A triage system for pre-admission questioning

Abstract

The application relates to a triage system for a pre-diagnosis stage, comprising an online triage platform, wherein a main entrance module is arranged for a patient to input information, perform triage interaction and use common functions of a hospital, triage operation is performed, an AI assistant window is arranged for the patient to input triage information, a reasoning model of a medical history collection module classifies chief complaint information of triage, an AI program executes a reasoning intention of a 1-step thinking path, a smooth thinking process is formed, and accurate chief complaint information is obtained; a triage model is arranged to process the chief complaint information, regular identification, jieba dictionary identification and model identification are performed, a cross-fusion mode is used, a logical and perfect multi-level auditing mechanism is realized, deviation of a triage result is effectively reduced, and correct guidance is provided for the patient to see a doctor.

Description

Technical Field

[0001] This invention belongs to the technical field of machine triage systems, and particularly relates to a triage system for the pre-diagnosis stage. Background Technology

[0002] With advancements in medical technology and increasing patient demands, hospital outpatient departments are facing growing pressure. An efficient triage system has become crucial for optimizing medical resources and improving patient experience. Traditional triage methods rely heavily on the experience of nurses or guides, resulting in low efficiency and susceptibility to subjective factors, failing to meet the high demands for accuracy and intelligence in modern healthcare.

[0003] With the rapid development of artificial intelligence technology, the application of AI in the medical field is gradually extending from assisted diagnosis to intelligent services throughout the entire process. As the first link in the medical process, the efficiency and accuracy of outpatient triage directly affect the patient's medical experience and the allocation of medical resources. At present, some hospitals have adopted electronic triage systems or simple intelligent triage devices. These systems usually collect patient symptom information through fixed questionnaires and make department recommendations using preset rules. At the same time, some advanced triage systems have begun to introduce natural language processing technology to initially achieve semantic understanding of patient symptoms.

[0004] However, existing triage systems lack a robust mechanism for reviewing triage results. Some systems rely solely on simple rule matching and lack multi-level review processes, which may lead to deviations in triage results and affect the correct guidance for patients seeking medical treatment.

[0005] Therefore, designing a triage system for the pre-diagnosis stage that can more effectively review triage results and improve triage accuracy has become an urgent technical problem to be solved. Summary of the Invention

[0006] To address the aforementioned technical issues, a triage and consultation system for the pre-diagnosis stage is provided, including an online consultation platform. The online consultation platform includes a main entry module, a medical history collection module, and a pre-visit module. The main entry module has a patient management window, which allows patients to input their information and synchronize it with the hospital's database. The AI ​​assistant window allows patients to input consultation information via text input and voice-to-text conversion, and interact with the medical history collection module. The function guide window displays frequently used function entries, which include links to existing functions in the hospital. The medical history collection module has a reasoning model that uploads the consultation information entered by the patient to the existing AI program. The AI ​​program then executes a 1-step thinking path to infer the patient's consultation intent, organizes the consultation intent to obtain the chief complaint information, and imports it into the pre-consultation module. The existing AI program is one of o3-mini, deepseek-r1, gpt-4o, and o1-preview. The pre-visit module features a guidance function. Based on the chief complaint information provided by the medical history collection module, this function displays triage redirection windows in the AI ​​assistant window. These windows include intelligent triage, department search, doctor search, appointment booking, and doctor recommendation. The intelligent triage window uses a triage model to analyze imported chief complaint and patient information and output recommended departments. The department search, doctor search, appointment booking, and doctor recommendation windows access frequently used functions via links. The department search and doctor search windows allow patients to input department and doctor names, which are then searched from the hospital database for relevant recommendations. The appointment booking window completes the appointment booking within the hospital system based on the recommended doctors. The doctor recommendation window uses the patient's historical doctor information in the hospital database to recommend suitable doctors or specialists within the same department, and then redirects the patient to the appointment booking window.

[0007] As a further improvement to this method, the diagnostic reasoning model obtains diagnostic intent reasoning based on the execution of a 1-step thinking path in an existing AI program. The reasoning steps include: s1.1, break down the consultation information to obtain the chief complaint information, and then classify the chief complaint information; s1.2, the AI ​​program analyzes the chief complaint information and categorizes and organizes the patient's current condition based on the chief complaint information; s1.3, the AI ​​program performs medical reasoning based on the patient's current condition and outputs candidate diseases or differential diagnoses that require further information; s1.4, refine the reply content based on the scenario instruction.

[0008] As a further improvement to this method, the chief complaint information is classified into consultation, medication consultation, triage, unclear intent, medication purchase and prescription, and report interpretation. The consultation process is abstractly divided into standard process and non-standard process. The standard process is based on the seven elements of consultation, including collecting medical information, diagnosis decision, treatment plan, and doctor-patient consultation.

[0009] As a further improvement to this method, the triage model includes a departmental knowledge base, which contains different department names and corresponding user keywords. The analysis steps of the triage model include... s2.1, obtain the patient's chief complaint information and patient information, directly perform regular expression recognition on the basic information in the chief complaint information and patient information, and perform jieba dictionary recognition and model recognition after the chief complaint information and basic information are segmented. The basic information includes the patient's age and gender. s2.2 combines the department knowledge base for entity matching, regular expression recognition, jieba dictionary recognition and model recognition to output the department classification results. The department classification results contain medical entity information corresponding to the department. After cross-fusion, the recognition results are output. s2.3 Select the corresponding department result based on the medical entity, then perform department mapping based on the patient's file, adjust the department result, and output the triage result.

[0010] As a further improvement to this method, the cross-fusion prioritizes the recognition result of regular expression recognition. If the identified medical entity is a unified entity, the recognition result of regular expression recognition is applied. If it is not a unified entity, if the length of the medical entity obtained by jieba dictionary recognition and model recognition is the same, the recognition result of jieba dictionary recognition is applied. If the length of the medical entity obtained by model recognition is greater than the recognition result of jieba dictionary recognition, the recognition result of model recognition is applied.

[0011] As a further improvement to this method, the regular expression recognition matches the chief complaint information and basic information with user keywords in the department knowledge base, and outputs the matching results as medical entities. The matching rule is to first perform precise matching on the two, and if there are no results in precise matching, then perform fuzzy matching. The fuzzy matching algorithm sets the edit distance. When the word length is greater than 3 and the first character is the same, the matching is performed in a way that allows 1 edit distance for 4 or 5 characters and 2 edit distances for more than 5 characters.

[0012] As a further improvement to this method, the regular expression recognition also includes a voting mechanism. When multiple medical entities are obtained through fuzzy matching, they are grouped and voted on according to their priority from high to low, with a depth of 1. If the votes are consistent, no result is output. The entity priority is as follows: [Department] > [Department-specific] > [Traditional Chinese Medicine] > [Disease] > [Allergy history] > [Surgery] > [Treatment] > [Symptoms, location, examination] > [Traditional Chinese Medicine drugs, location, examination, others, high priority configuration] > [Drugs (Western medicine + health care products), location, examination, others, high priority configuration] > [Examination, location] > [Location, high priority configuration] > [Other, high priority configuration].

[0013] As a further improvement to this method, the jieba dictionary recognition is based on the probabilistic graphical PGM-Naive Bayes model to identify medical entities. It assumes that entities are mutually independent and calculates the contribution of different entities to each department. The calculation formula is as follows: in, Describes the j-th type of entity. Let i represent the i-th type of department; the posterior probability of the department is calculated as follows: The model recognition is based on the BERT model McBERT to identify medical entities.

[0014] As a further improvement to this method, the mapping rules for the department mapping include, 1) Pediatrics Department Mapping: If the patient is under 14 years old, the results will be mapped to the pediatrics department. 2) Traditional Chinese Medicine (TCM) Department Mapping: If the chief complaint information includes TCM-related TCM or Chinese patent medicine CME, then the mapping will be executed. If the product details are TCM-related, then the mapping will be executed from andrology to TCM andrology. 3) Mapping of psychotropic drugs: If psychotropic drugs are mentioned in the chief complaint or product details, then the neurology and geriatrics departments will be mapped to the psychiatry department. 4) Gender mapping: If a male patient is assigned to the gynecology department, he will be mapped to the andrology department; if a female patient is assigned to the andrology department, she will be mapped to the gynecology department.

[0015] After adopting the above method, the main entry module allows patients to input information, interact with the doctor, and use commonly used hospital functions for consultation and triage. Patients input consultation information into the AI ​​assistant window, and the reasoning model of the medical history collection module categorizes the chief complaint information. The AI ​​program then executes a one-step thinking process based on this categorization, which, compared to multi-step thinking, facilitates a smoother thinking process and helps obtain accurate chief complaint information.

[0016] The triage model processes the chief complaint information, performing regular expression recognition, jieba dictionary recognition, and model recognition respectively. Using a cross-fusion approach, when the medical entities output by regular expression recognition are verified by both jieba dictionary recognition and model recognition, the medical entities are identified as unified entities. When the results of regular expression recognition are inconsistent, jieba dictionary recognition and model recognition mutually verify each other. If the length of the medical entities is consistent, the jieba dictionary recognition result is applied; if the model recognition result is longer, the model recognition result is applied. This achieves a logically sound multi-level verification mechanism. Furthermore, before outputting the triage results, a department mapping method is used to guide patients to specific departments based on age, gender, and medication history. This reduces basic errors that may occur when patients have similar complaints, effectively minimizing triage deviations and providing accurate guidance for patients seeking medical treatment. Attached Figure Description

[0017] Figure 1 The diagram shown is a schematic of the online consultation platform framework of this patent.

[0018] Figure 2 The diagram shown illustrates the classification effect of the main complaint information in this patent.

[0019] Figure 3 The diagram shown is a schematic of the BERT model structure of this patent. Detailed Implementation

[0020] like Figure 1-3 As shown, to solve the above-mentioned technical problems, a pre-diagnosis triage system includes an online consultation platform, which has a main entry module, a medical history collection module, and a pre-visit module. The main entry module has a patient management window, which allows patients to input their information and synchronize it with the hospital's database. The AI ​​assistant window allows patients to input consultation information via text input and voice-to-text conversion, and interact with the medical history collection module. The function guide window displays frequently used function entries, which include links to existing functions in the hospital. The medical history collection module has a reasoning model that uploads the consultation information entered by the patient to the existing AI program. The AI ​​program then executes a 1-step thinking path to infer the patient's consultation intent, organizes the consultation intent to obtain the chief complaint information, and imports it into the pre-consultation module. The existing AI program is one of o3-mini, deepseek-r1, gpt-4o, and o1-preview. The pre-visit module features a guidance function. Based on the chief complaint information provided by the medical history collection module, this function displays triage redirection windows in the AI ​​assistant window. These windows include intelligent triage, department search, doctor search, appointment booking, and doctor recommendation. The intelligent triage window uses a triage model to analyze imported chief complaint and patient information and output recommended departments. The department search, doctor search, appointment booking, and doctor recommendation windows access frequently used functions via links. The department search and doctor search windows allow patients to input department and doctor names, which are then searched from the hospital database for relevant recommendations. The appointment booking window completes the appointment booking within the hospital system based on the recommended doctors. The doctor recommendation window uses the patient's historical doctor information in the hospital database to recommend suitable doctors or specialists within the same department, and then redirects the patient to the appointment booking window.

[0021] The aforementioned consultation reasoning model is based on existing AI programs executing a 1-step thinking path to obtain consultation intent reasoning. The reasoning steps include: s1.1, break down the consultation information to obtain the chief complaint information, and then classify the chief complaint information; s1.2, the AI ​​program analyzes the chief complaint information and categorizes and organizes the patient's current condition based on the chief complaint information; s1.3, the AI ​​program performs medical reasoning based on the patient's current condition and outputs candidate diseases or differential diagnoses that require further information; s1.4, refine the reply content based on the scenario instruction.

[0022] The chief complaint information is categorized into consultation, medication consultation, triage, unclear intent, medication purchase and prescription, and report interpretation. The consultation process is abstractly divided into standard and non-standard processes. The standard process is based on the seven elements of consultation, including collecting medical information, making diagnostic decisions, treatment plans, and doctor-patient consultation.

[0023] The triage model has a departmental knowledge base, which contains different department names and corresponding user keywords. The analysis steps of the triage model include... s2.1, obtain the patient's chief complaint information and patient information, directly perform regular expression recognition on the basic information in the chief complaint information and patient information, and perform jieba dictionary recognition and model recognition after the chief complaint information and basic information are segmented. The basic information includes the patient's age and gender. s2.2 combines the department knowledge base for entity matching, regular expression recognition, jieba dictionary recognition and model recognition to output the department classification results. The department classification results contain medical entity information corresponding to the department. After cross-fusion, the recognition results are output. s2.3 Select the corresponding department result based on the medical entity, then perform department mapping based on the patient's file, adjust the department result, and output the triage result.

[0024] The cross-fusion prioritizes the recognition result of regular expression recognition. If the identified medical entity is a unified entity, the recognition result of regular expression recognition is applied. If it is not a unified entity, if the length of the medical entity obtained by jieba dictionary recognition and model recognition is the same, the recognition result of jieba dictionary recognition is applied. If the length of the medical entity obtained by model recognition is greater than the recognition result of jieba dictionary recognition, the recognition result of model recognition is applied.

[0025] The regular expression recognition matches the chief complaint information and basic information with user keywords in the department's knowledge base, and outputs the matching results as medical entities. The matching rule is to first perform a precise match between the two. If no result is found in the precise match, then a fuzzy match is performed. The fuzzy match algorithm sets an edit distance. When the word length is greater than 3 and the first character is the same, the matching is performed in a way that allows 1 edit distance for 4 or 5 characters and 2 edit distances for more than 5 characters.

[0026] The regular expression recognition also includes a voting mechanism. When multiple medical entities are obtained through fuzzy matching, they are grouped and voted on according to their priority from high to low, with a depth of 1. If the votes are consistent, no result is output. The entity priority is as follows: [Department] > [Department-specific] > [Traditional Chinese Medicine] > [Disease] > [Allergy history] > [Surgery] > [Treatment] > [Symptoms, location, examination] > [Traditional Chinese Medicine drugs, location, examination, others, high priority configuration] > [Drugs (Western medicine + health care products), location, examination, others, high priority configuration] > [Examination, location] > [Location, high priority configuration] > [Other, high priority configuration].

[0027] As a further improvement to this method, the jieba dictionary recognition is based on the probabilistic graphical PGM-Naive Bayes model to identify medical entities. It assumes that entities are mutually independent and calculates the contribution of different entities to each department. The calculation formula is as follows: in, Describes the j-th type of entity. Let i represent the i-th type of department; the posterior probability of the department is calculated as follows: The model recognition is based on the BERT model McBERT to identify medical entities.

[0028] The mapping rules for the department mapping include, 1) Pediatrics Department Mapping: If the patient is under 14 years old, the results will be mapped to the pediatrics department. 2) Traditional Chinese Medicine (TCM) Department Mapping: If the chief complaint information includes TCM-related TCM or Chinese patent medicine CME, then the mapping will be executed. If the product details are TCM-related, then the mapping will be executed from andrology to TCM andrology. 3) Mapping of psychotropic drugs: If psychotropic drugs are mentioned in the chief complaint or product details, then the neurology and geriatrics departments will be mapped to the psychiatry department. 4) Gender mapping: If a male patient is assigned to the gynecology department, he will be mapped to the andrology department; if a female patient is assigned to the andrology department, she will be mapped to the gynecology department.

[0029] The main entry module allows patients to input information, interact with doctors, and use commonly used hospital functions for consultation and triage. Patients input consultation information into the AI ​​assistant window, and the medical history collection module's reasoning model categorizes the chief complaint information. The AI ​​program then executes a one-step thinking process based on this categorization, which, compared to multi-step thinking, facilitates a smoother thinking process and helps obtain accurate chief complaint information.

[0030] The triage model processes the chief complaint information, performing regular expression recognition, jieba dictionary recognition, and model recognition respectively. Using a cross-fusion approach, when the medical entities output by regular expression recognition are verified by both jieba dictionary recognition and model recognition, the medical entities are identified as unified entities. When the results of regular expression recognition are inconsistent, jieba dictionary recognition and model recognition mutually verify each other. If the length of the medical entities is consistent, the jieba dictionary recognition result is applied; if the model recognition result is longer, the model recognition result is applied. This achieves a logically sound multi-level verification mechanism. Furthermore, before outputting the triage results, a department mapping method is used to guide patients to specific departments based on age, gender, and medication history. This reduces basic errors that may occur when patients have similar complaints, effectively minimizing triage deviations and providing accurate guidance for patients seeking medical treatment.

Claims

1. A triage and consultation system for the pre-diagnosis stage, characterized in that: This includes an online consultation platform, which includes a main entry module, a medical history collection module, and a pre-consultation module. The main entry module has a patient management window, which allows patients to input their information and synchronize it with the hospital's database. The AI ​​assistant window allows patients to input consultation information via text input and voice-to-text conversion, and interact with the medical history collection module. The function guide window displays frequently used function entries, which include links to existing functions in the hospital. The medical history collection module includes a reasoning model that uploads the patient's input consultation information to an existing AI program, which then performs a 1-step process. The thinking path infers the patient's consultation intent, organizes the consultation intent to obtain the chief complaint information, and imports it into the pre-consultation module; the existing AI program is one of o3-mini, deepseek-r1, gpt-4o, and o1-preview; The pre-visit module has a guidance function. Based on the chief complaint information provided by the medical history collection module, the guidance function displays a triage jump window in the AI ​​assistant window. The types of triage jump windows include intelligent triage, department search, doctor search, appointment registration and doctor recommendation. The intelligent triage window is equipped with a triage model that analyzes imported chief complaint information and patient information to output recommended departments. The department search, doctor search, appointment booking, and doctor recommendation windows access frequently used functions via links. The department search and doctor search windows allow patients to input department and doctor names, which are then retrieved from the hospital database for recommendations. The appointment booking window completes appointment booking within the hospital system based on recommended doctors. The doctor recommendation window uses the patient's historical doctor information in the hospital database to recommend suitable doctors or specialists within the same department, and then redirects the patient to the appointment booking window.

2. The triage and consultation system for the pre-diagnosis stage according to claim 1, characterized in that: The aforementioned consultation reasoning model is based on existing AI programs executing a 1-step thinking path to obtain consultation intent reasoning. The reasoning steps include: s1.1, break down the consultation information to obtain the chief complaint information, and then classify the chief complaint information; s1.2, the AI ​​program analyzes the chief complaint information and categorizes and organizes the patient's current condition based on the chief complaint information; s1.3, the AI ​​program performs medical reasoning based on the patient's current condition and outputs candidate diseases or differential diagnoses that require further information; s1.4, refine the reply content based on the scenario instruction.

3. The triage and consultation system for the pre-diagnosis stage according to claim 2, characterized in that: The chief complaint information is categorized into consultation, medication consultation, triage, unclear intent, medication purchase and prescription, and report interpretation. The consultation process is abstractly divided into standard and non-standard processes. The standard process is based on the seven elements of consultation, including collecting medical information, making diagnostic decisions, treatment plans, and doctor-patient consultation.

4. The triage and consultation system for the pre-diagnosis stage according to claim 1, characterized in that: The triage model has a departmental knowledge base, which contains different department names and corresponding user keywords. The analysis steps of the triage model include... s2.1, obtain the patient's chief complaint information and patient information, directly perform regular expression recognition on the basic information in the chief complaint information and patient information, and perform jieba dictionary recognition and model recognition after the chief complaint information and basic information are segmented. The basic information includes the patient's age and gender. s2.2 combines the department knowledge base for entity matching, regular expression recognition, jieba dictionary recognition and model recognition to output the department classification results. The department classification results contain medical entity information corresponding to the department. After cross-fusion, the recognition results are output. s2.3 Select the corresponding department result based on the medical entity, then perform department mapping based on the patient's file, adjust the department result, and output the triage result.

5. A triage and consultation system for the pre-diagnosis stage according to claim 4, characterized in that: The cross-fusion prioritizes the recognition result of regular expression recognition. If the identified medical entity is a unified entity, the recognition result of regular expression recognition is applied. If it is not a unified entity, if the length of the medical entity obtained by jieba dictionary recognition and model recognition is the same, the recognition result of jieba dictionary recognition is applied. If the length of the medical entity obtained by model recognition is greater than the recognition result of jieba dictionary recognition, the recognition result of model recognition is applied.

6. A triage and consultation system for the pre-diagnosis stage according to claim 5, characterized in that: The regular expression recognition matches the chief complaint information and basic information with user keywords in the department's knowledge base, and outputs the matching results as medical entities. The matching rule is to first perform a precise match between the two. If no result is found in the precise match, then a fuzzy match is performed. The fuzzy match algorithm sets an edit distance. When the word length is greater than 3 and the first character is the same, the matching is performed in a way that allows 1 edit distance for 4 or 5 characters and 2 edit distances for more than 5 characters.

7. A triage and consultation system for the pre-diagnosis stage according to claim 6, characterized in that: The regular expression recognition also includes a voting mechanism. When multiple medical entities are obtained through fuzzy matching, they are grouped and voted on according to their priority from high to low, with a depth of 1. If the votes are consistent, no result is output. The entity priority is as follows: [Department] > [Department-specific] > [Traditional Chinese Medicine] > [Disease] > [Allergy history] > [Surgery] > [Treatment] > [Symptoms, location, examination] > [Traditional Chinese Medicine drugs, location, examination, others, high priority configuration] > [Drugs (Western medicine + health care products), location, examination, others, high priority configuration] > [Examination, location] > [Location, high priority configuration] > [Other, high priority configuration].

8. A triage and consultation system for the pre-diagnosis stage according to claim 5, characterized in that: The jieba dictionary recognition method identifies medical entities based on a probabilistic graphical PGM-Naive Bayes model. It assumes that entities are independent of each other and calculates the contribution of different entities to each department using the following formula: , in, Describes the j-th type of entity. Let i represent the i-th type of department; the posterior probability of the department is calculated as follows: The model recognition is based on the BERT model McBERT to identify medical entities.

9. A triage and consultation system for the pre-diagnosis stage according to claim 4, characterized in that: The mapping rules for the department mapping include, 1) Pediatrics Department Mapping: If the patient is under 14 years old, the results will be mapped to the pediatrics department. 2) Traditional Chinese Medicine (TCM) Department Mapping: If the chief complaint information includes TCM-related TCM or Chinese patent medicine CME, then the mapping will be executed. If the product details are TCM-related, then the mapping will be executed from andrology to TCM andrology. 3) Mapping of psychotropic drugs: If psychotropic drugs are mentioned in the chief complaint or product details, then the neurology and geriatrics departments will be mapped to the psychiatry department. 4) Gender mapping: If a male patient is assigned to the gynecology department, he will be mapped to the andrology department; if a female patient is assigned to the andrology department, she will be mapped to the gynecology department.

Images