Intelligent pre-check management method based on AI

Abstract

The application relates to an AI-based intelligent pre-examination management method, and relates to the technical field of smart medical treatment.The method comprises the following steps: acquiring first target data and second target data of a target patient; performing natural language processing on the second target data, and extracting first target features, including key symptoms, severity modifiers and duration; determining a first correlation between any two first target features, and correcting the first target features according to the first correlation to obtain second target features; according to a second correlation between the second target features and the first target data, fusing the first target data and the second target features, generating third target features, and inputting the third target features into a pre-constructed pre-examination classification model to output a pre-examination classification result, and pushing the pre-examination classification result to a triage guidance system and / or a user terminal.The application can optimize medical resource allocation, improve the medical treatment experience of patients, and improve the accuracy and efficiency of pre-examination.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical treatment, in particular to an AI-based intelligent pre-examination management method. BACKGROUND

[0002] Outpatient pre-examination triage is the first and key link in the patient medical treatment process. The traditional pre-examination triage mode highly depends on nurses' manual inquiry and experience judgment. Through measuring basic vital signs (such as body temperature, blood pressure) and combining with patient complaints, the preliminary disease classification and emergency level classification are carried out according to the pre-examination triage standard, and the patients are guided to the appropriate department for treatment. However, the traditional outpatient pre-examination triage mode mainly depends on nurses to complete manually, and has limitations. The evaluation results are easily affected by personal experience and state, and are highly subjective. The consistency and accuracy are difficult to guarantee. During the peak period of treatment, the manual operation efficiency is low, and the process bottleneck is easily formed, which leads to a long waiting time of patients in line, reduces the medical treatment experience, and at the same time, this method can only obtain limited structured data such as body temperature and blood pressure. The rich symptom information in the patient complaint text lacks deep mining and analysis ability, and the emergency level cannot be fully evaluated. In addition, the traditional pre-examination is a one-time link, and lacks the continuous monitoring and dynamic early warning mechanism for the change of patient condition during the waiting period, which may delay the identification and treatment of critical patients, and has the risk of medical safety.

[0003] With the rapid development of AI (artificial intelligence) technology, especially natural language processing (NLP), machine learning and data mining technology, it provides a new method to solve the above problems. For example, information is entered using an electronic form, but such systems often only electronicize paper processes and do not truly realize intelligent analysis and decision support. Therefore, how to automatically and intelligently process patient information, improve the accuracy and efficiency of pre-examination, optimize the allocation of medical resources, and improve the patient medical treatment experience is a problem that needs to be solved. SUMMARY

[0004] Therefore, it is necessary to provide an AI-based intelligent pre-examination management method capable of improving the accuracy and efficiency of pre-examination in view of the above technical problems.

[0005] In a first aspect, an AI-based intelligent pre-examination management method is provided, which comprises:

[0006] obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe the vital signs of the target patient, and the second target data being used to describe the complaint text of the target patient;

[0007] performing natural language processing on the second target data to extract first target features, the first target features at least including key symptoms, severity modifiers and duration;

[0008] determine a first correlation relationship of any two of the key symptoms, the severity modifier and the duration, and correct the first target feature according to the first correlation relationship to obtain a second target feature;

[0009] fuse the first target data and the second target feature according to a second correlation relationship between the second target feature and the first target data to generate a third target feature;

[0010] input the third target feature into a pre-examination classification model pre-constructed to output a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and push the pre-examination classification result to a triage guide system and / or a user terminal of the target patient.

[0011] Optionally, the first target data of the target patient is obtained based on a pre-examination terminal device, and the first target data includes:

[0012] in response to receiving the identity of the target patient, receiving click operation complaint data and / or voice complaint data of the target patient based on a display interface and / or a voice input interface of the pre-examination terminal device;

[0013] performing conversion and association processing on the click operation complaint data and / or the voice complaint data to generate the complaint text, and associating the complaint text with the identity;

[0014] defining the complaint text after the association processing and the identity as the second target data of the target patient.

[0015] Optionally, the first target data of the target patient is obtained based on a pre-examination terminal device, and the first target data includes:

[0016] in response to receiving the identity of the target patient, determining historical medical data of the target patient and the second target data according to the identity;

[0017] combining the historical medical data and the second target data, controlling a target sensor module integrated on the pre-examination terminal device to measure vital signs of the target patient;

[0018] obtaining a first measurement signal determined based on the vital sign measurement result, and converting the first measurement signal into a second measurement signal in digital form;

[0019] associating the second measurement signal with the identity to obtain the first target data of the target patient.

[0020] Optionally, the second target data is subjected to natural language processing, and the first target feature is extracted, including:

[0021] The second target data is received, and the second target data is preprocessed to obtain third target data, and the preprocessing includes word segmentation and stop word removal;

[0022] Based on the pre-stored medical named entity in the medical knowledge database, a plurality of target medical named entities are identified and extracted from the third target data;

[0023] According to the modification relationship between the plurality of target medical named entities, the plurality of target medical named entities are standardized mapped to generate the first target feature.

[0024] Optionally, determining the first association relationship between any two of the key symptom, the severity modifier and the duration in the first target feature includes:

[0025] Based on the standardized mapping relationship of the first target feature, the target medical named entity as a key symptom and the target medical named entity as a severity modifier and / or duration are subjected to dependency syntax analysis to generate a dependency relationship tree structure;

[0026] Traverse the dependency relationship tree structure to obtain a direct dependency relationship with the key symptom as the governing word;

[0027] In response to the severity modifier and the duration both having a direct dependency relationship with the key symptom, it is defined that the key symptom and the severity modifier and the duration, and the severity modifier and the duration all have a first association relationship;

[0028] In response to the key symptom and the severity modifier and / or the duration not having a direct dependency relationship, it is determined whether there is a first association symbol between the relative position distance of the key symptom and the severity modifier and / or the duration, and the severity modifier and the duration in the text sequence of the second target data;

[0029] In response to the key symptom, the severity modifier and the duration all having the first association symbol, it is defined that the key symptom and the severity modifier and the duration, and the severity modifier and the duration do not have a first association relationship;

[0030] In response to the first association symbol existing between any two of the key symptom, the severity modifier and the duration, it is defined that the first association relationship does not exist between the first two target items;

[0031] in response to the first association symbol not existing between the key symptom, the severity modifier and the duration, determining whether the second association symbol exists between any two of the key symptom, the severity modifier and the duration;

[0032] in response to the second association symbol existing, defining a first association relationship existing between the two of the second target;

[0033] Optionally, according to the first association relationship, the first target feature is corrected to obtain a second target feature, including:

[0034] the first target feature existing the first association relationship is acquired, and based on a feature quantity and a feature attribute of the first target feature, a first correction parameter is determined;

[0035] in response to the first association relationship being an association between the severity modifier and the key symptom, according to a pre-set first mapping table, a standardized severity level corresponding to the severity modifier is determined, and a second correction parameter is determined in combination of the first correction parameter and the standardized severity level;

[0036] in response to the first association relationship being an association between the duration and the key symptom, according to a pre-set second mapping table, a standardized duration value corresponding to the duration is determined, and a third correction parameter is determined in combination of the first correction parameter and the standardized duration value;

[0037] in response to the first association relationship being an association between the duration and the severity modifier, according to a pre-set third mapping table, a standardized correction parameter corresponding to the first association relationship is determined, and a fourth correction parameter is determined in combination of the first correction parameter and the standardized correction parameter;

[0038] in response to the first association relationship existing the key symptom, a standard medical term code corresponding to the key symptom is determined;

[0039] the first target feature is corrected based on the second correction parameter, the third correction parameter, the fourth correction parameter and the standard medical term code to obtain a second target feature.

[0040] Optionally, according to a second association relationship between the second target feature and the first target data, the first target data and the second target feature are fused to generate a third target feature, including:

[0041] in response to an identity corresponding to the second target feature and the first target data being same, it is determined that the second target feature and the first target data exist a second association relationship;

[0042] In response to the existence of the second association relationship, the first target data is standardized to generate a first feature sub-vector, and the second target feature is standardized and / or vectorized to generate a second feature sub-vector;

[0043] The first feature sub-vector and the second feature sub-vector are spliced to generate a fusion feature vector;

[0044] The fusion feature vector is input into a pre-constructed neural network fusion model to output a target patient feature vector after fusion processing, and the target patient feature vector is defined as the third target feature.

[0045] Optionally, the third target feature is input into a pre-constructed pre-examination classification model to output a pre-examination classification result, which includes:

[0046] The third target feature is input into the pre-examination classification model, and the pre-examination classification model is a multi-task deep learning classification model;

[0047] The third target feature is transformed and information is extracted based on the bottom network of the multi-task deep learning classification model to obtain a fourth target feature;

[0048] The fourth target feature is input into a first output head and a second output head of the multi-task deep learning classification model, respectively, to output a pre-examination classification result containing the target department and the criticality level, the first output head is a department recommendation classifier, and the second output head is a criticality level classifier;

[0049] The multi-task deep learning classification model is trained by minimizing a joint loss function, and the joint loss function is composed of a cross-entropy loss of department recommendation and a cross-entropy loss of criticality level.

[0050] Optionally, the pre-examination classification result is pushed to a triage guidance system and / or a user terminal of the target patient, which includes:

[0051] The data of the pre-examination classification result is packaged into a standardized data exchange format message;

[0052] Based on a first communication interface, the data exchange format message is pushed to the triage guidance system;

[0053] Based on the identity of the target patient, a receiving address of the user terminal is determined, and based on a second communication interface and the receiving address, the data exchange format message is pushed to the user terminal of the target patient.

[0054] In a second aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0055] obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient;

[0056] performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration;

[0057] determining a first correlation relationship of any two of the first target features, and correcting the first target features according to the first correlation relationship to obtain second target features;

[0058] fusing the first target data and the second target features according to a second correlation relationship between the second target features and the first target data to generate third target features;

[0059] inputting the third target features into a pre-constructed pre-examination classification model to output a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guidance system and / or a user terminal of the target patient.

[0060] In a third aspect, a computer readable storage medium is provided, having a computer program stored thereon, wherein the computer program is executed by a processor to implement the following steps:

[0061] obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient;

[0062] performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration;

[0063] determining a first correlation relationship of any two of the first target features, and correcting the first target features according to the first correlation relationship to obtain second target features;

[0064] Based on the second correlation between the second target feature and the first target data, the first target data and the second target feature are fused to generate a third target feature;

[0065] The third target feature is input into the pre-constructed pre-examination classification model, and the pre-examination classification result is output. The pre-examination classification result includes at least the target department and the level of urgency. The pre-examination classification result is then pushed to the triage guidance system and / or the user terminal of the target patient.

[0066] Fourthly, a computer program product is provided, the computer program product comprising a computer program, which, when executed by a processor, performs the following steps:

[0067] The first target data and the second target data of the target patient are obtained based on the pre-screening terminal device. The first target data is used to describe the vital signs of the target patient, and the second target data is used to describe the chief complaint text of the target patient.

[0068] Natural language processing is performed on the second target data to extract the first target features, which include at least key symptoms, severity modifiers, and duration.

[0069] Determine a first association relationship between any two of the key symptoms, severity modifiers, and duration of the first target features, and modify the first target features based on the first association relationship to obtain a second target feature;

[0070] Based on the second correlation between the second target feature and the first target data, the first target data and the second target feature are fused to generate a third target feature;

[0071] The third target feature is input into the pre-constructed pre-examination classification model, and the pre-examination classification result is output. The pre-examination classification result includes at least the target department and the level of urgency. The pre-examination classification result is then pushed to the triage guidance system and / or the user terminal of the target patient.

[0072] The AI-based intelligent pre-examination management method comprises: obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient; performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration; determining a first correlation between any two of the key symptom, the severity modifier, and the duration, and correcting the first target features according to the first correlation to obtain second target features; fusing the first target data and the second target features according to a second correlation between the second target features and the first target data to generate third target features; and inputting the third target features into a pre-constructed pre-examination classification model to output a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guide system and / or a user terminal of the target patient. The AI-based intelligent pre-examination management method can improve the accuracy and consistency of triage, reduce human errors, shorten the waiting time of patients in line, alleviate the congestion of outpatient services, optimize the allocation of medical resources, improve the patient experience, and realize continuous risk assessment by deep mining of patient complaint information combined with dynamic vital sign monitoring, thereby improving the accuracy and efficiency of pre-examination. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 An application environment diagram of the AI-based intelligent pre-examination management method in one embodiment;

[0074] Figure 2 A flowchart of the AI-based intelligent pre-examination management method in one embodiment;

[0075] Figure 3 An internal structure diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0076] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0077] It should be understood that in the description of the present application, unless the context clearly requires otherwise, the terms "including", "containing" and the like, throughout the description, mean "comprising" rather than "consisting of" or "consisting exclusively of"; that is, the term "including, but not limited to" is meant.

[0078] It should also be understood that the terms "first", "second" and the like are used only for descriptive purposes and are not to be construed as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise stated, the meaning of "multiple" is two or more.

[0079] It should be noted that the terms "S1", "S2" and the like are only for the purpose of describing the steps and do not specifically refer to the order or sequence, nor are they used to limit the present application. They are only used to facilitate the description of the method of the present application and cannot be understood as indicating the order of the steps. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the scope of protection required by the present application.

[0080] The AI-based intelligent pre-examination management method provided by the present application can be applied to the application environment as shown in Figure 1 . Among them, the terminal 102 communicates with the data processing platform set on the server 104 through the network, wherein the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices, and the server 104 can be realized by an independent server or a server cluster composed of multiple servers.

[0081] In one embodiment, as shown in Figure 2 , an AI-based intelligent pre-examination management method is provided, which is applied to the terminal in Figure 1 for example, including the following steps:

[0082] S1: Obtain first target data and second target data of the target patient based on the pre-examination terminal device, wherein the first target data is used to describe the vital signs of the target patient, and the second target data is used to describe the chief complaint text of the target patient.

[0083] It should be noted that the pre-examination terminal device is a self-service or assisted intelligent hardware deployed in the outpatient department of a hospital, which is usually integrated with multiple medical sensors (such as infrared temperature measurement module, electronic sphygmomanometer arm, blood oxygen probe, etc.) for one-key measurement of vital signs. At the same time, it provides a touch screen interface to efficiently collect the patient's chief complaint symptoms through a guided menu or voice input function. All data is automatically digitized and uploaded to the background AI system for analysis, thereby realizing efficient and standardized pre-examination diversion. Vital signs refer to body temperature, blood pressure, heart rate, blood oxygen saturation, etc. When there is a problem in a certain part, such as severe problems in brain structure (such as cerebral hemorrhage, brain tumor, and significant increase in intracranial pressure), the involved vital signs can include elevated blood pressure, slowed heart rate, and abnormal respiratory rhythm, etc. The patient's chief complaint text refers to the patient's subjective description of their own disease or discomfort, which is usually presented in the form of unstructured text and can include key symptoms such as headache, cough, chest pain, etc., symptom attributes such as location-sunken temple-, nature-needle-like-, severity-unbearable-, duration-three days, etc., and related factors such as triggers-after activity, aggravating / relieving factors-improved after rest, etc.

[0084] S2: performing natural language processing on the second target data to extract first target features, the first target features including at least key symptoms, severity modifiers, and duration.

[0085] It should be noted that the severity modifier can include the above unbearable, etc., and the duration can include the above three days, etc. Natural language processing refers to NLP.

[0086] S3: determining a first association relationship between any two of the key symptoms, the severity modifier, and the duration, and correcting the first target features according to the first association relationship to obtain second target features.

[0087] It should be noted that the first association relationship is used to describe whether there is a connection between two features, and the second target features are determined by correcting the first target features.

[0088] S4: according to a second association relationship between the second target features and the first target data, performing fusion processing on the first target data and the second target features to generate third target features.

[0089] It should be noted that fusion processing refers to standardizing, vectorizing, etc. structured vital sign data (such as body temperature, blood pressure) and subjective unstructured text complaint data (such as symptom description), and then deeply integrating through a neural network model to generate a feature vector that can more accurately represent the overall condition of the patient.

[0090] S5: inputting the third target feature into a pre-constructed pre-check classification model, outputting a pre-check classification result, the pre-check classification result at least including a target department and a criticality level, and pushing the pre-check classification result to a triage guide system and / or a user terminal of the target patient.

[0091] It should be noted that the pre-check classification model is a multi-task deep learning classification model, including a first output head and a second output head, respectively used for outputting a target department and a criticality level.

[0092] In some embodiments, the first target data of the target patient is obtained based on a pre-check terminal device, including:

[0093] In response to receiving the identity of the target patient, the pre-check terminal device receives the click operation chief complaint data and / or the voice chief complaint data of the target patient based on a display interface and / or a voice input interface of the pre-check terminal device, wherein the identity is a unique identifier of the patient, which can be obtained through a medical insurance card or the like;

[0094] The click operation chief complaint data and / or the voice chief complaint data are processed by transformation and association to generate the chief complaint text, and the chief complaint text is associated with the identity, wherein for the click operation, the system dynamically combines the selected parts, symptoms, attributes and other elements into grammatically correct sentences based on a pre-set clinical logic model, for example, the head, pain and severe are associated to generate head severe pain, and for the voice data, the recognized text needs to be punctuated, segmented and normalized in medical terminology, then the system combines the text fragments generated in the two ways to obtain a complete and structured chief complaint text for subsequent analysis;

[0095] The chief complaint text after the association processing and the identity are defined as the second target data of the target patient.

[0096] In some embodiments, the first target data of the target patient is obtained based on a pre-check terminal device, including:

[0097] In response to receiving the identity of the target patient, the pre-check terminal device receives the click operation chief complaint data and / or the voice chief complaint data of the target patient based on a display interface and / or a voice input interface of the pre-check terminal device, wherein the identity is a unique identifier of the patient, which can be obtained through a medical insurance card or the like;

[0098] In combination with the historical visit data and the second target data, a target sensor module integrated in the pre-examination terminal device is controlled to measure vital signs of the target patient. If the historical visit data and the second target data match with high similarity, the same target sensor module of the pre-examination terminal device is called according to the patient historical examination data to perform examination measurement, such as, if the historical visit data is brain disease, blood pressure is measured, and the second target data is also related to brain disease, a blood pressure detector is called to perform examination measurement, and the like. If the historical visit data and the second target data match with low similarity, a responsive target sensor module is called according to the second target data, such as, if the historical data is brain disease, and this time is abdominal disease, the matching similarity is defined as low.

[0099] A first measurement signal determined based on the vital sign measurement result is acquired, and the first measurement signal is converted into a second measurement signal in digital form, that is, through analog-to-digital conversion, raw data is converted into a numerical value with clear medical meaning and unit, such as 36.5℃, 120 / 80 mmHg, and the like.

[0100] The second measurement signal is associated with the identity to obtain first target data of the target patient.

[0101] In some specific embodiments, the second target data is subjected to natural language processing, and the first target feature extraction includes:

[0102] The second target data is received, and pre-processing is performed on the second target data to obtain third target data. The pre-processing includes word segmentation and stop word removal. The word segmentation is a process of cutting continuous Chinese text sequences into single independent word units according to semantics, such as cutting my severe headache into my / severe / headache. The stop word removal is a process of filtering out a large number of words in the text that do not carry key semantic information, such as de, le, and the like.

[0103] Based on the pre-stored medical named entities in the medical knowledge database, a plurality of target medical named entities are identified and extracted from the third target data. The medical named entities can be words describing disease manifestations, words describing symptom severity, and words describing disease duration, and the like.

[0104] The plurality of target medical named entities are standardized mapped according to the modification relationship between the plurality of target medical named entities to generate the first target feature. The modification relationship refers to the semantic association between a symptom entity and its attribute (such as degree, position, and time), for example, in severe headache, severe modifies headache, indicating the severity of pain, which is standardized mapped.

[0105] In some embodiments, determining the first association between any two of the first target features of the key symptom, the severity modifier, and the duration includes:

[0106] performing dependency parsing on the target medical named entity as a key symptom and the target medical named entity as a severity modifier and / or duration based on the standardized mapping relationship of the first target features, to generate a dependency tree structure, it should be noted that dependency parsing refers to analyzing the grammatical dependency relationship between words in a sentence, regarding the sentence as a network connected by words through dependency relationship, i.e. a dependency tree structure, such as severe headache;

[0107] traversing the dependency tree structure to obtain a direct dependency relationship with the key symptom as the governing word, i.e. by traversing the dependency tree, finding a subtree with the symptom entity as the root node, and the modifier entity in the directly connected child node is determined to be associated with the symptom;

[0108] in response to the severity modifier and the duration both having a direct dependency relationship with the key symptom, defining the key symptom and the severity modifier and the duration, and the severity modifier and the duration both having a first association, i.e. if the severity modifier and the key symptom have a direct dependency relationship, the duration and the key symptom have a direct dependency relationship, then the severity modifier and the duration have a direct dependency relationship;

[0109] in response to the key symptom and the severity modifier and / or the duration not having a direct dependency relationship, determining whether the key symptom and the severity modifier and / or the duration, and the relative position distance between the severity modifier and the duration in the text sequence of the second target data have a first association, wherein the relative position distance refers to the closeness between two words or entities in the text, which is usually measured by the number of words separated, such as in the case of a patient having severe headache and mild cough, the relative position distance between severe and headache is very close, while the distance between severe and cough is farther (separated by 3 words of headache and mild), and the first association symbol is a separator, such as “,”, “.”, etc.;

[0110] in response to the key symptom, the severity modifier and the duration all having the first association symbol, defining the key symptom and the severity modifier and the duration, and the severity modifier and the duration not having a first association, i.e. if the three have the first association symbol, then there is no association;

[0111] In response to the first association symbol existing between any first two of the key symptom, the severity modifier and the duration, it is defined that there is no first association relationship between the first two, that is, as long as there is the first association symbol between the two, there is no association relationship between them;

[0112] In response to the first association symbol not existing between the key symptom, the severity modifier and the duration, it is determined whether there is a second association symbol between any second two of the key symptom, the severity modifier and the duration, wherein the second association symbol is an association symbol such as “and”, “as well as”, “and” and the like;

[0113] In response to the second association symbol existing, it is defined that there is a first association relationship between the second two.

[0114] In some specific embodiments, according to the first association relationship, the first target feature is modified to obtain a second target feature, which includes:

[0115] The first target feature having the first association relationship is obtained, and a first correction parameter is determined based on a feature quantity and a feature attribute of the first target feature, wherein the feature attribute refers to, for example, degree, part, time and the like, and the feature quantity refers to the number of features having the first association relationship, such as severe headache, the feature quantity being 2, severe headache for 3 days, the feature quantity being 3, and the like. The feature quantity and the feature attribute are standardized, that is, the data after the standardization determines the first correction parameter, which includes:

[0116]

[0117] wherein, is the first correction parameter, , and all represent weight coefficients, represents the nth feature attribute, represents the feature quantity;

[0118] In response to the first association relationship being the association between the severity modifier and the key symptom, a standardized severity level corresponding to the severity modifier is determined according to a pre-set first mapping table, and a second correction parameter is determined in combination with the first correction parameter and the standardized severity level, wherein the first mapping table is a mapping table pre-set and stored in a database, which can be directly called. For example, the standardized severity level of severe is high, which is quantified to 10 levels, the highest level being 10, and the like, which will not be described in detail. The second correction parameter is S=K×A / Y, wherein S is the second correction parameter, A is the standardized severity level, and Y is a first conversion coefficient.

[0119] In response to the first association relationship being an association between the duration and the key symptom, a standardized duration value corresponding to the duration is determined according to a pre-set second mapping table, and a third correction parameter is determined in combination of the first correction parameter and the standardized duration value, where the second mapping table is a mapping table pre-set and stored in a database and can be directly called, for example, the duration is 3 days, which is quantified as 72 hours, and the like, and will not be described herein again, and the third correction parameter is T = K x t x c, where T is the third correction parameter, t is the standardized duration value, and c is a second conversion coefficient;

[0120] In response to the first association relationship being an association between the duration and the severity modifier, a standardized correction parameter corresponding to the first association relationship is determined according to a pre-set third mapping table, and a fourth correction parameter is determined in combination of the first correction parameter and the standardized correction parameter, where the third mapping table is a mapping table pre-set and stored in a database and can be directly called, for example, the duration is 3 days, which is quantified as 72 hours, the standardized severity level of the severe is high, which is quantified as 10 levels, the standardized correction parameter is h = c1F1 + c2F2, h is the standardized correction parameter, c1 and c2 are weight coefficients, F1 and F2 are the standardized duration value and the standardized severity level respectively, and the like, and will not be described herein again, and the fourth correction parameter is W = K x h x g, where W is the fourth correction parameter, h is the standardized correction parameter, and g is a third conversion coefficient;

[0121] In response to the first association relationship existing the key symptom, a standardized medical terminology code corresponding to the key symptom is determined, where the standardized medical terminology code is a recognized and systematic code system for uniquely and unambiguously representing medical terminology, and the specific conversion method is a common method, which will not be described herein again, for example, the bellyache is mapped to a standardized code, that is, R10.9 in ICD-10 represents abdominal pain, and the like;

[0122] The first target feature is corrected based on the second correction parameter, the third correction parameter, the fourth correction parameter and the standardized medical terminology code to obtain a second target feature, that is, the first target feature is digitized or coded to generate a corresponding second target feature.

[0123] The first target data and the second target feature are fused to generate a third target feature according to a second association relationship between the second target feature and the first target data, including:

[0124] In response to the identity corresponding to the second target feature and the first target data being the same, it is determined that the second target feature and the first target data have a second association relationship;

[0125] In response to the existence of the second association relationship, the first target data is standardized to generate a first feature sub-vector, and the second target feature is standardized and / or vectorized to generate a second feature sub-vector;

[0126] The first feature sub-vector and the second feature sub-vector are spliced to generate a fusion feature vector;

[0127] The fusion feature vector is input into a pre-constructed neural network fusion model to output a target patient feature vector after fusion processing, and the target patient feature vector is defined as the third target feature.

[0128] In some embodiments, the third target feature is input into a pre-constructed pre-examination classification model to output a pre-examination classification result, which includes:

[0129] The third target feature is input into the pre-examination classification model, and the pre-examination classification model is a multi-task deep learning classification model, wherein the multi-task deep learning classification model can be generated based on a multi-layer perception machine (MLP) or a Transformer encoder.

[0130] The third target feature is subjected to feature transformation and information extraction based on a bottom network of the multi-task deep learning classification model to obtain a fourth target feature, wherein the bottom network refers to an initial part of a deep learning model responsible for extracting general and basic features from original input data, such as a convolutional layer, a recurrent neural network layer, or a Transformer layer. A multi-task head shares one bottom network to prevent overfitting. Feature transformation and information extraction are common methods, i.e., nonlinear transformation, and the specific process is not repeated here.

[0131] The fourth target feature is input into a first output head and a second output head of the multi-task deep learning classification model, respectively, to output a pre-examination classification result containing the target department and the criticality level. The first output head is a department recommendation classifier, and the second output head is a criticality level classifier. The first output head is a Softmax classifier that outputs a probability distribution vector including multiple candidate departments and the probability of being selected by the candidate department. The final recommended department is the department with the highest probability. The second output head is also a Softmax classifier that outputs a probability distribution corresponding to different criticality levels, such as 1st level - critical, 2nd level - critical, 3rd level - urgent, 4th level - less urgent, and 5th level - non-urgent. The final criticality level is the level with the highest probability.

[0132] The multi-task deep learning classification model is trained by minimizing a joint loss function composed of a cross-entropy loss of department recommendation and a cross-entropy loss of criticality level, wherein during training, a joint loss function combining the two task losses is used for training, such as L3 = r1L1 + r2L2, wherein L3 represents the total loss, L1 represents the loss of the department recommendation task, L2 represents the loss of the criticality level classification task, and r1 and r2 both represent weight hyperparameters.

[0133] In some embodiments, the pre-examination classification result is pushed to a triage guidance system and / or a user terminal of the target patient, including:

[0134] The data of the pre-examination classification result is packaged into a standardized data exchange format message, i.e., the data is packaged according to a preset data exchange standard, such as a JSON / XML format;

[0135] Based on a first communication interface, the data exchange format message is pushed to the triage guidance system, wherein the first communication interface is an application programming interface, etc., based on which the pre-examination classification result is pushed to a medical care platform;

[0136] Based on the identity of the target patient, a receiving address of the user terminal is determined, and based on a second communication interface and the receiving address, the data exchange format message is pushed to the user terminal of the target patient, i.e., according to the identity, a mobile terminal identity bound during patient filing, such as a mobile phone number, a WeChat public number, a small program user ID, etc., is queried, based on which the pre-examination classification result is pushed to the user terminal.

[0137] In the AI-based intelligent pre-examination management method, the method comprises: acquiring first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient; performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration; determining a first correlation between any two of the key symptom, the severity modifier, and the duration, and correcting the first target features according to the first correlation to obtain second target features; fusing the first target data and the second target features according to a second correlation between the second target features and the first target data to generate third target features; inputting the third target features into a pre-constructed pre-examination classification model to output a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guide system and / or a user terminal of the target patient. The application can improve the accuracy and consistency of triage, reduce human error, shorten the waiting time of patients in line, alleviate outpatient congestion, optimize the allocation of medical resources, improve the patient experience, and realize continuous risk assessment by deep mining of patient complaint information and dynamic vital sign monitoring, thereby improving the accuracy and efficiency of pre-examination.

[0138] It should be understood that, although Figure 2 The steps in the flowchart of the application are shown in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 2 At least part of the steps in the application can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0139] In one embodiment, a computer device, which can be a terminal, is provided, and its internal structure diagram can be as shown in Figure 3As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement an AI-based intelligent pre-examination management method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0140] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0141] In one embodiment, a computer device is provided, including a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the following steps:

[0142] S1: obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe the vital signs of the target patient, and the second target data being used to describe the chief complaint text of the target patient;

[0143] S2: performing natural language processing on the second target data to extract first target features, the first target features including at least key symptoms, severity modifiers and duration;

[0144] S3: determining a first association relationship of any two first target features among the key symptoms, the severity modifiers and the duration, and correcting the first target features according to the first association relationship to obtain second target features;

[0145] S4: according to a second association relationship between the second target features and the first target data, fusing the first target data and the second target features to generate third target features;

[0146] S5: inputting the third target feature into a pre-constructed pre-examination classification model, outputting a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guide system and / or a user terminal of the target patient.

[0147] In one embodiment, a computer readable storage medium is provided, and the computer readable storage medium has stored thereon a computer program, and the computer program is executed by a processor to implement the following steps:

[0148] S1: acquiring first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient;

[0149] S2: performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration;

[0150] S3: determining a first association relationship of any two first target features among the key symptom, the severity modifier, and the duration, and correcting the first target features according to the first association relationship to obtain second target features;

[0151] S4: fusing the first target data and the second target features according to a second association relationship between the second target features and the first target data to generate third target features;

[0152] S5: inputting the third target features into a pre-constructed pre-examination classification model, outputting a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guide system and / or a user terminal of the target patient.

[0153] In one embodiment, a computer program product is provided, and the computer program product includes a computer program, and the computer program is executed by a processor to implement the following steps:

[0154] S1: acquiring first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient;

[0155] S2: performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration;

[0156] S3: determining a first correlation relationship of any two of the first target features in the key symptoms, the severity modifiers and the duration, and correcting the first target features according to the first correlation relationship to obtain second target features;

[0157] S4: according to a second correlation relationship between the second target features and the first target data, fusing the first target data and the second target features to generate third target features;

[0158] S5: inputting the third target features into a pre-check classification model pre-constructed to output a pre-check classification result, the pre-check classification result at least including a target department and a criticality level, and pushing the pre-check classification result to a triage guide system and / or a user terminal of the target patient.

[0159] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM).

[0160] The technical features of the above embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0161] The above embodiments only express several implementation ways of the present application, and the description is more specific and detailed, but it should not be understood as a limitation to the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the scope of protection of the present application.

Claims

1. An AI-based intelligent pre-check management method, characterized in that, The method comprises: obtaining first target data and second target data of a target patient based on a pre-examination terminal device, the first target data being used to describe vital signs of the target patient, and the second target data being used to describe a chief complaint text of the target patient; performing natural language processing on the second target data to extract first target features, the first target features at least including a key symptom, a severity modifier, and a duration; determining a first association relationship between any two of the first target features of the key symptom, the severity modifier, and the duration, and correcting the first target features according to the first association relationship to obtain second target features; fusing the first target data and the second target features according to a second association relationship between the second target features and the first target data to generate third target features; inputting the third target features into a pre-constructed pre-examination classification model to output a pre-examination classification result, the pre-examination classification result at least including a target department and a criticality level, and pushing the pre-examination classification result to a triage guide system and / or a user terminal of the target patient; determining the first association relationship between any two of the first target features of the key symptom, the severity modifier, and the duration comprises: performing dependency syntax analysis on the target medical named entity as the key symptom and the target medical named entity as the severity modifier and / or the duration based on a standardized mapping relationship of the first target features to generate a dependency relationship tree structure; traversing the dependency relationship tree structure to obtain a direct dependency relationship with the key symptom as a governing word; in response to the severity modifier and the duration both having a direct dependency relationship with the key symptom, defining that the key symptom and the severity modifier and the duration, and the severity modifier and the duration all have the first association relationship; in response to the key symptom and the severity modifier and / or the duration not having a direct dependency relationship, determining whether there is a first association symbol between the relative position distance of the key symptom and the severity modifier and / or the duration in the text sequence of the second target data; in response to the key symptom, the severity modifier, and the duration all having the first association symbol, defining that the key symptom and the severity modifier and the duration, and the severity modifier and the duration do not have the first association relationship; in response to any two of the first target features of the key symptom, the severity modifier, and the duration having the first association symbol, defining that the two first target features do not have the first association relationship; in response to the key symptom, the severity modifier, and the duration all not having the first association symbol, determining whether there is a second association symbol between any two of the first target features of the key symptom, the severity modifier, and the duration. In response to the existence of the second association symbol, a first association relationship between two items of the first target feature is defined to exist; According to the first association relationship, the first target feature is corrected to obtain a second target feature, including: Obtaining the first target feature existing in the first association relationship, determining a first correction parameter based on the feature quantity and the feature attribute of the first target feature, and performing standardization processing on the feature quantity and the feature attribute, that is, the data after standardization processing determines the first correction parameter, including: wherein, is a first correction parameter, , and each represent a weight coefficient, represents an nth feature attribute, represents a feature quantity; In response to the first association relationship being an association between the severity modifier and the key symptom, determining a standardized severity level corresponding to the severity modifier according to a pre-set first mapping table, and determining a second correction parameter in combination with the first correction parameter and the standardized severity level, the second correction parameter being S=K×A / Y, where S is the second correction parameter, A is the standardized severity level, and Y is a first conversion coefficient; In response to the first association relationship being an association between the duration and the key symptom, determining a standardized duration value corresponding to the duration according to a pre-set second mapping table, and determining a third correction parameter in combination with the first correction parameter and the standardized duration value, the third correction parameter being T=K×t×c, where T is the third correction parameter, t is the standardized duration value, and c is a second conversion coefficient; In response to the first association relationship being an association between the duration and the severity modifier, determining a standardized correction parameter corresponding to the first association relationship according to a pre-set third mapping table, and determining a fourth correction parameter in combination with the first correction parameter and the standardized correction parameter, the fourth correction parameter being W=K×h×g, where W is the fourth correction parameter, h is the standardized correction parameter, and g is a third conversion coefficient; In response to the existence of the key symptom in the first association relationship, determining a standard medical terminology code corresponding to the key symptom; Based on the second correction parameter, the third correction parameter, the fourth correction parameter, and the standard medical terminology code, the first target feature is corrected to obtain a second target feature. 2.The AI-based intelligent pre-check management method of claim 1, wherein, Based on a pre-check terminal device, second target data of a target patient is obtained, including: In response to receiving an identity of the target patient, receiving click operation complaint data and / or voice complaint data of the target patient based on a display interface and / or a voice input interface of the pre-check terminal device; Performing conversion association processing on the click operation complaint data and / or the voice complaint data to generate a complaint text, and associating the complaint text with the identity; The complaint text after association processing and the identity are defined as the second target data of the target patient. 3.The AI-based intelligent pre-check management method of claim 2, wherein, Based on a pre-check terminal device, first target data of a target patient is obtained, including: In response to receiving an identity of the target patient, determining historical medical data and the second target data of the target patient according to the identity; In combination with the historical visit data and the second target data, a target sensor module integrated in the pre-examination terminal device controls to measure vital signs of the target patient; Obtaining a first measurement signal determined based on the vital sign measurement result, and converting the first measurement signal into a second measurement signal in digital form; Associating the second measurement signal with the identity to obtain the first target data of the target patient. 4.The AI-based intelligent pre-check management method of claim 3, wherein, The first target feature is extracted by performing natural language processing on the second target data, including: Receiving the second target data, pre-processing the second target data to obtain third target data, the pre-processing including word segmentation and stop word removal; Based on the pre-stored medical named entity in the medical knowledge database, a plurality of target medical named entities are identified and extracted from the third target data; According to the modification relationship between the plurality of target medical named entities, the plurality of target medical named entities are standardized mapped to generate the first target feature. 5.The AI-based intelligent pre-check management method of claim 4, wherein, The second target feature and the first target data are fused to generate a third target feature, including: In response to the identity corresponding to the second target feature and the first target data being the same, it is determined that the second target feature and the first target data have a second association relationship; In response to the existence of the second association relationship, the first target data is standardized processed to generate a first feature sub-vector, and the second target feature is standardized processed and / or vectorized processed to generate a second feature sub-vector; The first feature sub-vector and the second feature sub-vector are spliced to generate a fusion feature vector; The fusion feature vector is input into a pre-constructed neural network fusion model to output a target patient feature vector after fusion processing, and the target patient feature vector is defined as the third target feature. 6.The AI-based intelligent pre-check management method of claim 5, wherein, The third target feature is input into a pre-constructed pre-examination classification model to output a pre-examination classification result, including: The third target feature is input into the pre-examination classification model, and the pre-examination classification model is a multi-task deep learning classification model; Based on the bottom network of the multi-task deep learning classification model, the third target feature is feature-transformed and information-extracted to obtain a fourth target feature; The fourth target feature is input into a first output head and a second output head of the multi-task deep learning classification model respectively to output a pre-examination classification result containing the target department and the criticality level, the first output head being a department recommendation classifier, and the second output head being a criticality level classifier; Wherein, the multi-task deep learning classification model is trained by minimizing a joint loss function, and the joint loss function is composed of a cross-entropy loss of department recommendation and a cross-entropy loss of criticality level. 7.The AI-based intelligent pre-check management method of claim 6, wherein, The pre-examination classification result is pushed to a triage guidance system and / or a user terminal of the target patient, including: The data of the pre-examination classification result is packaged into a standardized data exchange format message; based on the first communication interface, push the data exchange format message to the triage guide system; based on the identity of the target patient, determine the receiving address of the user terminal, and based on the second communication interface and the receiving address, push the data exchange format message to the user terminal of the target patient.

Images