Intelligent method for real-time assessment and optimization of nutritional balance for stroke patients in ICU

Abstract

The present application relates to the field of nutritional balance evaluation, and discloses an ICU stroke patient nutritional balance real-time evaluation and optimization method based on artificial intelligence, comprising the following steps: collecting patient activity energy consumption through a wearable device, combining dietary intake and metabolic indicators to construct an individualized nutritional metabolism model, calculating nutritional balance degree in real time and generating a quantitative deviation index.Combining the quantitative deviation index and the target of the rehabilitation stage, a nutrition optimization scheme is generated through a reinforcement learning algorithm, the scheme is executed and trained, and the model parameters of the individualized nutritional metabolism model are automatically iterated to achieve the purpose of continuous optimization.The present application overcomes the pain points of traditional nutritional evaluation of ICU stroke patients, such as lag and dependence on experience-based decision-making, and realizes precise matching of nutritional supply and rehabilitation demand through the form of artificial intelligence, thereby shortening the stay time of ICU stroke patients in the ICU.

Description

A Real-Time Assessment and Optimization Method for Nutritional Balance in ICU Stroke Patients Based on Artificial Intelligence Technical Field

[0001] This invention relates to the field of nutritional balance assessment, and in particular to a method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence. Background Technology

[0002] Stroke, also known as cerebrovascular accident, is an acute brain injury caused by the sudden rupture (hemorrhagic stroke) or blockage of blood vessels in the brain, leading to ischemia and hypoxia in brain tissue (ischemic stroke). The core pathology of hemorrhagic stroke is the rupture of a cerebral blood vessel → hematoma compressing brain tissue, while the core pathology of ischemic stroke is the blockage of a cerebral blood vessel by a thrombus → necrosis of brain cells. Stroke can lead to hemiplegia, temporal and temporal impairment, loss of consciousness, and even death. Because stroke can cause metabolic disorders that worsen the condition, and can lead to functional impairment that hinders nutrient intake and nutritional imbalance that creates a vicious cycle, nutritional balancing is essential for stroke patients in the ICU. Compared with traditional clinical experience-based decision-making, artificial intelligence-based nutritional balancing has advantages in response speed, individualized accuracy, and multi-indicator collaborative decision-making, which can lead to reduced mortality, a sharp decrease in complications, and optimized resource allocation. For example, in a 65-year-old ischemic stroke patient, the AI ​​system, based on the early negative trend of NDI (-0.15) and the decline in BCAA utilization, adjusted the high-branched-chain amino acid formula 48 hours in advance, avoiding 12% muscle loss (dual-energy X-ray aspiration results), and ultimately shortening the mechanical ventilation time by 5 days. Therefore, a method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence is proposed. Summary of the Invention

[0003] This invention overcomes the shortcomings of existing technologies and provides a method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] The first aspect of this invention provides a method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence, comprising the following steps:

[0006] Wearable activity monitoring devices for ICU stroke patients are used to collect real-time personal status data of ICU stroke patients, and the personal status data of ICU stroke patients are combined to obtain time-series data of the target patient's personal status.

[0007] Temporal features of the target patient's individual status time series data are extracted using a temporal convolutional network, and a metabolic analysis model is constructed based on the temporal features. The real-time nutritional balance of the target patient is calculated based on the metabolic analysis model, and a quantitative deviation index is generated.

[0008] Nutrition optimization plans are generated based on the quantitative deviation index of the target patients and combined with reinforcement learning algorithms.

[0009] Furthermore, in a preferred embodiment of the present invention, the step of collecting real-time personal status data of ICU stroke patients through a wearable activity monitoring device, and combining the personal status data of ICU stroke patients to obtain time-series data of the target patient's personal status, specifically involves:

[0010] Identify ICU stroke patients who require nutritional balance assessment and designate them as target patients. At the same time, acquire wearable devices for detecting limb activity, heart rate, and exercise frequency in ICU stroke patients and designate them as target monitoring devices.

[0011] The target monitoring device is worn on the target patient and activated. A standard monitoring time is preset. During the standard monitoring time, the target monitoring device collects the target patient's limb activity, heart rate, and exercise frequency in real time.

[0012] An energy consumption algorithm is set up in the target monitoring device. Based on the energy consumption algorithm, combined with the real-time collection of the target patient's limb activity, heart rate and exercise frequency, the real-time activity energy consumption of the target patient is obtained.

[0013] The big data network is acquired, and the types, volumes, and cooking methods of the target patient's daily diet are determined. A nutrition database is retrieved from the big data network. The nutrition database includes the protein content, fat content, and carbohydrate content of different types of daily diets under different volumes and cooking methods, which are collectively referred to as dietary status.

[0014] Based on the nutrition database, the dietary status of the target patient is determined and labeled as the target dietary status, and the metabolic indicators of the target patient are obtained through the hospital's database.

[0015] The metabolic indicators, dietary status, and real-time activity energy consumption of the target patients were spatiotemporally aligned to obtain the time-series data of the individual status of the target patients.

[0016] Furthermore, in a preferred embodiment of the present invention, the step of extracting temporal features of the target patient's personal status time-series data through a temporal convolutional network, constructing a metabolic analysis model based on the temporal features, calculating the target patient's real-time nutritional balance based on the metabolic analysis model, and simultaneously generating a quantitative deviation index, specifically involves:

[0017] A data processing terminal is obtained, a temporal convolutional network is introduced into the data processing terminal, and the time-series data of the target patient's personal status is imported into the data processing terminal.

[0018] Temporal convolutional networks are used to perform temporal convolution on the time-series data of the target patient's personal status. The temporal convolution is used to extract time-series features corresponding to different time periods on the data, generating the diurnal energy consumption fluctuation curve, postprandial blood glucose response curve, and nitrogen balance trend curve of the target patient's personal status time-series data.

[0019] The diurnal energy consumption fluctuation curve, postprandial blood glucose response curve, and nitrogen balance trend curve of the target patient's personal status time series data are identified as the time series characteristics of the target patient's personal status.

[0020] Wavelet denoising and normalization are performed on the temporal features of the individual status of the target patient. At the same time, multiple interpolation is used to fill in the missing values ​​in the temporal features of the individual status of the target patient to obtain the fused features of the individual status of the target patient.

[0021] A dual-channel metabolic analysis training model was introduced, and the model parameters were iteratively optimized using a Bayesian update method to obtain an optimized dual-channel metabolic analysis training model.

[0022] The target patient's personal status fusion features were input into the optimized dual-channel metabolic analysis training model for training, generating the target patient's real-time nutritional balance.

[0023] Based on the target patient's real-time nutritional balance, the quantitative deviation index of the target patient is calculated.

[0024] Furthermore, in a preferred embodiment of the present invention, the introduction of a dual-channel metabolic analysis training model, and the iteration optimization of the model parameters of the dual-channel metabolic analysis training model through a Bayesian update method to obtain an optimized dual-channel metabolic analysis training model, specifically involves:

[0025] In the dual-channel metabolic training model, a Bayesian optimization framework is constructed, wherein the Bayesian optimization framework sets latent variable samples for different channels of the dual-channel metabolic training model.

[0026] Energy consumption latent variable samples were set in channel 1 of the dual-channel metabolic training model, and trace element metabolism latent variable samples were set in channel 2 of the dual-channel metabolic training model.

[0027] Within the Bayesian optimization framework, the likelihood functions of the latent variable samples in channel 1 and channel 2 are calculated respectively and labeled as the target likelihood function. The dual-channel Bayesian update training is then performed in conjunction with the target likelihood function.

[0028] Specifically, conjugate prior analytical training is performed in channel 1, and approximate Gaussian distribution training is performed in channel 2. After training, the distribution convergence of samples in channel 1 and channel 2 is calculated. When the distribution convergence reaches a preset range, the dual-channel Bayesian update training is stopped, and the optimized dual-channel metabolic analysis training model is output.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of inputting the target patient's personal state fusion features into the optimized dual-channel metabolic analysis training model for training, and generating the target patient's real-time nutritional balance, specifically involves:

[0030] In the optimized dual-channel metabolic analysis training model, channel 1 is the model analysis channel for analyzing macro-nutritional balance. Channel 1 is connected to the BiLSTM layer in the optimized dual-channel metabolic analysis training model. The BiLSTM layer is used to calculate the real-time basal energy consumption value of the target patient under the current dietary state by combining the target patient's personal state fusion features.

[0031] After obtaining the target patient's basal energy expenditure value from channel 1, the standard basal energy expenditure range of the target patient under the current dietary status is calculated by optimizing the dual-channel metabolic analysis training model.

[0032] If the real-time basal energy expenditure value is greater than the standard basal energy expenditure range, there is a risk of negative balance in the target patient output within the optimized dual-channel metabolic analysis training model. If the real-time basal energy expenditure value is less than the standard basal energy expenditure range, there is a risk of overfeeding in the target patient output within the optimized dual-channel metabolic analysis training model.

[0033] In optimizing the dual-channel metabolic analysis training model, channel 2 is the model analysis channel for performing micro-metabolic analysis on the fusion characteristics of the target patient's individual status;

[0034] Within channel 2, a nutritional metabolism map is constructed by analyzing the metabolic index characteristics of the target patient within the fusion features of the target patient's personal status. In the nutritional metabolism map, nodes represent the trace elements absorbed by the target patient, and edges represent the biochemical reaction relationships of different trace elements.

[0035] The nutrient metabolism graph is iteratively updated using a graph neural network. The iterative update of the graph neural network involves updating the content of trace elements represented by nodes in real time based on the biochemical reaction relationship of the edges, and calculating the metabolic efficiency and metabolic pathway of trace elements based on the content.

[0036] Based on the metabolic efficiency and pathways of trace elements, the target patient's nutritional balance is updated in real time in the optimized dual-channel metabolic analysis training model. That is, the target patient's real-time nutritional balance is calculated by combining the metabolic efficiency and pathways of trace elements with the target patient's real-time basal energy consumption under the current dietary state.

[0037] Furthermore, in a preferred embodiment of the present invention, the step of calculating the quantitative deviation index of the target patient based on the real-time nutritional balance of the target patient specifically involves:

[0038] Within the big data network, the standard nutritional balance corresponding to different rehabilitation stages of the target patient is obtained, and the standard nutritional balance weight of different rehabilitation stages is calculated.

[0039] The target patients are in different rehabilitation stages, including the acute phase, the muscle strength reconstruction phase, and the functional recovery phase.

[0040] Based on the standard nutritional balance weights for different rehabilitation stages, and combined with the real-time nutritional balance of the target patient, the difference between the real-time nutritional balance of the target patient and the standard value at different rehabilitation stages is calculated and calibrated as the real-time difference of nutritional balance.

[0041] Based on the real-time difference in nutritional balance at different stages of rehabilitation, an index for correcting nutritional balance, namely the quantitative deviation index, is calculated.

[0042] Furthermore, in a preferred embodiment of the present invention, the step of generating a nutrition optimization plan based on the quantitative deviation index of the target patient and in conjunction with a reinforcement learning algorithm specifically includes:

[0043] Within the big data network, the initial nutritional supplementation plan corresponding to different recovery stages of the target patient is obtained and marked as a nutritional optimization plan with undetermined content.

[0044] A reinforcement learning framework is constructed. Within the framework, based on the quantitative deviation index of the target patient, the content weight of the nutritional optimization plan with undetermined content is calculated, and the positive and negative values ​​of the quantitative deviation index are analyzed.

[0045] If the quantification deviation index is positive, a positive reward mechanism is triggered, wherein the positive reward mechanism sets the nutritional supplementation plan in the nutritional optimization plan with undetermined content as a priority;

[0046] If the quantification deviation index is negative, a negative penalty mechanism is triggered, wherein the negative penalty mechanism sets the nutritional conditioning scheme in the nutritional optimization scheme with undetermined content as a priority.

[0047] Based on the quantitative deviation index, combined with the content weight calculation of the nutritional optimization scheme to be determined, the nutritional optimization scheme to be analyzed is output. The contraindicated nutritional supplements for the target patients are identified through the hospital database. Based on the contraindicated nutritional supplements for the target patients, the corresponding ones are removed from the nutritional optimization scheme to be analyzed, and the target nutritional optimization scheme is obtained.

[0048] A second aspect of the present invention also provides an artificial intelligence-based real-time assessment and optimization system for nutritional balance in ICU stroke patients. The assessment and optimization system includes a memory and a processor. The memory stores assessment and optimization methods. When the processor executes the assessment and optimization methods, it performs the following steps:

[0049] Wearable activity monitoring devices for ICU stroke patients are used to collect real-time personal status data of ICU stroke patients, and the personal status data of ICU stroke patients are combined to obtain time-series data of the target patient's personal status.

[0050] Temporal features of the target patient's individual status time series data are extracted using a temporal convolutional network, and a metabolic analysis model is constructed based on the temporal features. The real-time nutritional balance of the target patient is calculated based on the metabolic analysis model, and a quantitative deviation index is generated.

[0051] Nutrition optimization plans are generated based on the quantitative deviation index of the target patients and combined with reinforcement learning algorithms.

[0052] This invention addresses the technical deficiencies in the prior art and offers the following advantages: By collecting patient activity energy consumption data through wearable devices and combining it with dietary intake and metabolic indicators, an individualized nutritional metabolism model is constructed. This model calculates nutritional balance in real time and generates a quantitative deviation index. Combining the quantitative deviation index with rehabilitation stage goals, a nutritional optimization plan is generated using a reinforcement learning algorithm. The plan is then trained and executed, while the model parameters of the individualized nutritional metabolism model are automatically iterated to achieve continuous optimization. This invention overcomes the pain points of traditional nutritional assessments for ICU stroke patients, which are often delayed and rely on experience-based decision-making. Through artificial intelligence, it achieves precise matching of nutritional supply and rehabilitation needs, shortening the time ICU stroke patients spend in the ICU. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0054] Figure 1 shows a flowchart of a method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence;

[0055] Figure 2 shows a flowchart of the method for calculating the real-time nutritional balance and quantitative deviation index of the target patient;

[0056] Figure 3 shows a program view of the AI-based real-time assessment and optimization system for nutritional balance in ICU stroke patients. Detailed Implementation

[0057] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0058] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0059] Figure 1 shows a flowchart of an AI-based real-time assessment and optimization method for nutritional balance in ICU stroke patients, including the following steps:

[0060] S102: Through wearable activity monitoring devices for ICU stroke patients, real-time personal status data of ICU stroke patients is collected, and the personal status data of ICU stroke patients is combined to obtain time-series data of the target patient's personal status.

[0061] S104: Using a temporal convolutional network, extract the temporal features of the target patient's personal status time series data, construct a metabolic analysis model based on the temporal features, calculate the target patient's real-time nutritional balance based on the metabolic analysis model, and generate a quantitative deviation index.

[0062] S106: Generate a nutrition optimization plan based on the quantitative deviation index of the target patient and in conjunction with a reinforcement learning algorithm.

[0063] Furthermore, in a preferred embodiment of the present invention, the step of collecting real-time personal status data of ICU stroke patients through a wearable activity monitoring device, and combining the personal status data of ICU stroke patients to obtain time-series data of the target patient's personal status, specifically involves:

[0064] Identify ICU stroke patients who require nutritional balance assessment and designate them as target patients. At the same time, acquire wearable devices for detecting limb activity, heart rate, and exercise frequency in ICU stroke patients and designate them as target monitoring devices.

[0065] The target monitoring device is worn on the target patient and activated. A standard monitoring time is preset. During the standard monitoring time, the target monitoring device collects the target patient's limb activity, heart rate, and exercise frequency in real time.

[0066] An energy consumption algorithm is set up in the target monitoring device. Based on the energy consumption algorithm, combined with the real-time collection of the target patient's limb activity, heart rate and exercise frequency, the real-time activity energy consumption of the target patient is obtained.

[0067] The big data network is acquired, and the types, volumes, and cooking methods of the target patient's daily diet are determined. A nutrition database is retrieved from the big data network. The nutrition database includes the protein content, fat content, and carbohydrate content of different types of daily diets under different volumes and cooking methods, which are collectively referred to as dietary status.

[0068] Based on the nutrition database, the dietary status of the target patient is determined and labeled as the target dietary status, and the metabolic indicators of the target patient are obtained through the hospital's database.

[0069] The metabolic indicators, dietary status, and real-time activity energy consumption of the target patients were spatiotemporally aligned to obtain the time-series data of the individual status of the target patients.

[0070] It should be noted that stroke patients need to have their daily activity levels calculated to determine their energy expenditure, combined with their dietary status and metabolic indicators, to assess their nutritional needs and whether their current nutritional intake is excessive or insufficient. Therefore, target monitoring devices collect real-time data on the target patient's limb activity, heart rate, and exercise frequency to calculate energy consumption; the types, volumes, and cooking methods of the target patient's daily diet are determined to calculate dietary status; and finally, daily real-time metabolic indicators, including 15 metabolic indicators such as blood glucose, prealbumin, and blood urea nitrogen, are recorded in the hospital database to construct a patient-specific biochemical time-series database.

[0071] Furthermore, in a preferred embodiment of the present invention, the step of generating a nutrition optimization plan based on the quantitative deviation index of the target patient and in conjunction with a reinforcement learning algorithm specifically includes:

[0072] Within the big data network, the initial nutritional supplementation plan corresponding to different recovery stages of the target patient is obtained and marked as a nutritional optimization plan with undetermined content.

[0073] A reinforcement learning framework is constructed. Within the framework, based on the quantitative deviation index of the target patient, the content weight of the nutritional optimization plan with undetermined content is calculated, and the positive and negative values ​​of the quantitative deviation index are analyzed.

[0074] If the quantification deviation index is positive, a positive reward mechanism is triggered, wherein the positive reward mechanism sets the nutritional supplementation plan in the nutritional optimization plan with undetermined content as a priority;

[0075] If the quantification deviation index is negative, a negative penalty mechanism is triggered, wherein the negative penalty mechanism sets the nutritional conditioning scheme in the nutritional optimization scheme with undetermined content as a priority.

[0076] Based on the quantitative deviation index, combined with the content weight calculation of the nutritional optimization scheme to be determined, the nutritional optimization scheme to be analyzed is output. The contraindicated nutritional supplements for the target patients are identified through the hospital database. Based on the contraindicated nutritional supplements for the target patients, the corresponding ones are removed from the nutritional optimization scheme to be analyzed, and the target nutritional optimization scheme is obtained.

[0077] It should be noted that the quantified deviation index is currently known. Since the quantified deviation index is used to provide accurate conditions for adjusting the nutrient content within the nutritional balance plan during the process of outputting the plan to patients, the first step is to obtain the nutritional optimization plan with the content to be determined. This nutritional optimization plan only contains the nutrients that the patient needs to supplement, and the content needs to be determined based on the quantified deviation index. The quantified deviation index includes positive and negative values. A positive value indicates that the patient needs nutritional supplementation, meaning that the current nutritional supplementation is insufficient and should be prioritized; conversely, a negative value indicates that nutritional adjustment is needed, meaning that the current nutritional intake is excessive and should be supplemented in batches or even with known excess nutrients. During the output of the nutritional optimization plan, it is also necessary to identify the patient's contraindicated nutrients to prevent potential risks, and finally output the target nutritional optimization plan.

[0078] Figure 2 shows a flowchart of the method for calculating the real-time nutritional balance and quantitative deviation index of the target patient, including the following steps:

[0079] S202: Using a temporal convolutional network, extract the temporal features of the target patient's personal status time series data, construct a metabolic analysis model based on the temporal features, calculate the target patient's real-time nutritional balance based on the metabolic analysis model, and generate a quantitative deviation index.

[0080] S204: Introduce a dual-channel metabolic analysis training model, and use Bayesian update to iteratively optimize the model parameters of the dual-channel metabolic analysis training model to obtain an optimized dual-channel metabolic analysis training model;

[0081] S206: Input the target patient's personal status fusion features into the optimized dual-channel metabolic analysis training model for training, and generate the target patient's real-time nutritional balance.

[0082] S208: Calculate the quantitative deviation index of the target patient based on the real-time nutritional balance of the target patient.

[0083] Furthermore, in a preferred embodiment of the present invention, the step of extracting temporal features of the target patient's personal status time-series data through a temporal convolutional network, constructing a metabolic analysis model based on the temporal features, calculating the target patient's real-time nutritional balance based on the metabolic analysis model, and simultaneously generating a quantitative deviation index, specifically involves:

[0084] A data processing terminal is obtained, a temporal convolutional network is introduced into the data processing terminal, and the time-series data of the target patient's personal status is imported into the data processing terminal.

[0085] Temporal convolutional networks are used to perform temporal convolution on the time-series data of the target patient's personal status. The temporal convolution is used to extract time-series features corresponding to different time periods on the data, generating the diurnal energy consumption fluctuation curve, postprandial blood glucose response curve, and nitrogen balance trend curve of the target patient's personal status time-series data.

[0086] The diurnal energy consumption fluctuation curve, postprandial blood glucose response curve, and nitrogen balance trend curve of the target patient's personal status time series data are identified as the time series characteristics of the target patient's personal status.

[0087] Wavelet denoising and normalization are performed on the temporal features of the target patient's personal state. At the same time, multiple interpolation is used to fill in the missing values ​​in the temporal features of the target patient's personal state, so as to obtain the fused features of the target patient's personal state.

[0088] It should be noted that the data processing terminal is the modeling terminal device used to construct models for analyzing patient personal status data. First, a temporal convolutional network is used to extract temporal features from the target patient's personal status time-series data to construct a metabolic analysis model. Modeling requires training with feature data; the extracted temporal features include diurnal energy consumption fluctuation curves, postprandial blood glucose response curves, and nitrogen balance trend curves. The extracted features may contain fluctuations and noise, requiring preprocessing, including denoising and missing value interpolation. Therefore, wavelet denoising and normalization are performed on the target patient's personal status time-series features, and multiple interpolation is used to fill in missing values ​​within the target patient's personal status time-series features, resulting in the target patient's personal status fusion features.

[0089] Furthermore, in a preferred embodiment of the present invention, the introduction of a dual-channel metabolic analysis training model, and the iteration optimization of the model parameters of the dual-channel metabolic analysis training model through a Bayesian update method to obtain an optimized dual-channel metabolic analysis training model, specifically involves:

[0090] In the dual-channel metabolic training model, a Bayesian optimization framework is constructed, wherein the Bayesian optimization framework sets latent variable samples for different channels of the dual-channel metabolic training model.

[0091] Energy consumption latent variable samples were set in channel 1 of the dual-channel metabolic training model, and trace element metabolism latent variable samples were set in channel 2 of the dual-channel metabolic training model.

[0092] Within the Bayesian optimization framework, the likelihood functions of the latent variable samples in channel 1 and channel 2 are calculated respectively and labeled as the target likelihood function. The dual-channel Bayesian update training is then performed in conjunction with the target likelihood function.

[0093] Specifically, conjugate prior analytical training is performed in channel 1, and approximate Gaussian distribution training is performed in channel 2. After training, the distribution convergence of samples in channel 1 and channel 2 is calculated. When the distribution convergence reaches a preset range, the dual-channel Bayesian update training is stopped, and the optimized dual-channel metabolic analysis training model is output.

[0094] It should be noted that the Bayesian optimization framework is a framework for updating model parameters using the Bayesian algorithm. The purpose of the Bayesian algorithm is to enhance the calibration capability of key parameters, such as the calibration capability of latent variables like trace element metabolism and nutrient consumption. By setting sample data, the model is optimized using Bayesian methods to ensure that the analysis results in the subsequent dual-channel metabolic analysis do not exhibit excessive bias, thereby enhancing accuracy. Different channels represent different variables. The likelihood functions of the latent variable samples in channel 1 and channel 2 are calculated. The likelihood functions serve as the conditional functions for Bayesian updates, and the model is trained independently for both channels, using conjugate prior analytical training and Gaussian distribution training, with the aim of improving the accuracy of patient nutritional balance assessment in clinical practice.

[0095] Furthermore, in a preferred embodiment of the present invention, the step of inputting the target patient's personal state fusion features into the optimized dual-channel metabolic analysis training model for training, and generating the target patient's real-time nutritional balance, specifically involves:

[0096] In the optimized dual-channel metabolic analysis training model, channel 1 is the model analysis channel for analyzing macro-nutritional balance. Channel 1 is connected to the BiLSTM layer in the optimized dual-channel metabolic analysis training model. The BiLSTM layer is used to calculate the real-time basal energy consumption value of the target patient under the current dietary state by combining the target patient's personal state fusion features.

[0097] After obtaining the target patient's basal energy expenditure value from channel 1, the standard basal energy expenditure range of the target patient under the current dietary status is calculated by optimizing the dual-channel metabolic analysis training model.

[0098] If the real-time basal energy expenditure value is greater than the standard basal energy expenditure range, there is a risk of negative balance in the target patient output within the optimized dual-channel metabolic analysis training model. If the real-time basal energy expenditure value is less than the standard basal energy expenditure range, there is a risk of overfeeding in the target patient output within the optimized dual-channel metabolic analysis training model.

[0099] In optimizing the dual-channel metabolic analysis training model, channel 2 is the model analysis channel for performing micro-metabolic analysis on the fusion characteristics of the target patient's individual status;

[0100] Within channel 2, a nutritional metabolism map is constructed by analyzing the metabolic index characteristics of the target patient within the fusion features of the target patient's personal status. In the nutritional metabolism map, nodes represent the trace elements absorbed by the target patient, and edges represent the biochemical reaction relationships of different trace elements.

[0101] The nutrient metabolism graph is iteratively updated using a graph neural network. The iterative update of the graph neural network involves updating the content of trace elements represented by nodes in real time based on the biochemical reaction relationship of the edges, and calculating the metabolic efficiency and metabolic pathway of trace elements based on the content.

[0102] Based on the metabolic efficiency and pathways of trace elements, the target patient's nutritional balance is updated in real time in the optimized dual-channel metabolic analysis training model. That is, the target patient's real-time nutritional balance is calculated by combining the metabolic efficiency and pathways of trace elements with the target patient's real-time basal energy consumption under the current dietary state.

[0103] It should be noted that Channel 1 is a model analysis channel for macro-nutrient balance analysis, specifically analyzing the energy intake and energy expenditure due to activity in the patient's dietary state. Channel 1 connects to a BiLSTM layer within the optimized dual-channel metabolic analysis training model. This BiLSTM layer is used to calculate the real-time basal energy expenditure of the target patient under the current dietary state, combining the individual status fusion features of the target patient. The patient's current state is determined based on energy expenditure, categorized as either overfeeding or negative balance. Different states require different treatments; for example, overfeeding requires stopping high-nutrient intake, while negative balance indicates insufficient energy intake, necessitating increased energy intake. Channel 2 is a model analysis channel for micro-metabolic analysis based on the individual status fusion features of the target patient. The content of ingested micronutrients affects human health. Micronutrients include, but are not limited to, leucine, glutamine, and vitamin D. Through graph neural network analysis, the metabolic efficiency and metabolic pathway analysis within the patient can be generated, thereby updating the patient's current nutritional balance by combining energy intake and expenditure with micronutrient metabolism to obtain an accurate patient application balance.

[0104] Furthermore, in a preferred embodiment of the present invention, the step of calculating the quantitative deviation index of the target patient based on the real-time nutritional balance of the target patient specifically involves:

[0105] Within the big data network, the standard nutritional balance corresponding to different rehabilitation stages of the target patient is obtained, and the standard nutritional balance weight of different rehabilitation stages is calculated.

[0106] The target patients are in different rehabilitation stages, including the acute phase, the muscle strength reconstruction phase, and the functional recovery phase.

[0107] Based on the standard nutritional balance weights for different rehabilitation stages, and combined with the real-time nutritional balance of the target patient, the difference between the real-time nutritional balance of the target patient and the standard value at different rehabilitation stages is calculated and calibrated as the real-time difference of nutritional balance.

[0108] Based on the real-time difference in nutritional balance at different stages of rehabilitation, an index for correcting nutritional balance, namely the quantitative deviation index, is calculated.

[0109] It should be noted that patients are in different stages of recovery, and the corresponding nutritional balance varies at each stage. Therefore, standard nutritional balance weights are calculated for each recovery stage. By mapping the target patient's real-time nutritional balance to the standard nutritional balance weights for different recovery stages, the difference between the patient's current nutritional balance and the standard value at each recovery stage can be immediately obtained. This yields a quantitative deviation index, which is used to provide accurate conditions for adjusting the nutritional content within the nutritional balance plan during the process of outputting nutritional balance plans for patients.

[0110] As shown in Figure 3, a second aspect of the present invention also provides a real-time assessment and optimization system for nutritional balance in ICU stroke patients based on artificial intelligence. The assessment and optimization system includes a memory 31 and a processor 32. The memory 31 stores assessment and optimization methods. When the assessment and optimization methods are executed by the processor 32, the following steps are implemented:

[0111] Wearable activity monitoring devices for ICU stroke patients are used to collect real-time personal status data of ICU stroke patients, and the personal status data of ICU stroke patients are combined to obtain time-series data of the target patient's personal status.

[0112] Temporal features of the target patient's individual status time series data are extracted using a temporal convolutional network, and a metabolic analysis model is constructed based on the temporal features. The real-time nutritional balance of the target patient is calculated based on the metabolic analysis model, and a quantitative deviation index is generated.

[0113] Nutrition optimization plans are generated based on the quantitative deviation index of the target patients and combined with reinforcement learning algorithms.

[0114] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence, characterized in that, Includes the following steps: Wearable activity monitoring devices for ICU stroke patients are used to collect real-time personal status data of ICU stroke patients. This data is then combined to obtain time-series data of the target patient's personal status. A temporal convolutional network is used to extract the temporal features of the target patient's personal status time-series data, and a metabolic analysis model is constructed based on these features. The real-time nutritional balance of the target patient is calculated based on this metabolic analysis model, and a quantitative deviation index is generated. Specifically, a data processing terminal is accessed, a temporal convolutional network is introduced into the data processing terminal, and the target patient's personal status time-series data is imported into the data processing terminal. The time-series data of the target patient's personal status is then subjected to temporal convolution through the temporal convolutional network. The temporal convolution involves extracting temporal features corresponding to different time periods from the data, generating diurnal energy consumption fluctuation curves, postprandial blood glucose response curves, and nitrogen balance trend curves of the target patient's personal status temporal data. These curves are then labeled as the target patient's personal status temporal features. Wavelet denoising and normalization are applied to these features, and multiple interpolation is used to fill in missing values, resulting in fused features of the target patient's personal status. A dual-channel metabolic analysis training model is introduced, and its parameters are iteratively optimized using a Bayesian update method. An optimized dual-channel metabolic analysis training model is obtained; the target patient's personal state fusion features are input into the optimized dual-channel metabolic analysis training model for training, generating the target patient's real-time nutritional balance; based on the target patient's real-time nutritional balance, the target patient's quantitative deviation index is calculated; wherein, the step of inputting the target patient's personal state fusion features into the optimized dual-channel metabolic analysis training model for training to generate the target patient's real-time nutritional balance specifically involves: in the optimized dual-channel metabolic analysis training model, channel 1 is the model analysis channel for analyzing macroscopic nutritional balance, and channel 1 is connected to a BiLSTM layer within the optimized dual-channel metabolic analysis training model, the BiLSTM layer being used to combine the target patient's personal state fusion features. The system calculates the real-time basal energy expenditure (BAE) of the target patient under the current dietary condition. After obtaining the BAE from Channel 1, the system optimizes the dual-channel metabolic analysis training model to calculate the standard BAE range for the target patient under the current dietary condition. If the real-time BAE is greater than the standard BAE range, the optimized dual-channel metabolic analysis training model indicates a risk of negative energy balance for the target patient. If the real-time BAE is less than the standard BAE range, the optimized dual-channel metabolic analysis training model indicates a risk of overfeeding for the target patient. In the optimized dual-channel metabolic analysis training model, Channel 2 is a model analysis channel for micro-metabolic analysis based on the integrated characteristics of the target patient's individual condition.Within channel 2, a nutritional metabolism map is constructed by analyzing the metabolic indicators of the target patient within the fusion features of the target patient's personal status. Nodes in the nutritional metabolism map represent trace elements absorbed by the target patient, and edges represent the biochemical reaction relationships of different trace elements. The nutritional metabolism map is iteratively updated using a graph neural network. This iterative update involves real-time updating of the content of trace elements represented by nodes based on the biochemical reaction relationships of the edges, and calculating the metabolic efficiency and metabolic pathway of the trace elements based on the content. Based on the metabolic efficiency and metabolic pathway of the trace elements, the target patient's real-time nutritional balance is updated in the optimized dual-channel metabolic analysis training model. Specifically, the real-time nutritional balance of the target patient is calculated by combining the metabolic efficiency and metabolic pathway of the trace elements with the target patient's real-time basal energy consumption under the current dietary state. Finally, a nutritional optimization plan is generated based on the target patient's quantitative deviation index using a reinforcement learning algorithm.

2. The method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence as described in claim 1, characterized in that, The process involves using a wearable activity monitoring device for ICU stroke patients to collect real-time personal status data and combining this data to obtain time-series data of the target patient's personal status. Specifically, this includes: identifying ICU stroke patients requiring nutritional balance assessment and designating them as target patients; acquiring a wearable device for detecting limb activity, heart rate, and movement frequency in ICU stroke patients and designating it as the target monitoring device; wearing and activating the target monitoring device on the target patient; pre-setting a standard monitoring time; and collecting real-time data on the target patient's limb activity, heart rate, and movement frequency using the target monitoring device within the standard monitoring time. An energy consumption algorithm is then set within the target monitoring device. Based on this algorithm, combined with the real-time collected data on the target patient's limb activity, heart rate, and movement frequency, the real-time activity energy consumption of the target patient is obtained. A big data network is acquired to determine the types, volumes, and cooking methods of the target patient's daily diet. A nutrition database is retrieved from the big data network, which includes the protein, fat, and carbohydrate content of different types of daily diets in different volumes and cooking methods, collectively referred to as dietary status. Based on the nutrition database, the dietary status corresponding to the target patient is determined and labeled as the target dietary status. The metabolic indicators of the target patient are obtained through the hospital's database. The metabolic indicators, dietary status, and real-time activity energy consumption of the target patient are spatiotemporally aligned to obtain the time-series data of the target patient's personal status.

3. The method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence as described in claim 1, characterized in that, The introduced dual-channel metabolic analysis training model is optimized through Bayesian updates by iteratively optimizing the model parameters. Specifically, a Bayesian optimization framework is constructed within the dual-channel metabolic training model, where latent variable samples are set for different channels of the model. Energy consumption latent variable samples are set in channel 1, and trace element metabolism latent variable samples are set in channel 2. Within the Bayesian optimization framework, the likelihood functions of the latent variable samples in channel 1 and channel 2 are calculated and labeled as target likelihood functions. Dual-channel Bayesian update training is then performed using these target likelihood functions. Conjugate prior analytical training is conducted in channel 1, and approximate Gaussian distribution training is conducted in channel 2. After training, the distribution convergence of samples in channel 1 and channel 2 is calculated. When the distribution convergence reaches a preset range, the dual-channel Bayesian update training is stopped, and the optimized dual-channel metabolic analysis training model is output.

4. The method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence as described in claim 1, characterized in that, The process of calculating the quantitative deviation index of the target patient based on the real-time nutritional balance of the target patient involves the following steps: Within a big data network, the standard nutritional balance corresponding to different rehabilitation stages of the target patient is obtained, and the standard nutritional balance weights for different rehabilitation stages are calculated. These different rehabilitation stages include the acute phase, muscle strength reconstruction phase, and functional recovery phase. Based on the standard nutritional balance weights for different rehabilitation stages, and combined with the target patient's real-time nutritional balance, the difference between the real-time nutritional balance of the target patient and the standard value in different rehabilitation stages is calculated and designated as the real-time difference in nutritional balance. Based on the real-time difference in nutritional balance for different rehabilitation stages, an index for correcting nutritional balance, i.e., the quantitative deviation index, is calculated.

5. The method for real-time assessment and optimization of nutritional balance in ICU stroke patients based on artificial intelligence as described in claim 1, characterized in that, The process of generating a nutritional optimization plan based on the quantitative deviation index of the target patient and in combination with a reinforcement learning algorithm is as follows: within a big data network, the initial nutritional supplementation plan corresponding to different recovery stages of the target patient is obtained and labeled as a nutritional optimization plan with undetermined content. A reinforcement learning framework is constructed. Within this framework, based on the quantitative deviation index of the target patient, the content weights of the nutritional optimization plans with undetermined content are calculated, and the positive and negative values ​​of the quantitative deviation index are analyzed. If the quantitative deviation index is positive, a positive reward mechanism is triggered, whereby the nutritional supplementation plan in the nutritional optimization plan with undetermined content is prioritized. If the quantitative deviation index is negative, a negative penalty mechanism is triggered, whereby the nutritional conditioning plan in the nutritional optimization plan with undetermined content is prioritized. Based on the quantitative deviation index and the content weight calculation of the nutritional optimization plans with undetermined content, the nutritional optimization plan to be analyzed is output. Contraindicated nutritional supplements for the target patient are identified through a hospital database, and corresponding supplements are removed from the nutritional optimization plans to be analyzed, resulting in the target nutritional optimization plan.

6. A real-time assessment and optimization system for nutritional balance in ICU stroke patients based on artificial intelligence, characterized in that: The evaluation and optimization system includes a memory and a processor. The memory stores an evaluation and optimization method program. When the evaluation and optimization method program is executed by the processor, it implements the evaluation and optimization method steps as described in any one of claims 1-5.

Images