An AI-assisted medical consultation method

Abstract

The present invention discloses an AI-assisted medical consultation method, which relates to the technical field of intelligent medical consultation. The present invention realizes personalized diagnosis by combining a dynamic weighted model with individual characteristics of patients. The electronic health records and medical history data of patients are no longer just used as static references, but directly affect the weight of each symptom in the diagnosis, and can dynamically adjust the importance of symptoms according to different patient groups. Through individualized dynamic weighting, it is possible to more accurately identify the high-risk symptoms of patients, and make reasonable diagnostic inferences based on their individual characteristics, thereby greatly reducing the possibility of misdiagnosis or missed diagnosis. At the same time, a symptom relationship map is introduced, which can visualize the interdependence between different symptoms. The symptom relationship map presents the correlation between symptoms in the form of nodes and edges. Doctors can quickly understand the overall symptom manifestations and potential causes of the patient through the map, thereby making a more efficient and accurate diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical consultation technology, and in particular to an AI-assisted medical consultation method. Background Art

[0002] Remote AI-assisted medical consultations usually include elements such as medical image analysis, medical record data mining, decision support, virtual medical assistants, and data privacy and security protection. This technology uses deep learning and big data analysis, combined with medical expertise, to provide doctors with fast and accurate medical judgment and diagnostic support, improve the accuracy and efficiency of medical diagnosis, and provide patients with a better treatment experience.

[0003] Most remote AI-assisted consultation systems help doctors diagnose patients' conditions in a data-driven way, assisting by matching symptoms with diseases, relying on symptom databases, and setting fixed weights for each symptom in the diagnostic model. However, some symptoms have completely different clinical significance in young people and the elderly, and fixed weights cannot reflect this difference. In addition, patients of different ages and genders react differently to the same symptoms. Remote consultations can easily overlook these contents, leading to misdiagnosis or missed diagnosis.

[0004] In addition, the diagnostic model of remote AI-assisted consultation is usually an independent symptom processing method, that is, each symptom is regarded as an isolated factor independent of other symptoms, but the symptoms of most diseases are interrelated, and the patient's symptoms may enhance or weaken each other in other diseases or under specific conditions. Traditional assisted consultation methods are prone to miss certain key linkage information. When multiple diseases have overlapping symptoms, it is difficult to accurately distinguish the root cause of each disease. Therefore, a remote AI-assisted consultation method is urgently needed to solve such problems. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides an AI-assisted medical consultation method to solve the problem that remote AI-assisted medical consultation methods lack dynamic consideration of individual patient characteristics, resulting in overly standardized diagnosis, ignoring the differences between different populations, simplifying the processing of symptom relationships, and being unable to effectively model the complex multi-layer dependencies between symptoms.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] The present invention provides an AI-assisted medical consultation method, which comprises:

[0009] Step S1, patient information collection,

[0010] Individual characteristics of patients were collected through their electronic health records and medical histories, including age, gender, past medical history, and medication history;

[0011] Step S2, dynamic symptom weighting based on individual characteristics,

[0012] Combined with the individual characteristics of the patients collected in step S1, each symptom is dynamically weighted. For example, for a 65-year-old male, the weights of chest pain and dyspnea are set higher, and these symptoms are more strongly correlated with cardiovascular disease in this group. A symptom weight model dynamically adjusted for the individual characteristics of the patients is constructed;

[0013] Step S3, constructing a symptom relationship map for the patient,

[0014] Based on the dynamic weighted result of step S2, combined with the real-time symptom manifestations of the patient during the remote consultation, an individualized symptom relationship map is generated;

[0015] Step S4, VR display of the symptom map,

[0016] Visually display the individualized symptom relationship map constructed in step S3 through VR technology;

[0017] Step S5, multi-symptom prediction,

[0018] Based on the symptom weights and symptom relationship map adjusted in step S4, multi-symptom disease prediction is performed through a machine learning model, combining the patient's individual characteristics and current symptoms to predict the most likely disease and its potential complications.

[0019] Furthermore, the symptom weight model in step S2 is constructed as follows:

[0020] Set each symptom i The base weight is w i 0 , the basic weight is modified according to the individual characteristics of the patient, and the individual characteristics are processed by category, including age A, gender G, medical history H and medication history M. The dynamically adjusted symptom weight w i for:

[0021]

[0022] Among them, w i Indicates symptoms i The final dynamic weight, α A , α G , α H , α M are the weight adjustment coefficients for age, gender, medical history and medication history, g A (A), g G (G), g H (H), g M (M) is the influence function of different individual characteristics.

[0023] Furthermore, in step S2, each influence function is defined according to different individual characteristics:

[0024] Age effect function g A (A): Where A represents the patient's age, A0 represents the reference age value of a certain disease (for example, the reference age of cardiovascular disease), and A max , A min Indicates the upper and lower limits at different age groups, for example, chest pain has little impact on the diagnosis of a 20-year-old patient, but is critical for a 70-year-old patient;

[0025] Gender influence function g G (G): If male, then g G (G) = β m If female, then g G (G) = β f , where G represents the patient's gender, β m , β f represents the gender effect coefficients for males and females, representing the specificity of symptoms in different genders;

[0026] Medical history influence function g H (H): Among them, H represents the patient's medical history, γ j Indicates the strength of association between a medical history and current symptoms. For example, a history of heart disease has a greater impact on the weight of chest pain. j (H) represents the influence function of a specific medical history, which is defined based on the probability of association between the specific medical history and symptoms;

[0027] Medication history influence function g M (M): g M (M) = ∑ k δ k ·m k (M), where M represents the patient's medication history, δ k Indicates the weight of the effect of a certain drug on symptoms. For example, long-term use of hypertension drugs will reduce the weight of hypertension-related symptoms. k (M) represents the specific impact function of medication on symptoms, reflecting the regulatory effect of medication on the patient's pathological response.

[0028] Furthermore, the symptom weight model construction method in step S2 also includes:

[0029] Add a regularization term Ω(w i ) Balanced adjusted symptom weights:

[0030]

[0031] Among them, Ω(w i ) represents the regularization term, λ represents the regularization coefficient, which controls the influence of the regularization term. The dynamic symptom weighted model can adjust the weight of each symptom in real time according to the individual characteristics of the patient, thereby improving the accuracy of diagnosis.

[0032] Furthermore, the graph structure includes the following types of symptom nodes:

[0033] Single symptom node, corresponding to the patient's independent symptoms. Single symptom node is the most basic node, directly corresponding to the patient's clinical manifestations. Through the single symptom node, the doctor can quickly understand the patient's current specific symptoms;

[0034] Composite symptom nodes combine strongly related symptoms into a cluster. For example, "fever + cough + dyspnea" is used as a composite symptom node to help doctors quickly identify possible causes and simplify the complexity of the graph.

[0035] Key symptom nodes are core symptoms that are highly indicative of a particular disease, e.g., chest pain is a key symptom of heart disease and jaundice is an indicative symptom of liver disease;

[0036] Auxiliary symptom nodes are secondary symptoms that are not indicative of disease alone, for example, nonspecific symptoms such as mild headache or fatigue;

[0037] Dynamic symptom nodes: symptom nodes that change in real time based on the patient's condition. For example, dyspnea may evolve into severe respiratory failure.

[0038] Complication nodes, i.e. nodes for potential complications, indicate complications that may be triggered by worsening symptoms, for example, severe infection may lead to sepsis;

[0039] The patient's individualized symptom relationship map combines the interdependence between symptoms with weighted results, prioritizes possible symptom associations for the patient, and improves the efficiency and accuracy of the consultation.

[0040] Furthermore, the method of constructing an individualized symptom relationship map in step S3 is:

[0041] According to the current symptoms of the patient, an initial set of symptom nodes is constructed. Each symptom S i Corresponding to a graph node v i , the node is combined with the dynamic weighted results of the symptoms to preliminarily set the node weight, w(v i )=w i

[0042] Among them, v i The picture shows the symptom Si The corresponding node, w(v i ) is the node v i The weight is taken from the symptom S obtained in step S2 i The dynamic weight w i ;

[0043] Symptoms i and S j The edge weight w(v i ,v j ) represents the correlation between two symptoms, and defines symptom S i and S j The correlation between ij for:

[0044] ρ ij =β ij ·(1+α A ·g A (A)+α G ·g G (G)+α H ·g H (H)), where ρ ij It is a symptom i and S j The higher the value, the stronger the correlation between the two symptoms. ij is the basic association, which indicates the basic association strength between two symptoms in common patients, based on epidemiological data, α A , α G , α H are the influence coefficients of age, gender and previous medical history, g A (A), g G (G), g H (H) is the corresponding feature influence function;

[0045] Through the correlation ρ ij Compute edge weights between symptoms:

[0046] Among them, w(v i ,v j ) is a symptom i and S j The edge weight between them, w(v i )、w(v j ) are nodes v i and v j The weight of the symptom S i and S j The dynamic weighted value of ij is the symptom association calculated in step S2;

[0047] Through the correlation ρ ij The edge weight between two symptoms is adjusted so that the edge weight reflects the association between the two symptoms in the current patient.

[0048] Furthermore, in remote consultation, the patient's symptoms may change over time or as the disease progresses. Therefore, the node weights in the graph need to be adjusted according to the real-time symptom manifestations. The method of constructing an individualized symptom relationship graph in step S3 also includes:

[0049] Assume that at the current time t, the symptom S i The performance intensity is s i (t), then node v i The weight is updated in real time as:

[0050] w(v i ,t)=w i ·s i (t), where s i (t) is the patient's symptom S at time t i The strength of w(v i ,t) is the symptom S at time t i Corresponding node v i Dynamic weights; dynamically update the weight of each node in the graph so that the graph can reflect changes in the patient's condition.

[0051] Furthermore, in most cases, multiple symptoms occur together, such as cough, fever, and dyspnea, which generally indicates a respiratory infection.

[0052] The method of constructing an individualized symptom relationship map in step S3 also includes:

[0053] Introducing Symptom Cluster Node V k , represents a combination node of multiple related symptoms, and the weight of the symptom cluster is defined as:

[0054] Among them, V k Represents a symptom cluster, which contains a set of related symptom nodes v i ,w(V k ) is the weight of the symptom cluster, indicating the importance of the combined symptom in the current condition, λ i It is a symptom i Contribution coefficient in the cluster;

[0055] Finally, the individualized symptom relationship graph G(V,E) consists of a node set V and an edge set E:

[0056]

[0057] Where V = {v1,v2,…,vn} is the set of all symptom nodes, E is the set of weighted edges between symptom nodes, where w(v i ,v j )>0 indicates symptom S i and S j There is a correlation between them;

[0058] Through this graph model, the patient's individual symptom association structure can be clearly displayed, and the weights of each node and edge can be dynamically updated in combination with real-time data.

[0059] Furthermore, each node is represented by a different size, color, or brightness to distinguish the importance and severity of the symptoms;

[0060] Doctors use the VR interface to interactively adjust symptom nodes:

[0061] The size indicates the importance of the symptom in this patient, the color indicates the type of symptom or its association with a specific disease, and the brightness indicates the severity or urgency of the symptom;

[0062] The visualization display method through VR technology is as follows:

[0063] Perform 3D visualization of nodes. Each symptom node is represented by a sphere of different sizes and colors in the VR environment. The size of the node is proportional to the importance or weight of the symptom, while the color reflects its association with a specific disease. For example, red indicates high risk and blue indicates low risk. Symptom cluster nodes are represented by larger nodes or specific color combinations, allowing doctors to quickly identify the cause of the symptom cluster.

[0064] Visualize the connections between nodes. The connections between symptom nodes, i.e., the edges, are represented by lines. The thickness of the line represents the strength of the association between symptoms, and the color of the line represents the urgency of the symptoms.

[0065] Through the immersive interaction of VR, doctors can more intuitively observe and analyze the interrelationships among patients’ symptoms and make dynamic adjustments based on real-time data.

[0066] Furthermore, in step S5, the multi-symptom disease prediction method is performed by the machine learning model as follows:

[0067] Based on the adjustment result in step S4, the importance of each symptom is obtained, and the relationship between the symptoms is represented in the form of a graph according to the correlation between the symptoms;

[0068] Through feature embedding technology, the individual characteristics of patients are used as model input features. Using machine learning models, the symptom weights are combined with the symptom relationship map to generate features that can express the patient's current condition.

[0069] The machine learning model uses the patient's symptom characteristics and individual characteristics to predict diseases. Each task corresponds to a potential disease. The model outputs the possibility of each disease and predicts potential complications through comprehensive analysis of the prediction results of different diseases.

[0070] The machine learning models include logistic regression, random forest, deep neural network, convolutional neural network, and long short-term memory network models.

[0071] The beneficial effects of the present invention are:

[0072] The present invention uses a dynamic symptom weighting model, combines the patient's age, gender, previous medical history and medication history, adjusts the weight of each symptom for different patients, and performs detailed processing based on the individual characteristics of each patient, thereby improving the accuracy of diagnosis.

[0073] The present invention constructs an individualized symptom relationship map to present the dependencies and mutual influences between symptoms in the form of a graph, so that each symptom is no longer isolated. The correlation between symptom nodes is expressed through edge weights, helping doctors understand the linkage effects between symptoms.

[0074] The present invention introduces a dynamic update mechanism to dynamically adjust the symptom weight and the symptom relationship map according to the real-time changing symptom manifestations of the patient. It can automatically adjust the weight of the symptom and promptly remind the doctor of the possible risk of serious complications, greatly improving the flexibility and timeliness of remote diagnosis.

[0075] The present invention introduces symptom cluster nodes and processes common symptom combinations as a node, thereby simplifying the process of doctors analyzing complex symptom combinations and allowing doctors to locate potential root causes of diseases more quickly through cluster weights.

[0076] The present invention uses VR technology to visualize the symptom relationship map, allowing doctors to view symptom nodes in different sizes, colors and brightness in three-dimensional space, intuitively presenting the importance and severity of symptoms, and improving the efficiency and accuracy of remote consultation. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0078] Figure 1 Schematic diagram of the AI-assisted medical consultation method of the present invention. DETAILED DESCRIPTION

[0079] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0080] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0081] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0082] Example 1, reference Figure 1 This embodiment provides an AI-assisted medical consultation method, comprising the following steps:

[0083] Step S1, patient information collection,

[0084] Individual characteristics of patients were collected through their electronic health records and medical histories, including age, gender, past medical history, and medication history;

[0085] Step S2, dynamic symptom weighting based on individual characteristics,

[0086] Combined with the individual characteristics of the patients collected in step S1, each symptom is dynamically weighted. For example, for a 65-year-old male, the weights of chest pain and dyspnea are set higher, and these symptoms are more strongly correlated with cardiovascular disease in this group. A symptom weight model dynamically adjusted for the individual characteristics of the patients is constructed;

[0087] The symptom weight model in step S2 is constructed as follows:

[0088] Set each symptom i The basic weight is The basic weight is modified according to the individual characteristics of the patient, and the individual characteristics are processed by category, including age A, gender G, medical history H and medication history M. The dynamically adjusted symptom weight w i for:

[0089]

[0090] Among them, w i Indicates symptoms i The final dynamic weight, α A , α G , αH , α M are the weight adjustment coefficients for age, gender, medical history and medication history, g A (A), g G (G), g H (H), g M (M) is the influence function of different individual characteristics;

[0091] In step S2, each influence function is defined according to different individual characteristics:

[0092] Age effect function g A (A): Where A represents the patient's age, A0 represents the reference age value of a certain disease (for example, the reference age of cardiovascular disease), and A max , A min Indicates the upper and lower limits at different age groups, for example, chest pain has little impact on the diagnosis of a 20-year-old patient, but is critical for a 70-year-old patient;

[0093] Gender influence function g G (G): If male, then g G (G) = β m If female, then g G (G) = β f , where G represents the patient's gender, β m , β f represents the gender effect coefficients for males and females, representing the specificity of symptoms in different genders;

[0094] Medical history influence function g H (H): g H (H) = ∑ j γ j ·h j (H), where H represents the patient’s medical history, γ j Indicates the strength of association between a medical history and current symptoms. For example, a history of heart disease has a greater impact on the weight of chest pain. j (H) represents the influence function of a specific medical history, which is defined based on the probability of association between the specific medical history and symptoms;

[0095] Medication history influence function g M (M): g M (M) = ∑ k δ k ·m k (M), where M represents the patient's medication history, δ k Indicates the weight of the effect of a certain drug on symptoms. For example, long-term use of hypertension drugs will reduce the weight of hypertension-related symptoms. k(M) represents the specific effect function of medication on symptoms, reflecting the regulatory effect of medication on the patient's pathological response;

[0096] The method of constructing the symptom weight model in step S2 also includes:

[0097] Add a regularization term Ω(w i ) Balanced adjusted symptom weights:

[0098]

[0099] Among them, Ω(w i ) represents the regularization term, λ represents the regularization coefficient, which controls the influence of the regularization term. The dynamic symptom weighting model can adjust the weight of each symptom in real time according to the individual characteristics of the patient, thereby improving the accuracy of diagnosis.

[0100] Specifically, personalized diagnosis can be achieved through the combination of dynamic weighted models and individual patient characteristics. The patient's electronic health records and medical history data, including age, gender, past medical history, medication history and other information, are no longer just static references, but directly affect the weight of each symptom in the diagnosis, overcoming the problem of fixed symptom weights in traditional remote consultations. The importance of symptoms can be dynamically adjusted according to different patient groups. Through individualized dynamic weighting, the patient's high-risk symptoms can be identified more accurately, and reasonable diagnostic inferences can be made based on their individual characteristics, thereby greatly reducing the possibility of misdiagnosis or missed diagnosis.

[0101] Step S3, constructing a symptom relationship map for the patient,

[0102] Based on the dynamic weighted result of step S2, combined with the real-time symptom manifestations of the patient during the remote consultation, an individualized symptom relationship map is generated;

[0103] The graph structure includes the following types of symptom nodes:

[0104] Single symptom node, corresponding to the patient's independent symptoms. Single symptom node is the most basic node, directly corresponding to the patient's clinical manifestations. Through the single symptom node, the doctor can quickly understand the patient's current specific symptoms;

[0105] Composite symptom nodes combine strongly related symptoms into a cluster. For example, "fever + cough + dyspnea" is used as a composite symptom node to help doctors quickly identify possible causes and simplify the complexity of the graph.

[0106] Key symptom nodes are core symptoms that are highly indicative of a particular disease, e.g., chest pain is a key symptom of heart disease and jaundice is an indicative symptom of liver disease;

[0107] Auxiliary symptom nodes are secondary symptoms that are not indicative of disease alone, for example, nonspecific symptoms such as mild headache or fatigue;

[0108] Dynamic symptom nodes: symptom nodes that change in real time based on the patient's condition. For example, dyspnea may evolve into severe respiratory failure.

[0109] Complication nodes, i.e. nodes for potential complications, indicate complications that may be triggered by worsening symptoms, for example, severe infection may lead to sepsis;

[0110] The patient's individualized symptom relationship map combines the interdependence between symptoms with weighted results, giving priority to the patient's possible symptom associations, improving the efficiency and accuracy of the consultation;

[0111] Specifically, the symptom relationship map was introduced to visualize the interdependence between different symptoms. Traditional diagnostic models often regard each symptom as an independent factor and ignore the potential linkage between different symptoms. However, the symptoms of many diseases are not isolated, and often manifest as complex combinations or interactions. The symptom relationship map presents the correlation between symptoms in the form of nodes and edges. Doctors can use the map to quickly understand the patient's overall symptom manifestations and potential causes, thereby making a more efficient and accurate diagnosis.

[0112] The method of constructing an individualized symptom relationship map in step S3 is:

[0113] According to the current symptoms of the patient, an initial set of symptom nodes is constructed. Each symptom S i Corresponding to a graph node v i , the node is combined with the dynamic weighted results of the symptoms to preliminarily set the node weight, w(v i )=w i

[0114] Among them, v i The picture shows the symptom S i The corresponding node, w(v i ) is the node v i The weight is taken from the symptom S obtained in step S2 i The dynamic weight w i ;

[0115] Symptoms i and S j The edge weight w(v i ,v j ) represents the correlation between two symptoms, and defines symptom S i and S j The correlation between ij for:

[0116] ρ ij =β ij ·(1+α A ·g A (A)+α G ·g G (G)+α H ·g H (H)), where ρ ij It is a symptom i and S j The higher the value, the stronger the correlation between the two symptoms. ij is the basic association, which indicates the basic association strength between the two symptoms in ordinary patients, based on epidemiological data, α A , α G , α H are the influence coefficients of age, gender and previous medical history, g A (A), g G (G), g H (H) is the corresponding feature influence function;

[0117] Through the correlation ρ ij Compute edge weights between symptoms:

[0118] Among them, w(v i ,v j ) is a symptom i and S j The edge weight between them, w(v i )、w(v j ) are nodes v i and v j The weight of the symptom S i and S j The dynamic weighted value of ij is the symptom association calculated in step S2;

[0119] Through the correlation ρ ij Adjust the edge weight between two symptoms so that the edge weight reflects the correlation between the two symptoms in the current patient;

[0120] In remote consultation, the patient's symptoms may change over time or as the disease progresses. Therefore, the node weights in the graph need to be adjusted according to the real-time symptom manifestations. The method of constructing an individualized symptom relationship graph in step S3 also includes:

[0121] Assume that at the current time t, the symptom S i The performance intensity is s i (t), then node v i The weight is updated in real time as:

[0122] w(v i ,t)=w i ·s i (t), where s i (t) is the patient's symptom S at time t i The strength of w(v i ,t) is the symptom S at time t i Corresponding node v i Dynamic weights; Dynamically update the weight of each node in the graph so that the graph can reflect changes in the patient's condition;

[0123] In most cases, multiple symptoms occur together, such as cough, fever, and difficulty breathing. This combination generally indicates a respiratory infection.

[0124] The method of constructing an individualized symptom relationship map in step S3 also includes:

[0125] Introducing Symptom Cluster Node V k , represents a combination node of multiple related symptoms, and the weight of the symptom cluster is defined as:

[0126] Among them, V k Represents a symptom cluster, which contains a set of related symptom nodes v i ,w(V k ) is the weight of the symptom cluster, indicating the importance of the combined symptom in the current condition, λ i It is a symptom i Contribution coefficient in the cluster;

[0127] Finally, the individualized symptom relationship graph G(V,E) consists of a node set V and an edge set E:

[0128]

[0129] Where V = {v1,v2,…,v n} is the set of all symptom nodes, E is the set of weighted edges between symptom nodes, where w(v i ,v j )>0 indicates symptom S i and S j There is a correlation between them;

[0130] This graph model can clearly display the individualized symptom association structure of patients, and dynamically update the weights of each node and edge in combination with real-time data;

[0131] Specifically, combined with the dynamic adjustment of real-time symptom data, the diagnosis results can be updated in real time according to changes in the patient's condition. During the remote consultation process, the patient's symptoms may change over time or as the disease progresses. Therefore, relying solely on static data often cannot fully reflect the patient's condition. By dynamically updating node weights and symptom association graphs, the judgment of the patient's condition can be adjusted at any time. This real-time nature improves the flexibility and adaptability of remote diagnosis.

[0132] Step S4, VR display of the symptom map,

[0133] Visually display the individualized symptom relationship map constructed in step S3 through VR technology;

[0134] Each node is represented by a different size, color, or brightness to distinguish the importance and severity of the symptoms;

[0135] Doctors use the VR interface to interactively adjust symptom nodes:

[0136] The size indicates the importance of the symptom in this patient, the color indicates the type of symptom or its association with a specific disease, and the brightness indicates the severity or urgency of the symptom;

[0137] The visualization display method through VR technology is as follows:

[0138] Perform 3D visualization of nodes. Each symptom node is represented by a sphere of different sizes and colors in the VR environment. The size of the node is proportional to the importance or weight of the symptom, while the color reflects its association with a specific disease. For example, red indicates high risk and blue indicates low risk. Symptom cluster nodes are represented by larger nodes or specific color combinations, allowing doctors to quickly identify the cause of the symptom cluster.

[0139] Visualize the connections between nodes. The connections between symptom nodes, i.e., the edges, are represented by lines. The thickness of the line represents the strength of the association between symptoms, and the color of the line represents the urgency of the symptoms.

[0140] Through VR's immersive interaction, doctors can more intuitively observe and analyze the interrelationships between patients' symptoms and make dynamic adjustments based on real-time data;

[0141] Specifically, multi-symptom clusters were introduced. Multi-symptom cluster nodes can effectively handle combinations of symptoms that occur simultaneously and point to a specific disease, simplifying the analysis of complex symptom combinations. By calculating the weights of symptom clusters, doctors can more easily identify the root causes that may cause these symptoms, rather than dealing with them one symptom at a time. The simplification greatly improves diagnostic efficiency.

[0142] Step S5, multi-symptom prediction,

[0143] Based on the symptom weights and symptom relationship graph adjusted in step S4, a machine learning model is used to predict multi-symptom diseases, and the most likely diseases and their potential complications are predicted in combination with the individual characteristics and current symptoms of the patient;

[0144] Specifically, the application of VR display technology has further improved doctors' understanding and operation of symptom relationship maps. Combined with VR technology, doctors can view patients' symptom relationship maps in three-dimensional form. The size, color and brightness of the nodes respectively indicate the importance, type and severity of the symptoms. The intuitive display method helps doctors better grasp the overall picture of the patient's condition, and optimize the diagnosis plan by interactively adjusting the symptom weight and correlation, making complex symptom correlation analysis easier to understand, helping doctors make faster and more accurate decisions during the diagnosis process.

[0145] In step S5, the multi-symptom disease prediction method is performed by the machine learning model as follows:

[0146] Based on the adjustment result in step S4, the importance of each symptom is obtained, and the relationship between the symptoms is represented in the form of a graph according to the correlation between the symptoms;

[0147] Through feature embedding technology, the individual characteristics of patients are used as model input features. Using machine learning models, the symptom weights are combined with the symptom relationship map to generate features that can express the patient's current condition.

[0148] The machine learning model uses the patient's symptom characteristics and individual characteristics to predict diseases. Each task corresponds to a potential disease. The model outputs the possibility of each disease. At the same time, through a comprehensive analysis of the prediction results of different diseases, it predicts potential complications. The machine learning model uses logistic regression, random forest, deep neural network, convolutional neural network, and long short-term memory network models.

[0149] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An AI-assisted medical consultation method, characterized in that: include, Step S1, patient information collection, collecting the patient's individual characteristics through the patient's electronic health record and medical history, the individual characteristics include age, gender, past medical history, and medication history; Step S2, dynamically weighting symptoms based on individual characteristics, combining the individual characteristics of the patient collected in step S1, dynamically weighting each symptom, and constructing a symptom weight model dynamically adjusted according to the individual characteristics of the patient; The symptom weight model in step S2 is constructed as follows: Set each symptom i The basic weight is The basic weight is modified according to the individual characteristics of the patient, and the individual characteristics are processed by category, including age A, gender G, medical history H and medication history M. The dynamically adjusted symptom weight w i for: Among them, w i Indicates symptoms i The final dynamic weight, α A , α G , α H , α M are the weight adjustment coefficients for age, gender, medical history and medication history, g A (A), g G (G), g H (H), g M (M) is the influence function of different individual characteristics; In step S2, each influence function is defined according to different individual characteristics: Age effect function g A (A): Among them, A represents the patient's age, A0 represents the reference age value of a certain disease, and A max , A min Indicates the upper and lower limits at different age levels; Gender influence function g G (G): If male, then g G (G) = β m If female, then g G (G) = β f , where G represents the patient's gender, β m , β f represents the gender effect coefficients for males and females, representing the specificity of symptoms in different genders; Medical history influence function g H (H): g H (H) = ∑ j γ j ·h j (H), where H represents the patient’s medical history, γ j Indicates the strength of association between a medical history and current symptoms, h j (H) represents the influence function of a specific medical history, which is defined based on the probability of association between the specific medical history and symptoms; Medication history influence function g M (M): g M (M) = ∑ k δ k ·m k (M), where M represents the patient's medication history, δ k Indicates the weight of the effect of a drug on symptoms, m k (M) represents the specific effect function of medication on symptoms, reflecting the regulatory effect of medication on the patient's pathological response; The method of constructing the symptom weight model in step S2 also includes: Add a regularization term Ω(w i ) Balanced adjusted symptom weights: Among them, Ω(w i ) represents the regularization term, λ represents the regularization coefficient, which controls the influence of the regularization term; Step S3, constructing a symptom relationship map for the patient, based on the dynamic weighted result of step S2 and combined with the real-time symptom manifestations of the patient's remote consultation, to generate an individualized symptom relationship map; The graph structure includes the following types of symptom nodes: Single symptom node, corresponding to the patient's independent symptom; Composite symptom nodes combine strongly related symptoms into a cluster; Key symptom nodes, core symptoms that are highly indicative in a particular disease; Auxiliary symptom nodes are minor symptoms that do not alone indicate disease; Dynamic symptom nodes, which change in real time based on the patient's condition; Complication nodes, i.e., nodes of potential complications, indicate complications that may be caused by the worsening of symptoms; The method of constructing an individualized symptom relationship map in step S3 is: According to the current symptoms of the patient, an initial set of symptom nodes is constructed. Each symptom S i Corresponding to a graph node v i , the node is combined with the dynamic weighted results of the symptom to preliminarily set the node weight, w(v i )=w i Among them, v i The picture shows the symptoms S i The corresponding node, w(v i ) is the node v i The weight is taken from the symptom S obtained in step S2 i The dynamic weight w i ; Symptoms i and S j The edge weight w(v i ,v j ) represents the correlation between two symptoms, and defines symptom S i and S j The correlation between ij for: ρ ij =β ij ·(1+α A ·g A (A)+α G ·g G (G)+α H ·g H (H)), where ρ ij It is a symptom i and S j The higher the value, the stronger the correlation between the two symptoms. ij is the basic correlation, which indicates the basic correlation strength between two symptoms in ordinary patients, α A , α G , α H are the influence coefficients of age, gender and previous medical history, g A (A), g G (G), g H (H) is the corresponding feature influence function; Through the correlation ρ ij Compute edge weights between symptoms: Among them, w(v i ,v j ) is a symptom i and S j The edge weight between them, w(v i )、w(v j ) are nodes v i and v j The weight of the symptom S i and S j The dynamic weighted value of ij is the symptom association calculated in step S2; Step S4, performing VR display on the symptom map, visually displaying the individualized symptom relationship map constructed in step S3 through VR technology; Step S5, multi-symptom prediction, based on the symptom weights and symptom relationship map adjusted in step S4, multi-symptom disease prediction is performed through a machine learning model, combining the patient's individual characteristics and current symptoms to predict the most likely disease and its potential complications.

2. The AI-assisted medical consultation method according to claim 1, characterized in that: The method of constructing an individualized symptom relationship map in step S3 also includes: Assume that at the current time t, the symptom S i The performance intensity is s i (t), then node v i The weight is updated in real time as: w(v i ,t)=w i ·s i (t), where s i (t) is the patient's symptom S at time t i The strength of w(v i ,t) is the symptom S at time t i Corresponding node v i Dynamic weight of .

3. The AI-assisted medical consultation method according to claim 2, characterized in that: The method of constructing an individualized symptom relationship map in step S3 also includes: Introducing Symptom Cluster Node V k , represents a combination node of multiple related symptoms, and the weight of the symptom cluster is defined as: Among them, V k Represents a symptom cluster, which contains a set of related symptom nodes v i ,w(V k ) is the weight of the symptom cluster, indicating the importance of the combined symptom in the current condition, λ i It is a symptom i Contribution coefficient in the cluster; Finally, the individualized symptom relationship graph G(V,E) consists of a node set V and an edge set E: Where V = {v1,v2,…,v n } is the set of all symptom nodes, E is the set of weighted edges between symptom nodes, where w(v i ,v j )>0 indicates symptom S i and S j There is a correlation between them.

4. The AI-assisted medical consultation method according to claim 3, characterized in that: Each node is represented by a different size, color, or brightness to distinguish the importance and severity of the symptoms; Doctors use the VR interface to interactively adjust symptom nodes: The size indicates the importance of the symptom in this patient, the color indicates the type of symptom, and the brightness indicates the severity of the symptom.

5. The AI-assisted medical consultation method according to claim 4, characterized in that: In step S5, the multi-symptom disease prediction method is performed by the machine learning model as follows: Based on the adjustment result in step S4, the importance of each symptom is obtained, and the relationship between the symptoms is represented in the form of a graph according to the correlation between the symptoms; Through feature embedding technology, the individual characteristics of patients are used as model input features. Using machine learning models, the symptom weights are combined with the symptom relationship map to generate features that can express the patient's current condition. The machine learning model uses the patient's symptom characteristics and individual characteristics to predict diseases. Each task corresponds to a potential disease, and the model outputs the possibility of each disease. At the same time, through a comprehensive analysis of the prediction results of different diseases, potential complications are predicted.

Images