A method and system for sleep apnea detection based on transcutaneous carbon dioxide monitoring

Abstract

The application discloses a sleep apnea detection method and system based on transcutaneous carbon dioxide monitoring, and belongs to the technical field of sleep apnea detection. The application integrates a miniaturized transcutaneous carbon dioxide detection module composed of a flexible optical sensor and a multi-channel PPG, a bioimpedance sensor and an accelerometer; a heating module is controlled by a PID algorithm to realize constant temperature, so that non-invasive transcutaneous carbon dioxide detection is ensured; a multi-sensor data fusion system is built to realize synchronous acquisition of physiological parameters; a low-power embedded processing unit is adopted to realize real-time processing and monitoring of physiological multi-modal data. The application also cooperates with a machine learning model and combines TcCO2 fluctuation, SpO2 drop rate and respiratory interval variability to realize real-time identification and early warning of sleep apnea events. The application realizes highly integrated and long-lasting continuous monitoring of respiratory health, and effectively improves the home screening capability of sleep breathing disorders.

Description

Technical Field

[0001] This invention belongs to the field of sleep apnea detection technology, and particularly relates to a method and system for sleep apnea detection based on transcutaneous carbon dioxide monitoring. Background Technology

[0002] Sleep apnea syndrome (OSA) is a common sleep-disordered breathing. Currently, the main method for detecting OSA is polysomnography (PSG). However, this method involves complex equipment that needs to be worn overnight, and the numerous tubing options lead to significant sleep disturbances and poor patient compliance. Although portable pulse oximeters are available on the market, oxygen saturation (SpO2) alone cannot fully reflect ventilation function and changes.

[0003] Existing technologies have several major pain points and shortcomings.

[0004] Lack of carbon dioxide monitoring: Most traditional wearable devices can only monitor blood oxygen saturation, while carbon dioxide concentration is the earliest and most critical indicator for diagnosing respiratory acidosis and hypoventilation. A simple drop in blood oxygen often lags behind the occurrence of apnea.

[0005] Non-invasive monitoring is challenging: traditional transcutaneous carbon dioxide monitoring devices are bulky, power-consuming, and require precise temperature control to promote local capillary dilation (typically maintained at 40℃~45℃) to obtain accurate signals; otherwise, accurate signals cannot be acquired. Currently, there is a lack of technical solutions for miniaturizing and highly integrating these devices into wrist-worn wearable devices.

[0006] High error rate of single indicator: Relying solely on the decrease in blood oxygen to predict apnea is easily affected by motion artifacts (such as turning over, hand movements, etc.) and cannot distinguish between central and obstructive apnea, resulting in a high false alarm rate in home screening. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention proposes a method and system for detecting sleep apnea based on transcutaneous carbon dioxide monitoring; aiming to achieve high-precision, low-interference, continuous home-based monitoring of sleep apnea.

[0008] The first aspect of this invention proposes a method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring, the method comprising: S1. Adjust the temperature of the sampling area to reach the target temperature; S2. At the target temperature, collect multimodal physiological data of the user from the self-sampling site; S3. Denoising and artifact removal are performed on the multimodal physiological data; S4. Extract multidimensional features from denoised and artifact-free multimodal physiological data; S5. Perform temporal alignment and fusion processing on multidimensional features to construct physiological feature vectors; S6. Detect sleep apnea based on physiological feature vectors and trigger an early warning.

[0009] According to the method of the first aspect of the present invention, in S1, the heating module is controlled to collect thermistor data of the sampling location to obtain the current temperature of the sampling location. A PID algorithm is used to calculate the deviation between the current temperature and the target temperature in real time. The duty cycle of dynamic pulse width modulation (PWM) is adjusted to smoothly transfer heat to the sampling location, so that the sampling location reaches the target temperature. Wherein: By stimulating the capillaries with heat, the diffusion rate of carbon dioxide through the stratum corneum is increased, thereby improving the quality of data acquisition while performing non-invasive detection.

[0010] According to the method of the first aspect of the present invention, in S2, multimodal physiological data of the user are collected using a sensing module, wherein the sensing module refers to a multimodal sensor array, including a flexible optical sensor, a multichannel PPG sensor, a bioimpedance sensor, and an accelerometer, and the collected analog signals are digitized and input to a processing unit; wherein: The flexible optical sensor emits an infrared beam with a specific wavelength of 4200nm to penetrate the skin surface of the sampling site and perform specific detection of transdermal carbon dioxide concentration. A multi-channel PPG sensor alternately emits red and infrared light to collect photoplethysmography (PPG) waves to monitor blood oxygen saturation and heart rate. Bioimpedance sensors measure the impedance changes of local tissues using alternating current and extract respiratory frequency waveforms; Accelerometers synchronously collect body motion information; Timestamp multimodal physiological data using a unified clock source.

[0011] According to the method of the first aspect of the present invention, in S3, multimodal physiological data are extracted, and an adaptive filtering algorithm is used, with the body motion information as a reference noise source, to eliminate the interference of motion artifacts on carbon dioxide concentration data and blood oxygen saturation data; wherein: When the target subject turns over or makes hand movements during sleep, the accelerometer outputs the corresponding motion vector; by deploying a least mean square (LMS) adaptive filter, the motion vector is subtracted from the disturbed physiological signal and the weights are adjusted to restore the physiological waveform.

[0012] According to the method of the first aspect of the present invention, in S4, the multidimensional features are extracted using a sliding window mechanism, wherein: Set a time sliding window to calculate the fluctuation trend of carbon dioxide concentration data, the rate of decrease of blood oxygen saturation data, and the variability of respiratory interval; Fluctuation trend extraction: Calculate the first derivative or linear regression slope of carbon dioxide concentration data within the window to quantify the carbon dioxide accumulation rate; Descent rate extraction: Identify local peaks and troughs in blood oxygen saturation data, calculate the ratio of drop depth to drop time to obtain the descent rate, and count the frequency of blood oxygen desaturation. Variation extraction: The peak-finding algorithm is used to identify the extreme points of adjacent cycles of the respiratory rate waveform, extract the time interval sequence, and calculate its standard deviation or coefficient of variation.

[0013] According to the method of the first aspect of the present invention, in S5, the extracted fluctuation trend features, decline rate features and variability features are aligned on the feature dimension based on a unified timestamp and concatenated into a one-dimensional or multi-dimensional feature vector, thereby constructing the physiological feature vector.

[0014] According to the method of the first aspect of the present invention, in S6, the physiological feature vector is input into a pre-deployed prediction model based on a machine learning algorithm to predict the confidence level of sleep apnea events within the current and future time windows; wherein: The learning algorithm refers to the learning algorithm based on the Long Short-Term Memory (LSTM) network or the random forest algorithm. The prediction model uses the time series dependencies to output the probability of obstructive sleep apnea occurring and triggers a vibration warning.

[0015] The method according to the first aspect of the present invention further includes: S7. Establish a personalized baseline database based on sleep physiological data from sleep apnea detection to dynamically adjust the warning threshold.

[0016] A second aspect of this invention provides a sleep apnea detection system based on transcutaneous carbon dioxide monitoring, the system comprising: The heating module is controlled to adjust the temperature of the sampling area to achieve the target temperature. The sensing module collects multimodal physiological data of the user at the self-sampling site under the target temperature; The preprocessing module performs denoising and artifact removal on the multimodal physiological data; The feature extraction module extracts multidimensional features from denoised and artifact-free multimodal physiological data; The alignment and fusion module performs temporal alignment and fusion processing on multi-dimensional features to construct physiological feature vectors. The detection module detects sleep apnea based on physiological feature vectors and triggers an early warning.

[0017] A third aspect of the present invention provides a wearable device, wherein the wearable device is configured with a detection system as described in the first aspect of the present invention to realize sleep apnea detection based on transcutaneous carbon dioxide monitoring.

[0018] In summary, this invention integrates a miniaturized transcutaneous carbon dioxide detection module composed of a flexible optical sensor and a multi-channel PPG, a bioimpedance sensor, and an accelerometer; it uses a PID algorithm to control the heating module to achieve constant temperature, ensuring accurate non-invasive transcutaneous carbon dioxide detection; it establishes a multi-sensor data fusion system to achieve synchronous acquisition of physiological parameters; and it employs a low-power embedded processing unit to achieve real-time processing and monitoring of multimodal physiological data. This invention also incorporates a machine learning model and combines TcCO2 fluctuations, SpO2 decline rate, and respiratory interval variability to identify and issue early warnings for sleep apnea events in real time. This invention achieves highly integrated, long-lasting continuous respiratory health monitoring, effectively improving the home-based screening capabilities for sleep-disordered breathing. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a sleep apnea detection method based on transcutaneous carbon dioxide monitoring according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The first aspect of this invention proposes a method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring, such as... Figure 1 As shown, the method includes: S1. Adjust the temperature of the sampling area to reach the target temperature; S2. At the target temperature, collect multimodal physiological data of the user from the self-sampling site; S3. Denoising and artifact removal are performed on the multimodal physiological data; S4. Extract multidimensional features from denoised and artifact-free multimodal physiological data; S5. Perform temporal alignment and fusion processing on multidimensional features to construct physiological feature vectors; S6. Detect sleep apnea based on physiological feature vectors and trigger an early warning.

[0023] According to the method of the first aspect of the present invention, in S1, the heating module is controlled to collect thermistor data of the sampling location to obtain the current temperature of the sampling location. A PID algorithm is used to calculate the deviation between the current temperature and the target temperature in real time. The duty cycle of dynamic pulse width modulation (PWM) is adjusted to smoothly transfer heat to the sampling location, so that the sampling location reaches the target temperature. Wherein: By stimulating the capillaries with heat, the diffusion rate of carbon dioxide through the stratum corneum is increased, thereby improving the quality of data acquisition while performing non-invasive detection.

[0024] According to the method of the first aspect of the present invention, in S2, multimodal physiological data of the user are collected using a sensing module, wherein the sensing module refers to a multimodal sensor array, including a flexible optical sensor, a multichannel PPG sensor, a bioimpedance sensor, and an accelerometer, and the collected analog signals are digitized and input to a processing unit; wherein: The flexible optical sensor emits an infrared beam with a specific wavelength of 4200nm to penetrate the skin surface of the sampling site and perform specific detection of transdermal carbon dioxide concentration. A multi-channel PPG sensor alternately emits red and infrared light to collect photoplethysmography (PPG) waves to monitor blood oxygen saturation and heart rate. Bioimpedance sensors measure the impedance changes of local tissues using alternating current and extract respiratory frequency waveforms; Accelerometers synchronously collect body motion information; Timestamp multimodal physiological data using a unified clock source.

[0025] According to the method of the first aspect of the present invention, in S3, multimodal physiological data are extracted, and an adaptive filtering algorithm is used, with the body motion information as a reference noise source, to eliminate the interference of motion artifacts on carbon dioxide concentration data and blood oxygen saturation data; wherein: When the target subject turns over or makes hand movements during sleep, the accelerometer outputs the corresponding motion vector; by deploying a least mean square (LMS) adaptive filter, the motion vector is subtracted from the disturbed physiological signal and the weights are adjusted to restore the physiological waveform.

[0026] According to the method of the first aspect of the present invention, in S4, the multidimensional features are extracted using a sliding window mechanism, wherein: Set a time sliding window to calculate the fluctuation trend of carbon dioxide concentration data, the rate of decrease of blood oxygen saturation data, and the variability of respiratory interval; Fluctuation trend extraction: Calculate the first derivative or linear regression slope of carbon dioxide concentration data within the window to quantify the carbon dioxide accumulation rate; Descent rate extraction: Identify local peaks and troughs in blood oxygen saturation data, calculate the ratio of drop depth to drop time to obtain the descent rate, and count the frequency of blood oxygen desaturation. Variation extraction: The peak-finding algorithm is used to identify the extreme points of adjacent cycles of the respiratory rate waveform, extract the time interval sequence, and calculate its standard deviation or coefficient of variation.

[0027] According to the method of the first aspect of the present invention, in S5, the extracted fluctuation trend features, decline rate features and variability features are aligned on the feature dimension based on a unified timestamp and concatenated into a one-dimensional or multi-dimensional feature vector, thereby constructing the physiological feature vector.

[0028] According to the method of the first aspect of the present invention, in S6, the physiological feature vector is input into a pre-deployed prediction model based on a machine learning algorithm to predict the confidence level of sleep apnea events within the current and future time windows; wherein: The learning algorithm refers to the learning algorithm based on the Long Short-Term Memory (LSTM) network or the random forest algorithm. The prediction model uses the time series dependencies to output the probability of obstructive sleep apnea occurring and triggers a vibration warning.

[0029] The method according to the first aspect of the present invention further includes: S7. Establish a personalized baseline database based on sleep physiological data from sleep apnea detection to dynamically adjust the warning threshold.

[0030] A second aspect of this invention provides a sleep apnea detection system based on transcutaneous carbon dioxide monitoring, the system comprising: The heating module is controlled to adjust the temperature of the sampling area to achieve the target temperature. The sensing module collects multimodal physiological data of the user at the self-sampling site under the target temperature; The preprocessing module performs denoising and artifact removal on the multimodal physiological data; The feature extraction module extracts multidimensional features from denoised and artifact-free multimodal physiological data; The alignment and fusion module performs temporal alignment and fusion processing on multidimensional features to construct physiological feature vectors. The detection module detects sleep apnea based on physiological feature vectors and triggers an early warning.

[0031] A third aspect of the present invention provides a wearable device, wherein the wearable device is configured with a detection system as described in the first aspect of the present invention to realize sleep apnea detection based on transcutaneous carbon dioxide monitoring.

[0032] First embodiment: Sleep apnea prediction method; the method includes the following steps: Data acquisition: Simultaneously acquire TcCO2 signals, SpO2 data, respiratory rate information, and body motion information collected by the accelerometer; Signal processing: An adaptive filtering algorithm is used to use the motion state collected by the accelerometer as reference noise to eliminate the interference of motion artifacts on the TcCO2 signal; Feature extraction: Multidimensional features are extracted from the denoised data, including the transcutaneous carbon dioxide concentration fluctuation trend, the rate of decrease in blood oxygen saturation and the variability of respiratory interval. The features of the denoised transcutaneous carbon dioxide (TcCO2) concentration fluctuation trend, the rate of decrease in blood oxygen saturation and the variability of respiratory interval are aligned and fused in time to form multimodal physiological data with multimodal features. Event identification and prediction: The features of the multimodal physiological data are used as input to a prediction model built based on machine learning algorithms to identify sleep apnea events within the current and future time windows and trigger warnings.

[0033] Among them, multimodal physiological features are extracted simultaneously using a sliding window mechanism; specifically including: Extraction of transdermal carbon dioxide concentration fluctuation trend features: Set a time sliding window, calculate the first derivative or linear regression slope of the denoised transdermal carbon dioxide signal within the window to obtain the fluctuation trend features of transdermal carbon dioxide concentration. Blood oxygen saturation decline rate feature extraction: Identify local peaks and troughs in the blood oxygen saturation signal, calculate the ratio of blood oxygen drop depth to drop time, and obtain the blood oxygen saturation decline rate; at the same time, count the frequency of blood oxygen desaturation events within a set time window. Respiratory interval variability feature extraction: Based on the acquired time-frequency waveform, the peak-finding algorithm is used to identify the extreme points of adjacent cycles, extract the time interval sequence between two adjacent respiratory extreme points, and calculate the standard deviation or coefficient of variation of the time interval sequence within the time sliding window to obtain the respiratory interval variability feature. Multimodal feature fusion: The extracted transcutaneous carbon dioxide concentration fluctuation trend feature, blood oxygen saturation decrease rate feature, and respiratory interval variability feature are time-aligned based on a unified timestamp and concatenated to construct a one-dimensional or multi-dimensional feature vector to form the multimodal physiological data.

[0034] The machine learning algorithm is selected from Long Short-Term Memory (LSTM) network or Random Forest algorithm.

[0035] The above methods include: establishing a personalized baseline database for users, continuously collecting long-term sleep physiological data of users, and dynamically adjusting the warning threshold for sleep apnea based on the user's historical physiological characteristics.

[0036] Second embodiment: Sleep apnea prediction system based on transcutaneous carbon dioxide monitoring.

[0037] The first processing module is configured to synchronously acquire the user's multimodal physiological data, including: transcutaneous carbon dioxide signals acquired by the miniaturized transcutaneous carbon dioxide sensor, blood oxygen saturation and heart rate signals acquired by the multi-channel PPG sensor, respiratory rate signals extracted by the bioimpedance sensor, and body movement information acquired by the accelerometer. A PID algorithm is used to control the heating module to maintain a constant temperature of 40℃~45℃ for precise signal acquisition.

[0038] The second processing module is configured to denoise and preprocess the multimodal physiological signals. The second processing module uses an adaptive filtering algorithm to eliminate motion artifacts from the percutaneous carbon dioxide signal by using the body motion information as a reference noise source.

[0039] The third processing module is configured to extract a multidimensional feature set for event judgment from the denoised and purified physiological waveform data. The multidimensional feature set includes: the trend of transcutaneous carbon dioxide concentration fluctuation, the rate of decrease in blood oxygen saturation, and the variability of respiratory interval.

[0040] The fourth processing module is configured to take the multimodal feature dataset as feature input, input it into the breathing event prediction model built based on machine learning algorithm for inference calculation, identify sleep apnea events in real time, and trigger an early warning.

[0041] The fifth processing module is configured to receive long-term multimodal physiological data and early warning records, synchronize the data to the cloud via wireless transmission (Bluetooth 5.0+), establish a personalized physiological baseline database for users, and dynamically calculate and update the early warning threshold based on the user's historical physiological characteristics.

[0042] The third processing module uses a sliding window mechanism to extract the multidimensional feature set, specifically configured as follows: Extraction of transdermal carbon dioxide concentration fluctuation trend features: Set a time sliding window, calculate the first derivative or linear regression slope of the denoised transdermal carbon dioxide signal within the window to obtain the fluctuation trend features of transdermal carbon dioxide concentration. Blood oxygen saturation decline rate feature extraction: Identify local peaks and troughs in the blood oxygen saturation signal, calculate the ratio of blood oxygen drop depth to drop time, and obtain the blood oxygen saturation decline rate; at the same time, count the frequency of blood oxygen desaturation events within a set time window. Respiratory interval variability feature extraction: Based on the acquired time-frequency waveform, the peak-finding algorithm is used to identify the extreme points of adjacent cycles, extract the time interval sequence between two adjacent respiratory extreme points, and calculate the standard deviation or coefficient of variation of the time interval sequence within the time sliding window to obtain the respiratory interval variability feature. Feature fusion construction: The extracted transcutaneous carbon dioxide concentration fluctuation trend feature, blood oxygen saturation decrease rate feature, and respiratory interval variability feature are time-aligned based on a unified timestamp and concatenated to construct a one-dimensional or multi-dimensional feature vector, forming the multimodal physiological data, and output to the fourth processing module.

[0043] Third embodiment: Smart bracelet; the smart bracelet includes: The sensor module is used to simultaneously acquire multimodal physiological data, including a flexible optical sensor, a multi-channel PPG sensor, a bioimpedance sensor, and an accelerometer. The transdermal carbon dioxide detection module includes a flexible optical sensor and a heating control unit. The heating control unit controls the temperature of the sampling site to be constant at 40℃~45℃ using a PID algorithm. Embedded processing unit for sensor data preprocessing, signal processing, and edge computing; The communication module supports Bluetooth 5.0 and above protocols for wireless data transmission and cloud synchronization.

[0044] The flexible optical sensor uses a specific wavelength of 4200nm infrared band to acquire optical detection signals related to transdermal carbon dioxide.

[0045] The multi-channel PPG sensor integrates red and infrared light sources to monitor blood oxygen saturation (SpO2) and heart rate. The respiratory rate parameter is used to further extract respiratory cycle and respiratory interval variability characteristics. The bioimpedance sensor measures the impedance changes of local tissues to extract the respiratory rate parameter. The embedded processing unit adopts a low-power MCU and PCB stacking design, and the embedded unit, sensor module and power module adopt a modular layout.

[0046] Fourth embodiment: Sleep apnea detection method based on transcutaneous carbon dioxide monitoring.

[0047] In S1: a modular hardware architecture is constructed, and the heating module is controlled to precisely regulate the temperature of the sampling area.

[0048] In some embodiments, a modular hardware architecture is constructed, including: adopting a PCB stacking design to highly integrate the sensor module, low-power microcontroller unit (MCU) and power management module in a modular layout; the heating control module includes: controlling the heating unit through a PID algorithm in the miniaturized transdermal carbon dioxide detection module to keep the sampling part of the target object constant in the range of 40°C to 45°C.

[0049] Specifically, a low-power MCU edge computing platform based on the ARM architecture can be built. After the device is started, the MCU reads the thermistor data of the skin contact surface, calculates the deviation between the current temperature and the target temperature in real time through a PID algorithm, and dynamically adjusts the duty cycle of the pulse width modulation (PWM) to smoothly transfer heat to the sampling site.

[0050] For example, the target constant temperature is set at 42℃. In the initial stage, a high duty cycle is output to rapidly heat the heating module, followed by a fine-tuning process to maintain a strictly constant temperature of 42℃ on the skin surface (fluctuation range ≤ ±0.2℃). This method promotes capillary arterialization through precise thermal stimulation, significantly increasing the diffusion rate of carbon dioxide through the stratum corneum, and significantly improving signal acquisition quality while ensuring non-invasive and safe detection.

[0051] S2: Synchronously acquire multimodal physiological data of the target object using a multimodal sensor array.

[0052] In some embodiments, synchronous acquisition using a multimodal sensor array includes: driving a flexible optical sensor, a multi-channel PPG sensor, a bioimpedance sensor, and an accelerometer to work simultaneously, and digitizing the acquired analog signals and storing them in a processing unit.

[0053] Specifically, the flexible optical sensor emits an infrared beam with a specific wavelength of 4200nm, penetrating the skin surface to achieve specific detection of transdermal carbon dioxide concentration (TcCO2); the multi-channel PPG sensor alternately emits red and infrared light to collect photoplethysmography (PPG) waves to monitor blood oxygen saturation (SpO2) and heart rate; the bioimpedance sensor measures the impedance changes of local tissues using a weak alternating current to extract respiratory rate waveforms; and the triaxial accelerometer synchronously collects body motion information. A unified clock source timestamps all data, ensuring seamless alignment between the data acquisition and processing systems.

[0054] S3: Acquire synchronously acquired multimodal physiological signals and perform denoising and artifact removal processing on them.

[0055] In some embodiments, acquiring and processing multimodal physiological signals includes: extracting TcCO2 signals, SpO2 data, respiratory rate information, and body motion information collected by an accelerometer; and using an adaptive filtering algorithm, with the body motion information as a reference noise source, to eliminate the interference of motion artifacts on TcCO2 signals and SpO2 signals.

[0056] Specifically, when the target subject turns over or makes hand movements during sleep, the accelerometer outputs a corresponding motion vector. By deploying a least mean square (LMS) adaptive filter at the edge of the MCU, this motion vector is subtracted from the disturbed physiological signal and the weights are adjusted to restore a pure physiological waveform, effectively avoiding false alarms caused by mechanical movement.

[0057] S4: A sliding window mechanism is used to analyze the denoised data and extract multidimensional features simultaneously.

[0058] In some embodiments, a sliding window mechanism is used to extract multidimensional features, including: setting a time sliding window and calculating the fluctuation trend of the denoised transcutaneous carbon dioxide signal, the rate of decrease of blood oxygen saturation, and the variability of respiratory interval, respectively.

[0059] For example, the sliding window length is set to 60 seconds, with a step size of 10 seconds.

[0060] Specifically: (1) Fluctuation trend extraction: Calculate the first derivative or linear regression slope of the TcCO2 signal within the 60-second window to quantify the carbon dioxide accumulation rate; (2) Descending rate extraction: Identify the local peaks and troughs of the SpO2 signal, calculate the ratio of the drop depth to the drop time to obtain the descending rate, and count the frequency of blood oxygen desaturation events during this period; (3) Variance extraction: The peak finding algorithm is used to identify the extreme points of adjacent cycles of the respiratory rate waveform, extract the time interval sequence, and calculate its standard deviation or coefficient of variation. This method captures the abnormality of ventilation and oxygenation status during apnea through multidimensional quantitative analysis of local fluctuation characteristics.

[0061] S5: Perform temporal alignment and fusion of the extracted multidimensional features to construct multimodal physiological data.

[0062] In some embodiments, constructing multimodal physiological data includes: aligning the fluctuation trend features, descent rate features, and respiratory interval variability features extracted in step S104 along the feature dimensions based on a unified timestamp, and concatenating them into a one-dimensional or multi-dimensional feature vector, which is then used as input to a machine learning prediction model. This step eliminates time delay errors between heterogeneous sensors, providing logically unified, high-quality data support for the machine learning prediction model.

[0063] S6: Input the fused multimodal physiological data into the prediction model to identify respiratory events and trigger early warnings.

[0064] In some embodiments, identifying breathing events includes: inputting the feature vector into a prediction model of a pre-deployed machine learning algorithm (such as a Long Short-Term Memory Network (LSTM) or a Random Forest algorithm) to infer the confidence of sleep apnea events within the current and future time windows.

[0065] Specifically, taking the LSTM model as an example, the model uses its cellular state units to process long-range dependencies in time series. If the model outputs that the probability of obstructive sleep apnea is greater than a preset threshold (e.g., 85%), a vibration alert is immediately triggered on the wristband, and the alert event is wirelessly transmitted to the mobile device or cloud for synchronization via Bluetooth 5.0 or higher.

[0066] S7: Establish a personalized baseline database based on long-term sleep physiological data and dynamically adjust the warning threshold.

[0067] In some embodiments, dynamically adjusting the warning threshold includes: continuously collecting and uploading long-term multimodal data of the user to the cloud, establishing a physiological baseline (such as basal blood oxygen level and basal metabolic carbon dioxide concentration) for the user through an unsupervised statistical algorithm, and dynamically issuing and updating the alarm trigger threshold in step S106 based on the characteristics of the historical baseline.

[0068] For example, if the baseline database shows that a user's average nighttime blood oxygen saturation is only 92%, the system will automatically adjust its warning logic, lowering the individual's absolute blood oxygen trigger threshold to 87%, or changing the trigger condition to a decrease of more than 4% relative to the baseline. This method breaks through the limitations of traditional medical devices with a single fixed threshold, truly achieving personalized health monitoring with high sensitivity and low false alarm rate.

[0069] Fifth embodiment: Sleep apnea detection system based on transcutaneous carbon dioxide monitoring.

[0070] First processing module: The first processing module synchronously collects the user's multimodal physiological data and performs precise constant temperature control.

[0071] In some embodiments, the simultaneous acquisition of multimodal physiological data and the implementation of isothermal control include: driving a miniaturized transcutaneous carbon dioxide sensor, a multi-channel PPG sensor, a bioimpedance sensor, and an accelerometer to operate synchronously; at the same time, the first processing module controls the heating module through a PID algorithm to keep the sampling site of the target object at a constant temperature in the range of 40°C to 45°C, so as to ensure high-quality acquisition of transcutaneous carbon dioxide signals.

[0072] Specifically, the first processing module is equipped with a high-precision clock synchronization mechanism. A miniaturized transdermal carbon dioxide sensor acquires transdermal carbon dioxide signals, a multi-channel PPG sensor acquires blood oxygen saturation and heart rate signals, a bioimpedance sensor extracts respiratory rate signals, and an accelerometer acquires body motion information. The first processing module uses a PID algorithm to calculate the difference between the current skin temperature and the target constant temperature in real time, and dynamically adjusts the output power of the heating module.

[0073] For example, the target constant temperature is set at 42°C. The first processing module drives the heating element to rapidly heat up and maintain a constant temperature of 42°C (±0.2°C), thereby arterializing the local capillaries. Simultaneously, the module uses a unified time reference and allows for timestamp alignment after data acquisition from different sensors at different sampling rates. This method provides a rigorous time reference and high-quality underlying data for subsequent multimodal data fusion.

[0074] Second processing module: The second processing module performs noise reduction and preprocessing on the multimodal physiological signals.

[0075] In some embodiments, denoising of multimodal physiological signals includes: the second processing module calling an adaptive filtering algorithm to use the body motion information acquired by the accelerometer as a reference noise source to eliminate the interference of motion artifacts on the transcutaneous carbon dioxide signal and blood oxygen saturation signal.

[0076] Specifically, the second processing module is equipped with a least mean square (LMS) or recursive least squares (RLS) adaptive filter. When a user moves their body at night, such as turning over, the system inputs the three-axis motion vector generated by the accelerometer into the filter as a noise reference sequence, dynamically calculates the filter weights, and adaptively cancels out baseline drift and waveform distortion caused by motion from the original mixed signal. This method significantly reduces the drastic fluctuations in physiological signals caused by unconscious body movements, fundamentally eliminating false positives caused by motion artifacts.

[0077] The third processing module: The third processing module uses a sliding window mechanism to extract multi-dimensional feature sets and perform temporal fusion.

[0078] In some embodiments, extracting and fusing multidimensional feature sets includes: within a set time sliding window, calculating in parallel the fluctuation trend of denoised transcutaneous carbon dioxide concentration, the rate of decrease in blood oxygen saturation, and the variability of respiratory interval, and splicing them to construct a multimodal physiological data feature vector.

[0079] For example, the time sliding window length is set to 60 seconds, and the step size is 10 seconds. The third processing module is specifically configured to execute the following sub-steps: (1) Extraction of fluctuation trend features: Calculate the least squares linear regression slope (or first derivative) of the denoised transdermal carbon dioxide signal within a 60-second window. If the slope is greater than zero, it indicates that carbon dioxide is in an accumulation state. (2) Extraction of the rate of decline feature: Identify the local peaks and troughs of the blood oxygen saturation signal within the window, calculate the ratio of the blood oxygen drop depth (e.g., a drop of 4%) to the drop time (e.g., 20 seconds) to obtain the rate of decline (0.2% / s); at the same time, count the frequency of blood oxygen desaturation events (i.e., ODI index) within the set time window. (3) Extraction of variation features: Based on the denoised respiratory frequency waveform, the peak detection algorithm is used to identify the extreme points of adjacent respiratory cycles, extract the time interval sequence between two adjacent respiratory extreme points (e.g., 3s in normal cases, and extended in abnormal cases), and calculate the standard deviation or coefficient of variation of the time interval sequence. (4) Feature fusion construction: The fluctuation trend, rate of decline and variability features calculated above are time-series aligned based on a unified timestamp and spliced ​​together to construct a multi-dimensional feature vector, which is then output to the fourth processing module. This method transforms the messy time series waveform into a quantitative matrix with high clinical representation significance.

[0080] Fourth processing module: The fourth processing module runs the respiratory event prediction model to perform inference calculations and triggers early warnings.

[0081] In some embodiments, performing inference calculations and triggering an early warning includes: the fourth processing module taking the multimodal physiological data feature vector as input and inputting it into a respiratory event prediction model constructed based on a machine learning algorithm to identify sleep apnea events in real time; and triggering a local or remote early warning when the output probability or severity reaches a set condition.

[0082] Specifically, the fourth processing module deploys a classification and early warning model constructed using a Long Short-Term Memory (LSTM) network or a random forest algorithm. Taking LSTM as an example, after receiving the fused feature vector, the model combines it with the historical temporal context state stored in its internal memory units to infer the confidence level of airway obstruction occurring within the current time window.

[0083] For example, if the fused feature vector shows an increased TcCO2 slope and a faster rate of SpO2 decline, the model infers a 92% probability of a sleep apnea event occurring. Since this probability exceeds a preset trigger threshold (e.g., 85%), the system immediately triggers a hardware motor to vibrate and wake the user to change position, effectively ensuring the patient's nighttime ventilation safety.

[0084] Fifth processing module: A personalized physiological baseline database is established through the fifth processing module, and the warning threshold is dynamically calculated and updated.

[0085] In some embodiments, dynamically calculating and updating the warning threshold includes: the fifth processing module receiving long-term multimodal physiological data and warning records, synchronizing them to a cloud server via a wireless transmission module supporting Bluetooth 5.0 and above protocols; establishing a user-specific physiological baseline database using unsupervised learning, and automatically adjusting the warning trigger parameters based on historical characteristics.

[0086] Specifically, the cloud system continuously analyzes the user's sleep data for multiple nights, and statistically analyzes their physiological characteristics such as basal blood oxygen levels and basal metabolic carbon dioxide concentration.

[0087] For example, if the fifth processing module detects that a user with a consistently low baseline blood oxygen saturation level has been maintained at around 91% for an extended period, the system will activate an adaptive adjustment mechanism. This will dynamically lower the absolute threshold for triggering a blood oxygen saturation alarm for that user from the standard 90% to 86% (or adjust it to a relative baseline decrease of ≥4%), and then send the updated threshold to the fourth processing module. This method achieves truly personalized health management, completely eliminating the frequent false alarms caused by fixed thresholds for specific patient groups.

[0088] Compared with the prior art, the first solution proposed in this invention has the following essential differences.

[0089] (1) Breaking through the limitations of single blood oxygen monitoring, significantly improving the foresight and comprehensiveness of early warning. Existing wearable sleep monitoring devices mainly rely on changes in blood oxygen saturation (SpO2) or heart rate to detect respiratory events, which is a lagging response to the results of apnea; while this invention introduces the trend of transcutaneous carbon dioxide (TcCO2) concentration change as the core feature, and models from the early stage of insufficient ventilation to achieve forward prediction of apnea events.

[0090] (2) Overcoming the bottleneck of motion artifact interference and significantly reducing the false alarm rate of home monitoring. In response to the pain points of signal baseline drift and motion artifacts caused by nighttime sleep turning over and hand waving, this invention innovatively introduces an adaptive filtering algorithm based on a triaxial accelerometer. By using body motion information as reference noise input to the adaptive filtering model, the invention achieves synchronous artifact removal processing of multimodal signals, significantly improving the detection stability in home scenarios.

[0091] (3) Achieving a balance between non-invasive detection, improved detection stability, and safe temperature control. Traditional transdermal carbon dioxide detectors are bulky and require extremely high temperatures. This invention uses a flexible optical sensor with a specific wavelength (4200nm infrared band) combined with a PID high-precision temperature control algorithm to precisely stabilize the wrist contact surface at 40℃~45℃. This solution effectively promotes local capillary arterialization, increases the diffusion rate of carbon dioxide gas to obtain high-quality weak signals, and avoids the risk of skin burns caused by excessively high temperatures, achieving non-invasive continuous and safe monitoring with medical-grade precision.

[0092] (4) Existing multimodal monitoring systems usually process each sensor signal independently or simply splice them together. However, this invention uses a time series alignment mechanism to couple and model the TcCO2 trend, SpO2 decrease rate and respiratory interval variability, thereby constructing a unified feature space that reflects the dynamic relationship between ventilation and oxygenation.

[0093] (5) Constructing an edge-cloud collaborative architecture to achieve personalized dynamic early warning for each user. This invention not only supports edge computing and real-time event recognition on the wristband (low-power MCUs such as ARM Cortex-M4), but also establishes a personalized physiological baseline database in the cloud through wireless transmission. The system can dynamically calculate and adjust the warning threshold for sleep apnea based on the user's long-term historical sleep characteristics (such as baseline heart rate, nighttime baseline blood oxygen, and metabolic baseline). This adaptive mechanism breaks the rigid setting of "fixed threshold" in traditional medical devices and achieves truly personalized and precise health management.

[0094] (6) Balancing high integration and long battery life, greatly improving patient compliance. Traditional polysomnography (PSG) involves numerous wiring and requires hospitalization, resulting in a poor patient experience. This invention achieves highly miniaturized integration of multimodal sensors (optical, impedance, acceleration) and heating modules through PCB stacking design. Combined with optimized underlying firmware architecture and low-power power management, it achieves a battery life of more than 7 days in continuous multi-parameter monitoring mode. The product is lightweight and comfortable, reducing sleep constraints for patients and greatly improving compliance with long-term home-based continuous monitoring for patients with chronic diseases.

[0095] In summary, this invention integrates a miniaturized transcutaneous carbon dioxide detection module composed of a flexible optical sensor and a multi-channel PPG, a bioimpedance sensor, and an accelerometer; it uses a PID algorithm to control the heating module to achieve constant temperature, ensuring accurate non-invasive transcutaneous carbon dioxide detection; it establishes a multi-sensor data fusion system to achieve synchronous acquisition of physiological parameters; and it employs a low-power embedded processing unit to achieve real-time processing and monitoring of multimodal physiological data. This invention also incorporates a machine learning model and combines TcCO2 fluctuations, SpO2 decline rate, and respiratory interval variability to identify and issue early warnings for sleep apnea events in real time. This invention achieves highly integrated, long-lasting continuous respiratory health monitoring, effectively improving the home-based screening capabilities for sleep-disordered breathing.

[0096] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring, characterized in that, The method includes: S1. Adjust the temperature of the sampling area to reach the target temperature; S2. At the target temperature, collect multimodal physiological data of the user from the self-sampling site; S3. Denoising and artifact removal are performed on the multimodal physiological data; S4. Extract multidimensional features from denoised and artifact-free multimodal physiological data; S5. Perform temporal alignment and fusion processing on multidimensional features to construct physiological feature vectors; S6. Detect sleep apnea based on physiological feature vectors and trigger an early warning.

2. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 1, characterized in that, In S1, the control heating module collects thermistor data from the sampling location to obtain the current temperature of the sampling location. A PID algorithm is used to calculate the deviation between the current temperature and the target temperature in real time. The duty cycle of dynamic pulse width modulation (PWM) is adjusted to smoothly transfer heat to the sampling location, ensuring that the sampling location reaches the target temperature. Wherein: By stimulating the capillaries with heat, the diffusion rate of carbon dioxide through the stratum corneum is increased, thereby improving the quality of data acquisition while performing non-invasive detection.

3. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 2, characterized in that, In S2, a sensing module is used to collect the user's multimodal physiological data. The sensing module refers to a multimodal sensor array, including a flexible optical sensor, a multi-channel PPG sensor, a bioimpedance sensor, and an accelerometer. The collected analog signals are digitized and then input to the processing unit. The flexible optical sensor emits an infrared beam with a specific wavelength of 4200nm to penetrate the skin surface of the sampling site and perform specific detection of transdermal carbon dioxide concentration. A multi-channel PPG sensor alternately emits red and infrared light to collect photoplethysmography (PPG) waves to monitor blood oxygen saturation and heart rate. Bioimpedance sensors measure the impedance changes of local tissues using alternating current and extract respiratory frequency waveforms; Accelerometers synchronously collect body motion information; Timestamp multimodal physiological data using a unified clock source.

4. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 3, characterized in that, In S3, multimodal physiological data are extracted, and an adaptive filtering algorithm is used, with the body motion information as a reference noise source, to eliminate the interference of motion artifacts on carbon dioxide concentration data and blood oxygen saturation data; wherein: When the target subject turns over or makes hand movements during sleep, the accelerometer outputs the corresponding motion vector; by deploying a least mean square (LMS) adaptive filter, the motion vector is subtracted from the disturbed physiological signal and the weights are adjusted to restore the physiological waveform.

5. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 4, characterized in that, In S4, a sliding window mechanism is used to extract the multidimensional features, wherein: Set a time sliding window to calculate the fluctuation trend of carbon dioxide concentration data, the rate of decrease of blood oxygen saturation data, and the variability of respiratory interval; Fluctuation trend extraction: Calculate the first derivative or linear regression slope of carbon dioxide concentration data within the window to quantify the carbon dioxide accumulation rate; Descent rate extraction: Identify local peaks and troughs in blood oxygen saturation data, calculate the ratio of drop depth to drop time to obtain the descent rate, and count the frequency of blood oxygen desaturation. Variation extraction: The peak-finding algorithm is used to identify the extreme points of adjacent cycles of the respiratory rate waveform, extract the time interval sequence, and calculate its standard deviation or coefficient of variation.

6. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 5, characterized in that, In S5, the extracted fluctuation trend features, decline rate features, and variability features are aligned on the feature dimension based on a unified timestamp and concatenated into a one-dimensional or multi-dimensional feature vector to construct the physiological feature vector.

7. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 6, characterized in that, In S6, the physiological feature vector is input into a pre-deployed prediction model based on a machine learning algorithm to predict the confidence level of sleep apnea events within the current and future time windows; wherein: The learning algorithm refers to the learning algorithm based on the Long Short-Term Memory (LSTM) network or the random forest algorithm. The prediction model uses the time series dependencies to output the probability of obstructive sleep apnea occurring and triggers a vibration warning.

8. The method for detecting sleep apnea based on transcutaneous carbon dioxide monitoring according to claim 7, characterized in that, The method further includes: S7. Establish a personalized baseline database based on sleep physiological data from sleep apnea detection to dynamically adjust the warning threshold.

9. A sleep apnea detection system based on transcutaneous carbon dioxide monitoring, characterized in that, The system includes: The heating module is controlled to adjust the temperature of the sampling area to achieve the target temperature. The sensing module collects multimodal physiological data of the user at the self-sampling site under the target temperature; The preprocessing module performs denoising and artifact removal on the multimodal physiological data; The feature extraction module extracts multidimensional features from denoised and artifact-free multimodal physiological data; The alignment and fusion module performs temporal alignment and fusion processing on multidimensional features to construct physiological feature vectors. The detection module detects sleep apnea based on physiological feature vectors and triggers an early warning.

10. A wearable device, characterized in that, The wearable device is configured with the detection system as described in claim 9 to realize sleep apnea detection based on transcutaneous carbon dioxide monitoring.

Images