiratory rate of 25 breaths per minute, and a heart rate of 100 beats per minute. These individual indicator values ​​can be directly obtained from sophisticated testing instruments, and the thresholds for emergency levels and corresponding indicator items are relatively clear, making it easy to classify emergencies based on individual performance indicators. Standard risk indicators are generally textual descriptions, also known as risk performance indicators, such as "cardiac arrest," "persistent chest pain," and "active or severe blood loss." This textual information may come from descriptions by patients or medical staff, and due to the richness of language, it is difficult to completely match the descriptions of standard risk indicators in the triage standards.

[0154] In related technologies, hospital information systems typically obtain numerical vital sign data of patients through instruments (such as thermometers, blood pressure monitors, etc.). After obtaining this data, medical staff manually input it into the HIS (Hospital Information System) interface. In the main complaint text box of the interface, multiple symptom descriptions are selected from a preset dictionary. For descriptions not in the dictionary, medical staff can manually type them in. Finally, medical staff manually classify the symptom to a level from 1 to 4 and manually input it into the HIS interface.

[0155] The HIS system in related technologies is based on a symptom lexicon. Due to the limited nature of the lexicon, it is difficult to accurately describe the current condition of the patient. In addition, the manual grading method requires medical staff to be proficient in the grading standards and have rich medical experience; otherwise, it is easy to cause unreasonable grading.

[0156] This application proposes a risk performance index-based triage and grading method based on unsupervised learning. Using deep learning, a network model capable of accurately extracting medical text information is trained using a large amount of medical text data as samples. In practical applications, medical staff can voice-input patient symptoms, and the backend service uses speech-to-text technology to obtain a textual description of the patient's condition. Medical staff can also input a textual description of patient symptoms, using this textual information as input to the network model to return triage and grading results. Replacing word selection from a dictionary with multi-sentence textual descriptions of the condition allows for a more accurate description of the patient's illness. The deep learning-based automatic triage and grading scheme reduces reliance on the professional knowledge of medical staff. Furthermore, by training the deep learning model using unsupervised learning, a large amount of manually labeled sample data is not required, thus reducing the workload of labelers.

[0157] This application's embodiments relate to the training of deep learning models, the comparison of text feature similarity, and forward inference for intelligent triage.

[0158] The forward inference process is as follows: A primary symptom database is pre-defined, and the text features of each statement in the corresponding accompanying symptoms are extracted and cached in the text feature database corresponding to this primary symptom. Through a speech-to-text interface, the real-time acquired patient condition description speech is converted into patient condition description text. A Named Entity Recognition (NER) network is used to identify symptom entities in the condition description text. After matching the primary symptom, the text features of the clause containing the accompanying symptom are extracted using a BERT network. The matching between the current text features and the accompanying symptom text feature database is calculated using a cosine similarity algorithm. The emergency level corresponding to the matched standard accompanying symptom statement is the current patient's emergency level. If there are multiple matches (e.g., matching level 1 "cardiac arrest" and level 3 "acute asthma"), the highest level (level 1) is used as the final output. The combined result of the text condition rating and the physiological performance index rating is used as the final classification result.

[0159] In this embodiment, a large amount of online medical condition description text is crawled, and a BERT model is trained unsupervised. The BERT model is then followed by an average pooling layer to extract textual features from the entire sentence. Cosine similarity is used to match the current patient's condition with text in a standard risk database, and the emergency level of the current patient is inferred based on the matching results. First, the main symptoms are extracted, and then accompanying symptoms are matched for grading. This embodiment infers the emergency level based on AI (Artificial Intelligence), without relying on the professional knowledge of medical personnel.

[0160] It should be noted that, in the embodiments of this application, if the above-mentioned method for determining the emergency level of a patient is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a mobile phone, tablet computer, desktop computer, personal digital assistant, navigator, digital phone, video phone, television, sensor device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0161] Based on the foregoing embodiments, this application provides a device for determining the emergency level of a patient. The device includes various modules, which can be implemented by a processor in the device; of course, they can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0162] Figure 12 This is a schematic diagram of the composition of a patient emergency level determination device according to an embodiment of this application, as shown below. Figure 12 As shown, the device 1200 includes a first determining module 1201, a classification module 1202, a filtering module 1203, and a second determining module 1204, wherein:

[0163] The first determining module 1201 is used to determine at least one first symptom of the patient;

[0164] Classification module 1202 is used to determine a first primary symptom and at least one first accompanying symptom from the at least one first symptom;

[0165] The filtering module 1203 is used to determine a target reference accompanying symptom from the at least one reference accompanying symptom based on a first similarity between each of the first accompanying symptom and at least one reference accompanying symptom in a preset first accompanying symptom library; the first accompanying symptom library is an accompanying symptom library corresponding to the first primary symptom.

[0166] The second determining module 1204 is used to determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

[0167] In one embodiment, the first determining module 1201 includes: a first acquiring submodule, configured to acquire the patient's condition description information; and a first determining submodule, configured to determine at least one first symptom of the patient based on the condition description information and a trained symptom recognition model.

[0168] In one embodiment, the first acquisition submodule includes: a first acquisition unit, configured to acquire the patient's speech describing the illness and convert the speech into text describing the illness; or, a second acquisition unit, configured to acquire the patient's text describing the illness; and a first determination submodule, configured to determine at least one first symptom of the patient based on the text describing the illness and a trained symptom recognition model.

[0169] In one embodiment, the classification module 1202 is configured to determine a first primary symptom and at least one first accompanying symptom from the at least one first symptom based on a second similarity between each of the first symptom and at least one of the reference primary symptoms in a preset primary symptom database.

[0170] In one embodiment, the classification module 1202 includes: a matching submodule, configured to determine a second similarity between each first symptom and each reference main symptom according to a preset matching order; a second determining submodule, configured to determine the reference main symptom corresponding to the highest second similarity as a candidate main symptom; a third determining submodule, configured to determine a first main symptom based on the candidate main symptom; and a fourth determining submodule, configured to determine other first symptoms besides the first main symptom as first accompanying symptoms.

[0171] In one embodiment, the reference main symptom includes a normative main symptom and associated non-normative main symptoms; the third determining submodule includes: a second determining unit, configured to determine the candidate main symptom as the first main symptom when the candidate main symptom is a normative main symptom; and a third determining unit, configured to determine the normative main symptom associated with the candidate main symptom as the first main symptom when the candidate main symptom is a non-normative main symptom.

[0172] In one embodiment, the second determining submodule is used to determine the reference main symptom with the highest matching order among the reference main symptoms corresponding to the highest second similarity as the candidate main symptom when there are multiple reference main symptoms corresponding to the highest second similarity.

[0173] In one embodiment, the filtering module 1203 includes: a fifth determining submodule, configured to determine a first text feature for each first accompanying symptom; a sixth determining submodule, configured to determine a second text feature for each reference accompanying symptom; a seventh determining submodule, configured to determine the similarity between each first text feature and each second text feature; and a filtering submodule, configured to determine the reference accompanying symptom corresponding to the second text feature with a similarity greater than a first threshold as the target reference accompanying symptom.

[0174] In one embodiment, the fifth determining submodule includes: an extraction unit, configured to extract a third text feature for each of the first accompanying symptoms using a trained feature extraction model; and a pooling unit, configured to perform average pooling on each of the third text features to obtain a first text feature corresponding to the first accompanying symptom.

[0175] In one embodiment, the apparatus further includes: an acquisition module for acquiring at least one historical medical condition text; a replacement module for replacing a portion of the text at a preset replacement position in each of the historical medical condition texts to obtain a corresponding historical medical condition sample; a first extraction module for extracting a fourth text feature of each of the historical medical condition samples using a feature extraction model; a second extraction module for extracting a fifth text feature at the preset replacement position in each of the historical medical condition texts using the feature extraction model; a third determination module for determining the character at the preset replacement position in each of the historical medical condition texts based on a text feature dictionary and each of the fifth text features; and an update module for updating the feature extraction model based on the character at each preset replacement position and the one-hot encoding of each preset replacement position.

[0176] It should be noted that, in the embodiments of this application, if the above-mentioned method for determining the emergency level of a patient is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a mobile phone, tablet computer, desktop computer, personal digital assistant, navigator, digital phone, video phone, television, sensor device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0177] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0178] Correspondingly, embodiments of this application provide a device, Figure 13 This is a schematic diagram of a hardware entity of the device according to an embodiment of this application, such as... Figure 13As shown, the hardware entity of the device 1300 includes a memory 1301 and a processor 1302. The memory 1301 stores a computer program that can run on the processor 1302. When the processor 1302 executes the program, it implements the steps in the method for determining the emergency level of a patient in the above embodiments.

[0179] The memory 1301 is configured to store instructions and applications executable by the processor 1302, and can also cache data to be processed or already processed by the processor 1302 and the various modules in the device 1300 (e.g., image data, audio data, voice communication data and video communication data), which can be implemented by flash memory or random access memory (RAM).

[0180] Correspondingly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the method for determining the emergency level of a patient provided in the above embodiments.

[0181] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects as the device embodiments. For technical details not disclosed in the storage medium and method embodiments of this application, please refer to the descriptions of the device embodiments of this application for understanding.

[0182] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0183] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0184] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0185] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional units in the embodiments of this application may all be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in a combination of hardware and software functional units.

[0186] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks. Alternatively, if the integrated units of this application are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the related technology, can be embodied in the form of software products. These computer software products are stored in a storage medium and include several instructions to cause computer devices (which may be mobile phones, tablets, desktops, personal digital assistants, navigators, digital phones, video phones, televisions, sensing devices, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0187] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several product embodiments provided in this application can be arbitrarily combined to obtain new product embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0188] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of determining an emergency level of a patient, characterized by, The method includes: Obtain the patient's condition description information; based on the condition description information and the trained symptom recognition model, determine at least one primary symptom of the patient; According to a preset matching order, determine the second similarity between each of the first symptoms and each of the reference main symptoms in at least one of the reference main symptoms in the preset main symptom database; determine the reference main symptom corresponding to the highest second similarity as a candidate main symptom; determine the first main symptom based on the candidate main symptom; and determine the other first symptoms besides the first main symptom as first accompanying symptoms. Determine a first text feature for each of the first accompanying symptoms; determine a second text feature for each of the reference accompanying symptoms in at least one reference accompanying symptom in the first accompanying symptom database; determine the similarity between each of the first text features and each of the second text features based on a cosine similarity algorithm; determine the reference accompanying symptom corresponding to the second text feature with a similarity greater than a first threshold as the target reference accompanying symptom; the first accompanying symptom database is an accompanying symptom database corresponding to the first primary symptom; The emergency level corresponding to the target reference accompanying symptoms is determined as the emergency level of the patient.

2. The method of claim 1, wherein, The process of obtaining the patient's condition description information includes: Obtain the patient's voice description of their condition and convert it into text; or, obtain the patient's text description of their condition. The step of determining at least one first symptom of the patient based on the description of the illness and the trained symptom recognition model includes: Based on the description of the illness and the trained symptom recognition model, at least one primary symptom of the patient is determined.

3. The method of claim 1, wherein, The reference main symptoms include standard main symptoms and associated non-standard main symptoms; The step of determining the first primary symptom based on the candidate primary symptoms includes: If the candidate primary symptom is a standard primary symptom, the candidate primary symptom will be determined as the first primary symptom. If the candidate primary symptom is an irregular primary symptom, the regular primary symptom associated with the candidate primary symptom is determined as the first primary symptom.

4. The method of claim 1, wherein, The step of identifying the reference main symptom corresponding to the highest second similarity as the candidate main symptom includes: If there are multiple reference main symptoms corresponding to the highest second similarity, the reference main symptom with the highest matching order among the reference main symptoms corresponding to the highest second similarity is determined as the candidate main symptom.

5. The method of claim 1, wherein, The determination of the first textual feature for each of the first accompanying symptoms includes: The trained feature extraction model is used to extract the third text features for each of the first accompanying symptoms; Each of the third text features is subjected to average pooling to obtain the first text feature corresponding to the first accompanying symptom.

6. The method of claim 5, wherein, The method further includes: Obtain at least one historical medical history text; Replace a portion of the text at a preset replacement position in each of the historical medical condition texts to obtain the corresponding historical medical condition sample; The fourth textual feature of each of the historical disease samples is extracted using a feature extraction model; The feature extraction model is used to extract the fifth text feature at the preset replacement position in each of the historical medical texts; Based on the text feature dictionary and each of the fifth text features, determine the word at the preset replacement position in each of the historical disease texts; The feature extraction model is updated based on the character at each preset replacement position and the one-hot encoding at each preset replacement position.

7. A device for determining the emergency level of a patient, characterized in that, The device includes: The first determining module includes: a first acquiring submodule, used to acquire the patient's condition description information; and a first determining submodule, used to determine at least one first symptom of the patient based on the condition description information and a trained symptom recognition model. The classification module includes: a matching submodule, configured to determine a second similarity between each of the first symptoms and at least one of the reference main symptoms in a preset main symptom database according to a preset matching order; a second determination submodule, configured to determine the reference main symptom corresponding to the highest second similarity as a candidate main symptom; a third determination submodule, configured to determine a first main symptom based on the candidate main symptom; and a fourth determination submodule, configured to determine other first symptoms besides the first main symptom as first accompanying symptoms. The filtering module is used to determine a first text feature for each of the first accompanying symptoms; determine a second text feature for each of the reference accompanying symptoms in at least one reference accompanying symptom in the first accompanying symptom library; determine the similarity between each of the first text features and each of the second text features based on a cosine similarity algorithm; determine the reference accompanying symptom corresponding to the second text feature with a similarity greater than a first threshold as the target reference accompanying symptom; the first accompanying symptom library is an accompanying symptom library corresponding to the first primary symptom; The second determining module is used to determine the emergency level corresponding to the target reference accompanying symptoms as the emergency level of the patient.

8. An apparatus comprising a memory and a processor, said memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for determining the emergency level of a patient as described in any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When executed by a processor, the computer program implements the steps in the method for determining the emergency level of a patient as described in any one of claims 1 to 6.

Images