Intelligent hospital guide system

Abstract

The invention discloses an intelligent hospital guide system, which belongs to the field of medical information and is characterized in that medical terms are converted into uncomfortable symptom description which is easy to recognize by a patient by deeply analyzing characteristics of common diseases and combining a natural language processing technology, so that the user friendliness of the system is improved; approximate Bayes inference is carried out by using cosine similarity, an improved C4.5 decision tree algorithm is combined, a decision tree is dynamically constructed to generate accurate question and answer logic, accurate matching and department recommendation of patient symptoms are realized, and the accuracy and efficiency of hospital guidance can be improved; in conclusion, the system provided by the invention effectively solves the defects of traditional manual hospital guide, improves the registration accuracy, and promotes the overall improvement of the medical service quality by reducing the hospital guide cost and improving the working efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of medical information technology, and more specifically, relates to an intelligent diagnosis guidance system. Background Art

[0002] In the current medical system, manual medical guidance faces problems such as communication barriers, inconsistent accuracy and high labor costs. As the difficulties faced by manual medical guidance become increasingly prominent, the research and development of intelligent medical guidance robots at home and abroad has gradually attracted widespread attention. At present, the development of intelligent medical guidance robots has made certain progress, mainly focusing on the use of artificial intelligence, the Internet of Things, motion control and other technologies to give robots functions such as medical guidance, epidemiological history screening, voice interaction, and navigation, so as to improve the convenience and efficiency of patients' medical treatment.

[0003] In China, the research focus of intelligent medical guidance robots is to integrate a variety of advanced technologies, such as robot disinfection, face recognition, voice interaction, autonomous positioning, path planning, etc. Through the application of these key technologies, intelligent medical guidance robots can identify the diagnosis and treatment link of patients, make reasonable predictions and provide corresponding medical guidance services, thereby improving the efficiency of patients' medical treatment, reducing the time of hospital stay, reducing the number of medical guidance workers, and relieving the work pressure of medical staff.

[0004] In foreign countries, the development of intelligent guidance robots focuses more on the field of robot-assisted medical care, such as robot-assisted surgery (RAS) and assistant robots for nursing staff. Among them, robot-assisted surgical systems have been widely used in some fields, such as the active robot system (ROBODOC) in hip replacement surgery, which can reduce the risk of surgical complications. As for assistant robots for nursing staff, they are mainly used to complete some simple and repetitive tasks, such as picking up objects and disinfection.

[0005] Intelligent medical guidance technologies at home and abroad are committed to improving the convenience and efficiency of patients' medical treatment, reducing the workload of medical staff, and optimizing the allocation of medical resources, thereby promoting the intelligent and digital development of medical services. Summary of the invention

[0006] In view of the above defects or improvement needs of the prior art, the present invention provides an intelligent medical guidance system, which aims to solve the shortcomings of traditional manual medical guidance and can quickly and accurately recommend medical departments according to the patient's condition, thereby improving the hospital's medical efficiency.

[0007] To achieve the above object, according to a first aspect of the present invention, an intelligent medical guidance system is provided, comprising:

[0008] The preprocessing module is used to establish the corresponding relationship between the treatment departments, disease sites and symptoms corresponding to various diseases, take all disease sites as a set Place, take the symptoms that are easy for patients to identify as a set Status, and construct an easily identifiable symptom set Feature, where Feature = Place × Status; input each disease into the corpus to obtain the sentences related to it and perform word unit decomposition, calculate the similarity between each word unit and each easily identifiable symptom in Feature, and determine the weight of each easily identifiable symptom according to the similarity, so as to form an easily identifiable symptom weight vector for each disease;

[0009] The interactive module is used to display the human-shaped area so that users can select the corresponding diseased area according to their own pain conditions;

[0010] The question-answering module is used to ask the user for confirmation information of the easily identifiable symptoms related to the onset site selected by the user to construct a target easily identifiable symptom weight vector; wherein, when the jth round of questioning is conducted to the user, the easily identifiable symptoms with the largest relative increase in information entropy are selected from the unselected easily identifiable symptoms related to the onset site and asked to the user; j≥0;

[0011] The treatment department confirmation module is used to calculate the similarity between the target easily identifiable symptom weight vector and the easily identifiable symptom weight vector of each disease, and take the department where the disease to which the easily identifiable symptom weight vector with the greatest similarity to the target easily identifiable symptom weight vector belongs as the recommended treatment department.

[0012] According to a second aspect of the present invention, there is provided an intelligent diagnosis guidance method, comprising:

[0013] S1, establish the corresponding relationship between the treatment department, the site of onset and the symptoms corresponding to various diseases, take all the sites of onset as the set Place, take the symptoms that are easy for patients to identify as the set Status, and construct the easily identifiable symptom set Feature, where Feature = Place × Status; input each disease into the corpus to obtain the sentences related to it and perform word unit decomposition, calculate the similarity between each word unit and each easily identifiable symptom in Feature, and determine the weight of each easily identifiable symptom according to the similarity, and form the easily identifiable symptom weight vector of each disease;

[0014] S2, displays a human-shaped area so that the user can select the corresponding diseased area according to his or her own pain condition;

[0015] S3, according to the onset site selected by the user, asking the user for confirmation information of the easily identifiable symptoms related to the onset site to construct a target easily identifiable symptom weight vector; wherein, when the j-th round of asking the user is performed, the easily identifiable symptoms with the largest relative increase in information entropy are selected from the unselected easily identifiable symptoms related to the onset site and asked to the user;

[0016] S4, respectively calculating the similarity between the target easily identifiable symptom weight vector and the easily identifiable symptom weight vector of each disease, and taking the department of the disease to which the easily identifiable symptom weight vector having the greatest similarity to the target easily identifiable symptom weight vector belongs as the recommended department for consultation.

[0017] According to a third aspect of the present invention, there is provided an electronic device, comprising: a computer-readable storage medium and a processor;

[0018] The computer-readable storage medium is used to store executable instructions;

[0019] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method as described in the second aspect.

[0020] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the method described in the second aspect.

[0021] According to a fifth aspect of the present invention, there is provided a computer program product, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the method as described in the second aspect is implemented.

[0022] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0023] 1. The system provided by the present invention, through in-depth analysis of the characteristics of common diseases, combined with natural language processing technology, converts medical terms into descriptions of discomfort symptoms that are easy for patients to identify, thereby improving the user-friendliness of the system; uses cosine similarity for approximate Bayes inference, combined with an improved C4.5 decision tree algorithm, dynamically constructs a decision tree to generate accurate question-answering logic, achieves accurate matching of patient symptoms and department recommendations, and can improve the accuracy and efficiency of medical guidance; wherein, the C4.5 algorithm is a classic algorithm for generating decision trees, which can select split attributes through information gain rate, overcoming the deficiency of the ID3 algorithm that tends to select attributes with multiple attribute values ​​as split attributes through information gain; at the same time, it can handle discrete and continuous attribute types, that is, discretize continuous attributes; it can perform pruning operations after constructing a decision tree; it can handle training data with missing attribute values. Considering that the C4.5 algorithm is a deterministic reasoning, in the present invention, when asking the patient "whether he has a fever", if the patient answers "no", the C4.5 algorithm will exclude all diseases that do not include "fever". Therefore, the present invention adopts the improved C4.5 algorithm to further improve and innovate the key technologies and applications of the intelligent medical guidance robot to meet the growing medical needs and challenges. Among them, similar to the C4.5 algorithm, the improved C4.5 algorithm used in the present invention still uses the incremental ratio of the conditional entropy expectation to be greedy, but the present invention adopts possibility reasoning, and the object of calculating entropy is probability distribution. In summary, the system provided by the present invention effectively solves the shortcomings of traditional manual medical guidance, improves the accuracy of registration, and promotes the overall improvement of medical service quality by reducing medical guidance costs and improving work efficiency.

[0024] 2. Furthermore, the system provided by the present invention aims at the problem of elderly patients’ weak understanding and learning ability, and uses the human figure to click on the pain points and voice consultation interaction to understand the disease status of elderly patients. Through big data analysis and the existing medical system database, the cause of the disease and the corresponding department are analyzed to achieve a high registration accuracy.

[0025] 3. Furthermore, the system provided by the present invention also includes an expansion module for calling the ChatGPT interface to provide reference diagnosis suggestions (auxiliary function). Through the ChatGPT interface, patients can obtain the diagnosis and treatment suggestions generated by it through voice broadcast, providing patients with additional reference information, allowing patients to understand the relevant treatment plans for the disease in advance, and also improving the targeted communication between patients and doctors during actual diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 One of the schematic diagrams of the intelligent medical guidance system provided by an embodiment of the present invention;

[0027] Figure 2The second schematic diagram of the intelligent medical guidance system provided by the embodiment of the present invention;

[0028] Figure 3 A schematic diagram of data flow of the intelligent medical guidance system provided by an embodiment of the present invention;

[0029] Figure 4 A pseudo code diagram of the Decomposition algorithm provided in an embodiment of the present invention;

[0030] Figure 5 A pseudo code diagram of the Vectorization algorithm provided in an embodiment of the present invention;

[0031] Figure 6 A pseudo code diagram of the decision tree question-answering logic provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] Due to the relative lack of medical knowledge among the majority of patients and the shortage of manual guidance resources, it is difficult for patients to quickly and accurately confirm the department to be treated. Based on this, the embodiment of the present invention provides an intelligent guidance system, such as Figure 1 As shown, including:

[0034] The preprocessing module is used to establish the corresponding relationship between the treatment departments, disease sites and symptoms corresponding to various diseases, take all disease sites as a set Place, take the symptoms that are easy for patients to identify as a set Status, and construct an easily identifiable symptom set Feature, where Feature = Place × Status; input each disease into the corpus to obtain the sentences related to it and perform word unit decomposition, calculate the similarity between each word unit and each easily identifiable symptom in Feature, and determine the weight of each easily identifiable symptom according to the similarity, so as to form an easily identifiable symptom weight vector for each disease;

[0035] An interactive module for displaying humanoid regions so that users can select the corresponding body part according to their own pain location;

[0036] Specifically, the humanoid area displayed in the interactive module is as follows Figure 2As shown in the figure, considering that the vision and comprehension reaction ability of elderly patients gradually decline, it is more difficult for them to read text, operate mobile phones and understand the meaning of symptom terms. Graphics are used to guide patients to register in a non-textual and intuitive way to solve the problem of elderly patients' difficulty in reading. Patients can narrow down the scope of the disease by clicking on the uncomfortable part.

[0037] The question-and-answer module is used to ask the user for confirmation information of easily identifiable symptoms related to the onset site selected by the user to construct a target easily identifiable symptom weight vector; wherein, when the j-th round of questioning is conducted to the user, the easily identifiable symptoms with the largest relative increase in information entropy are selected from the easily identifiable symptoms related to the onset site that have not been selected and asked to the user.

[0038] Further, considering that the accent problem is a common phenomenon in languages ​​everywhere. For example, in some dialects, it is not possible to clearly distinguish between "z" and "zh", "c" and "ch", "s" and "sh", front nasal and back nasal, etc., and it is more difficult for the elderly to overcome their pronunciation problems. The present invention obtains some information of the patient through k rounds of questions and answers (where k is dynamically determined according to the question and answer situation), and sets each question in the question and answer system to be in the form of "Do you think you have the characteristic attribute" and accompanied by three options of "yes", "no", and "uncertain", and further determines the department corresponding to the disease within the disease range narrowed by clicking on the human-shaped area.

[0039] The treatment department confirmation module is used to calculate the similarity between the easily identifiable symptom weight vector of each disease and the target easily identifiable symptom weight vector, and take the department where the disease to which the easily identifiable symptom weight vector with the greatest similarity to the target easily identifiable symptom weight vector belongs as the recommended treatment department.

[0040] The consultation department confirmation module is an auxiliary registration and consultation information window. This module determines the possible range of symptoms based on the human-shaped clicks and question-and-answer system, and selects the most appropriate target department by comprehensively considering the commonness of the disease and the degree of conformity with the description determined by the question and answer. Since the judgment of the disease has been evaluated for accuracy, it should be possible to give an effective probability assessment of the target department in combination with the prevalence of the disease. According to the level of this probability, the system should be able to realize direct registration, registration after inquiry and confirmation, or contact manual guidance service. Before registration is completed, the patient should be informed of the disease judgment made by the system, brief typical symptoms of the disease, and information about the target department to be registered. When the effective probability assessment value is not high, the patient should be asked whether to confirm. After registration is completed, the patient should be informed of the queue position, the information of the doctor sitting in the department, and the current diagnosis and treatment situation, so as to fully realize the intelligent registration service.

[0041] Furthermore, the system also includes an extension module for calling the ChatGPT interface to provide reference diagnosis suggestions (auxiliary functions): the patient's historical answers during the number-queuing process, the click information in the human area, and the system's disease judgment will be converted into natural language text through special processing. The generated text will describe the patient's basic situation and to a certain extent take into account the ambiguity in the patient's answer and the credibility of the system's own judgment. By calling the ChatGPT API, this text will be used to provide inquiry information to the ChatGPT that has been specially set up (informing ChatGPT in advance that he hopes to provide some medical advice), and then obtain the diagnosis and treatment suggestions generated by it and inform the patient through voice broadcast. This can provide additional reference information for patients, allowing patients to understand the relevant treatment plans for the disease in advance, and will also improve the pertinence of communication between patients and doctors during actual diagnosis. It is worth noting that in implementation, it will be emphasized to patients that the diagnostic suggestions are for reference only, because the information generated by ChatGPT itself may be misleading.

[0042] The algorithm adopted by the system provided by the present invention is described below.

[0043] 1. Data Collection

[0044] The collection of basic medical information is one of the primary tasks in the construction of the intelligent medical guidance system. The present invention collects a large amount of medical information based on the latest version of medical textbooks, including specific common symptoms, relevant specific departments, detailed anatomical divisions of the human body, etc. These data come from multiple channels, and in order to ensure the quality and accuracy of the data, the data are evaluated and screened.

[0045] A large number of common diseases were selected, and the corresponding departments, parts, and symptoms of these diseases were sorted out and organized into a disease database for use. The data example is shown in Table 1.

[0046] Table 1 Disease database table entry format

[0047]

[0048] 2. Data preprocessing

[0049] In order to apply disease-related information to the question-answering generation logic, it is necessary to convert highly professional medical terms into popular expressions that are understandable, perceptible, and judgeable to patients. The present invention applies natural language processing technology with rules and databases to solve the difficulty of large data volumes and difficulty in manual processing.

[0050] Patients who come to the doctor feel discomfort, which may or may not be specifically located, but most people can identify the following sensations and lesions:

[0051] (1) Visible injuries or lesions: skin damage, bone trauma, rash, etc.;

[0052] (2) Invisible physical pain: pain, itching, soreness, swelling, blockage, fatigue, etc.;

[0053] (3) Invisible mental illness: dizziness, irritability, etc.;

[0054] (4) Discomfort in special parts of the body: dizziness, nosebleed, runny nose, etc.

[0055] In the present invention, two specific sets are fixed:

[0056] Place: = {some common discomfort parts}, Status = {some specific types of discomfort that are easy for patients to identify}

[0057] In the present invention, each pair of "position" + "specific discomfort" is called an "easily identifiable symptom", and then all the "easily identifiable symptoms" are selected as the whole of the feature attributes, that is, Feature: = Place × Status, where × represents the Cartesian product between sets.

[0058] In fact, not all Cartesian product pairs are meaningful, and some meaningless pairs need to be removed in actual implementation. For example, assuming Place = {ear, nose}, Status = {hearing loss, reduced sense of smell}, the Cartesian product of the set Place and Status contains not only normal pairs, but also two abnormal pairs (ear, reduced sense of smell) and (nose, hearing loss), which do not have reasonable semantics and need to be removed. It is understandable that in most cases, the Cartesian product can efficiently generate a large number of meaningful pairs, such as Place = {arm, thigh, skin}, Status = {pain, itching, damage}, and there is no need to delete any elements at this time.

[0059] At present, based on the data collected above, each "disease" item can be decomposed into the "easily recognizable symptoms" it manifests, and this relationship is regarded as a mapping

[0060]

[0061] in is a subset of Feature, which is used to represent the set of “easily identifiable symptoms” that the disease can be decomposed into. is function, and satisfies here Used to reflect the relative importance of feature f among all the "easily identifiable symptoms" manifested by the disease.

[0062] For example, let disease s = heart failure, feature f1 = (throat, dry cough), f2 = (heart, accelerated or irregular beating), f3 = (head, dizziness), some medical knowledge and statistical experience derive the following relationship.

[0063] Decomposition:s→({f1,f2,f3},[f1→0.1,f2→0.5,f3→0.4])

[0064] This means that the disease "heart failure" is manifested in the three "easily identifiable symptoms" of "dry throat and cough", "rapid or irregular heartbeat", and "dizziness", with the latter two being the main ones, and the phenomenon of "rapid or irregular heartbeat" often occurs more frequently.

[0065] In order to support the calculation of this mapping relationship, it is necessary to collect and establish a so-called corpus database, which stores the bigrams.

[0066] (Disease name, a sentence describing a specific manifestation of the symptom)

[0067] By obtaining relevant information from mainstream medical information websites and downloading pre-trained medical-related word vector libraries, it is assumed that each string str can be Get its word vector representation Word2Vec(str) at the cost of .

[0068] The algorithm Decomposition receives the corpusDatabase and a specific disease s, and finally outputs the following: The pseudo code description of the algorithm is as follows Figure 4 As shown in Figure 1, this algorithm relies on a predefined constant, similarityCriticalValue, to determine whether the semantics overlap. In line 8, it is assumed that each inner product is positive.

[0069] It is understandable that the data collected in Section 1 above, although professional and comprehensive, is not easy to process by the program. The reasons include two aspects:

[0070] (1) For a specific disease such as "heart failure", the "symptom attributes" in the corresponding table may be descriptions such as "heart enlargement, increased ventricular volume at end-systole", etc. These descriptions are suitable for clinical diagnosis but not for self-diagnosis by patients.

[0071] (2) The degree of overlap between symptoms of different diseases is low. Six main symptoms were selected for each disease. However, in most cases, many symptoms of one disease do not appear in the symptoms of other diseases at all, which means that the classification is too detailed (considering the extreme case, if the symptoms corresponding to each disease do not overlap with the symptoms of other diseases, it means that each disease is described by some symptoms that are unique to it. Suppose the algorithm wants to determine the disease by asking "Does the patient have such and such symptoms?" At this time, each question can only determine whether the patient has a certain disease, that is, at most one wrong answer can be eliminated. However, assuming that half of the diseases have the common symptom of "fever", then as long as questions related to "fever" are asked, half of the wrong options can be eliminated. Obviously, the latter is more efficient. Therefore, it is hoped that the degree of overlap between symptoms of different diseases is high).

[0072] For these reasons, a number of "Easy Recognizable Symptoms" were selected, and the meaning of "Easy Recognizable Symptoms" was generally to solve the above two problems. We hope that these symptoms are easy to understand from the name, and on the other hand, we can cover all diseases as much as possible with as few "Easy Recognizable Symptoms" as possible. Obviously, a disease should naturally be associated with several different "Easy Recognizable Symptoms". Furthermore, a disease is often more likely to show some symptoms and more rarely show other symptoms, so the following format is used to represent a disease:

[0073]

[0074] The text separated by commas in square brackets is the "easy-to-identify symptoms", and the number below the text indicates the weight of the symptom. However, the data collected in Section 1 is numerous, and it would be time-consuming to manually convert a symptom into a weighted combination of "easy-to-identify symptoms" based on professional knowledge and evaluation, and it would not make full use of the data that has been collected. Therefore, the focus of data preprocessing is to build a "converter" that performs the above conversion. The converter is implemented through the Decomposition algorithm, which uses a pre-collected medical corpus to complete this conversion. The operation process is as follows:

[0075] (1) First, the algorithm receives a disease name s that needs to be converted as input;

[0076] (2) Next, search for all descriptive sentences about s in the corpus. These sentences will form a long paragraph.

[0077] (3) The algorithm then uses lemmatization to break down this long paragraph into individual lemmas;

[0078] (4) cyclically examine each word unit, analyze which "easy-to-identify symptoms" the word unit is more closely associated with, and assign bonus points to the corresponding "easy-to-identify symptoms" accordingly;

[0079] (5) After examining all the words, each "easily identifiable symptom" will receive a bonus score. By presetting a threshold, only those "easily identifiable symptoms" that are greater than the threshold will be considered to be related to the disease. Furthermore, the relative size of these scores also determines the weight of these "easily identifiable symptoms".

[0080] Example:

[0081] Demonstrate the process of algorithm operation for s=heart failure.

[0082] (1) First, the algorithm receives the disease “heart failure”;

[0083] (2) Next, searching the corpus for “description of heart failure” may yield the following sentences:

[0084] The main symptoms of heart failure include dyspnea, limited mobility and fluid retention. When the condition worsens, symptoms such as pulmonary congestion, shortness of breath, chest tightness, edema of both lower limbs, fatigue and dizziness may occur.

[0085] (3) The algorithm then tokenizes. After removing stop words, this might give the following token sequence:

[0086] Dyspnea, limited mobility, fluid retention, exacerbation, possible, pulmonary congestion, chest tightness, edema of both lower limbs, exertion, dizziness

[0087] (4) Next, examine each word in a loop. Assuming that the word being examined is "dyspnea", analyze the degree of association between each "easily identifiable symptom" and "dyspnea", which will be represented by a number from 0 to 1. Add this number to the "scorer" of the corresponding "easily identifiable symptom". For example, the degree of association between "dyspnea" and "dry cough in the throat" is 0.5, so the scorer corresponding to "dry cough in the throat" will be increased by 0.5.

[0088] (5) After examining all the words, filter the “easily identifiable symptoms” according to the algorithm description. Suppose that “dry cough in the throat”, “accelerated or irregular heartbeat”, and “dizziness” are selected, with scores of 10, 50, and 40 respectively. The weights of the three are set to 0.1, 0.5, and 0.4 respectively.

[0089] 3. Matching degree calculation

[0090] We will obtain some information about the patient through m rounds of questions and answers (where m is determined dynamically based on the situation of the questions and answers). Each question will be in the form of "Do you think you have characteristic attribute f?" and will be accompanied by three options: "Yes", "No", and "Unsure". The next section will introduce how the number of rounds m and the characteristic attribute f for each question are selected. Assuming that the user's answer has been obtained, it is regarded as a mapping

[0091]

[0092] where f1,…,f m is the characteristic attribute of the questions asked in rounds 1 to m, a j It is the answer about the jth round. The three choices "yes", "no" and "uncertain" will correspond to 1, 0 and 0.5 respectively. Extend the mapping to Feature with the default value 0 to get the mapping Answer.

[0093] The algorithm Vectorization receives the disease database illnessDatabase, the corpus database corpusDatabse and a specific disease i established above, and finally outputs a mapping of Feature→[0,1]. The pseudo code description of the algorithm is as follows Figure 5 As shown:

[0094] Now, assuming that the patient's answer has generated the final Answer, for each possible disease i, define the matching degree

[0095]

[0096] The department corresponding to the disease with the highest matching degree is the registration department that the intelligent medical guidance system should recommend.

[0097] It is understandable that in Section 1, all diseases have been represented as weighted combinations of “easily identifiable symptoms”, so the next tasks include:

[0098] Ask “wisely” questions: Ask the patient if he has a certain “easily recognizable symptom”;

[0099] “Accurate” judgment: Determine the department you should go to based on the answer results.

[0100] When the actual algorithm is running, the above two tasks need to be completed alternately. For the sake of convenience, first assume that we ask questions "randomly". That is, each time we randomly select an "easy-to-identify symptom" that has not been asked and ask the patient whether he has the symptom. After several rounds of answers, we have obtained a rough outline of what symptoms the patient has. The goal of the matching calculation is to calculate the probability of the patient suffering from each disease based on these answers. This is achieved through the algorithm Vectorization. The core idea of ​​the algorithm is to use cosine similarity to evaluate the matching degree.

[0101] A disease s can be represented as a series of weighted "easily identifiable symptoms" combinations f1,···,fm, where f1,···,fm is only a small part of the "easily identifiable symptoms". The remaining easily identifiable symptoms fm+1,···,fn are all taken into consideration and their corresponding weights are set to 0. Thus, a disease s is actually converted into an n-dimensional vector, where the first component is the weight corresponding to the "easily identifiable symptom" f1, and the second component is the weight corresponding to the "easily identifiable symptom" f2... And the patient's answer to the easily identifiable symptoms asked can also be converted into an n-dimensional vector according to the following rules:

[0102] Assume that “easily identifiable symptoms” fj has been asked, and assign 1, 0.5, or 0 to the jth component of the vector, depending on whether the patient answered “yes,” “don’t know,” or “no”;

[0103] Assuming that the “easily identifiable symptom” fj has not been asked, the jth component is set to 0.

[0104] Therefore, the algorithm flow is:

[0105] (1) First, the algorithm receives the questions asked and the patient’s answers as input;

[0106] (2) Next, the algorithm constructs the patient's answers into a vector according to the above method, and evaluates the cosine similarity between the vector and the vector corresponding to each disease one by one;

[0107] (3) For diseases that belong to the same department, the sum of their similarities will be used as the “score” of the department;

[0108] (4) The algorithm selects the department with the largest score as the answer.

[0109] Example:

[0110] Suppose a patient has heart failure and feels shortness of breath and dizziness. If we ask questions about the three "easily recognizable symptoms" of "fever", "rapid or irregular heartbeat", and "dizziness", the patient's answers are "no", "don't know", and "yes" respectively. At this time, the representation vector of the disease "heart failure" is:

[0111]

[0112] The patient's answer corresponds to the vector:

[0113]

[0114] The cosine similarity between them is 0.8971, so 0.8971 will be added to the scorer of "cardiovascular medicine" corresponding to "heart failure". After the above analysis and calculation for each disease, assuming that if the score on the scorer of "cardiovascular medicine" is the highest, the algorithm can make a recommendation and recommend that the patient register for "cardiovascular medicine".

[0115] 4. Dynamic generation of decision trees

[0116] The previous section describes how to determine the correct registration department after obtaining the patient's final answer. This section describes the algorithm for generating the question-answering logic (i.e., the question decision tree) used to obtain those answers. This algorithm is essentially a modified version of the C4.5 algorithm.

[0117] Remember PartialAnswer j The patient’s answer after the jth (j≥0) round of questions and answers, Answer j is the extended version, and its specific format is as described in the previous section. This answer will induce a mapping:

[0118]

[0119] This is a normalized function that can be used to calculate information entropy (p j )∶=-∑ i∈illness p j (i)log2 p j (i) Where:

[0120]

[0121] Fixed f∈Feature\dom(PartialAnswer j ) and q∈{0,0.5,1}, define:

[0122] Answer j (f,q) :PartialAnswer j ∪[f→q] extends to Feature

[0123] This will similarly induce the normalized function p′ f,qThe conditional entropy (p j |(f,q))∶=Entropy(p′ f,q ), conditional entropy expectation:

[0124]

[0125] It can be understood as the amount of information expected to be obtained after selecting the question feature attribute f.

[0126] Among them, Entropy(p′ f,q )=-∑ i∈illness p′ f,q (i)log2p′ f,q (i);

[0127] Answer j It is a vector generated by the user's first j rounds of answers (or actually a correspondence relationship), which corresponds each f in Feature to an item in {0, 0.5, 1}. The corresponding rule is:

[0128]

[0129] Fix f to be a feature attribute that has not been asked before, q∈{0,0.5,1}, then Answer j (f,q) is in the Answer j Based on , the value at f is modified to the vector (or corresponding relationship) obtained by q, and the corresponding rule becomes:

[0130]

[0131] For example, assuming that j = 2, Feature = {fever, dry throat, accelerated or irregular heartbeat, dizziness}, and assuming that the first two rounds of questions asked about the two symptoms "fever" and "dry throat", the corresponding vector of the answer is:

[0132]

[0133] In this case, there are:

[0134]

[0135] Evaluate the relative increase in information entropy:

[0136]

[0137] The idea of ​​the improved C4.5 algorithm used in the present invention is similar to that of the C4.5 algorithm. After the jth round of question-answering, the feature attribute with the largest EntropyGainRatio(j,·) value is greedily selected from the unselected feature attributes to ask questions. The pseudo code description of the question-answering logic is as follows: Figure 6 shown.

[0138] QA is not an independent algorithm, but a collection of process calls. PostQuestion(·) is a module that encapsulates feature attributes into questions and sends them to the user end. GetUserAnswer(·) is a module that obtains answers from the user end and correctly updates relevant global variables (especially Answer j ) is a module for calculating the relative increment of information entropy (which depends on the correct value of the global variable). QA is granted the permission to call these modules, which actually constitutes a process.

[0139] It is understandable that the purpose of this section is to make the questions more discriminative. For example, if the patient has already indicated that there is no trauma, you should not ask about leg fractures, but instead ask about a symptom that can effectively narrow the search scope among the "easy to identify symptoms" of no trauma. And this question must be based on the questions asked previously and the answers the patient has already given. Therefore, the process of building a decision tree is actually the following process of asking, analyzing, and re-asking:

[0140] (1) Choose the best “easily recognizable symptom” to ask based on the patient’s answers;

[0141] (2) After asking questions about the symptom, record the patient's answers and then return to step (1).

[0142] Similar to the C4.5 algorithm, the improved C4.5 algorithm used in the present invention also uses the expected increment ratio of conditional entropy to perform greedy. However, the C4.5 algorithm is a deterministic reasoning. If the patient answers "no" when asked "Do you have a fever?", the algorithm will directly exclude all diseases that do not include "fever". Therefore, the improved C4.5 algorithm used in the present invention uses possibility reasoning, and the object of entropy calculation is probability distribution.

[0143] The quality of a question depends on the "situation" after the question is asked. Specifically, the "situation" is nothing more than the probability that the patient should go to each department based on the current information. This probability can be reflected by the relative size of the score calculated by the algorithm in Section 3. We first propose an indicator to measure the two "situations". Observation:

[0144] Situation 1: The probabilities of the patient being admitted to the five undetermined departments are 0.2, 0.2, 0.2, 0.2, 0.2 respectively;

[0145] Situation 2: The probabilities of the patient being admitted to the five undetermined departments are 0.0, 0.1, 0.2, 0.3, and 0.4 respectively.

[0146] Obviously, we would rather see situation 2 than situation 1, because situation 2 obviously contains more information, while all we can do in the face of situation 1 is guesswork. The concept of "amount of information" is captured by information entropy in information theory. The core of the C4.5 algorithm is to choose the "easily identifiable symptom" that can bring the greatest "information entropy gain" to ask questions. Since it is impossible to predict the patient's possible answers before asking questions, it is necessary to weigh the average gains brought by the patient's three answers of "yes", "don't know", and "no". This concept is captured by the conditional entropy expectation in information theory.

[0147] That is, for each "easily identifiable symptom" that can be asked, calculate the relative increase in information entropy that can be brought about by asking the question in an average sense, and then select the question with the largest ratio to ask.

[0148] In summary, the system provided by the present invention integrates an intelligent medical guidance robot system in the existing registration machine equipment of the hospital, so that it is directly connected to the registration machine equipment to realize the offline intelligent humanized medical guidance mode. The system does not require additional hardware investment and can be directly applied to the existing facilities of the hospital, which is convenient and fast. Patients can directly access the intelligent medical guidance service through the registration machine equipment, and can enjoy the humanized medical guidance service without learning the operation of intelligent electronic products; the human-shaped clicking pain points and voice consultation interaction are adopted to interact with patients. Through voice questions and answers, the patient's physical discomfort symptoms are understood, and clear and straightforward options are provided for patients to choose. Through the symptoms, the disease condition is reversed, and the big data analysis and medical database are used to provide a high-accuracy registration service to meet the needs of patients of all ages; it can be deeply linked with the hospital, and the hospital's day-time appointment schedule and the field expertise of each expert physician are imported. According to the expert consultation situation and disease segmentation of the day, the registration priority service is realized to improve the registration efficiency of the hospital. At the same time, it helps hospitals to improve the scientific rationality of the consultation scheduling structure and reasonably allocate medical resources to the greatest extent.

[0149] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent medical guidance system, characterized in that: include: The preprocessing module is used to establish the corresponding relationship between the treatment departments, disease sites and symptoms corresponding to various diseases, take all disease sites as a set Place, take the symptoms that are easy for patients to identify as a set Status, and construct an easily identifiable symptom set Feature, where Feature = Place × Status; input each disease into the corpus to obtain the sentences related to it and perform word unit decomposition, calculate the similarity between each word unit and each easily identifiable symptom in Feature, and determine the weight of each easily identifiable symptom according to the similarity, so as to form an easily identifiable symptom weight vector for each disease; The interactive module is used to display the human-shaped area so that users can select the corresponding diseased area according to their own pain conditions; The question-answering module is used to ask the user for confirmation information of the easily identifiable symptoms related to the onset site selected by the user to construct a target easily identifiable symptom weight vector; wherein, when the jth round of questioning is conducted to the user, the easily identifiable symptoms with the largest relative increase in information entropy are selected from the unselected easily identifiable symptoms related to the onset site and asked to the user; j≥0; The treatment department confirmation module is used to calculate the similarity between the target easily identifiable symptom weight vector and the easily identifiable symptom weight vector of each disease, and take the department where the disease to which the easily identifiable symptom weight vector with the greatest similarity to the target easily identifiable symptom weight vector belongs as the recommended treatment department.

2. The system according to claim 1, characterized in that The question-answering system asks the following questions: whether there is an easily identifiable symptom related to the site of disease, and displays three options of "yes", "no" and "uncertain" for the user to choose.

3. The system according to claim 1, characterized in that The calculation formula of the relative increment of information entropy is: Among them, EntropyGainRatio(j,f) is the relative increase of information entropy when the easily identifiable symptom f is selected in the jth round of inquiry, and Entropy(p j )=-∑ i∈illness p j (i)log2p j (i), illness is a collection of various diseases, match(Answer j ,i) is the matching degree between the target easily identifiable symptom weight vector constructed based on the user's j-th round answer and the easily identifiable symptom weight vector of the i-th disease, Entropy(p j |(f,q))= Entropy(p′ f,q ),Entropy(p′ f,q )=-∑ i∈illness p′ f,q (i)log2p′ f,q (i), f∈Feature\dom(PartialAnswer j ), match(Answer j (f,q) ,i) is the matching degree between the target easily identifiable symptom weight vector constructed when the user's answer to the easily identifiable symptom f in the jth round is q and the easily identifiable symptom weight vector of the i-th disease, PartialAnswer j It is the set of easily identifiable symptoms related to the onset site that have been asked in the previous j rounds, and Feature is the set of all easily identifiable symptoms related to the onset site.

4. The system according to claim 1 or 2, characterized in that: The system also includes an expansion module for calling the ChatGPT interface to provide reference diagnosis suggestions.

5. An intelligent diagnosis guidance method, characterized in that: include: S1, establish the corresponding relationship between the treatment department, the site of onset and the symptoms corresponding to various diseases, take all the sites of onset as the set Place, take the symptoms that are easy for patients to identify as the set Status, and construct the easily identifiable symptom set Feature, where Feature = Place × Status; input each disease into the corpus to obtain the sentences related to it and perform word unit decomposition, calculate the similarity between each word unit and each easily identifiable symptom in Feature, and determine the weight of each easily identifiable symptom according to the similarity, and form the easily identifiable symptom weight vector of each disease; S2, displays a human-shaped area so that the user can select the corresponding diseased area according to his or her own pain condition; S3, according to the onset site selected by the user, asking the user for confirmation information of the easily identifiable symptoms related to the onset site to construct a target easily identifiable symptom weight vector; wherein, when the j-th round of asking the user is performed, the easily identifiable symptoms with the largest relative increase in information entropy are selected from the unselected easily identifiable symptoms related to the onset site and asked to the user; S4, respectively calculating the similarity between the target easily identifiable symptom weight vector and the easily identifiable symptom weight vector of each disease, and taking the department of the disease to which the easily identifiable symptom weight vector having the greatest similarity to the target easily identifiable symptom weight vector belongs as the recommended department for consultation.

6. An electronic device, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the method according to claim 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a processor to execute the method according to claim 5.

8. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to claim 5 is implemented.

Images