An ai-based large customer nutrition case rapid assessment system and method

Abstract

The application relates to the technical field of medical information, and discloses an AI-based rapid evaluation system and method for large customer nutrition cases, wherein the AI-based rapid evaluation method for large customer nutrition cases comprises the following steps: processing population statistical characteristic data by using a meta-learning network to generate a sampling basic model for a specific population; constructing a Bayesian decision network to calculate nutrition index information value and generate an index information cost ratio; constructing a sequential sampling strategy by using a Markov decision process according to the index information cost ratio, and performing adaptive index sampling; in the process of performing adaptive index sampling, calculating information gain rate to realize a dynamic termination decision and avoid over-sampling; and applying conditional probability inference to generate a complete nutrition evaluation result, so that a high-precision evaluation result can be obtained by measuring only a small number of indexes; and the application solves the resource limitation problem in the evaluation of large customer nutrition cases by using information value theory and meta-learning technology.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and more specifically, to an AI-based rapid assessment system and method for nutritional cases of large clients. Background Technology

[0002] In the nutrition management of large clients such as corporations and schools, medical institutions need to efficiently assess the nutritional status of the population to provide a scientific basis for health management. Traditional nutrition assessment methods require completing all tests for each member, usually including dozens of indicators, resulting in high testing costs, long processing times, and limited resources.

[0003] Existing nutritional assessment technologies suffer from the following main problems: they cannot achieve high-precision nutritional status assessment with limited detection indicators, especially lacking adaptability to individual differences among different populations; existing sampling methods cannot accurately quantify the balance between the value of indicator information and sampling costs, leading to unreasonable resource allocation; existing sparse sampling methods lack adaptability to group differences and cannot quickly adjust sampling strategies according to the characteristics of different populations; and in emergency situations, there is a lack of scientific decision-making methods to guide resource allocation, making it impossible to maximize the value of limited resources.

[0004] Therefore, there is a need for a rapid assessment method for nutritional cases of large clients that can overcome the above-mentioned technical problems and achieve efficient and accurate nutritional status assessment under limited resources. Summary of the Invention

[0005] This invention provides an AI-based rapid assessment system and method for nutritional cases of large clients, solving the technical problem in related technologies that cannot achieve high-precision nutritional assessment under limited resource conditions.

[0006] This invention provides an AI-based method for rapid assessment of nutritional cases of large clients, comprising the following steps:

[0007] Meta-learning networks are used to process population statistical feature data to generate a basic sampling model for a specific population.

[0008] Based on the sampling model, a Bayesian decision network is constructed to calculate the value of nutritional indicator information and generate the indicator information cost ratio. The information cost ratio is defined as the ratio of the expected information gain to the product of economic cost and time cost.

[0009] Based on the cost ratio of indicator information, a sequential sampling strategy is constructed using a Markov decision process to perform adaptive indicator sampling.

[0010] During the adaptive index sampling process, the information gain ratio is calculated to achieve dynamic termination decision and avoid oversampling.

[0011] Based on the sampling results after dynamic termination decision, conditional probability inference is applied to generate complete nutritional assessment results and provide an estimate of the reliability of the results.

[0012] In a preferred embodiment, the step of processing population statistical feature data using a meta-learning network includes:

[0013] Collect statistical characteristic data of the target population, including demographic characteristics, occupational characteristics, regional characteristics, historical nutritional status and their multidimensional data;

[0014] A meta-learning network is constructed, which adopts a two-layer structure. The outer network is responsible for learning the mapping function from population features to the parameters of the inner network, and the inner network is responsible for implementing the specific sampling strategy.

[0015] The meta-learning network is trained using a task-based meta-learning paradigm;

[0016] For new populations, the initial parameters of the sampling strategy for the new population are calculated using a pre-trained meta-learning network, and a sampling base model suitable for the new population is generated.

[0017] In a preferred embodiment, the step of constructing a Bayesian decision network to calculate the value of nutritional indicator information includes:

[0018] Construct a Bayesian network model to represent the probabilistic dependencies among nutritional indicators;

[0019] Learn the structure and parameters of Bayesian networks based on historical data and expert knowledge;

[0020] Calculate the expected information gain for each potential sampling indicator to quantify the information contribution of the expected information gain to the nutritional status assessment.

[0021] The information cost ratio is introduced to comprehensively consider the information value, economic cost, and time cost of the indicators.

[0022] In a preferred embodiment, the step of constructing a sequential sampling strategy using a Markov decision process includes:

[0023] A Markov decision process model is constructed to formalize the sequential sampling problem into a state transition process;

[0024] An improved Markov blanket information propagation algorithm is implemented to effectively utilize the conditional dependencies between indicators to predict the probability distribution of unmeasured indicators.

[0025] Based on the current state and information cost ratio, select the next optimal measurement indicator;

[0026] Perform measurements and update the status.

[0027] In a preferred embodiment, the Markov decision process model includes:

[0028] The state space represents the state of the currently measured set of indicators and their observations.

[0029] Action space represents the set of metrics that can be measured in the next step;

[0030] The state transition function represents the probability of transitioning to the next state after selecting a measurement index in one state;

[0031] The reward function, which reflects the net benefit brought by the measurement indicator, is defined as the information gain of the measurement indicator minus the measurement cost.

[0032] In a preferred embodiment, the step of calculating the information gain ratio to implement dynamic termination decision includes:

[0033] Define the information gain ratio to measure the marginal information gain brought about by each new measurement indicator;

[0034] Implement a threshold-based dynamic termination algorithm, setting two termination conditions: an information gain rate threshold condition and an evaluation accuracy condition. Sampling will stop when either condition is met.

[0035] Construct an assessment accuracy estimation model to predict the accuracy of the assessment results in real time under the current state;

[0036] The termination threshold is adaptively adjusted, and the termination condition parameters are dynamically set according to the urgency of the task and resource constraints.

[0037] In a preferred embodiment, the step of generating a complete nutritional assessment result using conditional probability inference includes:

[0038] The conditional probability distribution of unmeasured indicators is inferred using a probabilistic graphical model, and the complete distribution of nutritional indicators is inferred based on the state at the time when sampling has stopped.

[0039] By combining conditional probability distributions and expert rules, nutritional status assessment results are generated.

[0040] Calculate the reliability estimate of the evaluation results and quantify the uncertainty of the results;

[0041] Personalized nutrition recommendations are generated based on the assessment results and reliability estimates.

[0042] In a preferred embodiment, the reliability estimate of the evaluation result is calculated using the following formula:

[0043] ;

[0044] in, This represents a reliability estimate of the assessment results; Indicates the state Next target nutrient state variables The normalized conditional entropy is used to quantify the uncertainty of the nutrient status assessment results under the current sampling state; This indicates the state at the time of termination of sampling, including all measured indicators and their observations;

[0045] Normalized conditional entropy scales the original conditional entropy to... Within the interval, this makes reliability estimates more comparable;

[0046] The range of values ​​is A higher value indicates higher reliability, meaning lower uncertainty in the evaluation results.

[0047] In a preferred embodiment, the desired information gain is calculated using the following formula:

[0048] ;

[0049] in, Indicators Expected information gain; This represents the target nutritional state variable, which is the final nutritional status that needs to be assessed. Indicates the first One possible measurable nutritional indicator; express The entropy is used to measure the uncertainty of the target variable; Indicates that in the known under conditions The conditional entropy represents the condition of observing the index. Post-target variable Remaining uncertainty; Indicates about The mathematical expectation, that is, for A weighted average of all possible values; Indicates through measurement indicators The uncertainty that can be reduced is the amount of information obtained.

[0050] In a preferred embodiment, an AI-based rapid assessment system for large customer nutrition cases is used to execute an AI-based rapid assessment method for large customer nutrition cases, including:

[0051] The population feature processing module is used to process population statistical feature data using meta-learning networks to generate a sampling base model for a specific population.

[0052] The information value assessment module is used to construct a Bayesian decision network to calculate the value of nutritional indicator information and generate the indicator information cost ratio.

[0053] The adaptive sampling module is used to construct a sequential sampling strategy using a Markov decision process and execute adaptive index sampling.

[0054] The dynamic termination decision module is used to calculate the information gain rate to realize the dynamic termination decision of the sampling process;

[0055] The assessment results generation module is used to generate complete nutritional assessment results by applying conditional probability inference and to provide an estimate of the reliability of the results.

[0056] The beneficial effects of this invention are as follows:

[0057] By using a dynamic termination decision-making mechanism, optimal resource allocation is achieved, reducing the assessment and handling time in emergency situations from hours to minutes.

[0058] Meta-learning-based sampling strategies can automatically adapt to the characteristics of different population groups, reducing the data requirements for adapting new population groups.

[0059] By combining conditional probability inference and reliability estimation, it can provide highly accurate assessment results and reliable quantifications even under finite sampling conditions, providing comprehensive support for medical decision-making. Attached Figure Description

[0060] Figure 1 This is a flowchart of an AI-based rapid assessment method for nutritional cases of large clients according to the present invention. Detailed Implementation

[0061] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0062] At least one embodiment of the present invention discloses an AI-based method for rapid assessment of nutritional cases of large clients, such as... Figure 1 As shown, it includes the following steps:

[0063] Step 1: Use a meta-learning network to process population statistical feature data and generate a sampling base model for a specific population.

[0064] Specifically, the following steps are included:

[0065] Step 1.1: Collect statistical characteristic data of the target population, including demographic characteristics (such as age distribution and gender ratio), occupational characteristics (such as industry type and job nature), regional characteristics, historical nutritional status and other multidimensional data.

[0066] Step 1.2, construct the meta-learning network;

[0067] Meta-learning networks can learn the mapping relationship between the characteristics of different population groups and the optimal sampling strategy from historical data.

[0068] Meta-learning networks employ a two-layer structure:

[0069] The outer network is responsible for learning the mapping function from crowd features to the parameters of the inner network. ,in It is a meta-learning mapping function; It represents a population feature vector, which includes multidimensional demographic data such as age distribution, gender ratio, occupational characteristics, and regional characteristics; These represent the initial parameters of the inner network, which determine the weights and biases of the sampling strategy and are used to guide the subsequent sampling process.

[0070] The inner network is responsible for implementing specific sampling strategies, generating sampling strategy parameters for a specific population based on population characteristics, including parameters such as sampling priority, sampling order and termination conditions for each indicator. These parameters are adjusted individually according to the characteristics of different populations.

[0071] Step 1.3: Train the meta-learning network;

[0072] The task-based meta-learning paradigm is adopted, and its objective function is:

[0073] ;

[0074] in, These are the outer network parameters, which control how the meta-learning network generates inner network parameters based on population characteristics. The number of tasks for different groups of people used in the training; It is aimed at the first The inner network parameters after optimization for each individual population group determine the optimal sampling strategy for that specific population group. It is the first Test data from individual populations are used to evaluate the generalization performance of the sampling strategy on unseen data; It is a loss function that evaluates the quality of a sampling strategy and quantifies the effectiveness of the sampling strategy; Indexes representing specific population groups, ranging from 1 to ; Indicates outer layer network parameters Optimize to minimize the objective function; Indicates all The loss function of each individual group task is summed.

[0075] Step 1.4: Generate a basic sampling model for the new population based on the collected characteristics of the new population. The initial parameters of the sampling strategy for this population are calculated using the trained meta-learning network:

[0076] ;

[0077] in, This represents the initial parameters of the sampling strategy generated for the new population, including parameters such as the sampling priority, sampling order, and termination conditions for each indicator; This represents a meta-learning mapping function, used to transform population features into network parameters suitable for that specific population. This represents the feature vector of a new population, which includes multidimensional demographic data such as age distribution, gender ratio, occupational characteristics, and geographical characteristics.

[0078] This generates a basic sampling model suitable for this specific population, which will guide the subsequent sampling process.

[0079] The innovation of this step lies in using meta-learning techniques to enable the sampling strategy to quickly adapt to the characteristics of different population groups, solving the problem of traditional methods lacking the ability to adapt to group differences. Furthermore, the output sampling model incorporates population characteristic information, providing a personalized basis for subsequent information value assessment.

[0080] Step 2: Based on the sampling basic model, construct a Bayesian decision network to calculate the value of nutritional indicator information and generate the indicator information cost ratio. The information cost ratio is defined as the ratio of the expected information gain to the product of economic cost and time cost.

[0081] Specifically, the following steps are included:

[0082] Step 2.1: Construct a Bayesian network model to represent the probabilistic dependencies between nutritional indicators;

[0083] A Bayesian network is a directed acyclic graph. ,in This represents a set of nodes, corresponding to various nutritional indicators. This represents a set of directed edges, corresponding to the conditional dependencies between indices.

[0084] The joint probability distribution of a Bayesian network can be expressed as:

[0085] ;

[0086] in, This represents the joint probability distribution of all nutritional indicator variables; , , They represent the first , , Nutritional indicator variables, such as specific nutritional indicators like vitamin D, calcium, and iron; This indicates the total number of nutritional indicators; express The set of parent nodes, i.e., those directly affected A set of indicators; Indicates that in a given Under the condition of the parent node value The conditional probability distribution; Indicates from arrive The product operation of all conditional probabilities reflects the core characteristic of Bayesian networks that decompose joint distributions into products of conditional probabilities.

[0087] Step 2.2: Based on historical data and expert knowledge, learn the structure and parameters of the Bayesian network;

[0088] Network structure learning employs scoring-based methods, such as the BIC (Bayesian Information Criterion) or MDL (Minimum Description Length) criterion, to determine the optimal structure by evaluating the scores of different network structures. The BIC criterion evaluates network structures by balancing the model's log-likelihood with a model complexity penalty term, while the MDL criterion minimizes the number of bits required to describe the data based on coding theory. These scoring methods help select the structure that best reflects the true dependencies between nutritional indicators from among many possible network structures.

[0089] Parameter learning employs either maximum likelihood estimation or Bayesian estimation methods, and learns conditional probability tables. To accurately characterize the probabilistic dependencies among various nutritional indicators, among which... Indicates that under a given indicator parent node Value conditions The conditional probability distribution; maximum likelihood estimation estimates parameters by maximizing the likelihood function of the observed data, while Bayesian estimation introduces a prior distribution, providing a more robust estimate in the case of sparse data; these learning methods ensure that Bayesian networks can accurately capture the statistical dependencies between nutritional indicators;

[0090] Step 2.3: Calculate the Expected Information Gain (EIG) for each potential sampling indicator to quantify its information contribution to the nutritional status assessment.

[0091] ;

[0092] in, Indicators The expected information gain is used to quantify the information value that the indicator may bring. This represents the target nutritional state variable, which is the final nutritional status that needs to be assessed. Indicates the first One possible measurable nutritional indicator; express The entropy is used to measure the uncertainty of the target variable; Indicates that in the known under conditions The conditional entropy represents the condition of observing the index. Post-target variable Remaining uncertainty; Indicates about The mathematical expectation, that is, for A weighted average of all possible values; Indicates through measurement indicators The uncertainty that can be reduced is the amount of information obtained.

[0093] Step 2.4 introduces the Information Cost Ratio (ICR), which comprehensively considers the information value, economic cost, and time cost of the indicator:

[0094] ;

[0095] in, Indicators The information cost ratio is used to quantify the information value provided by this indicator per unit cost. Indicators The expected information gain measures the amount of information that can be obtained by measuring the metric. Indicates measurement index The economic costs; Indicates measurement index Time cost; Indicates measurement index The total cost is obtained by multiplying the economic cost and the time cost.

[0096] The innovation of this step lies in introducing information value theory to quantify the balance between the value and cost of indicators, thus solving the problem that traditional methods cannot scientifically allocate resources. Therefore, the output indicator information-cost ratio provides a scientific basis for subsequent adaptive sampling.

[0097] Step 3: Based on the cost ratio of indicator information, construct a sequential sampling strategy using a Markov decision process and execute adaptive indicator sampling.

[0098] Specifically, the following steps are included:

[0099] Step 3.1: Construct a Markov Decision Process (MDP) model to formalize the sequential sampling problem into a state transition process;

[0100] state space The state of the currently measured set of indicators and its observed values ​​is denoted as:

[0101] ;

[0102] in, It is a moment A set of indexes of measured indicators; Indicates a point in time The system status; Indicates the first One nutritional indicator variable; Indicators The actual observed value; Indicates index The set of indexes belonging to the measured indicators ;

[0103] Action space The set of indicators that may be measured next is denoted as:

[0104] ;

[0105] in, Indicates the state The following is a set of selectable actions; Indicates the nutritional indicators that can be measured; Indicates index Index set not belonging to the measured indicators That is, indicators that have not yet been measured;

[0106] State transition function Indicates the state Select measurement indicators Then, transition to state The probability of , where, Represents a probability function; Indicates the state at the next point in time; Indicates the current state; Indicates a point in time Choose the metric to measure; this function describes the dynamic characteristics of how the system state changes with the choice of action.

[0107] reward function Reflecting measurement indicators The net benefit derived is defined as the information gain of the indicator minus the measurement cost, where... Represents the reward function; Indicates the current state; Indicates the action chosen (the measured indicator); This function represents the new state after an action is performed; it quantifies the net value obtained by selecting a specific indicator for measurement in the current state, and is used to guide the choice of actions in the decision-making process.

[0108] Step 3.2: Implement the improved Markov blanket information propagation algorithm to effectively utilize the conditional dependencies between indicators to predict the probability distribution of unmeasured indicators.

[0109] For each indicator Define its Markov blanket ,include The parent node, child nodes, and other parent nodes of the child nodes;

[0110] According to the principle of conditional independence Given its Markov blanket, it is independent of the conditions of other nodes in the network, that is:

[0111] ;

[0112] in This indicates the nutritional indicator variable currently under consideration; Indicates except The set of all indicator variables outside of; This means that, given all other indicators are known... The conditional probability distribution; express A Markov blanket contains its parent node, child nodes, and other parent nodes of the child nodes; This indicates that, given only the known indices of its Markov blanket... The conditional probability distribution; this equation shows that knowing only the index values ​​in the Markov blanket is sufficient to completely determine The conditional probability is determined without considering other metrics in the network;

[0113] Using this property, after measuring some indicators, the conditional probability distribution of unmeasured indicators can be updated efficiently; this means that it is not necessary to collect all indicator data, but only to obtain the measured values ​​of key indicators, so as to make accurate inferences about other unmeasured indicators, thereby greatly improving sampling efficiency.

[0114] Step 3.3, based on the current state Based on the information cost ratio, select the next optimal measurement metric:

[0115] ;

[0116] in, Indicates a point in time The optimal measurement index selected; This indicates taking the parameter that makes the following expression reach its maximum value. ; Indicates action Belongs to the current state The set of optional actions, i.e., the set of indicators that have not yet been measured; This refers to the current state. Below, indicators The conditional information cost ratio is calculated using the following formula:

[0117] ;

[0118] in, Indicates the current state Below, indicators Conditional information cost ratio; Indicates a known state Under the conditions, measurement indicators Conditional expected information gain; Indicates measurement index The economic costs; Indicates measurement index Time cost; Indicates measurement index The overall cost.

[0119] and Indicates a known state Under the conditions, measurement indicators The conditional expected information gain; where, Indicates a known state Under the conditions, measurement indicators Conditional expected information gain, Indicates the current state Next target nutrient state variables The conditional entropy measures the uncertainty of assessing nutritional status under the current conditions; Indicates the current state And measured the indicators Post-target variable Conditional entropy; Indicates information about indicators The expected conditional entropy of possible values; the entire formula quantifies the measurement index in the current state. Uncertainty is expected to be reduced.

[0120] Step 3.4: Perform the measurement and update the status. Repeat steps 3.2 and 3.3 until the termination condition is met.

[0121] The innovation of this step lies in formalizing the sequential sampling problem as a Markov decision process and combining it with an improved Markov blanket algorithm to achieve efficient information propagation, ensuring that the indicator with the highest information value is selected for measurement at each step. Furthermore, this adaptive sampling strategy solves the problem of low efficiency in traditional fixed sampling schemes.

[0122] Step 4: During the adaptive index sampling process, calculate the information gain ratio to achieve dynamic termination decision and avoid oversampling;

[0123] Specifically, the following steps are included:

[0124] Step 4.1, define the Information Gain Ratio (IGR), which measures the marginal information gain brought about by each new measurement indicator:

[0125] ;

[0126] in, Indicates the first The information gain ratio of the step is used to measure the relative information gain of the current measurement metric compared to the previous measurement metric. , They represent the first Step and the first Indicators measured in steps; , They represent the measurement indicators respectively. and The previous state included and All measured indicators and their values ​​at any given time; Indicates the state Lower measurement index The resulting conditional expected information gain quantifies the amount of information provided by the newly added measurement indicators in the current step; Indicates the state Lower measurement index The resulting conditional expected information gain quantifies the amount of information provided by the measurement index in the previous step; the entire ratio reflects the changing trend of information acquisition efficiency. When the ratio is less than a certain threshold, it indicates that the marginal benefit of continuing sampling is diminishing, and the sampling process can be terminated.

[0127] Step 4.2: Implement a threshold-based dynamic termination algorithm, setting two termination conditions. Sampling will stop when either condition is met.

[0128] These two conditions are:

[0129] Information gain rate threshold condition: when When this occurs, it indicates that the information gain from the new measurement indicator is no longer economical relative to the cost, and sampling should be stopped. Indicates the first The information gain ratio of the step is used to measure the relative information gain of the current measurement metric compared to the previous measurement metric. It is a preset information gain rate threshold, which represents the minimum acceptable information gain rate. Below this value, it means that the marginal benefit of continuing to sample is too low.

[0130] Accuracy assessment conditions: when the current state The following is the estimated accuracy value of the assessment The preset goal has been achieved. At that time, that is Sampling can be terminated; among them, Indicates a point in time The system state includes the set of measured indicators and their observations; Indicates the current state The estimated accuracy of the assessment quantifies the accuracy of nutritional assessments based on measured indicators. This is a preset accuracy threshold, representing the minimum accuracy standard required for the evaluation. Reaching this value indicates that the evaluation results are sufficiently reliable.

[0131] Step 4.3: Construct an evaluation accuracy estimation model to predict the accuracy of the evaluation results in real time under the current state.

[0132] ;

[0133] in, Indicates the current state The estimated accuracy of the assessment is used to predict the accuracy of the assessment results obtained based on the set of measured indicators. It is an accuracy estimation function, a mathematical model used to map the current state to an accuracy estimate; Indicates a point in time The system status includes all measured indicators and their observations; It is the set of parameters of the function, including model parameters such as weights and biases, which can be obtained by training with historical data and used to adjust the model to adapt to the characteristics of different groups of people and evaluation tasks;

[0134] Step 4.4: Adaptively adjust the termination threshold, dynamically setting it based on the urgency of the task and resource constraints. and ;

[0135] In emergency situations (such as suspected food safety issues), increase and reduce To terminate sampling earlier and obtain results quickly;

[0136] When resources are plentiful and high accuracy is required, reduce And improve More thorough sampling is needed to improve the accuracy of the assessment.

[0137] The innovation of this step lies in introducing the concept of information gain ratio to achieve dynamic termination decisions in the sampling process, avoiding resource waste caused by oversampling while ensuring that the assessment results achieve the required accuracy. Therefore, this dynamic termination method is particularly suitable for nutritional assessments under resource constraints and emergency situations, achieving an optimal balance between resource input and assessment accuracy.

[0138] Step 5: Based on the sampling results after the dynamic termination decision, conditional probability inference is applied to generate complete nutritional assessment results and provide a reliability estimate of the results;

[0139] Specifically, the following steps are included:

[0140] Step 5.1: Use a probabilistic graphical model to infer the conditional probability distribution of the unmeasured index, based on the state at the time of termination of sampling. Infer the complete distribution of nutritional indicators;

[0141] The conditional probability distribution can be represented as:

[0142]

[0143] in, Indicates the state at the time of termination of sampling. The joint probability distribution of all nutritional indicators under the given conditions; , , They represent the first , , One nutritional indicator variable; This indicates the total number of nutritional indicators; This represents the set of measured indicators, that is, the set of nutritional indicator variables that have been measured at the time of termination of sampling; This represents the set of unmeasured indicators, that is, the set of nutritional indicator variables that have not been measured at the time of termination of sampling; This indicates the system state at the time of termination of sampling, including the specific observed values ​​of the measured indicators; It is the observed distribution of measured indicators, representing the known state. The probability distribution of the measured indicators; It is the conditional probability distribution of the unmeasured index, representing the probability distribution of the unmeasured index given the known values ​​of the measured index; the entire formula shows that the joint distribution of the complete index can be decomposed into the product of the measured part and the unmeasured part inferred from the measured values.

[0144] Step 5.2: Combine conditional probability distribution and expert rules to generate nutritional status assessment results;

[0145] Nutritional status assessment results It can be expressed as a function of measured and inferred indicators:

[0146] ;

[0147] in, This represents the final nutritional status assessment result and is the output of the assessment function. This represents a nutritional status assessment function, used to map measurement indicators and inferred indicators to the final assessment result; It represents the set of measured indicators, which includes all indicators and their values ​​actually measured during the sampling process; It is the inferred value of the unmeasured index, representing the estimated value of the unmeasured index calculated through the conditional probability model;

[0148] Step 5.3: Calculate the reliability estimate of the evaluation results and quantify the uncertainty of the results;

[0149] Result reliability score Defined as:

[0150] ;

[0151] in, The reliability score represents the evaluation result, and its value ranges from [value missing]. A higher value indicates a higher reliability of the evaluation results; It is in state Down The normalized conditional entropy is used to quantify the degree of uncertainty in the evaluation results; The nutritional status assessment results are the final assessment results generated based on measured and inferred indicators; This represents the system state at the time of sampling termination, containing observations of all measured metrics; normalization ensures that the conditional entropy value is mapped to... The interval makes the reliability score Also located Within the range; the lower the uncertainty of the evaluation result, the closer the normalized conditional entropy is to 0, and the higher the reliability score. The closer to 1;

[0152] Step 5.4: Based on the assessment results and reliability estimates, generate personalized nutrition recommendations;

[0153] Regarding the evaluation results By combining various nutritional indicators with an expert knowledge base, abnormal indicators and their potential causes can be identified.

[0154] Based on the severity and reliability of the abnormal indicators, intervention priorities are determined;

[0155] Generate targeted nutritional improvement recommendations and determine the confidence level of the recommendations based on reliability estimates.

[0156] The innovation of this step lies in combining conditional probability inference with expert rules to achieve an efficient mapping from limited sampling to a complete assessment. Simultaneously, it quantifies the uncertainty of the results through reliability estimation, providing more comprehensive information support for medical decision-making. Furthermore, this method, combining probabilistic inference and uncertainty quantification, solves the problem of traditional assessment methods struggling to guarantee reliability under limited data conditions.

[0157] Application example of this implementation method:

[0158] Based on the embodiments of this application, the application of this method will be described in detail below through a real-world scenario of assessing the nutritional status of enterprise employees.

[0159] A technology company needs to conduct nutritional status assessments for 2,000 employees to support its corporate health management program. Traditional methods require each employee to complete 35 nutritional indicators, totaling 70,000 tests, estimated to take 3 weeks and cost approximately 1.4 million yuan. The company wants to complete this assessment more efficiently with limited resources.

[0160] Implementation process example:

[0161] Meta-learning network application phase: First, collect the demographic characteristics of the company's employees, including: average age 32.5 years, male-to-female ratio 1.6:1, 85% in R&D or technical positions, 60% with sedentary habits, and 12% with irregular work and rest schedules. These characteristics are represented as a demographic feature vector. ;

[0162] Initial sampling strategy parameters are generated based on the characteristics of the population using a pre-trained meta-learning network. These parameters guided the initial sampling preferences, for example, assigning higher initial weights to indicators related to cervical spine problems and vision-related nutritional indicators common among IT professionals. Among them, This represents the initial sampling strategy parameter vector, which includes the sampling priority weights of each nutritional indicator. This represents a meta-learning network function used to map crowd features to optimal sampling strategy parameters; This represents a population feature vector, which includes statistical characteristics of the target population such as age, gender ratio, and occupational distribution.

[0163] Bayesian network modeling phase: Construct a Bayesian network containing 35 nutritional indicator nodes, where edges between nodes represent dependencies between indicators. For example, there is a strong dependency between vitamin D and calcium, so directed edges from vitamin D to calcium absorption are added to the network.

[0164] For the company's employee group, calculate the information cost ratio for each indicator.

[0165] For example, the EIG value of vitamin B12 is 0.42. EIG represents the expected information gain from measuring this indicator. The detection cost is 80 yuan and takes 15 minutes. Its ICR is calculated as follows: ICR represents the efficiency of information acquisition per unit of economic and time cost.

[0166] The EIG value for vitamin D is 0.38. EIG represents the expected information gain from measuring vitamin D. The test cost is 60 yuan and takes 10 minutes. Its ICR is calculated as follows: ICR represents the efficiency of information acquisition per unit of economic and time cost.

[0167] Based on the calculation results, the system determined the initial set of priority measurement indicators.

[0168] Adaptive sampling execution phase: The system starts with five basic indicators, including basal metabolic rate and hemoglobin, to construct the initial state. Then, based on the Markov decision process model, the next optimal measurement index is selected sequentially.

[0169] For example, after measuring basal metabolic rate (BMR), hemoglobin (Hb), and total cholesterol (TC), the current state is: .

[0170] The system calculates the conditional information cost ratios of the remaining indicators and finds that vitamin D has the highest ICR value in the current state, which is 0.00085. Therefore, vitamin D is selected as the fourth measurement indicator.

[0171] Dynamic Termination Decision Phase: As sampling progresses, the system continuously monitors changes in the Information Gain Rate (IGR). After measuring the 14th metric, the following is calculated: Below the preset threshold .

[0172] Meanwhile, the current estimate of the accuracy of the assessment It has exceeded the target accuracy. If one of the two termination conditions is met, the system decides to terminate the sampling process.

[0173] Evaluation Result Generation Phase: Based on the 14 measured indicators, the system uses a conditional probability model to infer the values ​​of the remaining 21 unmeasured indicators. For example, based on the measured values ​​of iron, hemoglobin, and transferrin, the system infers the distribution of ferritin levels and takes its conditional expectation as an estimate.

[0174] For each employee, the system generates a complete evaluation result based on 35 indicators and calculates a reliability score. For example, for one employee, the system calculates a reliability score. This indicates that the evaluation results have a high degree of credibility.

[0175] Based on the assessment results, the system identified common nutritional problems among the company's employees, such as vitamin D deficiency (68%) and magnesium deficiency (52%), and generated personalized nutritional improvement suggestions.

[0176] Technical effectiveness verification:

[0177] Improved sampling efficiency: The actual results of employee nutrition assessment after applying this method are as follows:

[0178] On average, each person only measured 13.4 indicators (38.3% of the total indicators), the assessment accuracy reached 92.7%, the assessment time was shortened from the original 3 weeks to 6.5 days (a reduction of 65.2%), and the total cost was reduced to 584,000 yuan (a saving of 58.3%).

[0179] Improved Adaptability to Different Departments: When applied to employees in different departments, this method demonstrates good adaptability. The system can automatically adjust the sampling strategy based on the characteristics of employees in different departments, as shown below:

[0180] A comparison of sampling strategies between the R&D department and the sales department:

[0181] Research and development department: Prioritize vision-related nutritional indicators and sedentary-related metabolic indicators;

[0182] Sales department: Prioritize energy metabolism indicators and immunity-related indicators;

[0183] Adapting to the data requirements of the new department: Only 15 samples are needed (85% less than traditional methods).

[0184] Emergency Response Effectiveness: In a suspected food safety incident, the system was able to complete a preliminary assessment of 100 employees who may have been exposed to the problematic food within 30 minutes, measuring an average of only 8 indicators per person, and accurately identifying 12 employees requiring further medical intervention. Before applying this method, similar situations would have taken 4 to 5 hours to complete the initial screening.

[0185] As can be seen from this practical application example, the AI-based rapid assessment method for nutritional cases of large clients provided in this application demonstrates significant technical effects in actual enterprise health management scenarios. It not only greatly improves assessment efficiency and reduces resource consumption, but also enhances the flexibility and adaptability of the assessment, especially in situations with limited resources and emergencies. The embodiments of the present invention have been described above, but these embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments based on the inspiration of these embodiments, all of which are within the protection scope of these embodiments.

Claims

1. A rapid assessment method for nutritional cases of large clients based on AI, characterized in that, Includes the following steps: Meta-learning networks are used to process population statistical feature data to generate a basic sampling model for a specific population. Based on the sampling-based model, a Bayesian decision network is constructed to calculate the information value of nutritional indicators and generate the information cost ratio. The information cost ratio is defined as the ratio of the expected information gain to the product of economic cost and time cost. The information cost ratio is calculated using the following formula: ,in, Indicators Information cost ratio This represents the expected information gain of the indicator. Indicates measurement index The economic cost, Indicates measurement index Time cost; Based on the cost ratio of indicator information, a sequential sampling strategy is constructed using a Markov decision process to perform adaptive indicator sampling. During the adaptive index sampling process, the information gain ratio is calculated to achieve dynamic termination decision and avoid oversampling. Based on the sampling results after dynamic termination decision, conditional probability inference is applied to generate complete nutritional assessment results and provide an estimate of the reliability of the results.

2. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The steps for processing population statistical feature data using meta-learning networks include: Collect statistical characteristic data of the target population, including demographic characteristics, occupational characteristics, regional characteristics, historical nutritional status and their multidimensional data; A meta-learning network is constructed, which adopts a two-layer structure. The outer network is responsible for learning the mapping function from population features to the parameters of the inner network, and the inner network is responsible for implementing the specific sampling strategy. The meta-learning network is trained using a task-based meta-learning paradigm; For new populations, the initial parameters of the sampling strategy for the new population are calculated using a pre-trained meta-learning network, and a sampling base model suitable for the new population is generated.

3. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The steps for constructing a Bayesian decision network to calculate the value of nutritional indicator information include: Construct a Bayesian network model to represent the probabilistic dependencies among nutritional indicators; Learn the structure and parameters of Bayesian networks based on historical data and expert knowledge; Calculate the expected information gain for each potential sampling indicator to quantify the information contribution of the expected information gain to the nutritional status assessment. The information cost ratio is introduced to comprehensively consider the information value, economic cost, and time cost of the indicators.

4. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The steps for constructing a sequential sampling strategy using a Markov decision process include: A Markov decision process model is constructed to formalize the sequential sampling problem into a state transition process; An improved Markov blanket information propagation algorithm is implemented to effectively utilize the conditional dependencies between indicators to predict the probability distribution of unmeasured indicators. Based on the current state and information cost ratio, select the next optimal measurement indicator; Perform measurements and update the status.

5. The AI-based rapid assessment method for nutritional cases of large clients according to claim 4, characterized in that, The Markov decision process model includes: The state space represents the state of the currently measured set of indicators and their observations. Action space represents the set of metrics that can be measured in the next step; The state transition function represents the probability of transitioning to the next state after selecting a measurement index in one state; The reward function, which reflects the net benefit brought by the measurement indicator, is defined as the information gain of the measurement indicator minus the measurement cost.

6. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The steps for calculating the information gain ratio to implement dynamic termination decision include: Define the information gain ratio to measure the marginal information gain brought about by each new measurement indicator; Implement a threshold-based dynamic termination algorithm, setting two termination conditions: an information gain rate threshold condition and an evaluation accuracy condition. Sampling will stop when either condition is met. Construct an assessment accuracy estimation model to predict the accuracy of the assessment results in real time under the current state; The termination threshold is adaptively adjusted, and the termination condition parameters are dynamically set according to the urgency of the task and resource constraints.

7. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The steps for generating complete nutritional assessment results using conditional probability inference include: The conditional probability distribution of unmeasured indicators is inferred using a probabilistic graphical model, and the complete distribution of nutritional indicators is inferred based on the state at the time when sampling has stopped. By combining conditional probability distributions and expert rules, nutritional status assessment results are generated. Calculate the reliability estimate of the evaluation results and quantify the uncertainty of the results; Personalized nutrition recommendations are generated based on the assessment results and reliability estimates.

8. The AI-based rapid assessment method for nutritional cases of large clients according to claim 7, characterized in that, The reliability estimate of the assessment results is calculated using the following formula: ； Where R represents the reliability estimate of the evaluation result; Indicates the state The normalized conditional entropy of the target nutrient state variable Y is used to quantify the uncertainty of the nutrient state assessment result under the current sampling state. This indicates the state at the time of termination of sampling, including all measured indicators and their observations; Normalized conditional entropy scales the original conditional entropy to the [0,1] interval, making reliability estimates more comparable; The value of R ranges from [0,1]. The larger the value, the higher the reliability, that is, the lower the uncertainty of the evaluation result.

9. The AI-based rapid assessment method for nutritional cases of large clients according to claim 1, characterized in that, The expected information gain is calculated using the following formula: ； in, Indicators The expected information gain; Y represents the target nutritional state variable, which is the final nutritional status that needs to be evaluated; Let Y represent the i-th possible measurable nutritional indicator; H(Y) represents the entropy of Y, used to measure the uncertainty of the target variable. Indicates that in the known The conditional entropy of Y under given conditions represents the conditional entropy of Y under observed index conditions. The remaining uncertainty of the target variable Y; Indicates about The mathematical expectation, that is, for A weighted average of all possible values; Indicates through measurement indicators The uncertainty that can be reduced is the amount of information obtained.

10. An AI-based rapid assessment system for nutritional cases of large clients, used to execute the AI-based rapid assessment method for nutritional cases of large clients as described in any one of claims 1-9, characterized in that, include: The population feature processing module is used to process population statistical feature data using meta-learning networks to generate a sampling base model for a specific population. The information value assessment module is used to construct a Bayesian decision network to calculate the value of nutritional indicator information and generate the indicator information cost ratio. The adaptive sampling module is used to construct a sequential sampling strategy using a Markov decision process and execute adaptive index sampling. The dynamic termination decision module is used to calculate the information gain rate to realize the dynamic termination decision of the sampling process; The assessment results generation module is used to generate complete nutritional assessment results by applying conditional probability inference and to provide an estimate of the reliability of the results.

Images