Intelligent health management method and system based on large model

Abstract

The invention provides an intelligent health management method and system based on a large model, and relates to the technical field of data processing, and the method comprises the steps: collecting a plurality of medical features of a patient and corresponding feature values; preprocessing the medical features; performing importance division on the medical features through a deep learning feature contribution propagation algorithm; determining important features for prediction based on the division result; combining an LSTM model and a hidden Markov model to construct a health state prediction model; inputting the important features and the corresponding feature values into a health state prediction model, and outputting the health state of the patient; according to the health state, rehabilitation suggestions are provided for the patient. Potential health problems or disease risks can be predicted in advance, so that early intervention is realized, and the probability of disease occurrence is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an intelligent health management method and system based on a large model. BACKGROUND

[0002] A large model generally refers to a very complex and large neural network model, which is composed of a large number of parameters and can learn from a large amount of data and make inferences and predictions. Intelligent health management refers to the use of intelligent technology, especially based on artificial intelligence, machine learning, data analysis, etc., to optimize the health management process of patients. The intelligent health management method based on a large model refers to collecting the health data of patients (such as body temperature, heart rate, etc.), processing and predicting the health status of these data using a large model algorithm, and providing targeted health management or rehabilitation suggestions.

[0003] The intelligent health management method based on a large model not only optimizes the process of health prediction and management, but also provides more accurate health guidance for individuals, ultimately improving the quality and efficiency of medical services and promoting more personalized and efficient health management.

[0004] However, the existing intelligent health management method has high computational resource requirements in the training process, low model transparency, poor interpretability, and difficulty in providing clear decision support for doctors and patients. In addition, the model may be affected by data bias, resulting in inaccurate results or being unsuitable for certain specific groups. At the same time, the health management system usually ignores individual differences, making it difficult to fully adapt to the personalized needs of different cultures, lifestyles, and health habits. SUMMARY

[0005] In view of the above deficiencies of the prior art, the purpose of the embodiments of the present application is to provide an intelligent health management method based on a large model, which can solve the technical problems of the existing intelligent health management method, which has high computational resource requirements in the training process, low model transparency, poor interpretability, and difficulty in providing clear decision support for doctors and patients, resulting in inaccurate results or being unsuitable for certain specific groups, and the health management system usually ignores individual differences, making it difficult to fully adapt to the personalized needs of different cultures, lifestyles, and health habits.

[0006] The first aspect of the embodiments of the present application proposes an intelligent health management method based on a large model, comprising:

[0007] S1: collecting a plurality of medical characteristics and corresponding characteristic values of a patient;

[0008] S2: preprocessing the medical characteristics;

[0009] S3: Use deep learning feature contribution propagation algorithm to divide the importance of pre-processed medical features;

[0010] S4: Based on the segmentation results, determine the important features for prediction;

[0011] S5: Combine the LSTM model and the hidden Markov model to build a health status prediction model;

[0012] S6: Input the important features and corresponding feature values ​​into the health status prediction model and output the patient's health status;

[0013] S7: Provide rehabilitation advice to patients based on their health status.

[0014] A second aspect of an embodiment of the present invention provides a large model-based intelligent health management system, comprising: a processor and a memory;

[0015] The memory stores programs or instructions that can be run on the processor. When the programs or instructions are executed by the processor, the steps of the large model-based intelligent health management method of the first aspect are implemented.

[0016] According to a third aspect of an embodiment of the present invention, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the large model-based intelligent health management method according to the first aspect are implemented.

[0017] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0018] In an embodiment of the present invention, first, multiple medical features of a patient and their corresponding eigenvalues ​​are collected, and the importance of the medical features is divided through a deep learning feature contribution propagation algorithm. Based on the division results, the important features used for prediction are determined, ensuring that the model focuses on important information. Then, a health status prediction model is constructed by combining the LSTM model and the hidden Markov model, further improving the reliability of the health status prediction. Finally, the important features and their corresponding eigenvalues ​​are input into the health status prediction model, and the patient's health status is output. Based on the health status, rehabilitation advice is provided to the patient, thereby ensuring personalized, precise, and real-time health management, improving medical efficiency and the accuracy of health management, providing individuals with more accurate health guidance, and ultimately improving the quality and efficiency of medical services, and promoting more personalized and efficient health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are only for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments of the present invention. It is clear that those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0020] Figure 1 This is a flow chart of a large-model-based intelligent health management method provided by an embodiment of the present invention;

[0021] Figure 2 This is a structural diagram of a large-model-based intelligent health management system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work should fall within the scope of protection of the present invention.

[0023] The following describes in detail the large model-based intelligent health management method provided by the embodiment of the present invention through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0024] Reference Manual Figure 1 , which shows a flow chart of a large-model-based intelligent health management method provided by an embodiment of the present invention.

[0025] The embodiment of the present invention provides an intelligent health management method based on a large model, which may include the following steps:

[0026] S1: Collect multiple medical characteristics of patients and their corresponding characteristic values.

[0027] Medical characteristics refer to various physiological and clinical indicators used to describe a patient's health status, such as body temperature, blood pressure, heart rate, and blood sugar. These characteristics can reflect the patient's physiological state. Feature values ​​are the specific numerical values ​​corresponding to the medical characteristics, such as a body temperature of 37°C or a blood pressure of 120 / 80 mmHg.

[0028] It is important to note that by collecting multi-dimensional health data from patients, we ensure a comprehensive health management process. This data can include health indicators at different levels, helping to establish a comprehensive health profile and providing a key basis for subsequent analysis and prediction.

[0029] In one possible implementation, the medical features specifically include: body temperature, blood pressure, heart rate, body fat percentage, respiratory rate, liver function, blood routine, blood glucose level, and blood oxygen saturation.

[0030] Among them, the body temperature represents the temperature of the human body, usually around 37°C, reflecting whether the body has fever or inflammatory response.

[0031] Among them, the blood pressure represents the pressure of blood on the blood vessel wall, usually represented by two values of systolic pressure and diastolic pressure, affecting cardiovascular health.

[0032] Among them, the heart rate represents the number of heartbeats per minute, used to assess heart function.

[0033] Among them, the body fat percentage represents the proportion of fat in the body, which is an important indicator for assessing body health and obesity.

[0034] Among them, the respiratory rate represents the number of breaths per minute, used to assess the health of the respiratory system.

[0035] Among them, the liver function is evaluated by checking indicators such as ALT and AST to assess the health of the liver and detect whether the liver is working normally. Blood routine is measured by measuring various components in the blood (such as red blood cells, white blood cells and platelets) to assess overall health, especially the immune system and blood system. Blood glucose level is a measure of glucose concentration in the blood, an important indicator for diabetes detection. Blood oxygen saturation represents the oxygen content in the blood, and low oxygen saturation usually indicates respiratory problems.

[0036] It should be noted that by collecting multiple important medical features, covering from basic physiological indicators (such as body temperature, blood pressure, heart rate) to biochemical and functional tests (such as liver function, blood glucose level and blood routine, etc.), the overall health status of the patient can be comprehensively evaluated. By combining these features, the overall health of the patient can be monitored in multiple dimensions and comprehensively. This multi-feature integration method can improve the accuracy of prediction, help doctors better understand and manage the health status of patients, and timely detect potential health problems, so as to make more appropriate intervention and treatment.

[0037] S2: Preprocess the medical features.

[0038] Among them, preprocessing refers to cleaning, transforming and normalizing data before further analysis or modeling, and preprocessing aims to improve data quality to ensure that it can adapt to subsequent analysis models.

[0039] In one possible implementation, preprocessing specifically includes: standardization processing and missing value processing.

[0040] It is necessary to point out that through standardization and missing value processing, the quality and consistency of the data are improved, making it more suitable for subsequent analysis and model training. Standardization processing avoids the influence of dimensional differences between different features on model performance, ensuring that the model can fairly evaluate the contribution of all features. Missing value processing ensures data integrity and avoids model bias or inaccurate predictions caused by missing data.

[0041] S3: Through the deep learning feature contribution propagation algorithm, the importance of the preprocessed medical features is divided.

[0042] Among them, the deep learning feature contribution propagation algorithm is a deep learning-based algorithm for evaluating the contribution of different features to the model output. It calculates the influence of each feature in the model to identify which features are most important to the final prediction. This algorithm can calculate the contribution of features layer by layer through propagation mechanism and backpropagation technology, and finally give a contribution score for each feature.

[0043] It is necessary to point out that through the deep learning feature contribution propagation algorithm, the influence of each medical feature on the health prediction model can be accurately evaluated. This algorithm not only helps to determine which features are most critical to the prediction result, but also eliminates redundant information, optimizes data input, and improves the efficiency and accuracy of prediction.

[0044] In one possible implementation, S3 specifically includes:

[0045] S301: Set the mean of each medical feature as the reference input.

[0046] Among them, the reference input refers to the mean of each medical feature. The mean is used as a benchmark input value to calculate the difference between each feature and it, helping to evaluate the contribution of the feature.

[0047] S302: According to the difference between the input medical feature and the reference input, calculate the contribution score of each medical feature:

[0048]

[0049] Among them, x i represents the i-th input medical feature, i = 1, 2, …, n, n represents the total number of medical features, C △xi△d represents the contribution score of the i-th input medical feature x i to the output difference △d, △x i represents the difference between the i-th input medical feature x i and the reference input, △d represents the output difference between the input medical feature and the reference input, i.e. the contribution score of the medical feature.

[0050] The contribution score is a measure of the influence of a certain input feature on the prediction result. By comparing the input feature with the reference input, the contribution score of each feature is calculated.

[0051] S303: Define the multiplier according to the contribution score:

[0052]

[0053] Where m ΔxΔd represents the multiplier, and △x represents the difference between the input medical feature x and the reference input.

[0054] Where the multiplier is a weight calculated according to the contribution score, used to adjust the contribution of the input medical feature to the output.

[0055] S304: Determine the contribution of the change in the input medical feature to the change in the output by the multiplier chain rule:

[0056]

[0057] Where y j represents the jth neuron of the intermediate layer, and △y j represents the jth target neuron y j the difference between the actual input and the reference activation value.

[0058] Where the multiplier chain rule is a method for calculating the gradient, which can help to calculate the specific contribution of each feature to the change in the output by backpropagation.

[0059] S305: Calculate the target reference activation value of the reference input based on the contribution:

[0060]

[0061] Where y 0 represents the reference activation value of the target neuron output, and f(.) represents the activation function of the neural network, represents the activation value of the 1st input medical feature, represents the activation value of the 2nd input medical feature.

[0062] Where the target reference activation value is the output value of the reference input after the activation function, which serves as the basis for calculating the contribution of the medical feature.

[0063] S306: Determine the output change by comparing the target reference activation value and the target output.

[0064] Where the output change is the difference between the target output and the reference output, reflecting the degree of change in the model output.

[0065] S307: Decompose the output change into positive contribution and negative contribution:

[0066] Δy = Δy + + Δy -

[0067]

[0068] where Δy represents the difference between the target neuron y under actual input and the reference activation value, Δy + represents the positive contribution, i.e., the increasing contribution of the input medical feature to the output, Δy - represents the negative contribution, i.e., the decreasing contribution of the input medical feature to the output, C ΔyΔd represents the contribution score of the output change to the target output, represents the contribution score of the positive contribution to the target output, represents the contribution score of the negative contribution to the target output.

[0069] where the positive contribution refers to the positive impact of the input medical feature on the output result, i.e., the contribution to increasing the output value, and the negative contribution refers to the negative impact of the input medical feature on the output result, i.e., the contribution to decreasing the output value.

[0070] S308: Determine the important features by combining the positive contribution, negative contribution, and the ranking of the contribution scores of each medical feature.

[0071] It should be noted that by evaluating the contribution of each medical feature to the prediction of health status, more accurate health management can be achieved. First, using the mean as the reference input can provide a clear benchmark for each feature, helping to judge the degree of change in actual data. Through the contribution score, multiplier, and chain rule, the role of each feature in the model can be understood in depth, and the impact of the feature on the output result can be accurately evaluated. Decomposing the output change into positive and negative contributions can help identify which features play a positive role in prediction and which may cause prediction bias, thereby further optimizing the model. Finally, by ranking these contributions, the most important features for health prediction can be selected, ensuring that the model focuses on the most critical information, thereby improving the accuracy and interpretability of the prediction.

[0072] S4: Based on the division result, determine the important features for prediction.

[0073] S5: Combine the LSTM model and the hidden Markov model to construct a health status prediction model.

[0074] where LSTM is a recurrent neural network (RNN) variant in deep learning, specifically designed to address the difficulty of standard RNNs in capturing long-term dependencies in long sequences. LSTM introduces a gating mechanism (including forget gate, input gate, and output gate) to control the flow of information, enabling it to effectively handle and remember long-term dependencies in sequence data. Hidden Markov Model (HMM) is a statistical model used to describe the process of state transitions in a system, with each hidden state corresponding to an observation value. HMM assumes that the state of the system changes over time and is only related to the previous state, satisfying the Markov property. HMM is widely used in time series analysis, especially suitable for inferring hidden states that cannot be directly observed from the observed results. Health status prediction model is a prediction model established based on patient health data, aiming to predict the health status of patients, usually including normal health, sub-health, and disease states. This model makes health predictions based on multi-dimensional health characteristics (such as blood pressure, heart rate, etc.) and historical data, providing rehabilitation recommendations and warnings.

[0075] It should be noted that by combining LSTM model and Hidden Markov Model to build a health status prediction model, the characteristics of time series data can be effectively utilized for dynamic health prediction. The introduction of LSTM model can handle the complex changes of patient health data over time, capture long-term dependencies, and ensure that the influence of historical data on the current health status is preserved. The Hidden Markov Model can further improve the accuracy of the prediction by modeling the hidden state transitions of the health status, especially when facing multiple possible health status transitions, it can more accurately evaluate the transition probabilities between different states.

[0076] S6: input the important features and corresponding feature values into the health status prediction model, and output the health status of the patient.

[0077] where the health status refers to the classification or prediction of the patient's health condition by the model after receiving the patient's health characteristics, such as whether the patient is in a normal health, sub-health, or disease state.

[0078] It should be noted that by inputting important features and their corresponding values into the health status prediction model, these key features can be directly used to make accurate health predictions. This process ensures that the model focuses on the most relevant information when evaluating health status, avoiding the interference of redundant data, and improving the efficiency and accuracy of the prediction. The selection of important features ensures that the model is not affected by irrelevant data during the health prediction process, thereby improving the reliability of the prediction results. In addition, through real-time prediction of health status, the system can respond quickly when the patient's health condition changes, providing support for personalized health management and rehabilitation recommendations.

[0079] In a possible implementation, S6 specifically includes:

[0080] S601: Output the hidden state of each important feature at the current time step through the LSTM model:

[0081] f t = σ(W f x t + U f h t-1 + b f )

[0082] i t = σ(W i x t + U i h t-1 + b i )

[0083] g t = tanh(W g x t + U g h t-1 + b g )

[0084] C t = f t * C t-1 + i t * g t

[0085] o t = σ(W o x t + U o h t-1 + b o )

[0086] h t = o t tanh(C t )

[0087] wherein f t represents the activation output vector of the forgetting gate at t, σ represents the sigmoid activation function, W f represents the weight matrix of the forgetting gate related to the current input, x t represents the input medical feature at t, h t-1 represents the hidden state at t-1, U f represents the weight matrix of the forgetting gate related to the hidden state at the previous moment, b f represents the bias term of the forgetting gate, i t represents the activation output vector of the input gate at t, W i represents the weight matrix corresponding to the input gate, and Ui represents the weight matrix of the input gate related to the hidden state at the previous moment, b i Represents the bias term of the input gate, g t represents the activation value of the candidate memory unit at time t, tanh represents the hyperbolic tangent function, W g Represents the weight matrix of candidate memory units related to the current input, C t-1 Indicates the unit state at time t-1, U g Represents the weight matrix of the unit state candidate related to the hidden state at the previous moment, b g The bias term representing the candidate cell state, C t Indicates the updated unit state at time t, o t represents the activation output vector of the output gate at time t, W o represents the weight matrix of the output gate associated with the current input, b o Represents the bias term of the output gate, h t represents the hidden state at time t, U o Represents the weight matrix of the output gate related to the hidden state at the previous moment.

[0088] Among them, the hidden state is the output of the LSTM model, which contains useful information about the current time step and passes it to the next time step to form a long-term memory of the time series data. Each time step has a hidden state.

[0089] S602: Combine the hidden states at each moment into an observation sequence by combining the adaptive weights related to time.

[0090] Among them, adaptive weights assign different weights to the hidden state of each time step so as to better reflect the relative importance of different time steps when combined.

[0091] It should be noted that adaptive weights can be dynamically adjusted based on time changes and biological rhythms (such as sleep cycles, daily activity patterns, etc.), ensuring that the model pays different attention to specific features in different time periods. This dynamic adjustment can more accurately reflect changes in a patient's health, especially in cases where health status is significantly affected by time (such as circadian rhythms, body temperature changes, etc.). The adaptive weight mechanism enables the model to dynamically adjust its focus on health data, providing more accurate, personalized, and flexible health status predictions.

[0092] The specific calculation formula of the adaptive weight is:

[0093] w t =θ·S(t)+(1-θ)·M(t)

[0094]

[0095] where w t denotes the weight of the hidden state at time t, denotes the weight adjustment coefficient, S(t) denotes the biological rhythm adjustment term at time t, M(t) denotes the mutation reinforcement term at time t, cos denotes the cosine function, t peak denotes the time of the peak value of the key physiological indicator, T cycle denotes the biological cycle, denotes the basic weight, t denotes the time index, denotes the circular constant, max denotes the maximum value, h t-1 denotes the hidden state at time t-1, ||.||2 denotes the Euclidean distance, h k denotes the hidden state at time k, h k-1 denotes the hidden state at time t-1, T denotes the total number of times.

[0096] It should be noted that the adaptive weight mechanism improves the flexibility of the model, which can adjust itself according to different input data characteristics and patient states, not only enhancing the adaptability of the model, but also providing a more reliable basis for subsequent health prediction.

[0097] S603: input the observation sequence into the hidden Markov model, and output the health state.

[0098] It should be noted that by combining the LSTM model and the hidden Markov model, the time series health data of the patient is processed in multiple steps to provide accurate health state prediction. LSTM can capture long-term dependencies in time series and process the influence of historical health data on the current health state, while dynamically adjusting memory information through forget gate, input gate and output gate. This allows the model to model the complex patterns of changes in the patient's health state over time. Combined with time-dependent adaptive weights, the hidden state can be weighted and adjusted according to the patient's biological rhythm, mutation events and other factors, thereby improving the accuracy and personalization of the prediction. Finally, by combining these hidden states into an observation sequence and inputting them into the hidden Markov model, the system can output the most likely health state. This combination method can provide more accurate health prediction and timely detection of potential health risks, thereby providing more targeted rehabilitation recommendations and personalized health management solutions for patients.

[0099] In one possible implementation, S603 specifically includes:

[0100] S6031: initialize the state transition matrix, emission matrix and initial state probability of the hidden Markov model.

[0101] The state transition matrix describes the probability of the system transitioning between different hidden states. Each row represents the probability of transitioning from one state to all other states. This matrix is the core part of the hidden Markov model and determines the evolution of the state. The emission matrix represents the probability of observing a particular observation value under a certain hidden state. It is used to calculate the likelihood of the system outputting an observation value given the state. The initial state probability represents the probability of the system being in a certain hidden state at the beginning. This is an important parameter when initializing the hidden Markov model and defines the state distribution at the initial time of the system.

[0102] S6032: Set the total probability of the observation sequence to 1.

[0103] S6033: Determine the forward probability through the initialized hidden Markov model:

[0104]

[0105] wherein α t (i) represents the forward probability of the system being in state s i at time t, α t-1 (j) represents the forward probability of the system being in state s j at time t-1, j = 1, 2, …, N, N represents the total number of states, a ji represents the transition probability from state s j to state s i , b i (O t ) represents the emission probability of observing observation value O i under state s t , i.e., the emission probability of the hidden state.

[0106] wherein the forward probability refers to the probability of the system being in a certain hidden state at a certain time given the observation sequence.

[0107] S6034: Determine the total likelihood value of the current time step through the forward algorithm according to the forward probability under the condition that the total probability of the observation sequence is set to 1:

[0108]

[0109] wherein P(O|λ) represents the total probability of the observation sequence given the observation sequence O and parameters λ, α T (i) represents the forward probability of the system being in state s i at time T, i = 1, 2, …, N, N represents the total number of states.

[0110] where the forward algorithm helps to compute the total probability of a given observation sequence by recursively computing the forward probabilities at each time step. The likelihood value refers to the probability of the observation sequence given the parameters of the hidden Markov model (such as the state transition matrix, emission matrix, and initial state probabilities). It is a measure of how well the model fits the observed data.

[0111] S6035: Define the computational formula of the maximum probability path according to the total likelihood value:

[0112] δ t (i) = max P(q1, q2,..., q t | O1, O2,..., OT i | λ) t

[0113] where δ t (i) represents the maximum probability path of the system being in state s i at time t, q t represents the hidden state at time t, and λ represents the parameter set of the hidden Markov model.

[0114] where the maximum probability path refers to the most likely sequence of hidden states given the observation sequence, calculated by the hidden Markov model. This path is usually calculated through recursive and backtracking algorithms.

[0115] S6036: Update the optimal path probability based on the computational formula through recursive method:

[0116] δ t (i) = max j [δ t-1 (j)a ji ]b i (O t )

[0117] where δ t-1 (j) represents the probability of the maximum probability path of the system being in state s j at time t-1.

[0118] S6037: Generate the optimal hidden state sequence through backtracking operation according to the optimal path probability:

[0119] q T = argmax j [δ T (j)]

[0120] where q T represents the state of the most likely state sequence at time T, argmax represents taking the maximum value, and δ T (j) represents the system being in state s​j the probability of the most probable path.

[0121] wherein the backtracking operation refers to tracing back along the path according to the optimal path probability obtained by recursion, and finally obtaining the most likely hidden state sequence. The hidden state sequence refers to the most likely state sequence obtained by the backtracking operation, which represents the possible hidden state of the system at each time step. In health management, these hidden states can represent different health states.

[0122] S6038: output the health state of the patient according to the optimal hidden state sequence.

[0123] It should be noted that through the multi-step calculation process of the hidden Markov model, the most likely health state sequence of the patient can be accurately inferred from the observation sequence. This process starts from initializing the state transition matrix and emission matrix of the model, ensuring that the model can make reliable state transitions between different time steps, and combining the forward algorithm to calculate the forward probability to evaluate the overall probability of the observation sequence. Through the recursive calculation of the maximum probability path and the backtracking operation, the system can accurately identify the most likely health state change path, and thus obtain the prediction result of the health state. This method is efficient in processing multi-step time series and health data with uncertainty, and can capture the dynamic changes of health state, discover potential health problems in time and provide accurate predictions.

[0124] S7: provide rehabilitation recommendations for the patient according to the health state.

[0125] wherein the rehabilitation recommendations are personalized health management or treatment plans provided based on the health state prediction results, aiming at the specific health conditions of the patient. Rehabilitation recommendations may include lifestyle adjustments, dietary recommendations, exercise programs, drug treatment, etc., aiming to improve the health state of the patient or prevent health deterioration.

[0126] It should be noted that by providing personalized rehabilitation recommendations for patients according to the health state prediction results, it ensures that each patient can obtain a targeted solution that meets their specific health conditions. This method combines health prediction with actual intervention, so that health management not only relies on the analysis results of data, but also can make precise intervention according to the unique needs of each patient. The individualization and real-time nature of rehabilitation recommendations enhance the patient's health management experience, enabling patients to obtain specific action guidelines for improving health in a timely manner, thereby improving the rehabilitation effect. At the same time, this health state-based recommendation generation method helps to prevent the occurrence or further deterioration of diseases, reducing the waste of medical resources and optimizing the rehabilitation process of patients. Ultimately, this method makes health management more efficient, scientific and targeted, helping to improve the overall health level and quality of life of patients.

[0127] In one possible implementation, the health status specifically includes: normal health, sub-health, and disease state.

[0128] Normal health refers to the individual being in good condition in terms of physiology, psychology, and social adaptation, meeting the medical standard of normal health, and generally having no significant disease symptoms or health problems. Sub-health refers to the individual having no obvious disease manifestations in physiology, but having certain health discomforts or functional disorders, such as fatigue, low immunity, susceptibility to cold, etc. Sub-health state is usually a precursor to disease, a state between health and disease. Disease state refers to the individual already having a certain disease, with obvious abnormalities in health status, usually showing certain symptoms or signs, and needing medical intervention and treatment.

[0129] It should be noted that by defining "normal health", "sub-health" and "disease state", a clear classification standard is provided for health management, which helps to more accurately assess and monitor the health status of individuals. By subdividing the health status into these three levels, the health risk of patients can be more flexibly judged, and appropriate health management strategies can be developed. The identification of sub-health state is particularly important, as it can serve as an early warning of disease, allowing potential health problems to be detected in a timely manner, and effective preventive measures to be taken to prevent the disease from worsening. Through this classification of health status, not only can patients understand their own health status, but also doctors or health management systems can be provided with clear intervention points to ensure that patients receive personalized and scientific health advice. This classification method improves the precision of health management, making prediction and intervention more targeted and effective.

[0130] In one possible implementation, the LSTM model is optimized by using the Zebra Optimization Algorithm.

[0131] The optimization specifically includes:

[0132] Initialize the population and set the initial parameters of the Zebra Optimization Algorithm, wherein the initial parameters include: population size, maximum number of iterations, and search range. Initialize the positions of all zebras, wherein the position of a zebra represents the parameters of the LSTM model:

[0133]

[0134] wherein z i+1 represents the position of the zebra at the i+1th iteration, z i represents the position of the zebra at the ith iteration, r represents a random number for enhancing diversity, N represents the population size, and z represents the position of the zebra.

[0135] Take the accuracy of the LSTM model as the fitness function to calculate the fitness value of each zebra.

[0136] According to the fitness value, the positions of the zebras are updated in combination with a Levy flight mechanism and an adaptive weight strategy:

[0137]

[0138] wherein, represents the new position of the ith zebra in the jth dimension after passing through the general stage P, w(t) represents the weight coefficient at the tth iteration, x ij represents the position of the ith zebra in the jth dimension, PZ i represents that the ith zebra is a pioneer zebra, I represents a coefficient indicating a guiding direction, represents the new position of the ith zebra in the jth dimension after passing through the foraging stage P1, x min represents the lower bound of the jth dimension variable, x max represents the lower bound of the jth dimension variable, s represents a Levy step length, e represents a natural constant, T represents a maximum number of iterations, u and v both represent random variables obeying a normal distribution, and β represents a Levy distribution index.

[0139] To prevent falling into a local optimum, the positions of the zebras are further updated by fusing a sine-cosine algorithm and a variable spiral search algorithm:

[0140]

[0141] wherein, S1 represents an escape strategy, S2 represents an attack strategy, represents the new position of the ith zebra in the jth dimension after passing through the defense stage P2, A represents a spiral coefficient, c1 represents a sine-cosine amplitude adjustment coefficient, c2 and c3 both represent random numbers, R represents a perturbation control constant, sin represents a sine function, and cos represents a cosine function.

[0142] The fitness values of the updated zebras are calculated.

[0143] When the fitness value of the updated zebra is greater than or equal to the fitness value of the current zebra, the position of the current zebra is updated. When the fitness value of the updated zebra is less than the fitness value of the current zebra, the position of the current zebra is kept unchanged.

[0144] The above steps are repeated until a maximum number of iterations is reached, and the optimal parameters of the optimized LSTM model are obtained, and the optimization of the LSTM model is completed.

[0145] It should be noted that the zebra optimization algorithm provides an efficient and globally searching method for the optimization of the LSTM model, which can improve the performance and stability of the model and avoid overfitting, and is particularly suitable for problems with a large parameter space and strong nonlinear relationship.

[0146] The technical scheme provided by the embodiment of the present application brings at least the following beneficial effects:

[0147] In the embodiment of the present application, first, a plurality of medical characteristics and corresponding characteristic values of a patient are collected, and a deep learning feature contribution propagation algorithm is used to divide the importance of the medical characteristics, and based on the division result, important features for prediction are determined, ensuring that the model pays attention to important information. Then, an LSTM model and a hidden Markov model are combined to construct a health state prediction model, further improving the reliability of health state prediction. Finally, the important features and corresponding characteristic values are input into the health state prediction model, and the health state of the patient is output, and rehabilitation suggestions are provided for the patient according to the health state, thereby ensuring personalized, precise and real-time health management, improving medical efficiency and health management accuracy, providing more accurate health guidance for individuals, ultimately improving medical service quality and efficiency, and promoting more personalized and efficient health management.

[0148] Referring to the accompanying drawings Figure 2 , a structure schematic diagram of an intelligent health management system based on a large model provided by an embodiment of the present application is shown.

[0149] The embodiment of the present application provides an intelligent health management system 20 based on a large model, which comprises a processor 201 and a memory 202.

[0150] The memory 202 stores programs or instructions executable on the processor 201, and the programs or instructions are executed by the processor 201 to realize the steps of the intelligent health management method based on a large model described above, and the same technical effects can be achieved. To avoid repetition, the present application will not be described again.

[0151] It should be understood that the processor 201 in the embodiment of the present application can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0152] It should also be understood that the memory 202 in embodiments of the present application can be volatile or nonvolatile memory, or can include both volatile and nonvolatile memory. Nonvolatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically EPROM (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), used as external cache. By way of example, and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DR RAM).

[0153] The above-described embodiments can be implemented in part or in whole through software, hardware (e.g., circuitry), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs can generate the flow or function according to the embodiments of the present application in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, such as from one website site, computer, server, or data center to another website site, computer, server, or data center through a wired (e.g., infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing a set of one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0154] It should be understood that the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0155] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0156] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, apparatuses, and units can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0157] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other manners. For example, the embodiments of the apparatus described above are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0158] The units described as separated components can or can not be physically separated, 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 purposes of the embodiments of the present application.

[0159] In addition, each functional unit in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated into one unit.

[0160] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts of the technical solutions that make contributions to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0161] The embodiments of the present application provide a readable storage medium, which includes: a program or instructions stored on the readable storage medium, the program or instructions are executed by a processor to implement the steps of the intelligent health management method based on a large model described above, and the same technical effects can be achieved. To avoid repetition, the present application will not be described again.

[0162] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present application, but not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application.

Claims

1. An intelligent health management method based on a large model, characterized in that: include: S1: Collect multiple medical characteristics of patients and their corresponding characteristic values; S2: preprocessing the medical features; S3: Use deep learning feature contribution propagation algorithm to divide the importance of pre-processed medical features; S4: Based on the segmentation results, determine the important features for prediction; S5: Combine the LSTM model and the hidden Markov model to build a health status prediction model; S6: Inputting the important features and corresponding feature values ​​into the health status prediction model, and outputting the health status of the patient; S7: Providing rehabilitation advice to the patient based on the health status.

2. The intelligent health management method based on a large model according to claim 1 is characterized in that: The medical characteristics specifically include: body temperature, blood pressure, heart rate, body fat percentage, respiratory rate, liver function, blood routine, blood sugar level and blood oxygen saturation.

3. The intelligent health management method based on a large model according to claim 1 is characterized in that: The preprocessing specifically includes: standardization processing and missing value processing.

4. The intelligent health management method based on a large model according to claim 1 is characterized in that: The S3 specifically includes: S301: Setting the mean of each of the medical features as a reference input; S302: Calculating a contribution score of each medical feature according to the difference between the input medical feature and the reference input; S303: defining a multiplier according to the contribution score; S304: Determine the contribution of the change in the input medical characteristics to the change in the output using the multiplier chain rule; S305: Calculating a target reference activation value of the reference input based on the contribution influence; S306: Determine the output change by comparing the target reference activation value and the target output; S307: Decomposing the output variation into positive contribution and negative contribution; S308: Determine the important feature by combining the positive contribution, the negative contribution, and the ranking of the contribution scores of the respective medical features.

5. The intelligent health management method based on a large model according to claim 1 is characterized in that: The S6 specifically includes: S601: Outputting the hidden state of each of the important features at the current time step through the LSTM model; S602: combining the hidden states into an observation sequence in combination with a time-related adaptive weight; S603: Input the observation sequence into the hidden Markov model, and output the health status of the patient.

6. The large model-based intelligent health management method according to claim 1, characterized in that: The S603 specifically includes: S6031: Initialize the state transfer matrix, emission matrix and initial state probability of the hidden Markov model; S6032: setting the total probability of the observation sequence to 1; S6033: Determine the forward probability through the initialized hidden Markov model; S6034: When the total probability of the observation sequence is 1, determine the total likelihood value of the current time step according to the forward probability using a forward algorithm; S6035: Define a calculation formula for the maximum probability path based on the total likelihood value; S6036: Based on the calculation formula, recursively update the optimal path probability; S6037: Generate an optimal hidden state sequence through backtracking operation according to the optimal path probability; S6038: Output the patient's health status based on the optimal hidden state sequence.

7. The large model-based intelligent health management method according to claim 1, characterized in that: The health status specifically includes: normal health, sub-health and disease status.

8. The large model-based intelligent health management method according to claim 1, characterized in that: Also includes: The LSTM model is optimized using the Zebra optimization algorithm.

9. An intelligent health management system based on a large model, characterized in that: include: processor and memory; The memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the large model-based intelligent health management method as described in any one of claims 1 to 8 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the large model-based intelligent health management method as described in any one of claims 1 to 8 are implemented.

Images