Organ specific aging assessment method integrating low oxygen load and physiological big data
By combining hypoxia experiments with physiological big data, an organ-specific aging assessment model was constructed, which solves the problem of insufficient accuracy in existing assessment methods, and realizes accurate assessment of organ aging and personalized risk warning, supporting health management and clinical intervention.
Patent Information
- Application Number
- CN202511650243.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-03-03
AI Technical Summary
Existing aging assessment methods are not accurate enough, making them difficult to widely apply in primary healthcare and large-scale population health management, and they cannot accurately reflect organ-specific aging states.
By combining hypoxia experiments and living environment monitoring with physiological big data, an organ-specific aging assessment model was constructed using machine learning and deep learning algorithms. Data was collected using a hypoxia chamber and smart wearable devices, and feature extraction and analysis were performed using neural networks and regression models to construct the organ aging assessment model.
It enables a comprehensive and accurate assessment of the aging status of various organs in the human body, improves the model's prediction accuracy, provides personalized risk alerts, and supports health management and clinical intervention.
Smart Images
Figure CN121601228A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical technology, specifically to an organ-specific aging assessment method that integrates hypoxia load and physiological big data. Background Technology
[0002] Aging is a complex, multifactorial process influenced by genetic and environmental factors, resulting in high heterogeneity in the rate and extent of aging, often leading to a discrepancy between biological age and chronological age. Aging is considered the root cause of many chronic diseases, such as cardiovascular disease, neurodegenerative diseases, and diabetes. Accurately assessing the degree of aging, especially organ-specific aging, is crucial for early identification of health risks, development of personalized health management plans, and slowing the aging process.
[0003] Currently, numerous methods exist for aging assessment, including bio-age assessment methods based on epigenetic markers, phenotypic indicators, and multi-omics data to reflect the degree of aging of the whole body or individual organs, and frailty index methods that comprehensively assess overall aging status from multiple dimensions such as physical, cognitive, psychological, sensory, nutritional, social, and disease aspects. However, these traditional methods have many limitations. Some models lack accuracy and are methodologically limited; their correlation with actual health outcomes is limited, making it difficult to directly predict the adverse consequences of aging; many indicators have limited applicability to specific populations, hindering large-scale adoption. For example, traditional bio-age algorithms are often difficult to widely apply in primary healthcare and large-scale population health management due to methodological limitations, weak correlation, and limited generalizability. Therefore, developing a more accurate, comprehensive, and widely applicable aging assessment method is urgently needed. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] To address the shortcomings of existing technologies, this invention provides an organ-specific aging assessment method that integrates hypoxia load and physiological big data to solve the aforementioned problems.
[0006] (II) Technical Solution
[0007] To achieve the above objectives, the present invention provides the following technical solution: an organ-specific aging assessment method integrating hypoxia load and physiological big data, comprising the following steps:
[0008] Data collection: Hypoxia load data are obtained through hypoxia experiments and living environment monitoring, while physiological big data such as physical examination reports and medical records are also collected;
[0009] Data processing and analysis: Cleaning and preprocessing the collected data, and using machine learning and deep learning algorithms for feature extraction and analysis;
[0010] Aging assessment model construction: An organ-specific aging assessment model is constructed by combining neural networks and regression models. It is trained and optimized using labeled datasets, and finally outputs the aging assessment results of each organ, including the organ aging index, the degree of aging difference relative to the average level of peers, and risk warning information.
[0011] Preferably, the hypoxia experiment specifically involves using a hypoxia chamber to simulate different degrees of hypoxia, allowing the subject to stay in the hypoxia environment for a specific period of time, and collecting changes in physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate in real time through physiological monitoring devices such as a blood oxygen saturation monitor and a heart rate monitor.
[0012] Preferably, living environment monitoring involves using wearable devices and environmental monitoring sensors to track changes in oxygen content in the daily living environment of the subjects over a long period of time, and to record the subjects' oxygen exposure time and degree in different locations.
[0013] Preferably, the physical examination report data includes basic vital signs such as height, weight, blood pressure, and body temperature; blood biochemical indicators such as blood glucose, blood lipids, liver function indicators, and kidney function indicators; routine blood indicators such as red blood cell count, white blood cell count, and platelet count; as well as electrocardiogram data and ultrasound examination data.
[0014] Preferably, the medical records include the subject's outpatient medical records and inpatient medical records, which record in detail their past medical history, diagnosis results, treatment plans, medication use, and the occurrence and development of the disease.
[0015] Preferably, during the data cleaning process, a small number of missing values are filled with mean, median, or predictive filling methods based on machine learning algorithms. For data records with a large number of missing values, it is considered whether to delete them based on the actual situation. Outliers are identified and processed by setting a reasonable data range and using statistical methods or anomaly detection algorithms based on machine learning.
[0016] Preferably, data preprocessing employs minimum-maximum normalization or Z-score normalization to standardize the data, and one-hot coding, label coding, or other methods to encode the categorical data.
[0017] Preferably, feature extraction and analysis employ dimensionality reduction algorithms such as principal component analysis and linear discriminant analysis to extract key features from high-dimensional data, and utilize deep learning models such as convolutional neural networks, recurrent neural networks and their variants to mine complex patterns and potential correlations in the data.
[0018] Preferably, the neural network uses models such as multilayer perceptron and deep belief network to learn and fuse features of hypoxia load data and physiological big data, and the regression model uses algorithms such as linear regression, ridge regression, and elastic network regression to establish a quantitative relationship between features and organ aging degree.
[0019] Preferably, the model training uses optimization algorithms such as stochastic gradient descent, Adagrad, Adadelta, and Adam to adjust the model parameters, and adopts cross-validation to evaluate the model performance, and selects the optimal model parameters and model structure.
[0020] Compared with existing technologies, this invention provides an organ-specific aging assessment method that integrates hypoxia load and physiological big data, and has the following beneficial effects:
[0021] This invention integrates multi-source physiological big data, including hypoxia experiments, living environment monitoring, physical examinations, and medical records, to achieve a comprehensive and accurate assessment of the aging status of various human organs. It utilizes medical hypoxia chambers and smart wearable devices to dynamically collect hypoxia load data, and combines advanced data cleaning, feature extraction, and deep learning methods to effectively improve data quality and model prediction accuracy. An innovative architecture combining neural networks and regression models enables precise quantitative prediction of organ aging indices, and provides a scientific basis for health management and clinical intervention through personalized risk alerts. The model demonstrates stable performance in actual testing, and the assessment results are highly consistent with expert annotations, exhibiting good practicality and promotional value.
[0022] This invention provides a scientific, precise, and personalized method for assessing organ-specific aging, effectively reflecting the impact of hypoxia on human aging and contributing to research on aging mechanisms and health management. This method combines multi-source data and artificial intelligence technology, improving the accuracy and practicality of aging assessment, and has broad application prospects and significant social value. Future optimization and expansion are possible, providing strong support for precision medicine and intelligent health management. Attached Figure Description
[0023] Figure 1 This is a schematic diagram of the method steps of the present invention;
[0024] Figure 2 Detailed diagram of the hypoxia experiment;
[0025] Figure 3 Detailed diagrams for the implementation of living environment monitoring;
[0026] Figure 4 Diagram of the aging assessment model architecture;
[0027] Figure 5 This is a flowchart of the data cleaning and preprocessing process. Detailed Implementation
[0028] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the invention, and not all of them. Unless otherwise specified, the embodiments and features described in this application can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0029] Furthermore, "multiple" refers to two or more. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the invention.
[0030] Please see Figure 1-5 The organ-specific aging assessment method of the present invention, which integrates hypoxia load and physiological big data, mainly includes three core steps: data acquisition, data processing and analysis, and aging assessment model construction. The specific implementation details of each step are as follows:
[0031] (I) Data Acquisition
[0032] The core objective of the data acquisition phase is to comprehensively and accurately acquire hypoxic load data and physiological big data, ensuring the integrity, timeliness, and reliability of the data, and providing a foundation for subsequent analysis and modeling.
[0033] 1. Data collection on hypoxia load
[0034] Hypoxia load data were collected through a combination of hypoxia experiments and environmental monitoring, reflecting the stress response of subjects in a controlled hypoxia environment and their long-term hypoxia exposure in daily life.
[0035] (1) Implementation of the hypoxia experiment
[0036] A medical-grade hypoxia chamber (model: HYC-III, oxygen control accuracy ±0.5%) was used to simulate different levels of hypoxia environment. The hypoxia concentration was set into 4 gradient groups (mild 18% O2, moderate 15% O2, severe 12% O2, and extremely severe 10% O2), while a normoxic control group (21% O2) was set up at the same time.
[0037] Inclusion criteria for subjects: aged 20-80 years, without serious cardiovascular, cerebrovascular, respiratory or mental diseases, and able to cooperate in completing the experiment; Exclusion criteria: pregnant women, breastfeeding women, patients with severe anemia with hemoglobin <90g / L, and patients in the acute infection phase.
[0038] Experimental Procedure: Subjects fasted for 12 hours and rested for 30 minutes before entering the chamber and wore monitoring equipment for 10 minutes to adapt. They stayed at each oxygen concentration gradient for 60 minutes in sequence, with a 30-minute rest between groups. Portable blood oxygen saturation monitor (CMS50D, accuracy ±2%), dynamic heart rate monitor (Polar H10, sampling frequency 1Hz), and respiratory rate monitor (RespSense RS-10) were used to collect index data at frequencies of once every 5 seconds, 1 second, and 10 seconds, respectively, and subjective discomfort symptoms were recorded simultaneously.
[0039] (2) Implementation of living environment monitoring
[0040] Using a smartwatch (Huawei Watch GT4 Pro, oxygen content measurement range 18%-25%, accuracy ±0.3%) and a fixed environmental oxygen content sensor (SEN0219, accuracy ±0.2%), subjects were continuously monitored for 3 months. The smartwatch was worn for at least 16 hours daily, recording environmental oxygen content, location, and activity status every hour; the sensor was installed in the subjects' main activity areas, recording oxygen content every 30 minutes, and the data was synchronized to the watch via Bluetooth. Finally, the oxygen exposure time, average oxygen content, minimum oxygen content, and fluctuation range in different locations were recorded.
[0041] 2. Physiological Big Data Collection
[0042] Physiological big data includes physical examination report data and medical record data, which are collected through a combination of hospital information systems and data provided voluntarily by the subjects. The data is anonymized and assigned a unique identifier to link all data.
[0043] (1) Collection of physical examination report data
[0044] Collect the subject's complete physical examination report from the past year. Key data includes:
[0045] Basic vital signs: height, weight, systolic blood pressure, diastolic blood pressure, body temperature, pulse;
[0046] Blood biochemical indicators: fasting blood glucose, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, direct bilirubin, creatinine, blood urea nitrogen, and uric acid;
[0047] Complete blood count indicators: red blood cell count, hemoglobin, white blood cell count, platelet count, neutrophil percentage, lymphocyte percentage;
[0048] Imaging and functional examination data: electrocardiogram (PR interval, QRS duration, etc.), echocardiography (LVEF, LVEDD, etc.), liver ultrasound (size, parenchymal echo, etc.), kidney ultrasound (cortical thickness, etc.), pulmonary function (FVC, FEV1, etc.).
[0049] (2) Medical record data collection
[0050] The subjects' outpatient and inpatient medical records from the past 5 years were retrieved. The core contents included: past medical history (name, diagnosis time, diagnostic basis), treatment plan (details of medication, surgery, and rehabilitation treatment), and the course of disease development (symptoms, changes in condition, complications, and follow-up examination results).
[0051] (II) Data Processing and Analysis
[0052] The core objective of data processing and analysis is to remove data noise, standardize data formats, and extract key features to provide high-quality data input for model building. Specifically, it includes three stages: data cleaning, data preprocessing, and feature extraction and analysis.
[0053] 1. Data Cleaning
[0054] (1) Handling missing values
[0055] Determination of missing percentage: <5% is considered a small number of missing values, and ≥30% is considered a large number of missing values;
[0056] Handling a small number of missing values: Normally distributed numerical data are filled with the mean, numerical data with extreme values are filled with the median, and indicators with strong correlations are filled with random forest prediction.
[0057] Handling of a large number of missing values: If the missing rate of key evaluation indicators is ≥30%, the record will be deleted. For missing non-key indicators, the record can be retained, and the indicator will be removed in subsequent modeling.
[0058] (2) Outlier handling
[0059] Outlier identification: For numerical data, reasonable ranges are set according to domain knowledge. For data without a clear range, the 3σ principle (normal distribution) or box plot method (non-normal distribution) is used. For time series data, the isolated forest algorithm (outlier probability ≥ 0.8) is used to identify abrupt outliers.
[0060] Outlier handling: Outliers caused by equipment malfunction or input errors are deleted and replaced with linear interpolation. Outliers representing the subject's true physiological state are retained and marked as "high-risk characteristics".
[0061] 2. Data Preprocessing
[0062] (1) Standardization of numerical data
[0063] Min-Max Normalization: Suitable for indicators with a well-defined distribution range, the formula is as follows Mapped to the [0,1] interval;
[0064] Z-score standardization: an index applicable to approximately normal distributions, the formula is... After standardization, the mean is 0 and the standard deviation is 1.
[0065] (2) Classification data coding
[0066] One-hot encoding: used for categorical data with no order relationship (such as disease type, exposure site), converting it into a binary vector;
[0067] Label encoding: used for categorical data with ordinal relationships (such as the degree of hypoxia exposure, the severity of disease), mapped to integers.
[0068] 3. Feature Extraction and Analysis
[0069] By combining dimensionality reduction algorithms with deep learning models, key features are screened and data correlations are discovered.
[0070] (1) Extracting main features using dimensionality reduction algorithm
[0071] Principal Component Analysis (PCA): For high-dimensional physiological indicators, they are converted into uncorrelated principal components, and the principal components with a cumulative variance contribution rate of ≥85% are selected as core features;
[0072] Linear Discriminant Analysis (LDA): Based on aging level labels, it maximizes inter-class distance and minimizes intra-class distance to extract discriminative features that distinguish different aging levels.
[0073] (2) Deep learning models uncover complex patterns
[0074] Convolutional Neural Network (CNN): A 3-layer CNN processes electrocardiogram data and extracts waveform features; a pre-trained ResNet-18 model is used for transfer learning to extract texture and morphological features of ultrasound images;
[0075] Recurrent Neural Network Variant (LSTM): Processes time-series hypoxia load data, with input being a 90-day time series (containing four features such as oxygen content and exposure duration), 64 neurons in the hidden layer, capturing long-term dependencies, and outputting a time-series feature vector.
[0076] The features extracted by the dimensionality reduction algorithm are fused with the features extracted by the deep learning model to form a comprehensive feature set.
[0077] (III) Construction of Aging Assessment Model
[0078] The architecture combines neural networks and regression models to achieve deep feature fusion and quantitative prediction of aging, which includes three stages: model structure design, model training and optimization, and evaluation result output.
[0079] 1. Model Structure Design
[0080] (1) Neural Networks: Feature Learning and Fusion
[0081] Multilayer Perceptron (MLP): The number of neurons in the input layer is equal to the dimension of the integrated feature set, with 3 hidden layers (64, 32, and 16 neurons respectively), the activation function is ReLU, and the fusion layer contains 16 neurons and performs batch normalization to achieve nonlinear fusion of hypoxia load features and physiological features, outputting a 16-dimensional fused feature vector.
[0082] Deep Belief Network (DBN): It consists of 3 Restricted Boltzmann Machines (RBMs) and 1 output layer. It learns low-level to high-level features layer by layer and outputs a standardized fused feature vector (with the same output dimension as MLP).
[0083] The feature vectors output by MLP and DBN are concatenated to form a 32-dimensional model input feature vector.
[0084] (2) Regression Model: Establishing Quantitative Relationships
[0085] Regression models were constructed for key organs such as the heart, liver, kidneys, and lungs:
[0086] Linear regression: suitable for scenarios where features are linearly correlated with the degree of aging;
[0087] Ridge regression: handles feature multicollinearity by constraining weights using L2 regularization;
[0088] Elastic Network Regression: Combining L1 and L2 regularization, it is suitable for high-dimensional feature scenarios, taking into account both feature selection and weight constraints.
[0089] The optimal regression model for each organ is selected through cross-validation.
[0090] 2. Model Training and Optimization
[0091] (1) Construction of labeled dataset
[0092] We recruited 1000 participants (aged 20-80 years, half male and half female, covering different hypoxia exposure backgrounds). Three experts with the title of associate chief physician or above used physiological big data, hypoxia load data, and clinical guidelines to label the aging index (0-1) of each organ and the average aging level of peers. The dataset was divided into training set, validation set, and test set in an 8:1:1 ratio.
[0093] (2) Model training process
[0094] Optimization algorithm: The Adam optimization algorithm is mainly used (learning rate 0.001, β1 = 0.9, β2 = 0.999, weight decay coefficient 1e-4). When the data distribution is unbalanced, the Adagrad algorithm is used.
[0095] Loss function: Mean Squared Error (MSE), formula is as follows
[0096] Cross-validation: 5-fold cross-validation, calculating the mean MSE and coefficient of determination (R²). 2 Select the optimal model parameters.
[0097] (3) Model optimization and adjustment
[0098] Overfit suppression: Add a Dropout layer (dropout rate = 0.2) to the hidden layer of the MLP and use early stopping (stop training if the validation set MSE does not decrease for 5 consecutive epochs);
[0099] Hyperparameter tuning: Grid search was used to optimize the number of neurons in the MLP hidden layer (32-128), regularization parameter λ (0.001-0.1), and Adam learning rate (0.0001-0.01). The validation set R was selected. 2 The largest combination of hyperparameters.
[0100] 3. Evaluation Results Output
[0101] The model receives preprocessed data, performs feature extraction, fusion, and quantitative prediction, and outputs aging assessment results for each organ:
[0102] Organ aging index: 0-1, accurate to 3 decimal places, 0.000-0.300 is "mild aging", 0.301-0.600 is "moderate aging", and 0.601-1.000 is "severe aging".
[0103] Degree of aging differences: Accurate to one decimal place, a positive number indicates aging faster than peers, and a negative number indicates aging slower than peers;
[0104] Risk warning information: Based on the aging index, degree of difference, and key risk characteristics, personalized warnings are generated to identify influencing factors and health advice.
[0105] (iv) Model Performance Verification
[0106] The model performance was validated using a test set of 100 participants. Evaluation metrics included MSE and R. 2 1. Assessment accuracy (consistency with expert annotation results):
[0107] Overall performance: The average MSE of each organ was 0.028, and the average R... 2 =0.88, average assessment accuracy =86.3%;
[0108] Organ-specific properties: Heart (MSE = 0.025, R 2=0.90, accuracy = 88.0%), liver (MSE = 0.030, R 2 =0.87, accuracy = 85.0%), kidney (MSE = 0.027, R 2 =0.89, accuracy = 87.0%), lung (MSE = 0.032, R 2 =0.86, accuracy =83.0%.
[0109] The verification results show that the model of the present invention has high prediction accuracy and stability, and can accurately reflect the true aging state of each organ.
[0110] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
[0111] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0112] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. An organ-specific aging assessment method integrating hypoxia load and physiological big data, characterized in that, Includes the following steps: Data collection: Hypoxia load data are obtained through hypoxia experiments and living environment monitoring, while physiological big data such as physical examination reports and medical records are also collected; Data processing and analysis: Cleaning and preprocessing the collected data, and using machine learning and deep learning algorithms for feature extraction and analysis; Aging assessment model construction: An organ-specific aging assessment model is constructed by combining neural networks and regression models. It is trained and optimized using labeled datasets, and finally outputs the aging assessment results of each organ, including the organ aging index, the degree of aging difference relative to the average level of peers, and risk warning information.
2. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The hypoxia experiment specifically involves using a hypoxia chamber to simulate different levels of hypoxia, allowing subjects to stay in the hypoxia environment for a specific period of time, and collecting real-time data on changes in physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate using physiological monitoring devices such as a blood oxygen saturation monitor and a heart rate monitor.
3. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The living environment monitoring uses wearable devices and environmental monitoring sensors to track changes in oxygen content in the daily life environment of the subjects over a long period of time, and records the subjects' oxygen exposure time and degree in different places.
4. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The physical examination report data includes basic vital signs such as height, weight, blood pressure, and body temperature; blood biochemical indicators such as blood glucose, blood lipids, liver function indicators, and kidney function indicators; routine blood indicators such as red blood cell count, white blood cell count, and platelet count; as well as electrocardiogram data and ultrasound examination data.
5. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The medical records include the subject's outpatient and inpatient medical records, which detail their past medical history, diagnosis results, treatment plans, medication use, and the course of the disease.
6. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: During the data cleaning process, a small number of missing values are filled using mean imputation, median imputation, or predictive imputation methods based on machine learning algorithms. For data records with a large number of missing values, it is considered whether to delete them based on the actual situation. Outliers are identified and processed by setting a reasonable data range and using statistical methods or anomaly detection algorithms based on machine learning.
7. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The data preprocessing employs minimum-maximum normalization or Z-score normalization to standardize the data, and uses one-hot encoding, label encoding, and other methods to encode the categorical data.
8. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The feature extraction and analysis employs dimensionality reduction algorithms such as principal component analysis and linear discriminant analysis to extract key features from high-dimensional data, and utilizes deep learning models such as convolutional neural networks, recurrent neural networks and their variants to mine complex patterns and potential correlations in the data.
9. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The neural network uses models such as multilayer perceptron and deep belief network to learn and fuse features from hypoxia load data and physiological big data. The regression model uses algorithms such as linear regression, ridge regression, and elastic network regression to establish a quantitative relationship between features and the degree of organ aging.
10. The organ-specific aging assessment method based on comprehensive hypoxia load and physiological big data according to claim 1, characterized in that: The model training uses optimization algorithms such as stochastic gradient descent, Adagrad, Adadelta, and Adam to adjust model parameters, and employs cross-validation to evaluate model performance, selecting the optimal model parameters and model structure.