Intelligent weight loss dynamic matching intervention method and decision-making system for obese patients in hospital based on multi-modal data fusion

Abstract

The invention discloses a hospital obese patient intelligent weight loss dynamic matching intervention method and decision making system based on multi-modal data fusion, and relates to the field of medical health, and the method comprises the following steps: S1, carrying out the real-time collection and fusion of multi-modal data; s2, dynamic metabolism modeling and plateau period breakthrough; s3, generating a hierarchical intervention strategy; s4, performing psychological-physiological linkage intervention; and S5, performing real-time safety control and feedback optimization. By collecting multi-dimensional data, constructing the model, accurately predicting the metabolic rate, the metabolic plateau period and the sudden eating tendency, dynamically adjusting the exercise intensity and the diet scheme according to the conditions of different patients, triggering a low-intensity exercise mode for the obese patients with cardiac insufficiency and calibrating diet, ensuring the exercise safety, and improving the exercise efficiency. Operation suggestions are generated on the basis of the standard, precise preoperative evaluation is conducted, postoperative rehabilitation is optimized in combination with nerve stimulation data, dangerous postures are monitored, the sport injury risk is reduced, and personalized, safe and efficient intelligent weight reduction intervention for obese patients is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical health, in particular to a hospital obese patient intelligent weight loss dynamic matching intervention method and decision system based on multi-modal data fusion. BACKGROUND

[0002] The obesity problem has evolved into a global major public health challenge, not only seriously affecting the physical health of individuals, leading to a significant increase in the incidence of diabetes, cardiovascular disease, hypertension and other chronic diseases, but also bringing a heavy burden to mental health, quality of life and social and economic development.

[0003] For example, the intelligent dynamic monitoring and active nutrition intervention method for obese people disclosed in CN113689934A includes the following steps: S1, overweight and obese people screening and health risk assessment; S2, intervention scheme; S3, monitoring; S4, nutrition intervention; S5, intervention result assessment.

[0004] In the prior art, when performing weight loss intervention on obese patients, there are dual limitations of single-modal data dependence and static intervention mechanism, and only general diet and exercise programs are developed based on basic physiological indicators, without fully considering individual metabolic differences and dynamic changes in metabolic state during weight loss, which cannot cope with the plateau problem caused by the decline in basal metabolic rate during weight loss, and it is difficult to maintain the weight loss effect in the long term. At the same time, obesity can cause different problems in the patient's body, and some obese patients with heart dysfunction have errors in exercise risk assessment, making it difficult to achieve precise and dynamic whole-cycle weight loss management. SUMMARY

[0005] The present application aims to provide a hospital obese patient intelligent weight loss dynamic matching intervention method and decision system based on multi-modal data fusion to solve the problem of not fully considering individual metabolic differences and dynamic changes in metabolic state during weight loss when performing weight loss intervention on obese patients, which makes it difficult to achieve precise and dynamic whole-cycle weight loss management.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: a hospital obese patient intelligent weight loss dynamic matching intervention method and decision system based on multi-modal data fusion, including the following steps:

[0007] S1, real-time acquisition and fusion of multi-modal data: collect and integrate multi-source data, construct a metabolic rate dynamic prediction model, and perform cross-modal correlation analysis;

[0008] S2, dynamic metabolic modeling and plateau breakthrough: perform cardiac function coupling analysis, optimize exercise intensity threshold, and predict metabolic plateau;

[0009] S3, hierarchical intervention strategy generation: for obese patients with heart failure, trigger low-intensity exercise mode and calibrate diet plan, generate surgery recommendations, conduct AI preoperative assessment, and combine neural stimulation system data for postoperative rehabilitation optimization;

[0010] S4, psychological-physiological interaction intervention: integrate voice emotion analysis results, cortisol levels and blood glucose fluctuation data to predict binge eating tendency and trigger intervention when warning;

[0011] S5, real-time safety control and feedback optimization: monitor exercise injury risk, dynamically optimize drug and nutrition synergy, and interface with hospital HIS system to control and feedback on the weight loss process of obese patients.

[0012] Preferably, in step S1, the real-time acquisition and fusion of multi-modal data includes the following steps:

[0013] S11, multi-source data dimension integration: collect multi-dimensional data including physiological indicators, exercise behavior data, psychological state data, medical image data and environmental parameters, and integrate multi-source data;

[0014] S12, cross-modal fusion: build a metabolic rate dynamic prediction model based on LSTM neural network, and combine multi-modal attention mechanism to perform cross-modal correlation analysis on physiological indicators.

[0015] Preferably, in step S11, the physiological indicator data acquisition includes basic metabolic data such as BMI, body fat rate, blood glucose fluctuation and cortisol level; exercise behavior data includes exercise trajectory and posture data; psychological state data includes psychological characteristics through emotional log text, voice emotion analysis and facial micro-expression recognition; medical image data includes metabolic characteristics of STN target area of heart and analysis of visceral fat distribution; environmental parameters include environmental data such as light intensity, temperature and humidity.

[0016] Preferably, in step S2, dynamic metabolic modeling and plateau breakthrough includes the following steps:

[0017] S21, cardiac function coupling analysis: based on echocardiogram segmentation technology, identify the metabolic state of STN target area, establish a cardiac energy consumption model, and optimize the exercise intensity threshold;

[0018] S22, metabolic plateau prediction: collect continuous time series data, use LSTM neural network to predict metabolic stagnation in advance, and execute heat cycle strategy after triggering the warning.

[0019] Preferably, in step S3, hierarchical intervention strategy generation includes the following steps:

[0020] S31, metabolic adaptive regulation: for patients with poor heart function, automatically trigger low-intensity exercise mode, and calibrate the diet plan.

[0021] S32, surgical decision support: build knowledge graph, automatically generate surgical recommendations according to patient physical condition, and call AI for preoperative assessment.

[0022] S33, postoperative rehabilitation optimization: combined with the data collected by the eight-point nerve stimulation system, dynamically adjust the postoperative electrical stimulation frequency, and optimize the nutrition supply strategy.

[0023] Preferably, in step S5, real-time safety control and feedback optimization includes the following steps:

[0024] S51, exercise injury warning and protection: deploy 3D posture recognition model, monitor dangerous posture in real time, terminate training and push correction guidance;

[0025] S52, drug-nutrition synergistic regulation: real-time analysis of the interaction between GLP-1 receptor agonists and high-protein diet, and dynamic adjustment of protein intake ratio;

[0026] S53, medical data collaboration: interface with hospital HIS system to obtain test data in real time, and trigger multidisciplinary consultation mechanism when abnormal fluctuations occur;

[0027] S54, multi-dimensional therapeutic effect index monitoring: integrate core indicators to generate dynamic evaluation report and adjust clinical plan.

[0028] The hospital obesity patient intelligent weight loss dynamic matching intervention decision system based on multi-modal data fusion includes a multi-modal data processing unit, a dynamic modeling unit, a hierarchical intervention module and a safety control unit;

[0029] The multi-modal data processing unit is responsible for data collection, cleaning, feature extraction and cross-modal fusion;

[0030] The dynamic modeling unit is used to build an LSTM neural network model to predict metabolic rate, identify plateau, and generate exercise intensity recommendations combined with heart function analysis;

[0031] The hierarchical intervention module is used for dynamic intervention of obese patients from metabolic regulation, surgical evaluation to postoperative rehabilitation;

[0032] The hierarchical intervention module includes a metabolic regulation module, a surgical decision module, a collaborative rehabilitation module and a psychological intervention module;

[0033] The metabolic regulation module is used to detect metabolic plateau, automatically generate daily intake diet plan, automatically calibrate diet structure, analyze muscle fatigue degree according to electromyographic signal, and dynamically adjust resistance training weight;

[0034] The surgery decision module generates surgery suggestions based on patient data and initiates AI preoperative assessment to predict postoperative complication probability and assist in anesthesia scheme formulation.

[0035] The synergistic rehabilitation module dynamically adjusts the frequency and pulse width of electrical stimulation based on the electromyographic signals collected by the eight-contact nerve stimulation system, optimizes the protein intake ratio in combination with postoperative blood glucose fluctuation data.

[0036] The psychological intervention module is used to fuse voice tone features, facial micro-expressions, and blood glucose fluctuations to identify multi-modal emotions, establish a binge eating warning model, and conduct binge eating tendency warning. When the warning is triggered, it pushes the mindfulness diet training audio and sends support reminders through the family end APP.

[0037] The safety control unit is used to monitor exercise conditions and nutrient intake in real time and adjust exercise intensity and dietary plans.

[0038] Preferably, the multi-modal data processing unit includes a cross-dimensional data access module, a feature extraction module, and a multi-modal attention fusion module.

[0039] The cross-dimensional data access module is used to collect multi-source data including physiological indicators, exercise data, psychological state, medical images, and environmental parameters in real time, and process the collected multi-source data.

[0040] The feature extraction module is used to analyze continuous metabolic data, identify the trend of basal metabolic rate changes, and extract metabolic plateau characteristics.

[0041] The multi-modal attention fusion module integrates text emotion logs, image expression features, and cortisol levels through an attention mechanism to generate a comprehensive feature vector of emotions and stress.

[0042] Preferably, the dynamic modeling unit includes a metabolic prediction and warning module and a simulation module.

[0043] The metabolic prediction and warning module is used to input time series data, build an LSTM neural network model, predict metabolic rate changes, trigger plateau warnings in advance, use neural networks to segment echocardiograms, and adjust exercise intensity thresholds.

[0044] The simulation module is used to simulate energy consumption trajectories of different dietary plans and exercise combinations based on measured resting energy consumption data from respiratory metabolism to predict body fat trends.

[0045] Preferably, the safety control unit includes a motion injury warning module and a dietary control module.

[0046] The motion injury early warning module monitors the dangerous posture in real time by constructing a whole body joint point model, monitors the quadriceps muscle oxygenation level through a near-infrared spectrometer, judges the muscle fatigue degree, and adjusts the motion posture and motion intensity;

[0047] The diet control module is used for establishing an interaction database of the GLP-1 receptor agonist and the high-protein diet, triggering a risk reminder and adjusting a diet scheme when the protein intake exceeds the recommended amount during the medication.

[0048] Compared with the prior art, the present application has the following beneficial effects:

[0049] In the present application, multi-modal data fusion is used to comprehensively collect physiological, motion, psychological and other multi-dimensional data, a model is constructed by combining algorithms, metabolic rate, metabolic plateau and binge eating tendency are accurately predicted, reliable basis is provided for intervention, motion intensity and diet scheme are dynamically adjusted according to different patient conditions, low-intensity motion mode is triggered for obese patients combined with cardiac insufficiency, and the diet is calibrated, motion safety is ensured, operation suggestions are generated based on the standard and accurate preoperative evaluation is carried out, postoperative rehabilitation is optimized by combining neural stimulation data, weight loss effect is improved, motion injury risk is reduced by monitoring dangerous posture, analyzing drug-nutrition interaction and connecting medical systems, abnormal response time is shortened, treatment effect is monitored in multiple dimensions and the scheme is dynamically adjusted, and individualized, safe and efficient intelligent weight loss intervention for obese patients is realized. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of the hospital obese patient intelligent weight loss dynamic matching intervention method based on multi-modal data fusion of the present application;

[0051] Figure 2 A system block diagram of the hospital obese patient intelligent weight loss dynamic matching intervention decision system based on multi-modal data fusion of the present application.

[0052] In the figure: 1, multi-modal data processing unit; 11, cross-dimension data access module; 12, feature extraction module; 13, multi-modal attention fusion module; 2, dynamic modeling unit; 21, metabolic prediction and early warning module; 22, simulation simulation module; 3, hierarchical intervention module; 31, metabolic regulation module; 32, operation decision module; 33, collaborative rehabilitation module; 34, psychological intervention module; 4, safety control unit; 41, motion injury early warning module; 42, diet control module. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.

[0054] Embodiment 1: Reference Figures 1-2 As shown: the present application proposes a hospital obese patient intelligent weight loss dynamic matching intervention method and decision system based on multi-modal data fusion, which realizes precise and dynamic whole-cycle weight loss management by integrating multi-dimensional data, constructing a dynamic model and a hierarchical intervention strategy.

[0055] I. Technical solution.

[0056] (I) Real-time acquisition and fusion of multi-modal data (S1): the data acquisition dimensions include physiological indicators, exercise behavior, psychological state, medical images and environmental parameters; the data fusion is based on a LSTM neural network to construct a metabolic rate dynamic prediction model, and a multi-modal attention mechanism is combined to perform cross-modal correlation analysis on text, images and physiological indicators; a graph neural network is used to construct a multi-modal data correlation graph, taking data features as nodes and causal relationships as edge weights, to improve semantic understanding ability.

[0057] (II) Dynamic metabolic modeling and plateau breakthrough (S2): cardiac function coupling analysis extracts STN target image features through echocardiogram segmentation technology, constructs an "image feature-energy consumption" mapping model, sets a motion intensity threshold based on a cardiac energy consumption model, and dynamically adjusts the intensity; metabolic plateau prediction collects time series data, derives features such as 7-day average heat intake and body fat rate weekly change rate, uses a LSTM neural network to predict metabolic arrest 3 weeks in advance, generates a ±300 kilocalorie fluctuation diet plan after triggering an early warning, and breaks through the metabolic plateau.

[0058] (III) Hierarchical intervention strategy generation (S3): metabolic adaptability regulation triggers a low-intensity exercise mode for obese patients with heart failure (LVEF < 50%), and calibrates the diet plan by connecting to the takeout platform; surgery decision support needs to construct a knowledge graph based on the "Obesity Diagnosis and Treatment Guidelines 2024", generate surgery recommendations, and call the AI preoperative assessment module; postoperative rehabilitation optimization dynamically adjusts the electric stimulation frequency and pulse width by combining the data of the eight-touch nerve stimulation system, optimizes the nutrition supply strategy, and improves the body fat loss speed by 15%.

[0059] (IV) Psychological-physiological linkage intervention (S4): Integrate voice emotion analysis results, cortisol levels and blood glucose fluctuation data to build a 72-hour correlation model between stress events and binge eating behavior, and push intervention reminders 24 hours in advance. The family side APP sends support reminders.

[0060] (V) Real-time safety control and feedback optimization (S5): Motion injury warning through the deployment of a 3D posture recognition model to monitor dangerous postures, terminate training and push correction guidance, and reduce the injury rate to below 8%; Drug-nutrition synergistic regulation through analysis of the interaction between GLP-1 receptor agonists and high-protein diets, dynamically adjusting the protein intake ratio (30%-40%), avoiding the risk of drug accumulation, and improving the body fat reduction speed by 15%; Medical data collaboration through interfacing with the hospital HIS system to obtain real-time test data, triggering multidisciplinary consultations when blood glucose fluctuates continuously; Efficacy index monitoring through the integration of body fat rate, metabolic rate, exercise safety and other core indicators to generate dynamic evaluation reports to support clinical program adjustments.

[0061] II. Technical principle analysis

[0062] (I) Multi-modal data fusion principle includes: cross-modal association captures metabolic patterns in time series data through LSTM neural networks, combines multi-modal attention mechanisms, and gives different weights to different data to achieve feature complementation; Graph neural network modeling treats physiological, imaging, psychological and other data as graph nodes, and represents the causal relationship between data through edge weights to build a semantic network and achieve deep semantic understanding of data.

[0063] (II) Dynamic metabolic modeling principle includes: LSTM neural network prediction uses LSTM to handle the long-term and short-term dependencies of time series data, learns the trend of metabolic rate changes, and identifies the metabolic plateau in advance to provide a time window for intervention; Heart function-exercise intensity coupling is based on the metabolic characteristics of the STN target area segmented by echocardiography to establish a heart energy consumption model, which links exercise intensity to myocardial oxygen consumption to achieve individualized exercise threshold setting and reduce the exercise risk of patients with heart failure.

[0064] (III) Graded intervention strategy principle includes: knowledge graph-driven decision-making builds a knowledge graph based on the "Obesity Diagnosis and Treatment Guidelines" to match patient data (BMI, liver function, etc.) with diagnosis and treatment rules to automatically generate surgery recommendations, achieving a combination of standardization and individualization; Neural stimulation and rehabilitation optimization adjusts gastrointestinal peristalsis through an eight-contact neural stimulation system, dynamically adjusts nutrition supply based on blood glucose fluctuation data, and uses the linkage of electrophysiological signals and metabolic indicators to accelerate postoperative rehabilitation.

[0065] (Tetra) Safety control and feedback principle includes: 3D posture recognition and motion protection builds a whole body joint model through computer vision technology, monitors the motion posture in real time, combines with near-infrared spectrometer to monitor muscle oxygenation level, realizes the "monitoring-early warning-adjustment" closed loop of sports injury; Drug-nutrition interaction analysis establishes the interaction database of GLP-1 receptor agonist and high-protein diet, dynamically adjusts the diet scheme based on pharmacokinetics and nutrition principles, avoids drug side effects and nutritional imbalance.

[0066] III. Core innovation points

[0067] 1. Multi-modal data fusion, breaking through the limitation of single mode, integrating physiological, psychological, image and other data, improving the intervention accuracy.

[0068] 2. Dynamic metabolic modeling, through LSTM and cardiac function analysis, real-time tracking of metabolic state, early prediction of plateau and adjustment of strategy.

[0069] 3. Grading and individualized intervention, for heart failure patients, postoperative rehabilitation and other scenes, provide differentiated intervention scheme, taking into account safety and effectiveness.

[0070] 4. Closed-loop safety control, through real-time monitoring, multidisciplinary data collaboration, shortens the abnormal response time, reduces the risk of exercise and drugs.

[0071] Example 2: Refer to Figure 1 As shown: hospital obese patient intelligent weight loss dynamic matching intervention method based on multi-modal data fusion, comprising the following steps:

[0072] S1, real-time acquisition and fusion of multi-modal data: collect and integrate multi-source data, build metabolic rate dynamic prediction model, and perform cross-modal correlation analysis;

[0073] In step S1, the real-time acquisition and fusion of multi-modal data includes the following steps:

[0074] S11, multi-source data dimension integration: collect multi-dimensional data including physiological indicators, motion behavior data, psychological state data, medical image data and environmental parameters, integrate multi-source data, physiological indicator data is monitored by wearable device in real time to obtain heart rate variability (HRV) and electromyographic signal, collect basic metabolism data such as BMI, body fat rate, blood glucose fluctuation and cortisol level; motion behavior data uses acceleration sensor and plantar pressure sensor to obtain motion trajectory and posture data (such as knee joint hyperextension angle), and combines electromyographic signal to analyze muscle fatigue degree; psychological state data analyzes psychological characteristics such as anxiety and binge eating tendency through emotion log text, voice emotion analysis (intonation, speech rate) and facial micro-expression recognition; medical image data uses echocardiogram segmentation technology (Dice coefficient 0.8269) to extract cardiac STN target area metabolism characteristics, and combines abdominal CT image to analyze visceral fat distribution; environmental parameters collect environmental data such as light intensity, temperature and humidity, and analyze the influence of environmental parameters on exercise tolerance (such as automatically reducing exercise intensity by 15% in high temperature environment);

[0075] S12, cross-modal fusion: based on LSTM neural network, a metabolic rate dynamic prediction model is constructed, multi-modal attention mechanism is combined, cross-modal correlation analysis is performed on text form emotion log, image form expression data and physiological indicators such as cortisol level, graph neural network (GNN) is used to construct multi-modal data correlation graph, physiological indicators, image features, psychological state and the like are taken as nodes, and the causal relationship between data is represented by edge weight, so as to improve the semantic understanding ability of data fusion.

[0076] S2, dynamic metabolism modeling and plateau breakthrough: cardiac function coupling analysis is performed, exercise intensity threshold is optimized, and metabolic plateau is predicted;

[0077] In step S2, dynamic metabolism modeling and plateau breakthrough includes the following steps:

[0078] S21, cardiac function coupling analysis: the echocardiogram is automatically segmented, the image features of the thalamic subnucleus (STN) target area are extracted, the segmentation accuracy reaches Dice coefficient 0.8269, 12 metabolism related parameters such as blood flow velocity and myocardial thickness of the STN target area are quantified by myocardial echocardiography technology, a “image feature-energy consumption” mapping model is constructed, patient echocardiogram image data (frame frequency ≥ 30 fps) and left ventricular ejection fraction (LVEF) indicators are input, based on the clinical statistical data of STN target area metabolism characteristics and myocardial oxygen consumption, a cardiac energy consumption model is established, and safe exercise intensity threshold is set for different patients according to the cardiac energy consumption model: when LVEF < 50%, the exercise intensity threshold is set to 60% of the maximum heart rate (age); when the oxygenation index < 50%, the intensity is automatically reduced by 10% by real-time monitoring of myocardial oxygenation level by near-infrared spectrometer (NIRS), and the exercise risk of patients with cardiac insufficiency is reduced.

[0079] S22, Metabolic plateau prediction: Collect continuous time series data, including daily dietary calories, exercise consumption (calculated by acceleration sensor), body fat rate (measured every week), basal metabolic rate (BMR, measured every month), and derive dynamic characteristics such as 7-day average calorie intake, body fat rate weekly change rate, etc. Based on LSTM neural network, 3 weeks in advance to warn about metabolic stagnation. After the warning is triggered, automatically generate a diet plan with a fluctuation of ±300 kcal per day (such as 1800 kcal one day and 2100 kcal the next day alternately).

[0080] S3, hierarchical intervention strategy generation: For obese patients with heart dysfunction, trigger low-intensity exercise mode and calibrate diet plan, generate surgery recommendations, conduct AI preoperative assessment, and optimize postoperative rehabilitation combined with neural stimulation system data.

[0081] In step S3, hierarchical intervention strategy generation includes the following steps:

[0082] S31, metabolic adaptive regulation: For patients with obesity combined with heart dysfunction and left ventricular ejection fraction less than 50%, automatically trigger low-intensity exercise mode. At the same time, interface with takeout platform data to calibrate the diet plan, with an error control within ±50 kcal / day.

[0083] S32, surgical decision support: According to the "Obesity Diagnosis and Treatment Guidelines 2024", build a knowledge graph. When the patient's BMI is greater than or equal to 35 and liver function is abnormal, automatically generate surgery recommendations such as sleeve gastrectomy, and call the AI preoperative assessment module, which has a target positioning error of less than 0.5 millimeters.

[0084] S33, postoperative rehabilitation optimization: Combined with the data collected by the eight-point neural stimulation system, dynamically adjust the postoperative electric stimulation frequency, which can be adjusted within the range of 1-130 Hz, and optimize the nutrition supply strategy, ultimately increasing the body fat loss rate by 15%.

[0085] S4, psychological-physiological linked intervention: Integrate voice emotion analysis results, cortisol levels, and blood glucose fluctuation data to establish a binge eating tendency warning model to predict binge eating tendencies. When the warning is triggered, start cognitive behavioral therapy for intervention. Use the Transformer time attention module to build a 72-hour association model between stress events and binge eating behaviors, and push intervention reminders 24 hours in advance to improve the effectiveness of psychological intervention.

[0086] S5, real-time safety control and feedback optimization: Monitor exercise injury risk, dynamically optimize drug and nutrition synergy, and interface with hospital HIS systems to perform real-time safety control and feedback optimization for obese patients during weight loss, ensuring intervention safety and improving response efficiency.

[0087] In step S5, real-time safety control and feedback optimization includes the following steps:

[0088] S51, sports injury warning and protection: deploy a 3D posture recognition model (accuracy 98.7%), real-time monitor dangerous postures such as knee joint hyperextension angle > 5°, lumbar lordosis angle > 30°, etc., immediately terminate training and push correction guidance, reduce sports injury rate to below 8%;

[0089] S52, drug-nutrition synergistic regulation: real-time analysis of the interaction of GLP-1 receptor agonists and high-protein diet, dynamic adjustment of protein intake ratio (30%-40% optimization), avoid drug accumulation risk, improve body fat loss speed by 15%;

[0090] S53, medical data collaboration: interface with hospital HIS system to obtain test data in real time, when blood glucose fluctuates > 3mmol / L for 3 consecutive days, trigger multidisciplinary consultation mechanism, abnormal state response time is shortened to within 4 hours;

[0091] S54, multi-dimensional therapeutic effect index monitoring: integrate body fat rate, metabolic rate, sports safety, psychological compliance and other core indicators to generate dynamic evaluation report, support clinical program adjustment (such as automatically upgrading intervention intensity when body fat reduction is less than 10% in 3 months).

[0092] Embodiment 3: Refer to Figure 2 As shown: hospital obesity patient intelligent weight loss dynamic matching intervention decision system based on multi-modal data fusion, including multi-modal data processing unit 1, dynamic modeling unit 2, hierarchical intervention module 3 and safety control unit 4;

[0093] The multi-modal data processing unit 1 is responsible for data acquisition, cleaning, feature extraction and cross-modal fusion;

[0094] The multi-modal data processing unit 1 includes a cross-dimensional data access module 11, a feature extraction module 12 and a multi-modal attention fusion module 13;

[0095] The cross-dimensional data access module 11 is used to collect physiological indicators (BMI, body fat rate, blood glucose, cortisol), sports data (acceleration, electromyographic signal, plantar pressure), psychological state (voice emotion features, emotion log text), medical images (echocardiogram segmentation data, Dice coefficient 0.8269) and environmental parameters in real time, filter noise points in sports data, and normalize image data to ensure the format of different modal data is uniform;

[0096] The feature extraction module 12 is used to analyze continuous metabolic data, identify basic metabolic rate (BMR) trend, and extract metabolic plateau features;

[0097] The multi-modal attention fusion module 13 integrates the text emotion log (such as the frequency of the keyword "anxiety"), the image expression feature (the angle of the drooping mouth corner), and the cortisol level through the attention mechanism to generate a "emotion-stress" comprehensive feature vector for binge eating tendency prediction.

[0098] The dynamic modeling unit 2 is used to construct an LSTM neural network model to predict the metabolic rate, identify the plateau period, and generate exercise intensity recommendations in combination with cardiac function analysis.

[0099] The dynamic modeling unit 2 includes a metabolic prediction and warning module 21 and a simulation module 22.

[0100] The metabolic prediction and warning module 21 is used to input time series data such as dietary calories, exercise consumption, and body fat rate, construct an LSTM neural network model, predict the metabolic rate change in the next two weeks, and trigger a plateau period warning three weeks in advance when the predicted metabolic rate decreases by more than 10%. The echocardiogram is segmented using a neural network, and a "myocardial energy consumption-exercise intensity" model is established based on the segmentation results. The exercise intensity threshold is adjusted in combination with the patient's age and body mass index (BMI) to reduce the risk of cardiac load.

[0101] The simulation module 22 is used to simulate the energy consumption trajectory of different dietary plans and exercise combinations based on the resting energy consumption data measured by the respiratory metabolism car to predict the three-month body fat change trend.

[0102] The hierarchical intervention module 3 is used to dynamically intervene in obese patients from metabolic regulation, surgery evaluation to postoperative rehabilitation to optimize weight loss effect and ensure safety.

[0103] The hierarchical intervention module 3 includes a metabolic regulation module 31, a surgery decision module 32, a collaborative rehabilitation module 33, and a psychological intervention module 34.

[0104] The metabolic regulation module 31 is used to detect the metabolic plateau period, automatically generate a daily intake diet plan, automatically calibrate the dietary structure by interfacing with the order data of the takeout platform, analyze muscle fatigue based on electromyographic signals, and dynamically adjust the resistance training weight (such as reducing the squat weight by 5% and automatically reducing the load by 10%) to avoid overtraining.

[0105] The surgery decision module 32 automatically generates a sleeve gastrectomy recommendation based on patient data and starts an AI preoperative evaluation of the double-electrode six-target point control surgery. It integrates echocardiographic cardiac function parameters and pulmonary function test data to predict the probability of postoperative complications (such as pulmonary embolism risk score) and assist in developing an anesthesia plan.

[0106] The synergistic rehabilitation module 33 dynamically adjusts the electric stimulation frequency (1-130 Hz) and pulse width according to the electromyographic signals collected by the eight-contact nerve stimulation system, promotes the recovery of gastrointestinal peristalsis, shortens the postoperative exhaust time, automatically switches the transition time of liquid diet to semi-liquid diet in combination with postoperative blood glucose fluctuation data, and optimizes the protein intake ratio in the 30%-40% interval;

[0107] The psychological intervention module 34 is used for fusing voice tone features, facial micro-expressions, and blood glucose fluctuations, identifying multi-modal emotions, establishing a binge eating warning model, and warning of binge eating tendency. When the warning is triggered, a mindfulness diet training audio (such as 5-minute breathing meditation guidance) is pushed, and a family end APP is sent for support reminders.

[0108] The safety control unit 4 is used for real-time monitoring of exercise conditions and nutrient intake, and adjusting exercise intensity and diet plan;

[0109] The safety control unit 4 includes a sports injury warning module 41 and a diet control module 42;

[0110] The sports injury warning module 41 monitors dangerous postures in real time by constructing a whole-body joint point model. Dangerous postures include knee joint hyperextension angle > 5°, lumbar lordosis angle > 30°, etc. The quadriceps muscle oxygenation level is monitored by a near-infrared spectrometer to determine muscle fatigue, and the exercise posture and exercise intensity are adjusted;

[0111] The diet control module 42 is used to establish an interactive database of GLP-1 receptor agonists and high-protein diets. When the patient's protein intake exceeds the recommended amount (>1.5g / kg / day) during medication, a risk reminder (such as increased risk of pancreatitis) is triggered, and the diet plan is adjusted.

[0112] The application first collects physiological indexes, motion behaviors, psychological states, medical images and environmental parameters and other multi-modal data through wearable devices, sensors and the like, and after cleaning and normalization, a metabolic rate dynamic prediction model and a multi-modal data correlation graph are constructed by using an LSTM neural network combined with a multi-modal attention mechanism and a graph neural network, cross-modal fusion is realized, then, the STN target region features are extracted by automatic segmentation of echocardiogram, a cardiac energy consumption model is constructed combined with myocardial echocardiography technology, a safe exercise intensity threshold is set, at the same time, the metabolic plateau is predicted 3 weeks in advance and a fluctuating diet plan is generated based on the LSTM neural network analysis of time series data, then, for obese patients with heart failure, a low-intensity exercise mode is triggered, a diet error is calibrated by connecting to a take-out platform, surgery suggestions are generated according to the knowledge graph and AI preoperative evaluation is performed, postoperative rehabilitation is optimized combined with neural stimulation system data, a binge eating tendency warning model is established by fusing voice emotion, cortisol and blood glucose data, an association model of stress events and binge eating behaviors is constructed and intervention reminders are pushed in advance, finally, dangerous postures are monitored through a 3D posture recognition model, drug and nutrition interaction is analyzed, test data are obtained by connecting to the hospital HIS system, multi-dimensional efficacy indicators are integrated to generate a dynamic evaluation report, and real-time safety control and feedback optimization are realized.

[0113] Although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to part of the technical features, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A hospital obese patient intelligent weight loss dynamic matching intervention method based on multi-modal data fusion, characterized in that, Includes the following steps: S1. Real-time acquisition and fusion of multimodal data: Acquire and integrate multi-source data, construct a dynamic prediction model for metabolic rate, and conduct cross-modal correlation analysis; S2. Dynamic metabolic modeling and plateau breakthrough: Perform cardiac function coupling analysis, optimize exercise intensity threshold, and predict metabolic plateau phase; S3. Generation of graded intervention strategies: For obese patients with heart failure, low-intensity exercise mode is triggered and diet plan is calibrated to generate surgical suggestions, conduct AI preoperative assessment, and optimize postoperative rehabilitation by combining data from the neurostimulation system. S4. Psychological-physiological linkage intervention: Integrate voice emotion analysis results, cortisol levels, and blood glucose fluctuation data to predict binge eating tendencies and intervene when warnings are triggered; S5. Real-time safety control and feedback optimization: By monitoring the risk of sports injuries, dynamically optimizing the synergistic effect of drugs and nutrition, and connecting with the hospital's HIS system, the weight loss process of obese patients can be controlled and feedback provided.

2. The intelligent weight loss dynamic matching intervention method for obese patients in a hospital based on multi-modal data fusion according to claim 1, characterized in that: In step S1, the real-time acquisition and fusion of multimodal data includes the following steps: S11. Multi-source data integration: Collect multi-dimensional data including physiological indicators, motor behavior data, psychological state data, medical imaging data, and environmental parameters, and integrate the multi-source data. S12, Cross-modal fusion: A dynamic prediction model for metabolic rate is constructed based on LSTM neural network, and combined with multimodal attention mechanism, cross-modal correlation analysis is performed on physiological indicators.

3. The intelligent weight loss dynamic matching intervention method for obese patients in hospitals based on multimodal data fusion according to claim 2, characterized in that: In step S11, physiological indicator data collection includes basal metabolic data such as BMI, body fat percentage, blood glucose fluctuation, and cortisol level; exercise behavior data includes movement trajectory and posture data; psychological state data is obtained through emotional log text, voice emotion analysis, and facial micro-expression recognition of psychological characteristics; medical imaging data is obtained by extracting metabolic characteristics of the cardiac STN target area and analyzing visceral fat distribution; environmental parameter collection includes environmental data such as light intensity, temperature, and humidity.

4. The intelligent weight loss dynamic matching intervention method for obese patients in hospitals based on multimodal data fusion according to claim 3, characterized in that: In step S2, dynamic metabolic modeling and plateau breakthrough include the following steps: S21. Cardiac Function Coupling Analysis: Based on echocardiographic segmentation technology, identify the metabolic state of the STN target area, establish a cardiac energy consumption model, and optimize the exercise intensity threshold. S22. Metabolic plateau prediction: Collect continuous time-series data and use an LSTM neural network to provide early warning of metabolic stagnation. Once the warning is triggered, execute a heat cycling strategy.

5. The intelligent weight loss dynamic matching intervention method for obese patients in hospitals based on multimodal data fusion according to claim 4, characterized in that: In step S3, the generation of the tiered intervention strategy includes the following steps: S31. Metabolic Adaptive Regulation: For patients with heart failure, it automatically triggers a low-intensity exercise mode and calibrates the diet plan. S32. Surgical Decision Support: Construct a knowledge graph to automatically generate surgical suggestions based on the patient's physical condition and call on AI for preoperative assessment; S33. Postoperative rehabilitation optimization: Based on the data collected by the eight-contact nerve stimulation system, the postoperative electrical stimulation frequency is dynamically adjusted, and the nutritional supplementation strategy is optimized.

6. The intelligent weight loss dynamic matching intervention method for obese patients in hospitals based on multimodal data fusion according to claim 5, characterized in that: In step S5, real-time safety control and feedback optimization includes the following steps: S51. Sports Injury Early Warning and Protection: Deploy a 3D posture recognition model to monitor dangerous postures in real time, terminate training, and push corrective guidance. S52. Drug-Nutrition Synergistic Regulation: Real-time analysis of the interaction between GLP-1 receptor agonists and high-protein diets to dynamically adjust the protein intake ratio. S53, Medical Data Collaboration: Connects to the hospital's HIS system to obtain test data in real time, and triggers a multidisciplinary consultation mechanism when abnormal fluctuations occur; S54. Multi-dimensional efficacy indicator monitoring: Integrate core indicators, generate dynamic evaluation reports, and adjust clinical protocols.

7. A hospital-based intelligent weight loss dynamic matching intervention decision-making system for obese patients based on multimodal data fusion, characterized in that: The method for intelligent weight loss dynamic matching intervention for obese patients in hospitals based on multimodal data fusion, as described in any one of claims 1-6, includes a multimodal data processing unit (1), a dynamic modeling unit (2), a graded intervention module (3), and a safety control unit (4). The multimodal data processing unit (1) is responsible for data acquisition, cleaning, feature extraction and cross-modal fusion; The dynamic modeling unit (2) is used to build an LSTM neural network model to predict metabolic rate, identify plateau phase, and generate exercise intensity suggestions in combination with cardiac function analysis. The graded intervention module (3) is used to provide dynamic intervention for obese patients from metabolic regulation, surgical assessment to postoperative rehabilitation; The graded intervention module (3) includes a metabolic regulation module (31), a surgical decision-making module (32), a collaborative rehabilitation module (33), and a psychological intervention module (34); The metabolic regulation module (31) is used to detect the metabolic plateau period, automatically generate a daily intake diet plan, automatically calibrate the diet structure, analyze the degree of muscle fatigue based on electromyography signals, and dynamically adjust the resistance training weight. The surgical decision module (32) generates surgical suggestions based on patient data and initiates AI preoperative assessment to predict the probability of postoperative complications and assist in the formulation of anesthesia plans; the collaborative rehabilitation module (33) dynamically adjusts the frequency and pulse width of electrical stimulation based on the electromyographic signals collected by the eight-point nerve stimulation system, and optimizes the protein intake ratio by combining postoperative blood glucose fluctuation data; the psychological intervention module (34) is used to integrate voice tone features, facial micro-expressions and blood glucose fluctuations to identify multimodal emotions, establish a binge eating warning model, and issue a binge eating tendency warning. When the warning is triggered, it pushes mindfulness eating training audio and links the family member's APP to send support reminders. The safety control unit (4) is used to monitor exercise status and nutrient intake in real time, and adjust exercise intensity and diet plan.

8. The intelligent weight loss dynamic matching intervention decision-making system for obese patients in hospitals based on multimodal data fusion according to claim 7, characterized in that: The multimodal data processing unit (1) includes a cross-dimensional data access module (11), a feature extraction module (12), and a multimodal attention fusion module (13); The cross-dimensional data access module (11) is used to collect multi-source data including physiological indicators, exercise data, psychological state, medical images and environmental parameters in real time, and to process the collected multi-source data. The feature extraction module (12) is used to analyze continuous metabolic data, identify the trend of basal metabolic rate changes, and extract features of metabolic plateau phase. The multimodal attention fusion module (13) integrates text emotion logs, image expression features and cortisol levels in a weighted manner through an attention mechanism to generate a comprehensive feature vector of emotion and stress.

9. The intelligent weight loss dynamic matching intervention decision-making system for obese patients in hospitals based on multimodal data fusion according to claim 7, characterized in that: The dynamic modeling unit (2) includes a metabolic prediction and early warning module (21) and a simulation module (22); The metabolic prediction and early warning module (21) is used to input time series data, construct an LSTM neural network model, predict changes in metabolic rate, trigger a plateau period early warning, use a neural network to segment echocardiograms, and adjust the exercise intensity threshold. The simulation module (22) is used to combine the resting energy consumption data measured by the respiratory metabolism vehicle to simulate the energy consumption trajectory of different diet plans and exercise combinations, and predict the trend of body fat changes.

10. The intelligent weight loss dynamic matching intervention decision-making system for obese patients in hospitals based on multimodal data fusion according to claim 7, characterized in that: The safety control unit (4) includes a sports injury early warning module (41) and a diet control module (42); The sports injury early warning module (41) constructs a whole-body joint model, monitors dangerous postures in real time, monitors the oxygenation level of the quadriceps femoris through a near-infrared spectrometer, determines muscle fatigue, and adjusts the exercise posture and exercise intensity. The diet control module (42) is used to establish an interactive database of GLP-1 receptor agonists and high-protein diets. When protein intake exceeds the recommended amount during medication, a risk warning is triggered and the diet plan is adjusted.

Images