Rabies exposure risk assessment method and device

Abstract

The invention discloses a rabies exposure risk assessment method and device, and the method comprises the steps: generating a patient information set according to the answer of a patient to a rabies exposure question; analyzing the patient information set by using a pre-trained question and answer analysis model, and acquiring wound visual content of the patient under the condition that an analysis result is greater than a preset threshold value; according to the visual content of the wound and the patient information set, the rabies exposure risk is evaluated, so that more scientific and accurate rabies exposure risk evaluation can be carried out, corresponding treatment suggestions can be provided, and patient medical records for reference of doctors can be generated, so that the working efficiency of the doctors can be improved, and safer, more accurate and faster medical services can be provided for the patients.

Description

Technical Field

[0001] The present invention relates to the medical field, and in particular to a rabies exposure risk assessment method and device. Background Art

[0002] Rabies is an animal-borne infectious disease caused by lyssavirus infection with symptoms mainly related to the central nervous system. The case fatality rate is almost 100%, and the number of rabies deaths worldwide is about 59,000 per year. It is a public health problem faced by the world. The World Health Organization (WHO) believes that post-exposure prophylaxis (PEP) is the most effective strategy for preventing rabies. Timely and standardized PEP can almost 100% prevent the disease. PEP mainly includes three aspects: first, timely and standardized wound treatment; second, rabies vaccination; third, the correct use of rabies passive immunization agents. Our country attaches great importance to standardized PEP.

[0003] However, there are some challenges in the implementation of this important prevention strategy. Medical resources are unevenly distributed in my country, and the technical levels of physicians engaged in rabies prevention vary. Although various regions have strengthened PEP training in recent years, some physicians, especially primary care physicians, still lack relevant knowledge and experience in rabies prevention, and there are misunderstandings in the risk assessment of rabies exposure, resulting in the continued existence of the problem of irregular PEP treatment. In addition, rabies exposure has obvious time distribution characteristics, especially in the summer and autumn seasons and in the afternoon to evening hours of the day, resulting in frequent queues in large diagnosis and treatment centers during these periods, patients waiting too long, and poor medical experience.

[0004] Therefore, there is an urgent need for a method to assist physicians in rabies exposure risk assessment and to improve the accuracy and efficiency of rabies prevention efforts, with a focus on enhancing the capabilities of medical professionals and optimizing the medical process. Summary of the invention

[0005] The present invention aims to provide a rabies exposure risk assessment method and device, which can solve the problems existing in the prior art such as inaccurate exposure risk assessment, irregular PEP treatment, low diagnosis and treatment efficiency during peak medical hours, and poor patient medical experience.

[0006] According to one aspect of the present invention, a method for assessing the risk of rabies exposure is provided, comprising: generating a patient information set based on the patient's answers to questions about rabies exposure; analyzing the patient information set using a pre-trained question-answer analysis model, and obtaining the visual content of the patient's wound when the analysis result is greater than a pre-set threshold; and assessing the risk of rabies exposure based on the visual content of the wound and the patient information set.

[0007] Preferably, the patient information set is analyzed using a pre-trained question-answer analysis model according to the following formula: R = M (I) = M (i1, i2, i3, i4..., i n ), where R is the analysis result, M is the pre-trained question-answering analysis model, {i1, i2, i3, i4..., i n} is the patient information set, and n is the number of answers in the patient information set.

[0008] Preferably, before using the pre-trained question-and-answer analysis model to analyze the patient information set, it also includes: obtaining medical knowledge input text information and general large language model pre-trained model information; performing feature extraction on the pre-processed medical knowledge input text information, constructing a feature knowledge base, and integrating it into the large language model; generating new questions that may be asked by patients, testing the medical diagnostic effect of the question-and-answer analysis model, and iteratively optimizing the question-and-answer analysis model based on the medical diagnostic effect.

[0009] Preferably, after obtaining the visual content of the patient's wound, the method further includes: processing the visual content of the wound according to a pre-trained visual analysis model to obtain a determination result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification : Where C is the number of categories, y c is the one-hot encoding of the true label, is the predicted probability of the visual analysis model for the category; when the wound type is a puncture wound or a laceration, the judgment result of the wound size is obtained according to the pre-trained visual analysis model, wherein the pre-trained visual analysis model also includes a loss function L MSE : Among them, x i is the actual wound size, is the predicted wound size, n is the number of samples; when the judgment result of the wound size does not meet the requirements, a re-shooting suggestion is given.

[0010] Preferably, when the judgment result of the wound size meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound location does not meet the requirements, a suggestion for reshooting is given; when the judgment result of the wound location meets the requirements, the steps of assessing the risk of rabies exposure are executed.

[0011] Preferably, after obtaining the visual content of the patient's wound, the method further includes: processing the visual content of the wound according to a pre-trained visual analysis model to obtain a determination result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification ; Where C is the number of categories, y c is the one-hot encoding of the true label, is the predicted probability of the visual analysis model for the category; when the wound type is a scratch or bite, the judgment result of the wound bleeding is obtained according to the pre-trained visual analysis model, where the pre-trained visual analysis model also includes the loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound bleeding situation does not meet the requirements, a re-shooting suggestion is given.

[0012] Preferably, when the judgment result of the wound bleeding condition meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound location does not meet the requirements, a re-shooting suggestion is given; when the judgment result of the wound location meets the requirements, the steps of assessing the risk of rabies exposure are executed.

[0013] Preferably, a judgment result on whether there are biological residues is obtained according to a pre-trained visual analysis model; when there are biological residues, the type of biological residues and the amount of biological residues R are determined, and the step of assessing the risk of rabies exposure is performed; the amount of biological residues R is calculated according to the following formula: R = k·(S d +∈) b ; where S d is the observed signal size of the organism remaining on the surface, ∈ is the error intensity, k is the proportional constant, b is the nonlinear coefficient, ∈, k, and b are obtained through neural network training.

[0014] Preferably, the risk of rabies exposure is assessed according to the following formula: Among them, α, β, γ are weight coefficients, f1, f2, f3 represent the characteristics of the wounding animal and the function of the wounding situation, the function of the severity of the wound, and the function of the patient's immune status, respectively; and among them, f1 is calculated according to the following formula: Wherein c1, c2, c3, c4, c5 are the weights of each factor; and wherein f2 is calculated according to the following formula: E=f(T, L, D, B, R), wherein E is the injury score, T is the wound type, L is the wound location, D is the wound size, B is the wound bleeding, and R is the biological residue; and wherein f3 is calculated according to the following formula:

[0015] According to another aspect of the present invention, a rabies exposure risk assessment device is provided, including: an information generation module, used to generate a patient information set based on the patient's answers to rabies exposure questions; a visual acquisition module, used to analyze the patient information set using a pre-trained question-answer analysis model, and when the analysis result is greater than a pre-set threshold, obtain the patient's wound visual content; and a risk assessment module, used to assess the rabies exposure risk based on the wound visual content and the patient information set.

[0016] The present invention provides a rabies exposure risk assessment method and device, the method comprising: generating a patient information set according to the patient's answers to rabies exposure questions; analyzing the patient information set using a pre-trained question-answer analysis model, and obtaining the patient's wound visual content when the analysis result is greater than a preset threshold; assessing the rabies exposure risk according to the wound visual content and the patient information set, thereby enabling a more scientific and accurate rabies exposure risk assessment, providing corresponding treatment suggestions, and generating patient medical records for physicians' reference, thereby improving the physician's work efficiency and providing patients with safer, more accurate, and faster medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0018] Figure 1 is a flow chart of a method for assessing rabies exposure risk according to an embodiment of the present invention;

[0019] Figure 2 A specific operation flow chart of the rabies exposure risk assessment method according to an embodiment of the present invention; and

[0020] Figure 3 2 is a diagram showing the structure of a rabies exposure risk device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] Reference will now be made in detail to various embodiments of the present invention, examples of which are shown in the accompanying drawings and described below. For ease of interpretation and precise definition in the appended claims, the terms "upper", "lower", "inner" and "outer" are used to describe features of the exemplary embodiments shown in the drawings with reference to their positions.

[0023] The embodiment of the present invention provides a method for assessing rabies exposure risk. Figure 1 As shown, the method includes: generating a patient information set based on the patient's answers to questions about rabies exposure; analyzing the patient information set using a pre-trained question-answer analysis model, and obtaining the patient's wound visual content when the analysis result is greater than a pre-set threshold; and assessing the risk of rabies exposure based on the wound visual content and the patient information set.

[0024] In the related art, the treatment of PEP faces the problems of unbalanced medical resources and different professional levels of doctors, which leads to non-standard treatment, low efficiency and poor patient experience, especially during peak hours. The embodiment of the present invention automatically collects patient information sets, uses pre-trained models for analysis, and obtains wound visual content when necessary to achieve rapid and accurate risk assessment, improves the work efficiency of doctors, optimizes the patient's medical experience, and provides an efficient and accurate new strategy for rabies prevention.

[0025] According to an embodiment of the present invention, the patient information set is analyzed using a pre-trained question-answer analysis model according to the following formula: R = M (I) = M (i1, i2, i3, i4..., i n ), where R is the analysis result, M is the pre-trained question-answering analysis model, {i1, i2, i3, i4..., i n} is the patient information set, and n is the number of answers in the patient information set.

[0026] According to an embodiment of the present invention, before using a pre-trained question and answer analysis model to analyze the patient information set, it also includes: obtaining medical knowledge input text information and general large language model pre-trained model information; performing feature extraction on the pre-processed medical knowledge input text information, constructing a feature knowledge base, and integrating it into the large language model; generating new questions that may be asked by patients, testing the medical diagnosis effect of the question and answer analysis model, and iteratively optimizing the question and answer analysis model according to the medical diagnosis effect.

[0027] The present invention improves the accuracy and efficiency of evaluation by automatically collecting patient information sets and using a pre-trained question-answer analysis model for analysis. The method also includes building a feature knowledge base to integrate into a large language model, generating new questions that patients may ask, testing the medical diagnosis effect of the question-answer analysis model, and performing iterative optimization.

[0028] According to an embodiment of the present invention, after obtaining the visual content of the patient's wound, the method further includes: processing the visual content of the wound according to a pre-trained visual analysis model to obtain a judgment result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification : Where C is the number of categories, yc is the one-hot encoding of the true label, is the predicted probability of the visual analysis model for the category; when the wound type is a puncture wound or a laceration, the judgment result of the wound size is obtained according to the pre-trained visual analysis model, wherein the pre-trained visual analysis model also includes a loss function L MSE : Among them, x i is the actual wound size, is the predicted wound size, n is the number of samples; when the judgment result of the wound size does not meet the requirements, a re-shooting suggestion is given.

[0029] In the related art, wound types are generally not classified in PEP, resulting in inadequate and inappropriate treatment of different types of wounds. The present invention classifies wound types into puncture wounds or lacerations and scratches and bites. Specifically, the obtained visual content of the wound is processed by using a pre-trained visual analysis model to determine the wound type.

[0030] Furthermore, based on the judgment results, the wound size is judged for puncture wounds or lacerations, and re-shooting is recommended if it does not meet the requirements, thereby improving the accuracy of wound assessment and improving the work efficiency of doctors. It is crucial to judge the wound size for puncture wounds or lacerations. These wounds are usually more serious, involving deep tissue damage, and the risk of rabies exposure is higher. Puncture wounds are caused by sharp objects piercing the skin and tissues. Generally, the entrance is small but the depth is deep. Common examples include cat bite wounds. Lacerations are caused by external forces tearing human tissues. The wound edges are uneven and torn. The wound depth and range are large, which can easily cause bleeding and infection. Common examples include dog bite wounds.

[0031] According to an embodiment of the present invention, when the judgment result of the wound size meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound location does not meet the requirements, a suggestion for reshooting is given; when the judgment result of the wound location meets the requirements, the steps of assessing the risk of rabies exposure are executed.

[0032] In the embodiment of the present invention, when the judgment result of the wound size meets the requirements, the model further judges the wound position to ensure accurate identification and positioning of the wound. If the judgment result of the wound position does not meet the requirements, it prompts to retake the wound to obtain a more accurate wound image. This step ensures the accuracy of the wound assessment and provides a reliable basis for subsequent rabies exposure risk assessment.

[0033] According to an embodiment of the present invention, after obtaining the visual content of the patient's wound, the method further includes: processing the visual content of the wound according to a pre-trained visual analysis model to obtain a judgment result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification ; Where C is the number of categories, y c is the one-hot encoding of the true label, is the predicted probability of the visual analysis model for the category; when the wound type is a scratch or bite, the judgment result of the wound bleeding is obtained according to the pre-trained visual analysis model, where the pre-trained visual analysis model also includes the loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound bleeding situation does not meet the requirements, a re-shooting suggestion is given.

[0034] In the related art, wound types are generally not classified in PEP, resulting in inadequate and inappropriate treatment of different types of wounds. The present invention classifies wound types into puncture wounds or lacerations and scratches and bites. Specifically, the obtained visual content of the wound is processed by using a pre-trained visual analysis model to determine the wound type.

[0035] Furthermore, based on the judgment results, the bleeding condition of scratches and bites is judged, and reshooting is recommended if it does not meet the requirements, thereby improving the accuracy of wound assessment and the work efficiency of doctors. For scratches or bites, it is important to judge the bleeding condition of the wound, because the bleeding condition can reflect the depth and severity of the scratch or bite wound, which has important clinical significance for the rabies exposure risk classification.

[0036] According to an embodiment of the present invention, when the judgment result of the wound bleeding condition meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, Indicates the predicted label; when the judgment result of the wound location does not meet the requirements, a re-shooting suggestion is given; when the judgment result of the wound location meets the requirements, the steps of assessing the risk of rabies exposure are executed.

[0037] In the embodiment of the present invention, when the judgment result of the wound bleeding situation meets the requirements, the model will further judge the wound location, thereby ensuring the accuracy of the wound assessment. For medical professionals, it can provide more accurate wound size and location information, improve medical efficiency, and ensure that patients receive timely and appropriate medical treatment.

[0038] According to an embodiment of the present invention, a judgment result on whether there is biological residue is obtained based on a pre-trained visual analysis model; when there is biological residue, the type of biological residue and the amount of biological residue R are determined, and the step of assessing the risk of rabies exposure is performed; the amount of biological residue R is calculated according to the following formula: R = k·(S d +∈) b ; where S d is the observed signal size of the organism remaining on the surface, ∈ is the error intensity, k is the proportional constant, b is the nonlinear coefficient, ∈, k, and b are obtained through neural network training.

[0039] In the related art, there is a lack of accurate judgment of wound biological residues in wound treatment after rabies exposure, especially when distinguishing different types of wounds such as puncture wounds, lacerations, scratches and bites, which leads to irregular wound treatment. The embodiment of the present invention can obtain a judgment result on whether there are biological residues. When there are biological residues, the type and amount of biological residues R are determined, and the steps of assessing the risk of rabies exposure are performed.

[0040] According to an embodiment of the present invention, the risk of rabies exposure is assessed according to the following formula: Among them, α, β, γ are weight coefficients, f1, f2, f3 represent the characteristics of the wounding animal and the function of the wounding situation, the function of the severity of the wound, and the function of the patient's immune status, respectively; and among them, f1 is calculated according to the following formula: Wherein c1, c2, c3, c4, c5 are the weights of each factor; and wherein f2 is calculated according to the following formula: E=f(T, L, D, B, R), wherein E is the injury score, T is the wound type, L is the wound location, D is the wound size, B is the wound bleeding, and R is the biological residue; and wherein f3 is calculated according to the following formula:

[0041] According to another embodiment of the present invention, a rabies exposure risk assessment device is provided. Figure 3 As shown, it includes: an information generation module, which is used to generate a patient information set based on the patient's answers to rabies exposure questions; a visual acquisition module, which is used to analyze the patient information set using a pre-trained question-answer analysis model, and when the analysis result is greater than a pre-set threshold, obtain the patient's wound visual content; a risk assessment module, which is used to assess the risk of rabies exposure based on the wound visual content and the patient information set.

[0042] In the related art, the treatment of PEP faces the problems of unbalanced medical resources and different professional levels of doctors, which leads to non-standard treatment, low efficiency and poor patient experience, especially during peak hours. The present invention automatically collects patient information sets, uses pre-trained models for analysis, and obtains wound visual content when necessary to achieve rapid and accurate risk assessment, improves the work efficiency of doctors, optimizes the patient's medical experience, and provides an efficient and accurate new strategy for rabies prevention.

[0043] Figure 2 FIG. 1 shows a specific operation flow chart of the rabies exposure risk assessment method according to an embodiment of the present invention. Figure 2 It can be seen that the present invention mainly includes three parts: intelligent question-answering system, intelligent recognition system and intelligent decision-making system.

[0044] 1. Intelligent Question Answering System

[0045] The intelligent question-and-answer system is an auxiliary medical question-and-answer system. The system uses artificial intelligence methods to build a knowledge base that includes rabies disease-related knowledge, diagnostic criteria, clinical manifestations and corresponding treatment recommendations, rabies prevention methods, etc. The knowledge base is input into the question-and-answer system as training data to participate in the training of the question-and-answer system. After the training is completed, the question-and-answer system will automatically generate questions for the intelligent question-and-answer system. After the patient's feedback, the question-and-answer system will generate a preliminary judgment on the patient.

[0046] (1) The data collection method is:

[0047] Medical dictionaries, manual review of medical information, medical records annotated by physicians, medical imaging materials, etc.;

[0048] (2) The training method is:

[0049] Step 1: Obtain medical knowledge input text information and obtain general large language model pre-training model information. Generally speaking, the large language pre-training model already contains all the general knowledge bases. Therefore, in this process, the input medical specialized information needs to be processed. The processing method is as follows: a. Obtain input medical professional knowledge information b. Pre-process the medical professional knowledge information. The pre-processing process includes question and answer setting, output generation, and diversified processing; c. Extract feature words and keywords and generate relevant knowledge base

[0050] Step 2: Extract features from the pre-processed patient input information, build a feature knowledge base, and integrate it into the large language model. By using the pre-processed information and fine-tuning the large language model, the patient's input data and the data in the large language model are used to construct feature values ​​to generate a new large language model weight library.

[0051] Step 3: Test the training effect. Artificially generate new questions that patients may ask, test the medical diagnosis effect of the large language model, and iteratively optimize the large language model based on the test results.

[0052] (3) Usage:

[0053] When a patient starts using the intelligent question-and-answer system, the system first generates a series of preliminary questions based on the medical knowledge questions entered by the patient. These questions are designed to collect key information about the patient's reason for this visit, the type of animal contact, the method of contact with animals, the patient's health status, previous rabies vaccination history, allergy history, etc. As the patient responds, the system will dynamically adjust the questioning method using natural language processing technology to gain a deeper understanding of the patient's specific situation.

[0054] Assume that the patient information set collected by the system is I = {i1, i2, i3, i4..., i n} where each i n Represents the patient's answer to the rabies exposure question. The system will analyze this information through the pre-trained model M and generate an exposure assessment score R, the formula is: R = M (I) = M (i1, i2, i3, i4......, i n )

[0055] This exposure assessment score R will be used to determine whether the patient has a risk of rabies exposure. The system sets a threshold T to distinguish between those with no risk of rabies exposure and those who need further examination:

[0056] If R≤T, the system will give a conclusion of “no risk of rabies exposure”. In this case, the system will advise the patient that no further PEP treatment is required and submit the relevant information to the physician for review to ensure that there are no potential problems that have been overlooked, thus completing the entire process.

[0057] In the case of R≤T, the intelligent question-and-answer system will also pay special attention to whether the patient belongs to the high-risk group for rabies. For example, laboratory workers engaged in rabies research, people who have long-term contact with wild animals or stray animals due to occupational reasons (such as veterinarians, animal rescuers), people who live in or go to areas with high incidence of rabies, and people who come into contact with rabies patients. If the patient belongs to these high-risk groups, the system will actively recommend pre-exposure prophylaxis (PrEP) to ensure that the patient is more fully protected in a potential risk environment. In addition, the system can also provide more detailed rabies-related popular science knowledge according to the needs of patients. This information includes the transmission route, clinical manifestations, severity after onset of rabies, and the importance and related methods of prevention. At the same time, the system will also encourage patients to consult professional physicians when they are unsure of their own situation to ensure the most appropriate medical advice and avoid ignoring any possible health risks. Through these measures, the intelligent question-and-answer system can not only give scientific advice under specific conditions, but also improve the public's awareness of rabies and comprehensively protect patient safety.

[0058] If R>T, the system believes that the patient cannot be ruled out as exposed and requires further diagnosis. At this point, the system will call the intelligent recognition system for more in-depth analysis and evaluation.

[0059] The intelligent recognition system includes more sophisticated professional image analysis to obtain more comprehensive risk assessment information, thereby providing more accurate medical advice.

[0060] In this way, the intelligent question-and-answer system can not only quickly determine whether a patient is at risk of rabies exposure, but can also flexibly call on other intelligent systems for in-depth analysis based on the complexity of the situation, ensuring that patients receive timely and accurate medical advice.

[0061] After determining that there is an exposure risk, if there is a skin wound, the intelligent recognition system will be used to further identify the wound, determine the severity of the wound, and then enter the intelligent decision-making system; if there is no skin wound, the intelligent question and answer system will be directly entered into the intelligent decision-making system based on the patient's condition.

[0062] 2. Intelligent recognition system

[0063] The intelligent recognition system is an image-based medical auxiliary recognition system. The system collects the patient's image information into the system through image sensors or information transmission interfaces. After collection, the system processes the image according to the neural network. After the processing is completed, detailed information about the patient's wound is obtained, including: (1) Wound type (puncture wounds, lacerations, scratches and bites) (2) Wound location (head and face, trunk, limbs) (3) Wound size (length, width, depth) (4) Wound bleeding (5) Biological residues

[0064] 1. Wound type (T): categorical variable, with numerical values ​​representing different types (e.g., puncture wound = 1, laceration = 2, scratch bite = 3)

[0065] 2. Wound location (L): Specifically, whether the head and face are included (1: included 0: not included); whether the trunk is included (1: included 0: not included); whether the limbs are included (1: included 0: not included)

[0066] 3. Wound size (D): including length, width and depth, expressed in L d , W d , H d Indicates (unit: cm)

[0067] 4. Wound bleeding (B): whether there is obvious bleeding, binary variable (1: bleeding, 0: no bleeding)

[0068] 5. Biological residues (R): Is there any biological residue, such as teeth, hair, saliva, etc. If there is biological residue, determine the type of biological residue and the amount of biological residue R: R=k·(S d +∈) b

[0069] Where S d is the observed signal size of the organism remaining on the surface, ∈ is the error intensity, k is the proportional constant, b is the nonlinear coefficient, ∈, k, and b are obtained through neural network training.

[0070] Define an injury score E to quantify the severity of the wound. This scoring formula will take all features into account and assign a weight to each feature:

[0071] The injury score E=f(T, L, D, B, R) is given based on the judgment of the neural network. At the same time, during this process, if the intelligent recognition system finds local deformity, suspected fracture, or suspected foreign body residue in the wound, it will ask the patient to undergo further imaging examinations. After the imaging examination results are output, they will be input into the intelligent decision-making system.

[0072] The training method of the neural network is:

[0073] Data collection: Since this type of medical image data is highly private and sensitive, unlike traditional collection methods, a large amount of desensitized and professionally labeled medical image data is required. The following methods can be used for collection: (1) Cooperate with hospitals or medical institutions to obtain de-identified medical image data. De-identification means removing all information that can identify the patient's identity; (2) Physician annotation: On the obtained images, the physician manually annotates the information to be identified in different dimensions of the image. (3) In addition, since the size information of the image needs to be obtained, it is also necessary to collect the camera parameters when the image is taken, such as aperture, focal length, etc.

[0074] Design loss function: In the intelligent recognition system, to output information of multiple dimensions, it is necessary to design different loss functions for different dimensions, which are described as follows:

[0075] For wound types, the loss function is designed

[0076] Where C is the number of categories, y c is the one-hot encoding of the true label, is the model's predicted probability for the class.

[0077] For the wound position, the loss function is designed as:

[0078] Among them, y represents the true label (1: included 0: not included), Represents the predicted label.

[0079] For wound size information, the loss function is designed as mean square error:

[0080] Among them, x i is the actual wound size, is the predicted wound size and n is the number of samples.

[0081] (4) For wound bleeding, the loss function is similar to the wound location.

[0082] Training neural network: The main network structure of this type of recognition system uses convolutional neural network as the backbone. Since the location and size of the wound need to be detected, target detection models (such as YOLO, SSD, FasterR-CNN, etc.) can be selected. These models can simultaneously predict the location and category of the target. The overall network is built in combination with the pre-processing and post-processing of the network; based on the diversity of the target data to be acquired, the neural network can be trained with a single-structure neural network word or multiple trainings, or with neural networks of different structures. Different loss functions can be set for different training tasks. These loss functions can include mean square error loss, cross entropy loss, etc. Therefore, overall, the structure of the neural network is a single-input multi-output or single-input multi-output network. At the same time, pre-trained models with medical image information (such as VGG, ResNet, etc.) can be used and fine-tuned on their basis to save training time and improve model performance.

[0083] At the same time, the input of the neural network can also be video. By adding timeline data, the video data can be analyzed to obtain specific target data.

[0084] How to use the system:

[0085] If the intelligent question-answering system needs to further determine the risk of rabies exposure and there are unhealed skin wounds, the intelligent recognition system will be called for processing. The processing flow of the intelligent recognition system is to first obtain images. The image acquisition rules of the intelligent recognition system are: ① Take and upload at least one clear photo about 15 cm away from the wound that can show the full picture of the wound; ② Apply appropriate force in the direction perpendicular to the long axis of the wound to open the wound to the maximum extent that the patient can tolerate, and then take and upload at least one clear photo as before. If there is a ruler, you can place the ruler and take the photo (the system screen can display the diagram of the open wound and the ruler placement method).

[0086] After taking a picture or video, the acquired picture is executed in multiple neural networks, the different attribute outputs of the neural network are obtained, and these attributes are transmitted to the intelligent decision-making system for the next decision.

[0087] 3. Intelligent Decision-making System

[0088] The intelligent decision-making system is a decision-making system that uses medical information as input. By obtaining information such as the characteristics of the animal that caused the injury, the injury situation (such as whether the animal actively attacked, or one dog injured multiple people, etc.), the patient's immune status and the severity of the wound, the intelligent decision-making system will make judgments and give two pieces of information: one is the patient's rabies exposure level, and the other is the patient's diagnosis and treatment recommendations, and generate medical records for the doctor's reference.

[0089] How the intelligent decision-making system works:

[0090] The input of the intelligent decision-making system consists of three parts: (1) characteristics of the wounding animal, (2) severity of the wound, and (3) the patient's immune status. The output of the intelligent decision-making system consists of two parts: (1) rabies exposure classification and (2) clinical treatment recommendations.

[0091] The risk assessment score S is the weighted sum of three main factors, each of which is represented by a specific function f1, f2, f3:

[0092] S = α×f1 (characteristics of the wounding animal, the circumstances of the wound) + β×f2 (wound severity) + γ×f3 (patient’s immune status)

[0093] Among them, α, β, and γ are weight coefficients used to indicate the proportion of each factor in the comprehensive evaluation. These weights can be determined by clinical data or expert opinions.

[0094] 1. Characteristics of the injured animal and the injury condition function f1

[0095] F1 mainly evaluates the possibility that the injured animal is infected with rabies virus, taking into account the following factors:

[0096] Type of animal that caused the injury (e.g. wild / stray animal = 1, owned animal = 0)

[0097] Aggressive behavior (active aggression = 1, aggression after provocation = 0)

[0098] The number of people injured by animals recently (if ≥ 3 people, the value is 1, otherwise 0)

[0099] Laboratory testing of the injured animal (testing confirmed rabies = 1, no testing or testing ruled out rabies = 0)

[0100] Animal-injury behavior (obvious behavioral abnormality = 1, no = 0)

[0101] The formula for defining f1 is: f1=c1×animal species+c2×aggression behavior+c3×number of injuries+c4×rabies status+c5×abnormal behavior

[0102] Where c1, c2, c3, c4, c5 are the weights of each factor.

[0103] 2. The wound severity function f2 is the output E of the intelligent recognition system.

[0104] 3. Patient immune status function f3

[0105] f3 Assess the patient's immune status, especially against rabies, including:

[0106] Rabies vaccination status (not fully vaccinated = 1, fully vaccinated = 0)

[0107] Whether the patient has severe immunosuppression (yes = 1, no = 0)

[0108] The formula for defining f3 is: f3=e1×rabies vaccination status+e2×immune function status, where e1 and e2 are the weights of each factor.

[0109] The risk assessment score S is assigned a point system with a total of 10 points. The higher the score, the higher the risk of rabies exposure. The threshold for exposure classification is obtained through neural network training.

[0110] The exposure classification threshold given by the neural network and the calculated risk assessment score can give the final rabies exposure classification. The rabies exposure classification is combined with the specific situation obtained by the intelligent question-answering system, and the intelligent decision-making system will give specific diagnosis and treatment recommendations.

[0111] It should be noted that this system does not have the authority to prescribe and cannot replace doctors in making diagnostic conclusions. However, as an auxiliary means of diagnosis and treatment for doctors, it can provide diagnosis and treatment suggestions for doctors to refer to and selectively adopt.

[0112] After the doctor makes a judgment, if he finds that the diagnosis conclusion given by this system is wrong, the relevant diagnosis data will be automatically added to the database as a bad case for subsequent training.

[0113] Example 1

[0114] The patient came to the hospital because his right calf was scratched by a dog. The working mode of this system is:

[0115] 1. Intelligent Question Answering System

[0116] The intelligent question-answering system automatically generates questions. During the communication between the patient and the intelligent question-answering system, the intelligent question-answering system absorbs the following content:

[0117] The patient came to the clinic because he was scratched on his right calf by a dog 15 hours ago; the dog that caused the injury was a family dog; it was not an active attack; it had only injured one person recently; the dog was not tested for rabies virus infection; the patient washed the wound with water and disinfected with alcohol by himself after the injury; the patient had a healthy body and no history of chronic diseases; he was bitten on his right thumb by a dog one year ago, and underwent wound treatment, rabies immunoglobulin, and one dose of rabies vaccine at Hepingli Hospital, but did not follow the instructions to receive subsequent vaccines; he has no history of allergies.

[0118] Based on the above information, the exposure assessment score R is calculated. R>T, and the assessment is that the possibility of exposure cannot be ruled out and further judgment is needed.

[0119] 2. Intelligent recognition system

[0120] After taking a photo of the wound and inputting it into the intelligent recognition system, the following conclusions were drawn:

[0121] Wound type (T): scratches and bites = 3

[0122] Wound location (L): Does it include the head and face: Not included = 0, Does it include the trunk: Not included = 0, Does it include the limbs: Included = 1

[0123] Wound size (D): length 7cm, width 0.1cm, depth 0.1cm

[0124] Bleeding from wound (B): No bleeding = 0

[0125] Biological residues (R): No residue = 0

[0126] 3. Intelligent Decision-making System

[0127] The intelligent decision-making system calculated the risk assessment score S and determined that the patient was exposed to rabies at level II. At the same time, the intelligent decision-making system gave the following diagnosis and treatment recommendations based on the patient's information:

[0128] Diagnosis: Dog scratch

[0129] Treatment: 1. The patient was exposed to rabies at level II again. Because he did not complete the full course of rabies vaccination last time, this time he was treated the same as the first level II exposure to rabies. 2. Use professional flushing equipment to flush the wound and disinfect it with iodine. Advise the patient to keep the wound clean and dry, and change the wound dressing in the trauma emergency department after 2 days. 3. Rabies vaccine should be administered using the 5-needle method or the 2-1-1 method, and the patient should be advised to receive subsequent doses of vaccine according to the schedule. 4. Follow up if any discomfort occurs.

[0130] Example 2

[0131] A patient comes to the hospital for treatment for a cat bite. Here is how this system works:

[0132] 1. Intelligent Question Answering System

[0133] The intelligent question-answering system automatically generates questions. During the communication between the patient and the intelligent question-answering system, the intelligent question-answering system absorbs the following content:

[0134] The patient came to the clinic because he was bitten on his right thumb by a cat one hour ago; the cat that caused the injury was a stray cat; it did not actively attack; it had only injured one person recently; the cat was not tested for rabies virus infection; it did not treat the injury on its own; it had a history of diabetes and poor blood sugar control; it had been fully vaccinated with rabies vaccine due to a dog bite two years ago; and it had no history of allergies.

[0135] Based on the above information, the exposure assessment score R is calculated. R>T, and the assessment is that the possibility of exposure cannot be ruled out and further judgment is needed.

[0136] 3. Intelligent recognition system

[0137] After taking a photo of the wound and inputting it into the intelligent recognition system, the following conclusions were drawn:

[0138] Wound Type (T): Puncture Wound = 1

[0139] Wound location (L): Does it include the head and face: Not included = 0, Does it include the trunk: Not included = 0, Does it include the limbs: Included = 1

[0140] Wound size (D): length 0.2cm, width 0.2cm, depth 0.5cm

[0141] Bleeding condition of wound (B): Bleeding = 1

[0142] Biological residues (R): No residue = 0

[0143] 3. Intelligent Decision-making System

[0144] The intelligent decision-making system calculated the risk assessment score S and determined that the patient was exposed to rabies at level III. At the same time, the intelligent decision-making system gave the following diagnosis and treatment recommendations based on the patient's information:

[0145] Diagnosis: Cat bite

[0146] Treatment: 1. Re-exposure to level III rabies; 2. Use professional flushing equipment to flush the wound, disinfect with iodine, and ask the patient to keep the wound clean and dry, and change the wound dressing in the trauma emergency department after 2 days; 3. Strengthen vaccination with 2 doses of rabies vaccine using the 0-3 method, and ask the patient to receive subsequent doses of vaccine according to the schedule; 4. Take oral antibiotics to prevent wound infection; 5. Seek medical attention immediately if you feel unwell.

[0147] The present invention integrates artificial intelligence technology to develop a set of phased automated processing procedures for rabies prevention and treatment clinics, which significantly improves the work efficiency of doctors and optimizes the medical experience of patients. The present invention can make medical services more efficient and humane, and can also accurately evaluate wound types (such as puncture wounds or lacerations, scratches and bites), locations, bleeding conditions, and biological residues, and provide patients with more scientific medical advice.

[0148] The solution of the patent of this invention deeply considers the practicality and feasibility of the system, and goes beyond the simple algorithm level. By designing a deep neural network algorithm and closely integrating it with the hardware environment and medical practice scenarios, the comprehensiveness and practicality of the solution are ensured. In addition, the present invention is specially optimized for the emergency treatment scenario of rabies, and can quickly provide accurate medical advice to gain valuable treatment time for patients.

[0149] The above embodiments are only examples for clearly explaining the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description, and these various changes, modifications, substitutions and deformations derived from the principles and spirit of the present invention still fall within the scope of protection of the present invention.

Claims

1. A method for assessing rabies exposure risk, characterized in that: include: A patient information set was generated based on the patient's responses to the rabies exposure questions; Analyzing the patient information set using a pre-trained question-answer analysis model, and obtaining the visual content of the patient's wound when the analysis result is greater than a pre-set threshold; Based on the visual content of the wound and the patient information set, the risk of rabies exposure was assessed.

2. The method according to claim 1, characterized in that: The patient information set is analyzed using a pre-trained question-answering analysis model according to the following formula: R=M(I)=M(i1,i2,i3,i4……,i n ) Wherein, R is the analysis result, M is the pre-trained question-answering analysis model, {i1,i2,i3,i4……,i n } is the patient information set, and n is the number of answers in the patient information set.

3. The method according to claim 2, characterized in that Before using the pre-trained question-answering analysis model to analyze the patient information set, it also includes: Obtain medical knowledge input text information and general large language model pre-training model information; Perform feature extraction on the preprocessed medical knowledge input text information, build a feature knowledge base, and integrate it into the large language model; Generate new questions that may be asked by patients, test the medical diagnosis effect of the question-answer analysis model, and iteratively optimize the question-answer analysis model based on the medical diagnosis effect.

4. The method according to claim 3, characterized in that After obtaining the visual content of the patient's wound, the method further includes: According to the pre-trained visual analysis model, the visual content of the wound is processed to obtain a judgment result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification : Where C is the number of categories, y c is the one-hot encoding of the true label, is the predicted probability of the category by the visual analysis model; When the wound type is a puncture wound or a laceration, a judgment result of the wound size is obtained according to a pre-trained visual analysis model, wherein the pre-trained visual analysis model also includes a loss function L MSE : Among them, x i is the actual wound size, is the predicted wound size, n is the number of samples; When the determination result of the wound size does not meet the requirement, a suggestion for reshooting is given.

5. The method according to claim 4, characterized in that When the judgment result of the wound size meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, represents the predicted label; When the determination result of the wound position does not meet the requirements, a suggestion for reshooting is given; When the determination result of the wound location meets the requirements, the step of assessing the risk of rabies exposure is performed.

6. The method according to claim 3, characterized in that After obtaining the visual content of the patient's wound, the method further includes: According to the pre-trained visual analysis model, the visual content of the wound is processed to obtain a judgment result of the wound type, wherein the pre-trained visual analysis model includes a loss function L classification : Where C is the number of categories, y c is the one-hot encoding of the true label, is the predicted probability of the category by the visual analysis model; When the wound type is a scratch or bite mark, a judgment result of the wound bleeding is obtained according to the pre-trained visual analysis model, wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, represents the predicted label; When the judgment result of the wound bleeding condition does not meet the requirements, a suggestion for reshooting is given.

7. The method according to claim 6, characterized in that When the judgment result of the wound bleeding condition meets the requirements, the judgment result of the wound position is obtained according to the pre-trained visual analysis model; wherein the pre-trained visual analysis model also includes a loss function L binary : Among them, y represents the true label, represents the predicted label; When the determination result of the wound position does not meet the requirements, a suggestion for reshooting is given; When the determination result of the wound location meets the requirements, the step of assessing the risk of rabies exposure is performed.

8. The method according to any one of claims 4 to 7, characterized in that: According to the pre-trained visual analysis model, the judgment result of whether there are biological residues is obtained; When there are biological residues, determine the type and amount of biological residues R, and perform the steps to assess the risk of rabies exposure; The biological residue R is calculated according to the following formula: R=k·(S d +∈) b Where S d is the observed signal size of the biological residue on the surface, ∈ is the error intensity, k is the proportional constant, b is the nonlinear coefficient, ∈, k, and b are obtained through neural network training.

9. The method according to claim 8, characterized in that The risk of rabies exposure was estimated according to the following formula: Among them, α, β, and γ are weight coefficients, and f1, f2, and f3 represent the characteristics of the wounding animal and the function of the wounding situation, the function of the severity of the wound, and the function of the patient's immune status, respectively; And among them, f1 is calculated according to the following formula: Among them, c1, c2, c3, c4, c5 are the weights of each factor; And wherein f2 is calculated according to the following formula: E=f(T,L,D,B,R), wherein E is the injury score, T is the wound type, L is the wound location, D is the wound size, B is the wound bleeding, and R is the biological residue; And among them, f3 is calculated according to the following formula: Among them, e1 and e2 are the weights of each factor.

10. A rabies exposure risk assessment device, characterized in that: include: An information generation module, for generating a patient information set based on the patient's answers to the rabies exposure questions; A visual acquisition module, used to analyze the patient information set using a pre-trained question-answer analysis model, and to acquire the visual content of the patient's wound when the analysis result is greater than a pre-set threshold; A risk assessment module is used to assess the risk of rabies exposure based on the visual content of the wound and the patient information set.

Images