A system for managing allergies in children

Abstract

The application relates to the field of intelligent medical treatment, in particular to a child allergy management system. The system comprises a risk early warning module for acute allergy time risk early warning; the risk early warning module comprises a data acquisition module for acquiring basic data of a to-be-tested person, including physiological index data, environment data and medical record data; a risk assessment module for inputting the basic data into a risk assessment model for assessment to obtain a risk result; the construction process of the risk assessment model is as follows: acquiring a basic data set of the to-be-tested person and a risk category label; inputting the data set and the risk category label into a neural network for training to obtain the risk assessment model; wherein the neural network is based on an oscillation period for parameter initialization, and the initialization parameters are calculated through the product of a weight adjustment factor, the oscillation period, a sine term or a cosine term of the oscillation period and a correlation parameter of the neural network. The application has good clinical value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, in particular to a child allergy management system, device, program product and computer readable storage medium. BACKGROUND

[0002] As a global health problem with high incidence, the risk of allergy is affected by a variety of complex factors, including physiological indicators of patients, environmental conditions, medical history and genetic factors. With the intensification of environmental pollution and climate change, the incidence of allergy-related diseases is on the rise, and the child population has become the main constituent group of synchronous increase in incidence and prevalence. Since allergic reactions can rapidly develop into serious health threats, especially triggering anaphylactic shock and other emergencies, it is crucial to assess, manage and effectively warn the risk of allergy in a timely and accurate manner. However, the management of child allergic diseases faces various challenges, such as the problem of patients not visiting doctors in time, single method of allergy cognitive education, and poor treatment compliance; at the physician level, due to the imperfect training mechanism of allergy specialists, physicians face the problems of insufficient diagnostic ability and incomplete standardized management process. Although the standardized diagnosis and treatment level of child allergic diseases has been significantly improved through the continuous efforts of several generations of pediatricians in China, the diagnosis rate and control rate are still not ideal and still face many challenges. Secondly, in the task of allergy risk warning, the existing artificial intelligence methods rely too much on local optimization in training the model, lack effective global optimization strategies, resulting in reduced accuracy of the model in dealing with complex features. In addition, the single weight initialization method limits the exploration ability in the initial stage, which easily causes training convergence difficulties of the model in complex medical tasks. SUMMARY

[0003] In view of the above problems, the present application provides a child allergy management system, which specifically comprises:

[0004] The risk warning module is used for acute allergy risk warning, and comprises:

[0005] The data acquisition module is used for acquiring the basic data of the testee, including physiological indicator data, environmental data and medical record data.

[0006] The risk assessment module is used for inputting the basic data into a risk assessment model for assessment to obtain a risk result, wherein the risk result comprises one or more of the following: no allergy risk, low allergy risk, moderate allergy risk and high allergy risk.

[0007] The construction process of the risk assessment model is as follows:

[0008] The basic data set of the testee and the label of the risk category are acquired.

[0009] The data set and the label of the risk category are input into a neural network for training to obtain a risk assessment model.

[0010] The neural network is initialized based on the oscillation period, and the initialization parameters are calculated by the product of the weight adjustment factor, the oscillation period, the sine term or the cosine term of the oscillation period, and the correlation parameter of the neural network.

[0011] During the construction process of the risk assessment model, the oscillation period is adaptively adjusted to optimize the iteration process of the neural network. When the loss of the loss function fluctuates sharply in the iteration process, the iteration is performed after adjusting the oscillation amplitude.

[0012] After the iteration parameter is updated in the construction process of the risk assessment model, a random disturbance is added to optimize the parameter update range of the neural network. The size of the random disturbance is dynamically adjusted by the number of iterations of the training.

[0013] The construction of the evaluation model also includes data augmentation. The data set and the label of the risk category are augmented to obtain augmented data, and the augmented data are input into a neural network model for training to obtain a risk assessment model. The data augmentation is performed by a generative adversarial network. The loss function calculation of the generative adversarial network includes an adversarial loss and a feature comparison loss. The calculation of the adversarial loss includes random noise data, feature data of allergic medical data, and the calculation of the feature comparison loss includes allergic medical data.

[0014] The feature data of the allergic medical data includes one or more of the following: statistical features, semantic features, image features, time series features, and specific medical indicator features. The allergic medical data includes one or more of the following: heart rate, blood pressure, air quality index, pollen concentration, allergic history, drug type, age, gender, temperature and humidity index, and family allergic history.

[0015] The generative adversarial network further includes a distribution adaptive evaluation mechanism, including a diversity loss and a quality evaluation loss. The quality evaluation loss is a quality difference measure obtained by comparing the mean and variance between the generated allergic medical data and the real allergic medical data. The diversity loss is calculated by a diversity loss factor and a generated sample.

[0016] The system further includes one or more of the following:

[0017] Early identification module: used for early identification and screening of allergic disease related symptoms;

[0018] Sensitive monitoring module: used for recording detection data of allergic remission period;

[0019] Sensitive detection and intelligent reading module: used for intelligent interpretation of allergen detection report;

[0020] Emergency plan module: for pushing the processing scheme of the acute allergic event or the risk of allergic event;

[0021] Health management module: for the health management of allergic children related to allergic and non-allergic diseases.

[0022] The purpose of the present application is to provide a risk assessment method for allergic reactions, comprising:

[0023] Obtaining the basic data of the subject, including physiological index data, environmental data, medical record data;

[0024] The basic data is input into the risk assessment model for evaluation to obtain a risk result, which includes one or more of the following: no allergic risk, low allergic risk, moderate allergic risk, high allergic risk;

[0025] The construction process of the risk assessment model is:

[0026] Obtaining the basic data set of the subject and the label of the risk category;

[0027] The data set and the label of the risk category are input into the neural network for training to obtain the risk assessment model;

[0028] The neural network is based on the oscillation period to initialize the parameters, and the initialization parameters are calculated by the product of the weight adjustment factor, the oscillation period, the sine term or the cosine term of the oscillation period, and the correlation parameters of the neural network.

[0029] The oscillation period is adaptively adjusted to optimize the iteration process of the neural network in the construction process of the risk assessment model, and the iteration is performed after adjusting the oscillation amplitude when the loss of the loss function fluctuates sharply in the iteration process.

[0030] The random disturbance is added to optimize the parameter update range of the neural network after the iteration parameter is updated in the construction process of the risk model, and the size of the random disturbance is dynamically adjusted by the number of iterations of the training.

[0031] The construction of the evaluation model also includes data augmentation; the data set and the label of the risk category are augmented to obtain augmented data, and the augmented data is input into the neural network model for training to obtain the risk assessment model; the data augmentation is augmented by the generative adversarial network, and the loss function calculation of the generative adversarial network includes the adversarial loss and the feature comparison loss, the adversarial loss calculation includes the random noise data, the feature data of the allergic medical data, and the feature comparison loss calculation includes the allergic medical data.

[0032] The feature data of the allergic medical data includes one or more of statistical features, semantic features, image features, time series features, and specific medical index features.

[0033] The generative adversarial network further includes a distribution adaptive evaluation mechanism, including a diversity loss and a quality evaluation loss, the quality evaluation loss being a quality difference measurement by comparing the mean and variance between the generated allergic medical data and the real allergic medical data, and the diversity loss being calculated by a diversity loss factor and the generated sample.

[0034] The present application aims to provide a computer device including a memory, a processor and a computer program or instructions stored on the memory, which are executed by the processor to realize the above-mentioned risk assessment method for allergic occurrence.

[0035] The present application aims to provide a computer readable storage medium having a computer program or instructions stored thereon, which are executed by the processor to realize the above-mentioned risk assessment method for allergic occurrence.

[0036] Advantages of the present application:

[0037] 1. In the allergic risk warning task, an oscillation period optimized neural network is used, and the oscillation period and random disturbance are used for parameter initialization, the initial distribution of neural network weight and bias is optimized, and the exploration ability of the model to the feature space is increased. Moreover, the oscillation period is combined with a dynamic adjustment mechanism, the period amplitude is adapted according to the loss function fluctuation, the global search ability in training is improved, and local optimum is avoided. In addition, the dynamic adjustment of local bias is added in the parameter optimization process, so that more depth feature interaction information is integrated into the neural network weight update.

[0038] 2. In the allergic risk warning task, a weight update combined with random disturbance and dynamic feedback is used, random disturbance is used after each weight update, and gradually decays with the training, so as to ensure wider exploration of the parameter space in the early stage and stability in the convergence stage. At the same time, the dynamic feedback factor is used to optimize the weight update step according to the importance difference of different features, so that the model is more suitable for the complexity and diversity of allergic medical data.

[0039] 3. In the allergy risk warning task, the nonlinear feature adaptation ability is enhanced, and in the error back propagation, the local deviation of the gradient information coupled with the oscillation period is improved, so that the neural network can learn the nonlinear features of the allergy medical data. In addition, the model's ability to adapt to complex feature interaction patterns is significantly enhanced, allowing it to capture subtle allergy factor correlations and further improve prediction accuracy.

[0040] 4. In the allergy risk warning task, the existing data augmentation method generates medical data with low quality, and the generated samples lack semantic consistency, which can easily lead to poor model training results. In addition, the diversity between samples is insufficient, which cannot effectively cover the problem of various types of allergic reactions. In the allergy risk warning task, a generative adversarial network is used to augment medical data, using the mechanism of generator and discriminator to generate medical data from random noise combined with feature embedding, thereby solving the problem of insufficient allergy-related medical data samples. In addition, the generator uses feature contrast loss to ensure that the generated data and the real data are consistent in the high-level semantic space. Feature contrast is optimized based on medical data feature embedding (such as heart rate, air quality index, etc.), ensuring the reliability of the generated data in a medical context. At the same time, the discriminator combines diversity loss to evaluate the difference between samples, ensuring the diversity of generated data and effectively avoiding the problem of single pattern. Moreover, the discriminator uses quality evaluation loss to evaluate the data quality based on the difference between the mean and variance of the generated data and the real data, improving the numerical consistency of the generated data.

[0041] 5. In the allergy risk warning task, a dynamic feature control mechanism is used, and the generator adjusts the noise distribution and feature embedding method adaptively to expand the coverage of the generated samples, allowing them to reflect a wider variety of allergic reaction types and symptoms. The discriminator uses a distribution adaptive evaluation mechanism to guide the generator to generate high-quality medical data that is superior to traditional methods in terms of data feature distribution and semantic space.

[0042] 6. The present application integrates the conditions of each stage of allergy and proposes a child allergy management system, including early identification of allergy, allergy monitoring, allergy detection and intelligent reading, allergy risk warning, allergy emergency plan, and allergy management outside the hospital, to enhance personalized treatment for each child's unique needs and provide effective management throughout the child's illness. The whole process of information service for dynamic monitoring, accurate diagnosis, intelligent evaluation, intelligent screening, standard treatment and follow-up plan of children's allergic diseases realizes closed-loop management. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to make the technical solutions in the embodiments of the present application clearer, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0044] Figure 1 The management system for children's allergies provided by the embodiments of the present application is shown in the schematic diagram.

[0045] Figure 2 The architecture of the management system for children's allergies provided by the embodiments of the present application is shown in the schematic diagram.

[0046] Figure 3 The application interface of the "user end" application provided by the embodiments of the present application is shown in the schematic diagram.

[0047] Figure 4 The application interface of the "physician end" application provided by the embodiments of the present application is shown in the schematic diagram.

[0048] Figure 5 The risk assessment method for allergies provided by the embodiments of the present application is shown in the schematic diagram.

[0049] Figure 6 The risk assessment device for allergies provided by the embodiments of the present application is shown in the schematic diagram. DETAILED DESCRIPTION

[0050] In order to make the technical solutions in the embodiments of the present application clearer, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0051] In some processes described in the specification and claims of the present application and the above-mentioned drawings, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed or in parallel without the order in which they appear in this text. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and these operations can be executed in sequence or in parallel. It should be noted that the "first", "second", etc. in this text are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence. "First" and "second" are not of different types.

[0052] Figure 1 The management system for children's allergies provided by the embodiments of the present application is shown in the schematic diagram, specifically including:

[0053] The risk warning module is used for acute allergy risk warning; the risk warning module includes:

[0054] Data acquisition module: obtain the basic data of the testee, including physiological index data, environmental data, medical record data;

[0055] In one specific embodiment, the model training data for the risk warning of allergy occurrence of the present application is obtained from multi-source data, including but not limited to the following categories:

[0056] 1) Patient physiological index data: such as heart rate, blood pressure, respiratory rate, etc.

[0057] 2) Environmental data: such as air quality index (PM2.5, PM10), temperature and humidity, allergen concentration, etc.

[0058] 3) Medical record data: such as allergy history, drug use record, hospitalization record, etc.

[0059] 4) Demographic information: such as age, gender, residence, etc.

[0060] In one embodiment, the attributes of the data include: Ra1 is heart rate (bpm), Ra2 is blood pressure (mmHg), Da1 is air quality index (AQI), Da2 is pollen concentration (ppm), Pa1 is allergy history (classified as none, mild, moderate, severe), Pa2 is drug type (classified as antihistamines, steroids, etc.), Xa1 is age (years), Xa2 is gender (classified as male, female), Da3 is temperature and humidity index (combined features), and Pa3 is family allergy history (classified as none, present).

[0061] It should be noted that the present embodiment is only to illustrate one data format and category of the present application, and in actual application, the number of attributes of the data is usually more than 10, and the number of attributes of the data may reach dozens or even hundreds.

[0062] Further, the collected data is labeled, and the labeling method of the present application is artificial labeling. In one embodiment, the labeling categories include: no allergy risk, low allergy risk, moderate allergy risk, and high allergy risk, a total of 4 categories.

[0063] In the present embodiment, the 5 example data are as follows:

[0064]

[0065] The present application adopts the Word2Vec algorithm for vectorization processing of the text. The Word2Vec algorithm is a commonly used vectorization algorithm in the art, which scans the text to be vectorized according to a pre-set large-scale corpus, and represents each word as a one-hot encoding vector, the dimension of the one-hot encoding vector being equal to the size of the vocabulary in the corpus.

[0066] In an embodiment, the construction of the evaluation model further comprises data augmentation; the data set and the label of the risk category are subjected to data augmentation to obtain augmented data, and the augmented data is input into a neural network model to obtain a risk evaluation model; the data augmentation is performed by a generative adversarial network, a loss function of the generative adversarial network comprises an adversarial loss and a feature contrast loss, the adversarial loss comprises random noise data and feature data of the allergic medical data, and the feature contrast loss comprises the allergic medical data.

[0067] In an embodiment, the feature data of the allergic medical data comprises one or more of, but is not limited to, the following: statistical features, semantic features, image features, time series features, and specific medical indicator features; and the allergic medical data comprises one or more of, but is not limited to, the following: heart rate, blood pressure, air quality index, pollen concentration, allergic history, type of drug used, age, gender, temperature and humidity index, and family allergic history.

[0068] In an embodiment, the generative adversarial network further comprises a distribution adaptive evaluation mechanism, comprising a diversity loss and a quality evaluation loss, the quality evaluation loss is a quality difference measurement by comparing the mean and variance between the generated allergic medical data and the real allergic medical data; and the diversity loss is calculated by a diversity loss factor and generated samples.

[0069] In a specific embodiment, in the task of the present application, the collection, labeling and preprocessing of allergic related medical training data are time-consuming and laborious, and insufficient training samples can easily lead to poor generalization ability of the model and affect the accuracy of the model.

[0070] To solve the problem of limited model generalization ability caused by insufficient original allergic related medical data, the collected and labeled allergic related medical training data are augmented by a generative adversarial network, and a feature contrast loss and a diversity loss are used in the training process of the generative adversarial network to improve the quality and diversity of the generated allergic related medical data.

[0071] Specifically, the training process of the generative adversarial network is as follows:

[0072] 1) The parameters of the generator and the discriminator of the generative adversarial network are initialized, the generator generates preliminary synthetic data using random noise and feature information of the original allergic related medical data, and the discriminator distinguishes between real allergic related medical data and synthetic allergic related medical data, and at the same time promotes the generator to gradually improve the quality and diversity of the synthetic allergic related medical data through adversarial training, let be the initial weight of the generator, The initial weight of the discriminator is initialized randomly, and the initialized parameter follows a normal distribution with a mean of 0 and a variance of a unit matrix.

[0073] 2) The generator generates allergy-related medical data by combining random noise and feature embedding vectors, and inputs the synthetic allergy-related medical data into the discriminator, aiming to minimize the discrimination error of the discriminator and enhance the diversity and realism of the allergy-related medical data through an adaptive mechanism, thereby solving the problem of lack of diversity in the generated allergy-related medical data caused by single noise input. The loss function of the generator includes an adversarial loss and a feature contrast loss, wherein the feature contrast loss performs high-level semantic contrast between the generated allergy-related medical data and the real allergy-related medical data in the form of perception loss, and the generated allergy-related medical data is consistent with the real allergy-related medical data at the high-level semantic level. The calculation method of the adversarial loss of the generator is represented as:

[0074]

[0075] In the formula, is the adversarial loss of the generator; is a random noise vector, which changes the distribution of the input noise or adjusts the generation strategy, such as using different noise vectors or feature embeddings, so that the generated data can cover a wider range of allergic reactions and symptoms, thereby increasing the coverage of the data set; is a feature embedding extracted from real allergy-related medical data, such as a high-dimensional representation obtained from original medical data through a feature extraction method, including statistical features (such as mean, variance, etc.), semantic features (case text embedding extracted by a language model such as BioBERT), image features (medical image feature vector extracted by a convolutional neural network such as ResNet), time series features (reflecting the trend or periodicity of allergic reaction dynamics), and specific medical indicators (such as IgE antibody level, skin scratch test results, etc.); is a generator network function, is a discriminator network function, represents expectation. Preferably, can be set as a random vector following a uniform distribution or a normal distribution.

[0076] and the calculation method of the feature contrast loss of the generator is represented as:

[0077]

[0078] In the formula, is the feature contrast loss of the generator, is a weight coefficient for adjusting the loss, is the output of a feature extraction network (such as a multi-layer convolutional neural network), are true allergy-related medical data, and attributes include Ra1 is heart rate (bpm), Ra2 is blood pressure (mmHg), Da1 is air quality index (AQI), Da2 is pollen concentration (ppm), Pa1 is allergy history (classified as none, mild, moderate, severe), Pa2 is drug type used (classified as antihistamines, steroids, etc.), Xa1 is age (years), Xa2 is gender (classified as male, female), Da3 is temperature and humidity index (combined features), Pa3 is family allergy history (classified as none, exists); is an L2 norm, which is used here to measure the difference between the generated allergy-related medical data and the true allergy-related medical data in the feature space. Preferably, is set to 0.3.

[0079] Further, in order to achieve dynamic adaptive control of the weight coefficient for adjusting the loss, an adjustment factor based on the maximum root mean square error is adopted, which applies greater punishment to the case of greater distance in the early stage of training, and gradually weakens as the quality of generation improves, and the calculation method is represented as:

[0080]

[0081] In the formula, is an initial weight factor, is a distance measure between the generated allergy-related medical data and the true allergy-related medical data, is the maximum distance that occurs during training. Preferably, is set to 0.3.

[0082] Based on this, the feature comparison loss function helps to ensure the reliability and practicality of the generated data in medicine by comparing the high-level semantic similarity of the generated allergy-related medical data and the true data in the feature space, and the generated data can better simulate the key attributes of the true data, so as to more accurately reflect the allergy phenomenon in actual application; the quality evaluation loss function makes the discriminator evaluate the consistency of the data by comparing the statistical characteristics (such as mean and variance) of the generated data and the true data, ensuring that the generated data is consistent with the true data in numerical value, which helps to improve the accuracy and reliability of the model in processing real-world data.

[0083] 3) While the discriminator performs binary classification on whether the input allergy-related medical data is real, it also analyzes the differences between the generated allergy-related medical data and the real allergy-related medical data based on statistical distribution analysis. By adopting a distribution adaptive evaluation mechanism, it can effectively evaluate and guide the improvement of the quality and diversity of the generated allergy-related medical data, thereby solving the problem of the generated allergy-related medical data having a single pattern or overfitting. The discriminator combines diversity loss with quality assessment loss, where the quality assessment loss measures the difference between the distributions by comparing the mean and variance of the generated allergy-related medical data and the real allergy-related medical data.

[0084] The discriminator's quality assessment loss is calculated as follows:

[0085]

[0086] In the formula, For the quality assessment loss of the discriminator, The weighting coefficients for the quality assessment loss of the discriminator. and These represent the mean and variance of the generated allergy-related medical data, respectively. and These represent the mean and variance of real allergy-related medical data, respectively. This is the L2 norm, used here to measure the numerical difference between the mean and variance. Preferably, Set it to 0.3.

[0087] Based on this, the diversity loss ensures that there are certain differences between the generated samples, which not only avoids overfitting of the model, but also enhances the generalization ability of the model. In the task of augmenting allergy-related medical data, taking into account the individual differences between cases helps to enhance the diversity of the augmented samples.

[0088] Furthermore, to encourage the generation of allergy-related medical data that varies across different samples, the discriminator, with the aid of diversity loss, imposes additional constraints on the generation of allergy-related medical data. This guides the generator to output as many diverse samples as possible while generating high-quality allergy-related medical data. The diversity loss is calculated as follows:

[0089]

[0090] In the formula, For the loss of diversity, As the diversity loss weighting factor, Indicates the first One generated sample, Indicates the first One generated sample, is the number of generated allergy-related medical data. Preferably, is set to 0.3.

[0091] 4) Iterative training of the generator and the discriminator, both of which gradually learn to generate higher-quality and more diverse synthetic allergy-related medical data in the confrontation, and the discriminator gradually learns to more accurately determine the true and false in the confrontation, thereby solving the problem that pure one-way training is easy to cause the network to be difficult to converge or fall into mode collapse, and the calculation method of the weight update of the generator and the discriminator is represented as:

[0092]

[0093]

[0094] wherein, is a parameter update operation, is a generative adversarial network learning rate. Preferably, is set to 0.001.

[0095] 5) Repeat the above steps until the preset stopping iteration condition is met, that is, the model training is completed. In an embodiment, the preset stopping iteration condition is to reach the preset maximum number of iterations, and preferably, the preset maximum number of iterations is set to 1000 times.

[0096] After the training of the allergy-related medical data augmentation model is completed, the trained allergy-related medical data augmentation model is used to increase the number of samples. In an embodiment, the original collected samples are 800, the allergy-related medical data augmentation model generates 200 samples, and the expanded allergy-related medical data set contains 1000 samples.

[0097] The risk assessment module: input the basic data into the risk assessment model for evaluation to obtain a risk result, the risk result including one or more of the following: no allergy risk, low allergy risk, moderate allergy risk, high allergy risk;

[0098] The construction process of the risk assessment model is:

[0099] Obtain the basic data set of the subject to be tested and the label of the risk category;

[0100] Input the data set and the label of the risk category into the neural network for training to obtain the risk assessment model;

[0101] Wherein, the neural network is based on the oscillation period for parameter initialization, and the initialization parameters are calculated by the product of the weight adjustment factor, the oscillation period, the sine term or the cosine term of the oscillation period, and the associated parameters of the neural network.

[0102] In one embodiment, the parameters include weights and biases, the initialization of the weights is calculated by the product of the weight adjustment factor, the oscillation period, the sine term of the oscillation period, and the associated parameters of the neural network to obtain the initialized weights, and the initialization of the biases is calculated by the product of the weight adjustment factor, the oscillation period, the cosine term of the oscillation period, and the associated parameters of the neural network to obtain the initialized biases.

[0103] In one embodiment, the oscillation period is adaptively adjusted in the risk assessment model construction process to optimize the iterative process of the neural network, and after the adjustment of the oscillation amplitude, the iteration is performed when the loss of the loss function fluctuates sharply in the iteration process.

[0104] In one embodiment, the risk model construction process adds random disturbance to the parameter update range of the neural network after the iteration parameter update, and the size of the random disturbance is dynamically adjusted by the number of iterations of the training.

[0105] In one specific embodiment, the present application uses a neural network algorithm based on oscillation period optimization as a feature extraction model for allergic related medical data, and extracts features from allergic related medical data. In order to solve the technical problems that the neural network optimization is easy to fall into local optimal solution and difficult to efficiently explore the parameter space, the dynamic oscillation method is used to update the weights and biases of the neural network, and the global search is realized by combining the random disturbance and adaptive adjustment method, so that the neural network can get rid of the dependence on local gradient in the training process. Gradient descent method, realize global optimization or approximate global optimization.

[0106] Specifically, the training process of the neural network algorithm based on oscillation period optimization is as follows:

[0107] 1) The parameters of the neural network are initialized. Unlike the conventional random initialization method, the present application adopts an initialization method based on oscillation theory, sets the oscillation period and disturbance amplitude, improves the exploration ability in the initial stage and the optimization efficiency in the subsequent stage, and in the feature extraction task of allergic related medical data, the data may have high nonlinearity and complex mode. By using the oscillation period and random disturbance amplitude, the diversity of the initial weight distribution can be improved, and the ability of the neural network to explore the feature space can be enhanced. For the complex correlation in allergic related medical data, the convergence difficulty that may be caused by traditional initialization methods can be effectively avoided, and the efficiency and effect of optimization can be improved. The initialization method is represented as:

[0108]

[0109] In the formula, is the initial weight of the neural network, is the initial adjustment factor of the weight of the neural network, is a random number uniformly distributed from -1 to 1, is a sine term obtained from the initial oscillation period of the neural network, is the initial oscillation period of the neural network; is a correlation parameter of the number of connections of neurons, is the number of connections of neurons. Preferably, set , is a random number uniformly distributed from 0 to 1, i.e. the oscillation period is randomly distributed between 0 and ; set , .

[0110] Further, the bias initialization of the neural network not only considers random factors, but also relates to the weighted average of the input mode of the neuron, expressed as:

[0111]

[0112] wherein, is the initial bias of the neural network, is the bias initialization adjustment factor of the neural network, is a cosine term obtained from the initial oscillation period of the neural network, is a correlation parameter of the sum of the initial weights and the bias of the neural network, is the sum of all initial weights corresponding to the current neuron and the input connection, is the initial weight corresponding to the current neuron and the input connection. Preferably, set , .

[0113] 2) In the iterative process of neural network training, the oscillation period is adaptively adjusted according to the change of the loss function value of the neural network and the complexity of the neural network, and the oscillation amplitude is moderately increased when the loss function fluctuates sharply, so as to accelerate the search process and avoid premature convergence to a local minimum. In the training process, the oscillation period is adjusted dynamically according to the fluctuation of the loss function, and the oscillation amplitude is increased when the loss fluctuates sharply, which helps to jump out of the local optimal solution, especially in the optimization difficulty caused by uneven sample distribution in allergic related medical data. The adjustment mechanism of the oscillation period improves the global search ability of the model, so that it can better capture the global features of allergic related medical data, while avoiding premature convergence to a suboptimal solution, expressed as:

[0114]

[0115] wherein, is the loss function value of the neural network at the the oscillation period variation in the next iteration, the first oscillation variation adjustment parameter, the second oscillation variation adjustment parameter, the dynamic adjustment factor of the neural network in the first iteration, the loss function value of the neural network in the first iteration, the dynamic adjustment factor of the neural network, the initial weight corresponding to the connection between the current neuron and the first input, , .

[0116] Further, according to the oscillation period variation of the neural network, the oscillation period is updated, which is represented as:

[0117]

[0118] wherein, the oscillation period of the neural network in the first iteration, the oscillation period of the neural network in the first iteration, the oscillation period of the neural network in the first iteration.

[0119] 3) In the process of error back propagation, the local bias of the neural network is calculated, which considers both the gradient of the loss function and the coupling effect of the oscillation period and the bias of the adjacent layer, and more dynamic information is integrated into the parameter update to improve the adaptability of the neural network to nonlinear and deep structures. The local bias is calculated by combining the oscillation period and the gradient information, so that the parameter update contains more dynamic information, thereby improving the adaptability of the neural network to nonlinear features in the allergic related medical data. In particular, for the complex feature interaction relationship in the allergic related medical data, the learning ability of the network to deep feature patterns is enhanced. The calculation method of the local bias is represented as:

[0120]

[0121] wherein, the local bias of the neural network in the first iteration, the partial derivative of the loss function with respect to the current weight of the neuron in the first iteration, the weight of the neural network in the first iteration, the current neuron and the first input in the first The weights corresponding to each input connection This is the adjustment factor for bias propagation. The parameter is the correlation parameter between the neural network's bias and the sum of its initial weights. For neurons in the first The iteration and the The sum of all weights corresponding to each input connection. Preferably, set... .

[0122] 4) Updating the weights and biases, and integrating local biases, enables more flexible dynamic adjustment of neural network parameters, thereby enhancing the fitting ability to complex allergy-related medical data. By employing oscillation period influence and feedback adjustment factors in the updating of weights and biases, parameter optimization can adapt to the differences in the importance of different features in allergy data, ensuring the robustness of the neural network in complex medical data environments and improving the accuracy of feature extraction, especially the ability to identify subtle correlations between elusive allergens. This is expressed as:

[0123]

[0124] In the formula, For the neural network in the first Weights for the next iteration For the neural network in the first Weights for the next iteration The step size for updating the weights of the neural network. For the neural network in the first Local deviation of the next iteration For the neural network in the first The oscillation period of the next iteration As a feedback adjustment factor, For neurons in the first The iteration and the The sum of all weights corresponding to each input connection. This is the index for feedback adjustment. Preferably, Set to 0.2, Set to 0.5. Set to 2.

[0125] Furthermore, the bias update is also affected by the oscillation period, and the calculation method is expressed as follows:

[0126]

[0127] In the formula, For the neural network in the first The bias of the next iteration. For the neural network in the first The bias of the next iteration. is a step size for bias update, is an adjustment factor in bias update. Preferably, is set to 0.0001, is set to 0.05.

[0128] 5) After the update of weights and biases in each iteration, in order to enhance the search ability for global minimum, a random disturbance is added to the updated weights of the neural network, and the disturbance size changes with the training round to maintain a larger search range in the early stage and reduce fluctuations in the later stage. When dealing with allergic related medical data, it can alleviate the local extremum problem of feature space, improve the generalization performance of neural network, ensure that the extracted medical features are more representative, and the calculation method is represented as:

[0129]

[0130] In the formula, is a parameter update operation, is the disturbance amplitude of the neural network weight in the th iteration, is a random number uniformly distributed from -1 to 1.

[0131] Further, the disturbance amplitude gradually decreases with the training, and the calculation method is represented as:

[0132]

[0133] In the formula, is the initial disturbance amplitude, is the disturbance decay adjustment factor, is the current iteration number. Preferably, is set to 0.001, is set to 0.4.

[0134] 6) Determine whether the training of the neural network is converged, observe the change of the loss function after continuous multiple iterations, if the change of the loss function is less than a certain threshold within the preset iteration round, terminate the training, and the change of the loss function of the neural network is calculated as:

[0135]

[0136] In the formula, is the loss function change of the neuron in the th iteration, is the loss function value of the neural network in the th iteration, is the loss function value of the neural network in the th iteration.

[0137] If the loss value is lower than a preset threshold value in multiple rounds of iterations and the training of the neural network is determined to be converged, and the iteration is stopped. Preferably, the preset threshold value for stopping the iteration is 0.005.

[0138] After the training of the neural network is completed, the trained neural network is used to extract features from the allergic medical data. Further, the data after the feature extraction is input into a preset Softmax function to calculate class probabilities, and the class with the maximum class probability is taken as the class of the allergic risk warning, such as the classes including no allergic risk, low allergic risk, medium allergic risk, and high allergic risk, a total of 4 classes.

[0139] In an embodiment, the system performs risk warning and gives a treatment plan based on the risk assessment result.

[0140] In an embodiment, the risk warning module further includes one or more of the following: a data expansion module, a warning notification module, the data expansion module is used to input the data in the collection module into the data expansion module to obtain expanded data, and the expanded data is input into the risk assessment module to obtain a risk result; the warning notification module is used to perform warning notification based on the risk result of the risk assessment module.

[0141] In an embodiment, the system further includes a data storage module, which is used to store the data in the data collection module in layers, and the layered storage includes hot data storage and cold data storage. The hot data is real-time access data, and the cold data is long-term storage data. In an embodiment, the system further includes a user interaction module, which includes a user end and a doctor end. The user can view real-time health status and risk assessment results, obtain personalized health management suggestions or popular science content, record allergic attacks or allergic drug use, intelligently interpret allergic results, conduct daily detection for allergic relief, perform allergic risk warning, allergic acute attack preplan, and allergic child health management through the user end. The doctor can view the risk assessment result, develop a personalized health management plan, push a list of high-risk patients, construct a health record, perform intelligent screening, allergic precise assessment, follow-up management, and hierarchical diagnosis and treatment through the doctor end.

[0142] In an embodiment, the system further includes a security and privacy protection module, which is used to limit access permission through encryption.

[0143] In an embodiment, the system further includes a system integration and interface module, which is used to obtain real-time medical data of patients, real-time meteorological data, real-time air quality index, and real-time pollen concentration.

[0144] In one embodiment, the early identification module: for early identification screening of allergic disease related symptoms; early screening model training through child clinical data to obtain early screening model. Allergic asthma is a common asthma phenotype in children, accounting for more than 80% of all asthma phenotypes. There is a natural process of occurrence and development of asthma, and children often take atopic dermatitis (eczema) as the first disease, and more than 2 / 3 of atopic dermatitis patients develop allergic rhinitis, and 1 / 3 of atopic dermatitis patients develop asthma. The key to asthma management is four early: early detection, early diagnosis, early treatment, and early management. Therefore, based on the allergic disease history of children, early risk factors, lung function test, allergen test results, an early screening model of the risk of asthma in children is constructed, implanted in the system, and the doctor can make a decision according to the uploaded information of the child through the prediction model to give the child and the parents appropriate guidance and suggestions, so as to realize the "four early management" of allergic asthma.

[0145] Sensitive and slow monitoring, that is, home self-detection and record during the allergic remission period. Parents can log in through the patient end entrance of the system, record the daily use of drugs and acute effect of drugs for the child in the health center interface, and can also record the test results of the child every time, and evaluate the scores of asthma control evaluation scale, rhinitis symptom evaluation scale, skin evaluation scale, etc. every month. In addition, the lung function of the child can be dynamically monitored by a portable lung function instrument, and the wheezing sound can be monitored by a wearable device such as an electronic stethoscope, real-time collection of children's health information, timely grasp of the allergic control situation and possible risk factors. At the same time, the system will automatically remind the use of drugs and monitoring, and push related allergic health popular science knowledge, enhance the initiative of parents and children in drug use, and guide parents to conduct home self-management of allergic children.

[0146] Sensitive detection and intelligent reading module, that is, intelligent interpretation of allergen detection report. The realization of this function relies on the allergen intelligent interpretation report system implanted in the "sensitive child management" system, and parents can upload the allergen detection report in the "my allergen" module. The system can record the detection results of different time allergens, and intelligently interpret and analyze the allergen detection report, and present the corresponding allergen avoidance management guidance in real time, so that parents can know the interpretation result of the allergen in advance after diagnosis.

[0147] Sensitive and emergency plan module, that is, the treatment plan for responding to acute allergic events or allergic events with risk. This function is a sensitive risk warning system that can assess the risk of allergic attack that may occur in children, capture the symptoms of children such as wheezing, choking, and other respiratory symptoms, skin rash, urticaria, and other skin symptoms, vomiting, diarrhea, and other digestive symptoms, and severe allergic reactions involving multiple systems. When the symptoms and the degree of attack are different, personalized emergency treatment guidance suggestions are pushed.

[0148] The sensitive external health management module is used for health management of allergic children related to allergic and non-allergic diseases. Parents can search for relevant popular science knowledge about daily care, diet, vaccination, etc. of allergic children in the system, or leave a message in the background.

[0149] In one embodiment, the system further comprises a data preprocessing module, which includes a data calculation module (for data calculation), a data cleaning module (for data processing), and a data management module (for data structured query, addition, deletion, modification, and other data applications). After the data collection module collects data, the data is processed by the data preprocessing module and then sent to the risk assessment module; or after the data preprocessing, the data is expanded by the data expansion module and then sent to the risk assessment model; or the data is expanded first, then processed by the data preprocessing module, and then sent to the risk assessment model.

[0150] The system further comprises a data mining module, which mines data through integrity rules, label features, statistical analysis, and data portraits to obtain mined data. The data preprocessing module processes the data using the data mining module to obtain mined data, or the data expansion module expands the data and the original data to obtain mined data.

[0151] The system further comprises a data service module, which trains the mined data through a data model module to obtain a data model (risk assessment model).

[0152] The system further comprises a data monitoring module (for real-time monitoring of data changes), a scale pushing module (for pushing allergy-related scales), and a service API module.

[0153] In one embodiment, the functions of the data storage module include:

[0154] 1) Hierarchical data storage: data is divided into hot data and cold data according to the type and access frequency of the data:

[0155] 2) Hot data: real-time access data, such as the latest monitoring indicators of patients;

[0156] 3) Cold data: long-term saved historical data, such as allergic attack records and medical records.

[0157] In addition, the data storage module supports multiple data types, including:

[0158] 1) Structured data: such as patient basic information and allergen detection reports;

[0159] 2) Semi-structured data: such as symptom records and medication records;

[0160] 3) Unstructured data: such as uploaded pictures, PDF format detection reports.

[0161] In one specific embodiment, the early warning and notification module: according to the risk score, set the early warning level, and push the notification, the push channel includes: through the short message, WeChat applet, email and other ways to inform the patient or his family in real time.

[0162] In addition, for high-risk cases, the system can trigger a phone reminder or manual customer service intervention.

[0163] In one specific embodiment, the user interaction module: on the patient / family side, support to view real-time health status and risk assessment results, and support to manually enter symptoms, medication records and other information, supplement data collection, in addition, you can also get personalized health management advice and popular science content.

[0164] On the doctor side, support to browse the patient's risk assessment results and health trend chart, while supporting to receive the high-risk patient list pushed by the system, and to intervene, and to be able to make personalized health management plan for the patient.

[0165] In one specific embodiment, the security and privacy protection module: the transmission data adopts TLS encryption, and the storage and management ensure that only authorized personnel can access related data.

[0166] In one specific embodiment, the system integration and interface module: access HIS, LIS and other hospital information systems, and obtain real-time patient medical data, in addition, through API, real-time access to weather data, air quality index, pollen concentration and other environmental information.

[0167] The standardized interface provides RESTful API, which supports connection with third-party applications to realize data sharing.

[0168] In one specific embodiment, the handling scheme for acute allergic events that have occurred or have the risk of occurring. This function is for the allergic risk warning system to assess the risk of allergic attack or allergic reaction that may occur in children, according to the onset of different symptoms and the degree of onset, and push personalized emergency treatment guidance suggestions.

[0169] In another specific embodiment, the children's allergy management system is divided into basic resource layer, data resource layer, data management layer, system application layer and user layer, and each layer communicates through standardized interface to ensure the flexibility and scalability of the system. Figure 2The base resource layer provides the hardware and network infrastructure required for system operation, including network devices, computing devices, storage devices, and terminal devices. This layer supports the deployment and operation of Spring Boot microservices, ensuring high availability and elasticity of the system. In addition, the base resource layer integrates information security components such as web application firewalls, web tamper-proofing, and antivirus gateways to protect the system from external attacks. The data resource layer is mainly responsible for data generation, collection, and storage. This layer uses Postgres as the main database, supporting complex queries and transaction processing to ensure data consistency and integrity. At the same time, Redis is used as a cache database to accelerate the reading speed of frequently accessed data and improve the response efficiency of the system. The data resource layer also implements data interfaces with external systems to support real-time synchronization and batch import of data.

[0170] The data management layer is built on top of the data resource layer and is responsible for the integration, cleaning, transformation, and loading (ETL) of data such as physical examination data, questionnaire data, risk models, and indicator data. This layer uses the domain-driven design (DDD) method to encapsulate business logic in domain models, ensuring that the system's business rules are clear and easy to maintain. The data management layer also implements fine-grained data permission management based on the Spring Security framework to ensure the security and compliance of data access. Specifically, it includes data preprocessing modules, data mining modules, and data service modules; the data system application layer is directly oriented towards users and provides rich functional applications. This layer is built based on Spring Cloud, supporting service discovery, configuration management, circuit breakers, and load balancing, ensuring reliable communication between microservices and high availability of the system. The system application layer also implements DDD-based business logic such as "Internet +", artificial intelligence, and big data analysis, especially for the specialized features of children's allergic diseases, providing a closed-loop management solution. In addition, this layer implements user identity authentication and access authorization to ensure the security and privacy of user data.

[0171] The user layer provides an intuitive user interface and interactive experience, supporting multi-terminal access including PCs, smartphones, and tablets. This layer communicates with the system application layer through RESTful APIs, ensuring fast response to user requests and real-time updating of data. The design of the user layer focuses on user experience, providing a simple and easy-to-use interface and smooth operation process.

[0172] In one specific embodiment, the allergic disease of children is a chronic inflammatory disease, which requires a long-term, continuous, standardized treatment management process, including screening, monitoring, evaluation, diagnosis and treatment, intervention, follow-up and other steps. A closed-loop management system should be established to enhance the interaction between the hospital, society, family doctor, Internet, child and parent, so as to enhance the personalized treatment for the unique needs of each child and effectively manage the child throughout the process. By designing a whole-process effective management process and applying it to daily home detection and health guidance, the transformation of children's allergic diseases from 'disease diagnosis and treatment center' to 'health promotion center' and 'active health' is promoted, so as to achieve the goal of controlling disease, improving life quality and improving prognosis. Based on this, the present application develops an allergic child intelligent chronic disease management tool based on the WeChat platform, that is,'sensitive child management' (occurrence of allergic risk assessment system). The user interface and the doctor interface are as shown in Figure 3 、 Figure 4

[0173] Figure 5 A risk assessment method for occurrence of allergy is provided for the embodiments of the present application, comprising:

[0174] Obtaining the basic data of the testee, including physiological index data, environmental data and medical record data;

[0175] The basic data is input into the risk assessment model for evaluation to obtain a risk result, which includes one or more of the following: no allergic risk, low allergic risk, moderate allergic risk and high allergic risk;

[0176] The construction process of the risk assessment model is:

[0177] Obtaining the basic data set of the testee and the label of the risk category;

[0178] The data set and the label of the risk category are input into the neural network for training to obtain the risk assessment model;

[0179] The neural network is based on the oscillation period to initialize the parameters, and the initialization parameters are calculated by the product of the weight adjustment factor, the oscillation period, the sine term or the cosine term of the oscillation period, and the correlation parameters of the neural network.

[0180] In one embodiment, the oscillation period is adaptively adjusted to optimize the iteration process of the neural network in the construction process of the risk assessment model. When the loss of the loss function fluctuates sharply in the iteration process, the iteration is performed after adjusting the oscillation amplitude.

[0181] In one embodiment, the construction process of the risk assessment model adds random disturbance to optimize the parameter update range of the neural network after updating the iteration parameters, and the size of the random disturbance is dynamically adjusted by the number of iterations of the training.​

[0182] In an embodiment, the construction of the evaluation model further comprises data augmentation; the data set and the label of the risk category are subjected to data augmentation to obtain augmented data, and the augmented data is input into a neural network model to obtain a risk evaluation model; the data augmentation is performed by a generative adversarial network, a loss function of the generative adversarial network comprises an adversarial loss and a feature contrast loss, the calculation of the adversarial loss comprises random noise data, feature data of the allergic medical data, and the calculation of the feature contrast loss comprises the allergic medical data. The feature data of the allergic medical data comprises one or more of the following: statistical features, semantic features, image features, time series features, and specific medical indicator features; and the allergic medical data comprises one or more of the following: heart rate, blood pressure, air quality index, pollen concentration, allergic history, drug type, age, gender, temperature and humidity index, and family allergic history.

[0183] In an embodiment, the generative adversarial network further comprises a distribution adaptive evaluation mechanism, comprising a diversity loss and a quality evaluation loss, the quality evaluation loss is used to measure the quality difference between the generated allergic medical data and the real allergic medical data by comparing the mean and variance; and the diversity loss is calculated by a diversity loss factor and generated samples.

[0184] The present disclosure further provides a computer program product or system, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned risk assessment method for allergic reaction.

[0185] Figure 6 The risk assessment device for allergic reaction provided by the embodiments of the present disclosure specifically comprises:

[0186] a memory and a processor; the memory is used to store program instructions; and the processor is used to call the program instructions, so as to implement any one of the above-mentioned risk assessment methods for allergic reaction.

[0187] The present disclosure further provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements any one of the above-mentioned risk assessment methods for allergic reaction.

[0188] The verification result of the verification embodiment shows that assigning inherent weights to the indications can improve the performance of the method compared with the default setting. It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here. In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the units is only a logical function division. There can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms. The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme. In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can be a physically independent unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of software functional units. Those skilled in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, which can include read only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0189] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiment methods can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, and the above-mentioned medium storage can be read only memory, magnetic disk or optical disk, etc.

[0190] The computer device provided by the present application has been described in detail above. For those skilled in the art, according to the idea of the embodiment of the present application, there will be changes in specific implementation and application range. In view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A system for managing allergy in children, characterized by, The application relates to an early warning system for allergic diseases. The system comprises: an early warning module for acute allergic reaction risk warning; the early warning module comprises: a data acquisition module for acquiring basic data of a subject, including physiological index data, environmental data and medical record data; a risk assessment module for inputting the basic data into a risk assessment model for assessment to obtain a risk result; the construction process of the risk assessment model is as follows: acquiring a basic data set of a subject and a label of a risk category; inputting the data set and the label of the risk category into a neural network for training to obtain a risk assessment model; wherein, the neural network is based on an oscillation period for parameter initialization, and the initialization parameters are calculated through the product of a weight adjustment factor, the oscillation period, a sine term or a cosine term of the oscillation period and a correlation parameter of the neural network; in the construction process of the risk assessment model, the oscillation period is used for self-adaptive adjustment to optimize the iteration process of the neural network; when the loss of a loss function fluctuates sharply in the iteration process, the oscillation amplitude is adjusted, and then iteration is performed; in the process of error back propagation, the oscillation period and gradient information are combined to calculate the local deviation of the neural network, the local deviation considers the gradient of the loss function and the coupling effect of the oscillation period and the deviation of a neighboring layer, and dynamic information is integrated in parameter updating.

2. The management system for allergies in children according to claim 1, characterized in that, after the iteration parameter updating in the construction process of the risk assessment model, random disturbance is added to optimize the parameter updating range of the neural network, and the size of the random disturbance is dynamically adjusted through the iteration number of training.

3. The management system for allergies in children according to claim 1, characterized in that, the construction of the assessment model also comprises data expansion; after the data set and the label of the risk category are expanded, expanded data are obtained, the expanded data are input into a neural network model for training to obtain a risk assessment model; the data expansion is performed through a generative adversarial network, a loss function calculation of the generative adversarial network comprises an adversarial loss and a feature comparison loss, the calculation of the adversarial loss comprises random noise data, feature data of allergic medical data, and the calculation of the feature comparison loss comprises the allergic medical data.

4. The management system for allergies in children according to claim 3, characterized in that, the feature data of the allergic medical data comprises one or more of the following: statistical features, semantic features, image features, time series features and specific medical index features; the allergic medical data comprises one or more of the following: heart rate, blood pressure, air quality index, pollen concentration, allergic history, drug type, age, gender, temperature and humidity index and family allergic history.

5. The management system for allergies in children according to claim 3, characterized in that, the generative adversarial network further comprises a distribution self-adaptive evaluation mechanism, including a diversity loss and a quality evaluation loss, the quality evaluation loss is used for quality difference measurement by comparing the mean and variance between generated allergic medical data and real allergic medical data; the diversity loss is calculated through a diversity loss factor and generated samples.

6. The management system for allergies in children according to claim 1, characterized in that, the system further comprises one or more of the following: an early identification module for early identification screening of allergic disease related symptoms; an allergic remission monitoring module for detection data recording in an allergic remission period; an intelligent interpretation module for intelligent interpretation of allergen detection reports. The acute sensitive plan module is used to push the processing scheme of the acute allergic event or the allergic event at risk; The sensitive health management module is used for the health management of allergic children and non-allergic diseases.

7. A method for assessing the risk of developing an allergy, characterized in that It includes: Obtaining the basic data of the testee, including physiological index data, environmental data, medical record data; The basic data is input into the risk assessment model for evaluation to obtain a risk result, which includes one or more of the following: no allergic risk, low allergic risk, moderate allergic risk, high allergic risk; The construction process of the risk assessment model is: Obtaining the basic data set of the testee and the label of the risk category; The data set and the label of the risk category are input into the neural network for training to obtain the risk assessment model; Wherein, the neural network is based on the oscillation period to initialize the parameters, and the initialization parameters are calculated by the product of the weight adjustment factor, the oscillation period, the sine term or the cosine term of the oscillation period, and the correlation parameter of the neural network; The oscillation period is adaptively adjusted in the construction process of the risk assessment model to optimize the iteration process of the neural network, and the iteration is performed after adjusting the oscillation amplitude when the loss of the loss function fluctuates sharply in the iteration process; In the process of error back propagation, the oscillation period and the gradient information are combined to calculate the local deviation of the neural network, which considers the gradient of the loss function and the coupling effect of the oscillation period and the adjacent layer deviation, and integrates the dynamic information in the parameter update.

8. A computer device comprising a processor, a memory and a computer program or instructions stored on the memory, characterized in that, The computer program or instruction is executed by the processor to realize the risk assessment method of allergic reaction in claim 7.

9. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instruction is executed by the processor to realize the risk assessment method of allergic reaction in claim 7.

Images