Multi-mode health monitoring and data analysis method based on smart watch

Abstract

The invention provides a multi-modal health monitoring and data analysis method based on a smart watch, and relates to the technical field of health monitoring. Wrist and upper arm posture data and respiratory rate data collected by the smart watch are integrated, and data fusion and risk analysis are performed in combination with a deep learning model and a logistic regression model; the accuracy and the real-time performance of suffocation risk assessment of obese patients are effectively improved; by introducing a weighted correction mechanism, the initial risk score is optimized, so that the final score more comprehensively reflects the sleep posture change and the breathing state; when the risk score reaches the preset threshold value, through multiple alarm mechanisms of vibration, sound and mobile terminal notification, an alarm is rapidly triggered, it is ensured that a guardian responds in time, the accident risk of a patient is reduced, and through comprehensive utilization of multi-modal data, the monitoring accuracy and the early warning timeliness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of health monitoring technology, and specifically to a multimodal health monitoring and data analysis method based on a smart watch. Background Art

[0002] With the increasing prevalence of obesity among young people, especially among primary and secondary school students, obesity-related health issues are receiving increasing attention. Among them, sleep apnea, as a common and serious health problem, urgently needs to be diagnosed and intervened in a timely manner through effective monitoring and early warning methods. Based on big data analysis and machine learning algorithms, smart watches can not only provide basic health data records, but also predict health status and risk warnings; different sleeping positions can significantly affect the degree of openness of the airway, thereby affecting the frequency and severity of sleep apnea. For example, lying on the back may increase the risk of airway obstruction, while lying on the side can help keep the airway open. Therefore, accurately monitoring and analyzing students' breathing conditions in different sleeping positions is of great significance for the timely identification and prevention of sleep apnea;

[0003] In the prior art, the publication number is CN116612621A, and the name is a watch health monitoring system and monitoring method, which relates to the field of health monitoring technology, including a data acquisition module, an analysis module, a comparison module, a threshold adjustment module, and an early warning module; collecting environmental data and user health index data, establishing an environmental assessment coefficient for environmental parameters, and establishing a health assessment coefficient for health index parameters, by comparing the environmental assessment coefficient with the environmental assessment threshold, judging the threshold of the environment, and issuing an early warning for an unsatisfactory environment, then comparing the health assessment coefficient with the health assessment coefficient, and issuing an early warning prompt to users who do not meet the health indicators, and finally correcting the health assessment threshold according to the environment to improve the recognition accuracy of the monitoring system, effectively prevent the occurrence of dangerous situations, and improve the practicality of the monitoring system.

[0004] Although existing smart watches have made significant progress in health monitoring, they still have many shortcomings, which restrict their application in high-risk groups such as patients with apnea monitoring.

[0005] Reliance on a single data source: Existing technologies often rely on a single data source, such as using only heart rate or respiratory rate for risk assessment, lacking comprehensive analysis of multimodal data. Because the risk of apnea varies significantly depending on sleeping posture, reliance on a single data source results in inaccurate and unreliable risk assessments, making it difficult to fully reflect a student's health status.

[0006] Insufficient algorithmic accuracy: Current smartwatch algorithms for sleep posture monitoring are still in their infancy. They cannot accurately distinguish complex sleep posture changes and lack in-depth correlation analysis of abnormal breathing rates under different sleeping positions. This makes it difficult for existing systems to provide timely and accurate warnings of apnea risks, especially in obese students, where such risk assessment is even more complex.

[0007] Alarm Mechanism Lag: The existing system's alarm mechanism is lagging, preventing guardians or students from using multiple alarm methods (such as vibration, sound, and mobile notifications) to promptly alert them when risks occur. This delay not only reduces the effectiveness of risk control but also may miss the optimal opportunity for intervention, making it more difficult to control health risks.

[0008] Lack of personalized risk score optimization: The existing monitoring system lacks an optimization mechanism for different sleeping positions and individual differences in the process of risk score generation. It cannot fully reflect the changes in individual sleep apnea risk under different sleeping positions, resulting in insufficient comprehensiveness and accuracy of risk scores.

[0009] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0010] The purpose of the present invention is to provide a multimodal health monitoring and data analysis method based on a smart watch to solve the problems raised in the above background technology.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] The multimodal health monitoring and data analysis method based on smart watches includes the following steps:

[0013] Step S1: Continuously collect the target patient's body posture data and respiratory rate data at different preset time periods every night, use the sleeping posture classification algorithm based on the deep learning model to perform sleeping posture analysis on the body posture data, and combine the sleeping posture analysis results with the respiratory rate data to generate a sleeping posture dataset;

[0014] Step S2: Comparing and analyzing the sleeping posture data sets within different preset time periods to generate sleeping posture change factors corresponding to each preset time period;

[0015] Step S3: Collecting the target patient's sleeping posture data set in the current preset time period and inputting it into the logistic regression model for multimodal data fusion and analysis to generate an initial asphyxia risk score for the target patient in the current preset time period;

[0016] Step S4: Introducing the sleeping posture change factor in the current preset time period into the initial asphyxia risk score for correction to obtain a final asphyxia risk score;

[0017] Step S5: Compare the final suffocation risk score with the preset threshold value, and call the early warning module based on the judgment result to issue a real-time alarm to the guardian's mobile terminal.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. By integrating multimodal posture data and respiratory rate data from the wrist and upper arm, and using deep learning models and logistic regression models for data fusion and analysis, the accuracy and real-time performance of asphyxia risk assessment have been significantly improved;

[0020] 2. Combining a weighted correction mechanism with a multiple alarm system, this method not only more comprehensively reflects changes in a patient's sleeping posture and respiratory status, but also ensures that guardians can respond promptly through various forms of alarms such as vibration, sound, and mobile notifications when the risk score reaches a preset threshold, greatly improving the safety of obese patients. It demonstrates significant innovation and application potential in the field of patient health monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0022] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0023] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0024] Example 1:

[0025] See also Figure 1 , the present invention provides a technical solution:

[0026] A multimodal health monitoring and data analysis method based on a smartwatch is applied to multimodal health monitoring of obese patients during sleep. The specific steps include:

[0027] Step S1: Continuously collect the target patient's body posture data and respiratory rate data at different preset time periods every night, use the sleeping posture classification algorithm based on the deep learning model to perform sleeping posture analysis on the body posture data, and combine the sleeping posture analysis results with the respiratory rate data to generate a sleeping posture dataset;

[0028] Further explanation: The sleeping posture category evaluation index is used to evaluate the degree of deviation of supine, side-lying and prone postures;

[0029] The body posture data includes first posture data obtained by measuring the target patient's wrist through the smart watch, and second posture data obtained by measuring points on the target patient's upper arm; the collected first posture data and second posture data are transmitted to the central data processing unit in real time via wireless transmission and stored in a time series data format;

[0030] 1.1) Continuously collecting first posture data of the target patient's wrist at a sampling frequency of 100 Hz using a three-axis accelerometer and a three-axis gyroscope in the smartwatch;

[0031] The collection of the first posture data and the second posture data is performed on two arms of the target patient respectively;

[0032] The first posture data includes: wrist spatial position and angle change data measured when the target patient is wearing the smartwatch and in a side-lying, supine, or prone position;

[0033] It should be noted that:

[0034] 1.11) The description of the wrist spatial position is as follows:

[0035] Establish a fixed three-dimensional space coordinate system (X, Y, Z):

[0036] The X-axis is the horizontal direction of the bed surface; the Y-axis is the height direction perpendicular to the bed surface; the Z-axis is the horizontal direction parallel to the bed surface and perpendicular to the X-axis;

[0037] The wrist acceleration data is collected at a sampling frequency of 100 Hz using the built-in three-axis accelerometer of the smartwatch. x (t), A y (t), A z (t)}.

[0038] The spatial position of the wrist is calculated using the following formula:

[0039]

[0040] Among them, (P x , P y , P z ) represent the spatial positions of the wrists, which are obtained by two integrations;

[0041] The sampling results are specifically described as 3D point cloud data using the wrist trajectory over time, recorded in the form of a time series:

[0042] P(t)={(P x , P y , P z ) t |t=1,2,...,N}

[0043] The sampling frequency is 100 Hz, and the sampling duration is determined according to the monitoring period; P(t) represents the spatial position of the wrist at the sampling time point t; N is the number of sampling time points in each preset time period;

[0044] 1.12) Quantitative description of wrist angle changes:

[0045] The angular velocity data is collected by the three-axis gyroscope built into the smart watch {G x (t),G y (t), G z (t)};

[0046] Calculate the cumulative rotation angle of the wrist in each dimension based on the angular velocity:

[0047]

[0048] Among them, (θ x ,θ y ,θ z ) are the rotation angles of the wrist in the X-axis, Y-axis, and Z-axis directions respectively;

[0049] During the target patient's sleep, the angle changes of the wrist on the X, Y, and Z axes are recorded respectively.

[0050] The second posture data includes: a longitudinal height difference index and an angle deviation index between a measurement point on the upper arm and a smartwatch worn on the other hand of the target patient when the target patient is in a side-lying, supine, or prone position;

[0051] It should be noted that: 1.13) Quantitative description of the longitudinal height difference index:

[0052] Definition: The height difference index refers to the difference between the spatial height of the upper arm measurement point and the vertical height of the smartwatch worn on the other hand.

[0053] The calculation formula for collecting height difference indicators is:

[0054]

[0055] in, is the Y-axis height of the upper arm measurement point, which is obtained by a three-axis accelerometer fixed at the upper arm position and two integral calculations;

[0056] is the Y-axis height of the smartwatch worn on the other hand, using P in the first posture data y The value is determined; arm and wrist are the position indexes of the wrist where the smart watch is worn on the other hand of the upper arm measurement point;

[0057] The sampling frequency is 100 Hz, and the height data of the upper arm and wrist are recorded simultaneously;

[0058] At each time point t, calculate the longitudinal height difference index H between the two diff (t).

[0059] The results are stored as a time series:

[0060] H diff (t) = {H diff (t1), H diff (t2), ...}

[0061] The above sequence indicates that a continuous height difference curve is generated during the monitoring period;

[0062] 1.14) Quantitative description of angle deviation index:

[0063] The angle deviation index is defined as the degree of deviation of the spatial angle between the upper arm measurement point and the smartwatch worn on the other hand in the current posture state.

[0064] Using the three-dimensional coordinate data of two points (P x , P y , P z ) arm ,(P x , P y , P z ) wrist Calculate the angle deviation index α:

[0065]

[0066] in,

[0067] \|·\| represents the vector modulus in centimeters (cm); the angle deviation index α is in degrees; Represents the displacement vector from the origin of the bed surface to the target patient's upper arm measurement point; The displacement vector representing the origin of the bed surface to the wrist wearing point of the target patient's smartwatch;

[0068] This embodiment uses the arccos function to map the ratio of the dot product of the vector to the module-length product to the angle deviation index α;

[0069] The specific calculation process of the angle deviation index is as follows:

[0070] The sampling frequency is 100 Hz, and the coordinate data of the upper arm and wrist are recorded in real time.

[0071] Calculate the angular deviation index α from each other at each time point and output the time series form:

[0072] α(t)={α(t1), α(t2),...}

[0073] Where α(t) is the angle deviation index at sampling time point t;

[0074] The following is the range of angle deviation index values ​​actually measured for the target patient in this embodiment;

[0075] Supine position α=85°~100°;

[0076] Side-lying position α=45°~135°;

[0077] Prone state α=0°~30°.

[0078] The first posture data is input into a sleeping posture classification algorithm based on a deep learning model to comprehensively determine the sleeping posture of the target patient; the specific logic includes:

[0079] Ensure that the posture data of the wrist and upper arm are fully synchronized in time by using timestamp alignment;

[0080] Apply a low-pass filter to the collected first posture data to remove high-frequency noise; standardize the data of each axis to make its mean 0 and variance 1;

[0081] Deep learning model: uses convolutional neural network (CNN) or long short-term memory network (LSTM) as the basic architecture, combined with the temporal characteristics of sensor data.

[0082] Model training: The model was trained using a pre-labeled sleeping posture dataset (including supine, side, and prone postures), with a training set ratio of 80% and a validation set ratio of 20%.

[0083] Model optimization: By adjusting the learning rate and batch size, using the cross entropy loss function and Adam optimizer, the model is trained until the loss function converges.

[0084] Posture classification: The preprocessed first posture data is input into the trained deep learning model to output the current sleeping posture category. The sleeping posture categories include supine, side, and prone, and are recorded as h∈{h1, h2, h3}, where h1, h2, and h3 represent supine, side, and prone, respectively.

[0085] Further explanation: Based on the sleeping posture category output by the sleeping posture classification algorithm, feature comparison is performed with predefined standard sleeping posture features to generate a sleeping posture category evaluation index for characterizing the degree of sleeping posture deviation;

[0086] For the target patient in this embodiment, the generation of the sleeping posture category evaluation index is based on the corresponding posture data collection of multiple times of the same type of sleeping posture. The specific logic includes:

[0087] By using a weighted formula to calculate the deviation scores of all sampling time points in the current preset time period, and summarizing and averaging these deviation scores, an evaluation index for the sleeping posture category is obtained;

[0088] Deviation score (t) = |height difference index|×a1+|angle deviation index|×a2

[0089]

[0090] Wherein, a1 and a2 are the weights of the height difference index and the angle deviation index, respectively; the values ​​of a1 and a2 are selected in the interval (0, 1), and a1 + a2 = 1; the deviation score (t) represents the deviation score at the t-th time sampling point in the current preset time period; N is the number of sampling time points in the preset time period; in this embodiment, the values ​​of a1 and a2 are 0.6 and 0.4, respectively;

[0091] The target patient is in sleeping posture h during the current preset time period j, and the sleeping posture category evaluation index is marked as S h，j ;

[0092] 1.2) Using a photoplethysmography (PPG)-based respiratory rate sensor to collect the target patient's respiratory rate data in real time at a sampling frequency of 100 Hz; this embodiment requires the PPG sensor to be worn on the target patient's abdomen or chest to ensure good contact between the sensor and the skin to avoid motion interference;

[0093] The collected respiratory rate data is transmitted to the central data processing unit in real time via wireless transmission and stored in a time series data format;

[0094] The collected respiratory rate data is processed through signal filtering and peak detection algorithms to extract respiratory rate indicators;

[0095] Signal filtering was performed by applying a Butterworth low-pass filter with a cutoff frequency set at 5 Hz to remove high-frequency noise, and the low-frequency drift in the respiratory signal was eliminated by the differential method.

[0096] The peak detection algorithm uses a moving window method to detect the peak position in the PPG signal to identify the respiratory cycle. In this embodiment, the moving window size is set to 2 seconds and the minimum peak interval is set to 3 seconds to avoid false detection.

[0097] Identify the time intervals between consecutive peaks and calculate the duration of each respiratory cycle (in seconds);

[0098] The respiratory rate index is defined as: respiratory rate index = 60 / respiratory cycle duration; the unit is beats per minute (BPM).

[0099] Calculate the mean, standard deviation, minimum, and maximum variation of the respiratory rate index of the target patient in the current preset time period j when the patient is in sleeping position h;

[0100] The sleeping posture category evaluation indicators and breathing rate indicators generated in different preset time periods every night are combined to form the final sleeping posture dataset.

[0101] It should be noted that this embodiment divides the sleep time each night into multiple preset time periods, with each 15-minute period being a preset time period;

[0102] Sleeping posture category evaluation indicators: including the sleeping posture category evaluation indicators for supine, side, and prone sleeping within each preset time period;

[0103] Respiratory rate index: including the average value, standard deviation, minimum and maximum variation range of respiratory rate index in each preset time period;

[0104] The above indicators are organized into a structured data table, where each row corresponds to a preset time period, including a time period identifier, a sleeping posture category evaluation indicator, and a breathing rate indicator.

[0105] Data storage: The final sleeping posture dataset is stored in a secure database for subsequent analysis.

[0106] Example 2:

[0107] Further explanation based on Example 1:

[0108] Step S2: Comparing and analyzing the sleeping posture data sets within different preset time periods to generate sleeping posture change factors corresponding to each preset time period;

[0109] Further explanation: 2.1) Statistical analysis is performed on the evaluation indicators of sleeping posture categories within different preset time periods, and the occurrence frequency and change trend of supine, side-lying, and prone postures within each preset time period are calculated;

[0110] The implementation details are as follows:

[0111] Frequency calculation: Count the number of times the supine, side-lying, and prone positions occur within each preset time period, and calculate the proportion of each position.

[0112] Change trend analysis: Calculate the change trend of each posture throughout the sleep cycle and observe the frequency of posture changes through time series analysis.

[0113] Data presentation: Frequency and change trends are expressed as percentage change rates to ensure comparability across different time periods;

[0114] 2.2) Perform statistical analysis on the respiratory rate indicators within different preset time periods, and calculate the mean value, standard deviation, and minimum and maximum variation ranges of the respiratory rate indicators within each preset time period;

[0115] The implementation details are as follows:

[0116] Average calculation: Calculate the arithmetic mean of the respiratory rate index within each preset time period, in BPM.

[0117] Standard deviation calculation: Calculate the standard deviation of the respiratory rate index within each preset time period to reflect the degree of fluctuation of the respiratory rate index.

[0118] Variation calculation: Calculate the difference between the maximum and minimum values ​​of the respiratory rate index within each preset time period, in BPM.

[0119] The above indicators are used as quantitative respiratory rate characteristics;

[0120] 2.3) Combine the analysis results of step 2.1) and step 2.2) and use multivariate regression analysis to determine the sleeping posture change factor of the preset time period j, recorded as Y j ;

[0121] The implementation details are as follows:

[0122] A multivariate linear regression model is used, and the model form is:

[0123] Y=β0+β1X1+β2X2+β3X3+…+β n X n +∈

[0124] Among them, Y is the sleeping posture change factor, X1, X2, ..., X n is the evaluation index of each sleeping posture category and the respiratory rate index, β i is the regression coefficient, ∈ is the error term; Y j The value range is limited to the interval (0,1);

[0125] The independent variables of the multivariate linear regression model included: supine frequency, lateral frequency, prone frequency, mean value, standard deviation, and minimum and maximum variation of respiratory rate index;

[0126] The dependent variable of the multivariate linear regression model was the sleeping posture change factor, which was a quantitative indicator reflecting the comprehensive sleeping posture stability.

[0127] The model training process of the multivariate linear regression model is as follows:

[0128] The first 80% of the sleeping posture dataset is used as the training dataset to train the regression model.

[0129] The regression coefficients were estimated by the least squares method to optimize the model fit.

[0130] The last 20% of the sleeping posture dataset was used as a validation dataset to evaluate the prediction accuracy of the regression model and calculate the coefficient of determination and mean square error.

[0131] If the model performs well, the coefficient of determination R 2 ≥0.8, when the mean square error (MSE) is low, the effectiveness of the model is confirmed;

[0132] Using the trained regression model, the sleeping posture category evaluation index and respiratory rate index in each preset time period are substituted into the model to calculate the corresponding sleeping posture change factor;

[0133] The numerical range of the sleeping posture change factor is standardized based on the model training results to ensure the comparability of data across time periods.

[0134] Example 3:

[0135] Further explanation based on Example 2:

[0136] Step S3: Collecting the target patient's sleeping posture data set in the current preset time period and inputting it into the logistic regression model for multimodal data fusion and analysis to generate an initial asphyxia risk score for the target patient in the current preset time period;

[0137] Further explanation: 3.1) The logistic regression model is selected as the algorithmic framework for multimodal data fusion and analysis; based on its efficiency and interpretability in risk assessment.

[0138] The implementation details are as follows:

[0139] The logistic regression model has low computational complexity and is suitable for real-time risk assessment; the model parameters can explain the impact of each variable on the risk of asphyxia, facilitating clinical application.

[0140] Classify suffocation risks into “high risk” and “low risk” categories;

[0141] The logistic regression model is represented as:

[0142]

[0143] Where P(Y=1) represents the initial asphyxia risk score, which is used to assess the probability of high asphyxia risk of the target patient, X1, X2,..., X n is the input variable of the multivariate regression analysis method in step 2.3), β i is the regression coefficient;

[0144] The training content of the logistic regression model is as follows:

[0145] Training set: input variables corresponding to the annotated asphyxiation risk status (high risk, low risk) in historical data;

[0146] Test set: An independent dataset used for model validation to ensure the generalization ability of the model.

[0147] Data format: The input variables are organized into a structured data matrix, where each row represents a sample and each column represents an input variable.

[0148] 3.3) Use historical data to train a logistic regression model to ensure it can effectively predict the initial apnea risk score. Implementation details are as follows:

[0149] Model training: Use the fully connected logistic regression algorithm and the maximum likelihood estimation method to fit the model parameters.

[0150] L2 regularization is applied during training to prevent overfitting, and the regularization parameter is set to 0.01.

[0151] Model Validation: Accuracy: ≥85%. Precision: ≥80%. Recall: ≥80%. F1 score: ≥0.8; Area under the ROC curve (AUC): ≥0.9.

[0152] Cross-validation: 5-fold cross-validation was used to evaluate model stability and generalization ability;

[0153] Parameter adjustment: adjust model hyperparameters based on verification results to optimize model performance;

[0154] Model Validation: Only after the model passes all validation metrics is it validated for use in generating an initial asphyxia risk score.

[0155] 3.4) The sleeping posture data set in the current preset time period is input into the trained logistic regression model to calculate and generate the initial asphyxia risk score P (Y=1) of the target patient in the current preset time period.

[0156] The implementation details are as follows:

[0157] Data input: Input the sleeping posture change factors and related respiratory rate indicators of the current preset time period into the logistic regression model;

[0158] Compute the linear combination of the logistic regression model:

[0159] Z=β0+β1X1+β2X2+…+β n X n

[0160] Apply the Sigmoid function to convert to probability value:

[0161]

[0162] The initial apnea risk score was determined as a percentage value of P(Y=1), ranging from 0% to 100%;

[0163] The generated initial asphyxia risk score and the corresponding time period identifier are stored in the database for subsequent correction and analysis.

[0164] Example 4:

[0165] Further explanation based on Example 3:

[0166] Step S4: Introducing the sleeping posture change factor in the current preset time period into the initial asphyxia risk score for correction to obtain a final asphyxia risk score;

[0167] Further explanation: 4.1) Design a weighted correction mechanism based on the sleeping posture change factor Y of the current preset time period j j The initial apnea risk score P(Y=1) was weighted and adjusted to ensure that the score could reflect the real-time changes in sleeping posture.

[0168] Weighted correction mechanism, the specific logic includes:

[0169] According to expert experience and data analysis results, the initial suffocation risk score P (Y = 1) and the sleeping posture change factor Y are set. j The weights in the total score are δ1 and δ2 respectively; the sleeping posture change factor Y in this embodiment is j The weight δ was set to 0.4, and the initial asphyxia risk score weight δ1 was set to 0.6;

[0170] P1(Y=1)=P(Y=1)×δ1+(Y j ×δ2)

[0171] Among them, P1 (Y = 1) is the final corrected asphyxia risk score;

[0172] The expert group dynamically adjusts the weight ratio based on actual monitoring data and clinical feedback to optimize the accuracy and sensitivity of the score;

[0173] The above implementation details are as follows:

[0174] Ensure that the final asphyxia risk score P1 (Y = 1) is between 0% and 100%, and handle abnormal values ​​by limiting the minimum and maximum values;

[0175] The final suffocation risk score is stored in the database together with the time period identifier for further analysis and real-time monitoring.

[0176] The final asphyxia risk score and the corresponding time period identifier are stored in the database and provided to the user interface or alarm system through the interface.

[0177] Embodiment 5:

[0178] Further explanation based on Example 4:

[0179] Step S5: Compare the final suffocation risk score with the preset threshold value, and call the early warning module based on the judgment result to issue a real-time alarm to the guardian's mobile terminal.

[0180] Further explanation: Real-time alerts include but are not limited to vibration alerts, mobile notifications, and sound alerts;

[0181] 5.1) Set the preset threshold of the final asphyxia risk score to determine the alarm triggering conditions;

[0182] In this embodiment, the preset threshold is set at 80%. When the final asphyxia risk score reaches or exceeds 80%, it is determined to be a high-risk state and an alarm is triggered.

[0183] This embodiment sets a threshold adjustment mechanism that allows the preset threshold to be dynamically adjusted between 70% and 90% through a mobile application based on the patient's health status and guardian needs, thereby achieving personalized monitoring;

[0184] 5.2) Compare the final asphyxia risk score with the preset threshold in real time;

[0185] Continuously monitor the final asphyxia risk score within each preset time period and automatically compare it with the preset threshold;

[0186] If the final asphyxia risk score is ≥80%, the system determines it as a "high risk" state; otherwise, it is determined to be a "low risk" state;

[0187] Status recording: Record each comparison result in the database for subsequent analysis and monitoring.

[0188] 5.3) When a "high risk" status is detected, the early warning module is called to perform real-time alarm operations and send various forms of alarms to the guardian's mobile terminal;

[0189] The early warning module includes a vibration alarm, an audio alarm and a mobile notification system.

[0190] Description of vibration alarm:

[0191] Frequency and duration: 3 vibrations per second, duration of 30 seconds.

[0192] Vibration signals are sent through a smartwatch or standalone vibration device worn on the patient.

[0193] Description of the sound alarm:

[0194] Decibel and duration: emits a high-frequency sound alarm of 85 decibels, which lasts for 30 seconds.

[0195] The sound signal is emitted through the built-in speaker of the smart watch or an independent sound alarm device.

[0196] Instructions for mobile notifications:

[0197] Notification content: Send a push notification containing "Patient has a high risk of suffocation, please check immediately" to the guardian's smartphone.

[0198] Notification method: Instant push notifications via the accompanying mobile app, supporting vibration and sound prompts.

[0199] Multiple alarms: Ensures that vibration alarms, sound alarms, and mobile notifications are triggered simultaneously, improving the visibility and audibility of alarms and ensuring timely responses from guardians.

[0200] 5.4) Ensure the synchronization of the early warning module and the mobile device to ensure the real-time and reliability of the alarm.

[0201] The implementation details are as follows:

[0202] Use low-latency wireless communication technology such as Bluetooth 5.0 or Wi-Fi 6 to ensure that the data transmission delay between the smartwatch and the mobile device does not exceed 200 milliseconds;

[0203] The AES-256 encryption protocol is used to protect the security of alarm information and prevent data leakage or tampering.

[0204] Dual-channel transmission is set to prevent the alarm from not being delivered in time due to failure of a single communication. In this embodiment, the dual-channel transmission is Bluetooth and Wi-Fi;

[0205] Alarm confirmation: After the guardian receives the alarm on the mobile terminal, he / she clicks the confirmation button to send a confirmation signal to the early warning module to prevent repeated alarms.

[0206] 5.5) Record each alarm event for subsequent analysis and system optimization.

[0207] The implementation details are as follows:

[0208] The timestamp, final suffocation risk score, triggered alarm type, and guardian confirmation status of each alarm are recorded in the database.

[0209] Regularly analyze alarm event logs, evaluate the accuracy and frequency of alarm triggering, and optimize threshold settings and alarm mechanisms.

[0210] Generate weekly or monthly reports for reference by guardians and medical professionals to help assess patients' sleep safety.

[0211] Example 6:

[0212] The following is a specific implementation example of the "sleeping posture dataset" technical solution. This example demonstrates, through actual data, the innovation and advantages of the technical solution of introducing the sleeping posture change factor in the current preset time period into the initial suffocation risk score for correction, obtaining a final suffocation risk score, and comparing the final score with the preset threshold. Based on the judgment result, the early warning module is called to issue a real-time alert to the guardian's mobile terminal.

[0213] To verify the effectiveness and innovation of the "Sleeping Posture Dataset" technical solution, six healthy patients were selected as test subjects, named Patient A, Patient B, Patient C, Patient D, Patient E, and Patient F. Each patient wore a smartwatch equipped with a three-axis accelerometer and a three-axis gyroscope (for wrist posture data collection), a synchronous measurement sensor mounted on the upper arm (for upper arm posture data collection), and a respiratory rate sensor based on photoplethysmography (PPG). Sleep monitoring was performed for two consecutive nights.

[0214] During the experiment, the smartwatch continuously collected first-posture data from each patient's wrist at a sampling frequency of 100Hz, while the upper arm measurement sensor collected second-posture data at the same sampling frequency. The PPG sensor also recorded the patient's respiratory rate data in real time at a sampling frequency of 100Hz. All collected data was transmitted in real time to a central data processing unit via wireless transmission technology (such as Bluetooth 5.0) and stored in a time series data format.

[0215] The collected first and second posture data are input into a sleep posture classification algorithm based on a deep learning model to comprehensively determine the patient's current sleeping position (supine, side, or prone). The sleeping posture category is then compared with predefined standard sleeping posture characteristics to generate a sleeping posture category evaluation index that represents the degree of deviation from the sleeping posture. The respiratory rate data is processed through signal filtering and a peak detection algorithm to extract indicators such as the respiratory rate mean, standard deviation, and variation.

[0216] Next, the sleeping posture category evaluation index and respiratory rate index within a preset 15-minute time period were combined to form the final sleeping posture dataset. Through multivariate regression analysis, the sleeping posture category evaluation index and respiratory rate index were converted into sleeping posture change factors, which were then weighted and modified in the initial suffocation risk score to obtain the final suffocation risk score. The preset threshold was set at 80%. When the final suffocation risk score reached or exceeded this threshold, the system automatically activated the early warning module, sending a real-time alert to the guardian's mobile device.

[0217] This trial analyzed sleep data from six patients over two nights to validate the effectiveness and superiority of this technology in apnea risk assessment and real-time alarming. Data analysis showed that incorporating corrections for sleep posture variation factors significantly improved the accuracy of risk scores, and the real-time alarm system was able to promptly identify high-risk conditions and ensure patient safety.

[0218] The following table shows the initial apnea risk score, sleep position variation factor, and final apnea risk score for six trial patients over different pre-set time periods, along with the alarm triggering status after comparison with pre-set thresholds. The data demonstrates that incorporating the sleep position variation factor into the score significantly improved the accuracy and timeliness of the apnea risk score. After the correction, some patients reached or exceeded the pre-set threshold, successfully triggering the alarm.

[0219] Table 1 Feasibility study of the final asphyxia risk score:

[0220]

[0221]

[0222] Initial apnea risk score (%): an unadjusted risk score calculated based on the sleeping position classification algorithm and respiratory rate indicators.

[0223] Sleeping posture change factor: A factor reflecting the degree of sleeping posture change obtained through multivariate regression analysis.

[0224] Final suffocation risk score (%): The score obtained by weighting and correcting the initial risk score and the sleeping posture change factor according to the weight. The formula is: final score = initial score × 0.6 + change factor × 0.4.

[0225] Preset threshold (%): The preset risk alarm threshold is set to 80%.

[0226] Alarm triggering: The decision on whether to trigger an alarm is based on the comparison between the final score and the threshold.

[0227] As can be seen from the tabular data, by introducing the sleeping posture change factor into the initial asphyxia risk score for correction, the final score more comprehensively reflects the patient's sleeping posture changes and respiratory rate status. In most cases, the corrected final score is improved compared to the initial score, especially in some time periods of patient D and patient B, the final score successfully reached or exceeded the preset threshold of 80%, triggering an alarm. This shows that the introduction of the sleeping posture change factor helps to improve the accuracy and timeliness of asphyxia risk assessment, can more effectively identify high-risk conditions, and ensure that guardians can take timely countermeasures. In addition, through multiple test data verifications, the system can maintain high reliability and consistency in different time periods and different individual patients, highlighting the innovation and practicality of this technical solution in the field of patient asphyxia risk monitoring.

[0228] 1. Multimodal data fusion: Combining wrist and upper arm posture data with respiratory rate data, and using deep learning and logistic regression models for comprehensive analysis, significantly improves the accuracy and reliability of asphyxia risk assessment.

[0229] 2. Real-time and Multiple Alarms: 100Hz high-sampling-rate data acquisition and low-latency wireless communication technology enable real-time monitoring and instant alarms. Simultaneous triggering of multiple alarm modes (vibration, sound, and mobile notifications) ensures that guardians can respond quickly to multi-sensory stimulation, safeguarding the patient's life.

[0230] 3. Customizable thresholds and dynamic adjustments: The preset thresholds can be dynamically adjusted (70% to 90%) based on the patient's specific situation and guardian needs, enabling personalized monitoring and improving the system's adaptability and practicality.

[0231] 4. Data security and system reliability: Adopting AES-256 encrypted transmission protocol and dual-channel communication mechanism (Bluetooth 5.0 and Wi-Fi 6) to ensure the security of data transmission and high reliability of the system, avoiding false alarms and missed alarms.

[0232] 5. Intelligent risk score correction: By introducing a sleeping posture change factor to perform weighted correction on the initial score, the system can more comprehensively reflect the patient's sleeping posture changes and respiratory status, improving the dynamic adaptability of risk assessment.

[0233] 6. Data recording and analysis: Each alarm event is recorded in the database to facilitate subsequent data analysis and system optimization, and continuously improve the system's prediction accuracy and response efficiency.

[0234] This embodiment can not only effectively identify and predict the risk of suffocation, but also ensure that the guardian can respond in time through multiple alarm forms, significantly improving the safety of patients, and has broad application prospects and commercial value.

[0235] It should be noted that: All calculation formulas in this application document use regression analysis including but not limited to machine learning algorithms to deeply analyze the relevant parameters collected and identify their natural trends and relationships. Use professional software, such as Python's Scikit-learn library or R language, to automatically generate mathematical models that match the data. Then, objectively evaluate the performance of the model through methods such as cross-validation, and combine continuous feedback and optimization to ensure that the created formula truly reflects the inherent laws of the data, thereby ensuring its effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula are dimensionally non-dimensionalized within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless technical means include but are not limited to Min-Max Normalization and Z-Score standardization;

[0236] The technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0237] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0238] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0239] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multimodal health monitoring and data analysis method based on a smartwatch, characterized in that: Multimodal health monitoring for obese patients during sleep, including the following steps: Step S1: Continuously collect the target patient's body posture data and respiratory rate data at different preset time periods every night, use the sleeping posture classification algorithm based on the deep learning model to perform sleeping posture analysis on the body posture data, and combine the sleeping posture analysis results with the respiratory rate data to generate a sleeping posture dataset; Step S2: Comparing and analyzing the sleeping posture data sets within different preset time periods to generate sleeping posture change factors corresponding to each preset time period; Step S3: Collecting the target patient's sleeping posture data set in the current preset time period and inputting it into the logistic regression model for multimodal data fusion and analysis to generate an initial asphyxia risk score for the target patient in the current preset time period; Step S4: Introducing the sleeping posture change factor in the current preset time period into the initial asphyxia risk score for correction to obtain a final asphyxia risk score; Step S5: Compare the final suffocation risk score with the preset threshold value, and call the early warning module based on the judgment result to issue a real-time alarm to the guardian's mobile terminal.

2. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 1, characterized in that: The sleeping posture category evaluation index is used to evaluate the degree of deviation of supine, side-lying and prone postures; The body posture data includes first posture data measured at the wrist of the target patient by the smart watch, and second posture data measured at a measurement point on the upper arm of the target patient; The collection of the first posture data and the second posture data is performed on two arms of the target patient respectively; The first posture data includes: wrist spatial position and angle change data measured when the target patient is wearing the smartwatch and in a side-lying, supine, or prone position; The second posture data includes: a longitudinal height difference index and an angle deviation index between a measurement point on the upper arm and a smartwatch worn on the other hand of the target patient when the target patient is in a side-lying, supine, or prone position; The first posture data is input into a sleeping posture classification algorithm based on a deep learning model to comprehensively determine the sleeping posture of the target patient; the sleeping posture categories include supine, side, and prone, and are recorded as h∈{h1, h2, h3}, where h1, h2, and h3 represent supine, side, and prone, respectively.

3. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 2, characterized in that: Based on the sleeping posture category output by the sleeping posture classification algorithm, feature comparison is performed with predefined standard sleeping posture features to generate a sleeping posture category evaluation index used to characterize the degree of sleeping posture deviation; By using a weighted formula to calculate the deviation scores of all sampling time points in the current preset time period, and summarizing and averaging these deviation scores, an evaluation index for the sleeping posture category is obtained; Deviation score (t) = |height difference index|×a1+|angle deviation index|×a2 Where a1 and a2 are the weights of the height difference index and the angle deviation index, respectively; the values ​​of a1 and a2 are selected in the interval (0, 1), and a1 + a2 = 1; the deviation score (t) represents the deviation score at the t-th time sampling point in the current preset time period; N is the number of sampling time points in the preset time period; The target patient is in sleeping posture h during the current preset time period j, and the sleeping posture category evaluation index is marked as S h，j ; The collected respiratory rate data is processed through signal filtering and peak detection algorithms to extract respiratory rate indicators; The respiratory rate index is defined as: respiratory rate index = 60 / respiratory cycle duration; Calculate the mean, standard deviation, minimum, and maximum variation of the respiratory rate index of the target patient in the current preset time period j when the patient is in sleeping position h; The sleeping posture category evaluation indicators and breathing rate indicators generated in different preset time periods every night are combined to form the final sleeping posture dataset.

4. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 3, characterized in that: Perform statistical analysis on the evaluation indicators of sleeping posture categories within different preset time periods, and calculate the frequency and change trend of supine, side-lying, and prone postures within each preset time period; Perform statistical analysis on the respiratory rate indicators within different preset time periods, and calculate the mean value, standard deviation, and minimum and maximum variation ranges of the respiratory rate indicators within each preset time period; The multivariate regression analysis method is used to determine the sleeping posture change factor of the preset time period j, which is recorded as Y j .

5. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 4, characterized in that: The logistic regression model is selected as the algorithm framework for multimodal data fusion and analysis; Classify suffocation risks into "high risk" and "low risk" categories; The logistic regression model is represented as: Where P(Y=1) represents the initial asphyxia risk score, X1, X2, ..., X n is the input independent variable of the multivariate regression analysis method, and βi is the regression coefficient; Use historical data to train a logistic regression model; The sleeping posture data set in the current preset time period is input into the trained logistic regression model to calculate and generate the initial asphyxia risk score P (Y=1) of the target patient in the current preset time period.

6. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 5, characterized in that: Design a weighted correction mechanism based on the sleeping posture change factor Y of the current preset time period j j Perform weighted adjustment on the initial asphyxia risk score P(Y=1); Weighted correction mechanism, the specific logic includes: Set the initial suffocation risk score P (Y = 1) and the sleeping posture change factor Y j The weights in the total score are δ1 and δ2 respectively; P1(Y=1)=P(Y=1)×δ1+(Y j ×δ2) Among them, P1(Y=1) is the final asphyxia risk score after correction.

7. The multimodal health monitoring and data analysis method based on a smartwatch according to claim 6, characterized in that: Real-time alerts include but are not limited to vibration alerts, mobile notifications, and sound alerts; Setting preset thresholds for the final suffocation risk score to determine alarm triggering conditions; Compare the final asphyxia risk score with the preset threshold in real time; Continuously monitor the final asphyxia risk score within each preset time period and automatically compare it with the preset threshold; If the final asphyxia risk score is ≥ the preset threshold, the system will determine it as a "high risk" state; Otherwise, it is judged as "low risk"; When a "high risk" state is detected, the early warning module is called to perform real-time alarm operations and send various forms of alarms to the guardian's mobile terminal; The early warning module includes a vibration alarm, an audio alarm and a mobile notification system.

Images