Non-contact sleep respiration monitoring method, equipment and system
By acquiring signals through non-contact fiber optic sensors and PPG sensors, and combining blind source separation and fuzzy inference systems, the limitations of existing sleep breathing monitoring devices in medical institutions have been overcome, enabling accurate determination of sleep breathing events and index calculation at home.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF GUANGZHOU MEDICAL UNIV (GUANGZHOU RESPIRATORY CENT)
- Filing Date
- 2026-04-08
- Publication Date
- 2026-05-12
AI Technical Summary
Existing sleep apnea monitoring devices (PSG) are only suitable for medical institutions such as hospitals, and require several electrodes to be attached to the patient, which affects comfort and may change sleep habits, leading to inaccurate monitoring results.
A non-contact method was adopted to acquire chest and abdominal vibration signals through fiber optic sensors, combined with pulse wave signals and blood oxygen saturation values collected by PPG sensors. Blind source separation and template enhancement were performed to construct a multi-feature fuzzy inference system, determine the event type, and calculate the sleep apnea-hypopnea index.
It enables undisturbed sleep breathing monitoring in a home environment, accurately identifies sleep breathing events, and automatically calculates the sleep apnea-hypopnea index, improving the accuracy and comfort of monitoring.
Smart Images
Figure CN122004789A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of sleep monitoring technology, and in particular relates to a non-contact sleep breathing monitoring method, device and system. Background Technology
[0002] Sleep apnea-hypopnea syndrome (SAHS) is a clinical syndrome characterized by recurrent apnea and / or hypopnea during sleep, caused by various factors, leading to hypoxemia, hypercapnia, or sleep disruption, and resulting in a series of pathophysiological changes. As the condition progresses, complications such as pulmonary hypertension, cor pulmonale, respiratory failure, hypertension, arrhythmia, and cerebrovascular accidents may occur. Sleep apnea includes central, obstructive, and mixed types. Continuous monitoring of sleep apnea-hypopnea can not only diagnose the condition and provide a basis for subsequent treatment, but also dynamically assess the patient's condition and evaluate treatment effectiveness during the treatment process.
[0003] The industry uses the AHI (Apnea-Hypopnea Index) to indicate the severity of sleep apnea-hypopnea syndrome. It represents the number of pauses and hypopnea per minute. Generally, an AHI < 5 is considered normal, 5 ≤ AHI < 15 is mild, 15 ≤ AHI < 30 is moderate, and AHI ≥ 30 is severe. Currently, the gold standard in the industry is polysomnography (PSG), which involves continuously and synchronously recording multiple parameters throughout the night, including electroencephalography (EEG) (to analyze sleep structure), electrooculography (EOG), mandibular electromyography (EMG), airflow and respiratory movements through the mouth and nose, electrocardiogram (ECG), blood oxygen saturation, snoring, limb movements, and body position. Professional technicians monitor the sleep throughout the night, and the results are manually interpreted the following day to generate a sleep report. However, the application scenarios of PSG are limited, and it is only suitable for medical institutions such as hospitals and clinics, and requires the participation of professional personnel. In addition, the monitoring process requires attaching several electrodes to the patient, which leads to poor comfort. This will change the patient's sleep habits, and the patient needs to stay overnight on the medical bed, which will affect some people who are more sensitive to the sleep environment. All of these factors may cause the monitoring results to not reflect the patient's true situation. Therefore, the demand for home-use sleep monitoring devices that can be operated by the patient themselves has been created, in order to achieve convenient and accurate monitoring of the patient's sleep breathing in the home environment. Summary of the Invention
[0004] The purpose of this invention is to provide a non-contact sleep apnea monitoring method, device, and system, which aims to solve the problem that existing sleep apnea monitoring devices (PSG) are only suitable for medical institutions such as hospitals and clinics, and that the monitoring process requires attaching several electrodes to the patient, which may result in the monitoring results not reflecting the patient's true condition.
[0005] In a first aspect, the present invention provides a non-contact sleep apnea monitoring method, the method comprising: S101. The chest vibration signal and the abdominal vibration signal are acquired by the first fiber optic sensor placed under the chest and the second fiber optic sensor placed under the abdomen of the supine subject, respectively, and the pulse wave signal and blood oxygen saturation value are collected from the fingers of the supine subject simultaneously by the PPG sensor. S102. After blind source separation of chest vibration signal and abdominal vibration signal, respiratory source signal, cardiac source signal and body motion noise source signal are extracted respectively. The cardiac source signal is templated and enhanced to obtain the heartbeat event time sequence. S103. Based on respiratory source signals, pulse wave signals, body motion noise source signals, blood oxygen saturation values, chest vibration signals, and abdominal vibration signals, calculate the respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body motion energy index, respiratory effort index, and chest-abdominal contradiction index. S104. Construct a multi-feature fuzzy reasoning system to determine the event type based on five features: respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body energy index, respiratory effort index, and chest-abdominal contradiction index. S105. Obtain the sleep period label for each time period based on the heartbeat event time sequence, respiratory source signal and body energy index, calculate the total sleep time based on the sleep period label, and count the total number of effective respiratory events that occurred within the total sleep time interval based on the event markers. Calculate the sleep apnea-hypopnea index based on the total number of effective respiratory events and the total sleep time.
[0006] In a second aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the non-contact sleep breathing monitoring method as described above.
[0007] Thirdly, the present invention provides a non-contact sleep apnea monitoring device, comprising: Communication port; A processor for executing computer instructions, including steps for performing the non-contact sleep apnea monitoring method as described; Internal communication bus, used for communication between systems; The memory is configured to store data and instructions. Input / output components are configured to support data input / output.
[0008] Fourthly, the present invention provides a non-contact sleep apnea monitoring system, comprising: One or more first fiber optic sensors are used to acquire chest vibration data of the object; One or more second fiber optic sensors are used to acquire abdominal vibration data of the object; One or more PPG sensors are used to synchronously acquire pulse wave data of the object; and As described above, this is a non-contact sleep breathing monitoring device.
[0009] This invention acquires chest vibration signals and abdominal vibration signals respectively using a first fiber optic sensor placed below the chest and a second fiber optic sensor placed below the abdomen of a supine subject. Simultaneously, a PPG sensor collects pulse wave signals and blood oxygen saturation values from the fingers of the supine subject. After blind source separation of the chest and abdominal vibration signals, respiratory, cardiac, and body motion noise signals are extracted. The cardiac signal is templated and enhanced to obtain a heartbeat event time sequence. A multi-feature fuzzy inference system is constructed to determine the event type. The sleep apnea-hypopnea index is calculated based on the total number of effective respiratory events and total sleep time. This application enables the identification of sleep apnea and hypopnea events in a undisturbed home environment and achieves automatic calculation of the sleep apnea-hypopnea index. Attached Figure Description
[0010] Figure 1 This is a flowchart of the non-contact sleep breathing monitoring method provided in Embodiment 1 of the present invention; Figure 2 This is a structural block diagram of the non-contact sleep breathing monitoring device provided in Embodiment 3 of the present invention; Figure 3 This is a schematic diagram of the non-contact sleep breathing monitoring system provided in Embodiment 4 of the present invention. Detailed Implementation
[0011] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0012] To illustrate the technical solution described in this invention, specific embodiments are described below.
[0013] Example 1: Please see Figure 1 The non-contact sleep apnea monitoring method provided in Embodiment 1 of the present invention includes the following steps. It should be noted that if substantially the same results are obtained, the non-contact sleep apnea monitoring method of the present invention is not based on... Figure 1 The sequence of processes shown is limited.
[0014] S101. The chest vibration signal and the abdominal vibration signal are acquired by the first fiber optic sensor placed under the chest and the second fiber optic sensor placed under the abdomen of the supine subject, respectively. At the same time, the pulse wave signal and blood oxygen saturation value are collected from the fingers of the supine subject by the PPG sensor.
[0015] In Embodiment 1 of the present invention, the length of both the first and second fiber optic sensors is at least 40 cm to match the width of most human bodies, and the width of both the first and second fiber optic sensors is at least 5 cm. Both sensors continuously acquire raw vibration signals at a sampling rate of at least 100 Hz. The PPG sensor is integrated into a finger sleeve, which is worn on the finger of a supine subject to simultaneously acquire pulse wave signals and blood oxygen saturation values.
[0016] In Embodiment 1 of the present invention, the non-contact sleep apnea monitoring method further includes the following steps: A unified microsecond-level high-precision hardware timestamp is applied to the chest vibration signal, abdominal vibration signal, and pulse wave signal to ensure strict synchronization of the chest vibration signal, abdominal vibration signal, and pulse wave signal, providing a timing basis for subsequent multimodal signal fusion.
[0017] S102. After blind source separation of chest vibration signal and abdominal vibration signal, respiratory source signal, heartbeat source signal and body motion noise source signal are extracted respectively. The heartbeat source signal is templated and enhanced to obtain the heartbeat event time sequence.
[0018] In Embodiment 1 of the present invention, S102 specifically includes: S1021. Preprocess and filter the chest vibration signal and abdominal vibration signal to remove power frequency interference and baseline drift, and then perform ensemble empirical mode decomposition (EEMD) to obtain a set of intrinsic mode functions (IMF) for each.
[0019] S1022. For both chest and abdominal vibration signals, calculate the average instantaneous frequency and normalized energy of each IMF component in each group of intrinsic mode functions. Select components that simultaneously satisfy the following conditions: average instantaneous frequency within the range of 0.1 Hz to 10 Hz (to cover the fundamental and harmonic frequencies of respiration and heartbeat, as well as body motion components) and normalized energy greater than 1%, to form the chest candidate component set. and abdominal candidate component set .
[0020] S1023. Construct a variable step-size recursive least squares (VSS-RLS) adaptive filter for the chest candidate component set. and abdominal candidate component set Filtering is performed. Specifically, the candidate chest component set is... and abdominal candidate component set The sum of the absolute values of the amplitudes of all components at each moment is used as the reference noise signal. This is used to comprehensively characterize global high-frequency vibrations and instantaneous large-amplitude motions. A reference noise signal is used. As reference noise input, the chest candidate component set IMF_C and the abdominal candidate component set... The synthesized signal is used as the main input for parallel adaptive filtering, with the step size factor μ based on... Instantaneous power adaptive adjustment: ,in, The initial step size, The attenuation coefficient; Finally, the preliminarily denoised chest-shaped mixed signal is output. Mixed signals from the abdomen .
[0021] S1024. The chest vibration signal is bandpass filtered at 0.1-0.5Hz to obtain a low-frequency respiratory reference signal, and the chest mixed signal is... Abdominal mixed signal Together with the low-frequency respiratory reference signal, they constitute the observation matrix X(t). X(t) uses the FastICA algorithm based on maximizing negative entropy to output several independent components with the same dimension as the observation matrix. The cross-correlation coefficient between each independent component and the low-frequency respiratory reference signal is calculated, and the component with the highest cross-correlation coefficient is identified as the respiratory source signal S_resp(t). The power of each component in the 0.8-3Hz frequency band is calculated, and the component with the highest power is identified as the heartbeat source signal S_heart(t). The remaining main components are regarded as the body motion noise source signal S_motion(t).
[0022] S1025. The heartbeat source signal is enhanced using a template, including: bandpass filtering (e.g., 0.8-3Hz) of the heartbeat source signal S_heart(t), and detecting candidate heartbeat peaks using an adaptive thresholding method. A fixed-duration signal segment is extracted centered on each peak point, and a typical heartbeat waveform template is generated through cluster analysis (e.g., K-means). All signal segments are cross-correlated with the template, and segments with high correlation (e.g., cross-correlation coefficient greater than 0.8) are retained as qualified segments. Subsampling is then performed based on the cross-correlation peaks to obtain a high-precision heartbeat event time sequence (i.e., the precise time point of each heartbeat R-peak). All qualified segments are aligned according to the heartbeat event time sequence, and coherent averaging is performed, ultimately outputting an enhanced heartbeat source signal S_heart_enhanced(t) with a significantly improved signal-to-noise ratio. The heartbeat event time sequence will be used to calculate heart rate variability characteristics.
[0023] S103. Based on respiratory source signals, pulse wave signals, body motion noise source signals, blood oxygen saturation values, chest vibration signals, and abdominal vibration signals, calculate the respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body motion energy index, respiratory effort index, and chest-abdominal contradiction index.
[0024] In Embodiment 1 of the present invention, S103 specifically includes: S1031. Perform a Hilbert transform on the respiratory signal to obtain the respiratory signal envelope Env_resp(t) representing the amplitude change. Calculate the preset percentile of Env_resp(t) within a sliding time window (the window can be set, e.g., 5 minutes) (any value between the 90th and 99th percentiles can be set according to actual needs), and dynamically update it (e.g., every 30 seconds) as the current respiratory baseline. Calculate the moving average of the blood oxygen saturation value using the same sliding time window, and dynamically update it (e.g., every 30 seconds) as the blood oxygen saturation baseline.
[0025] Calculate the respiratory amplitude drop ratio (R_drop): R_drop = [B_resp(t) - Env_resp(t)] / B_resp(t), which is the percentage decrease of the current respiratory envelope value relative to its dynamic baseline, where B_resp(t) is the respiratory baseline and Env_resp(t) is the respiratory signal envelope.
[0026] Blood oxygen saturation drop (S_drop): S_drop = B_spo2(t) - SpO2(t), which is the absolute value of the current blood oxygen value relative to its dynamic baseline, where B_spo2(t) is the blood oxygen saturation baseline and SpO2(t) is the blood oxygen saturation value.
[0027] S1032. Perform a 3Hz high-pass filter on the body motion noise source signal S_motion(t) and the chest vibration signal V_chest(t). Calculate the root mean square value of each filtered signal within a short time window (e.g., every 2 seconds) to obtain the body motion noise source signal energy E_motion(t) and the original high-frequency vibration energy E_hf(t), and then fuse them. The kinetic energy of the original fusion body is E_raw(t) = sqrt(α*E_motion(t)). 2 +β* E_hf(t) 2), where α and β are weighting coefficients and satisfy α + β = 1 (e.g., α = 0.7, β = 0.3). Calculate the 99th percentile of E_raw(t) within a past time window (e.g., 10 minutes) as the normalization baseline. Divide the current E_raw(t) by the normalization baseline to obtain the body motion energy index A(t). A(t) > 1 indicates the presence of significant body motion.
[0028] S1033. Perform 0.1-0.5Hz bandpass filtering on the chest vibration signal V_chest(t) and the abdominal vibration signal V_abd(t), and calculate the Hilbert envelope to obtain the chest respiratory envelope Env_c(t) and the abdominal respiratory envelope Env_a(t).
[0029] Calculate the instantaneous respiratory effort amplitude: Eff_inst(t) = sqrt(Env_c(t)) 2 + Env_a(t) 2 ).
[0030] Calculate the median of Eff_inst(t) over a sliding time window (e.g., 5 minutes) as the dynamic baseline Eff_base(t).
[0031] The final respiratory effort index is Eff_idx(t) = Eff_inst(t) / Eff_base(t). Here, Eff_idx(t) > 1.2 indicates increased respiratory effort.
[0032] S1034. Perform Hilbert transform on the 0.1-0.5Hz bandpass filtered signals of the chest vibration signal V_chest(t) and the abdominal vibration signal V_abd(t) to obtain the instantaneous phases φ_c(t) and φ_a(t).
[0033] Calculate the instantaneous phase difference Δφ(t) = φ_c(t) - φ_a(t) and normalize it to the interval [-π, π].
[0034] Within a decision window (e.g., 10 seconds), the percentage of sampling points where |Δφ(t)| > (2π / 3) radians (i.e., 120 degrees) is counted. This percentage is the chest-abdominal paradox index P(t). The threshold of 120 degrees is based on clinical observations, where significant chest-abdominal paradoxical breathing typically manifests as near-opposite (180-degree) movements; setting this threshold effectively captures moderate to severe asynchrony. P(t) > 30% indicates the presence of significant chest-abdominal paradoxical breathing.
[0035] S104. Construct a multi-feature fuzzy reasoning system to determine the event type based on five features: respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body energy index, respiratory effort index, and chest-abdominal contradiction index.
[0036] S104 specifically includes: S1041. Define an output fuzzy variable, the event comprehensive confidence level, with a value range of [0, 1], and define a fuzzy set {extremely low, low, medium, high, extremely high} and the corresponding membership function for it.
[0037] S1042. Input the respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body energy index, respiratory effort index, and chest-abdominal contradiction index into a predefined membership function to obtain their membership degree to fuzzy linguistic variables such as "extremely low", "low", "medium", "high", and "extremely high". For example, a trapezoidal function can be used as the membership function.
[0038] S1043. Construct a fuzzy rule base containing multiple if-then rules. The rules can be adjusted according to different groups of people. The rule form is: "If (input condition) then (event type and comprehensive confidence is X)", where X is a fuzzy linguistic value in the output variable, such as "high".
[0039] For example: Rule 1: If the respiratory amplitude drop ratio R_drop is "extremely high", the body energy index A(t) is "low", and the respiratory effort index E_idx(t) is "low", then the event is judged as "central apnea" with a comprehensive confidence level of "high".
[0040] Rule 2: If the respiratory amplitude drop ratio R_drop is "extremely high", the body energy index A(t) is "low", the respiratory effort index E_idx(t) is "high", and the chest-abdomen paradox index P(t) is "high", then the event is judged as "obstructive sleep apnea" with a comprehensive confidence level of "high".
[0041] Rule 3: If the initial stage of an event satisfies the characteristics of Rule 2 (obstructive), and the subsequent stage of the event satisfies the characteristics of Rule 1 (central), then the event is a "mixed apnea" with a "high" confidence level.
[0042] Rule 4: If the respiratory amplitude decrease is "high" (R_drop), the blood oxygen saturation decrease is "medium" (S_drop), and the body energy index A(t) is "low", then the event is judged as "hypoventilation" with a comprehensive confidence level of "high".
[0043] S1044. For each rule in the fuzzy rule base, calculate the minimum membership degree of all its antecedent (i.e., if part) conditions, and use this minimum as the trigger strength of the rule. Each rule activates the corresponding output fuzzy set of its consequent (then part) based on its trigger strength (e.g., "central apnea_high confidence"). The output fuzzy sets of all activated rules are then maximized and aggregated into a total output fuzzy set.
[0044] S1045. Calculate the centroid of the total output fuzzy set using the centroid method, mapping it to a scalar between 0 and 1, which is the overall confidence score C of this event. An event is officially adopted by the system only if its duration is greater than or equal to 10 seconds and its overall confidence score C is greater than or equal to a threshold (the threshold can be customized, for example, 0.65). The specific type of the adopted event (such as central apnea, obstructive apnea, mixed apnea, hypoventilation) is determined by the event type indicated by the rule with the highest trigger strength.
[0045] The "fuzzification-reasoning-defuzzification" process in Embodiment 1 of this invention is mainly to solve the uncertainty of physiological signals and the fuzziness of clinical auxiliary interpretation, overcome the "cliff effect" of rigid thresholds, and enhance the robustness and fault tolerance of the system.
[0046] S105. Obtain the sleep period label for each time period based on the heartbeat event time sequence, respiratory source signal and body energy index, calculate the total sleep time based on the sleep period label, and count the total number of effective respiratory events that occurred within the total sleep time interval based on the event markers. Calculate the sleep apnea-hypopnea index based on the total number of effective respiratory events and the total sleep time.
[0047] Specifically, including: S1051. Divide the signals monitored throughout the night, including the heartbeat event time sequence {t_k}, respiratory source signal S_resp(t), and body energy index A(t), into continuous preset time periods (e.g., 30 seconds). Extract the heart rate variability features, body movement features, respiratory features, and event markers for each time period. Input these features into a hidden Markov model for automatic staged sleep, and output the sleep stage label for each time period (awake W, light sleep N1, deep sleep N2, REM sleep). Heart rate variability characteristics: Based on the heartbeat event time sequence obtained in step S1025, calculate the time domain and frequency domain characteristics within this time period, such as: root mean square of the difference between adjacent RR intervals, and the ratio of low-frequency to high-frequency power.
[0048] Body motion characteristics: Based on the body motion energy index A(t), calculate the average body motion energy during this period.
[0049] Respiratory characteristics: Based on the respiratory source signal S_resp(t), the average respiratory rate and respiratory amplitude variability during this period are calculated.
[0050] Event marker: Whether the period contains a valid breathing event determined in step S104. A valid breathing event refers to an adopted event that has been determined in step S1045, has a duration of ≥10 seconds, and has a comprehensive confidence score C ≥ the threshold (e.g., the threshold is 0.65), including central apnea, obstructive apnea, mixed apnea, and hypoventilation.
[0051] S1052. In the sleep period tag sequence, search backward from the starting point and determine the start time of the first period in the first consecutive non-awake period (i.e., sleep period) as the time of falling asleep. Search backward from the ending point and determine the end time of the last sleep period before the last consecutive wake period as the time of waking up.
[0052] S1053. Determine the total sleep time (TST) based on the determined sleep onset and wake-up times. Calculate the apnea-hypopnea index (AHI) by counting the total number of effective breathing events occurring within the total sleep time (TST) interval based on event markers: AHI = Total number of effective breathing events / Total sleep time (TST) (hours).
[0053] Example 2: Embodiment 2 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the non-contact sleep breathing monitoring method provided in Embodiment 1 of the present invention.
[0054] Example 3: Embodiment 3 of the present invention provides a non-contact sleep apnea monitoring device. Figure 2 The diagram shown is a structural block diagram of a non-contact sleep breathing monitoring device 200.
[0055] The non-contact sleep apnea monitoring device 200 may include a communication port 201 connected to a network thereto for data communication. The non-contact sleep apnea monitoring device 200 may also include a processor 203, which, in the form of one or more processors, is used to execute computer instructions. These computer instructions may include, for example, steps of performing the non-contact sleep apnea monitoring method described in Embodiment 1 of the present invention.
[0056] In some examples, processor 203 may include one or more hardware processors, such as microcontrollers, microprocessors, reduced instruction set computers (RISC), application-specific integrated circuits (ASICs), graphics processing units (GPUs), central processing units (CPUs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs), advanced RISC machines (ARMs), programmable logic devices (PLDs), and any circuit or processor capable of performing one or more functions, or any combination thereof.
[0057] The non-contact sleep apnea monitoring device 200 may include an internal communication bus 205 for inter-system communication, a memory 207 configured to store data and instructions, and program instructions stored in the memory 207 in other types of non-transitory storage media that are executed by the processor 203. The methods and / or processes of this application can be implemented as program instructions. The non-contact sleep apnea monitoring device 200 also includes an input / output component 209 configured to support data input / output. For example, the test subject or other data acquisition personnel can input data into the non-contact sleep apnea monitoring device 200 via the input / output component 209 using an input device (e.g., a keyboard, touchscreen), such as the test subject's age, gender, height, weight, etc. The non-contact sleep apnea monitoring device 200 can also output data to an output device (e.g., a monitor, printer, etc.) via the input / output component 209.
[0058] It should be understood that, for ease of description, only one processor is described in the non-contact sleep apnea monitoring device 200 of this application. However, it should be noted that the non-contact sleep apnea monitoring device 200 of this application may also include multiple processors. Therefore, the operation and / or method steps disclosed in this application may be executed by one processor as described in this application, or they may be executed jointly by multiple processors. For example, if the processor 203 of the non-contact sleep apnea monitoring device 200 of this application executes steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors in information processing (e.g., the first processor executes step A, the second processor executes step B, or the first and second processors jointly execute steps A and B).
[0059] Example 4: like Figure 3 As shown, Embodiment 4 of the present invention provides a non-contact sleep apnea monitoring system, comprising: One or more first fiber optic sensors 301 are used to acquire chest vibration data of the object; One or more second fiber optic sensors 302 are used to acquire abdominal vibration data of the object; One or more PPG sensors 303 are used to synchronously acquire pulse wave data of the object; and The non-contact sleep breathing monitoring device 200 provided in Embodiment 3 of the present invention.
[0060] This invention acquires chest vibration signals and abdominal vibration signals respectively using a first fiber optic sensor placed below the chest and a second fiber optic sensor placed below the abdomen of a supine subject. Simultaneously, a PPG sensor collects pulse wave signals and blood oxygen saturation values from the fingers of the supine subject. After blind source separation of the chest and abdominal vibration signals, respiratory, cardiac, and body motion noise signals are extracted. The cardiac signal is templated and enhanced to obtain a heartbeat event time sequence. A multi-feature fuzzy inference system is constructed to determine the event type. The sleep apnea-hypopnea index is calculated based on the total number of effective respiratory events and total sleep time. This application enables the identification of sleep apnea and hypopnea events in a undisturbed home environment and achieves automatic calculation of the sleep apnea-hypopnea index.
[0061] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A non-contact sleep apnea monitoring method, characterized in that, The method includes: S101. The chest vibration signal and the abdominal vibration signal are acquired by the first fiber optic sensor placed under the chest and the second fiber optic sensor placed under the abdomen of the supine subject, respectively, and the pulse wave signal and blood oxygen saturation value are collected from the fingers of the supine subject simultaneously by the PPG sensor. S102. After blind source separation of chest vibration signal and abdominal vibration signal, respiratory source signal, cardiac source signal and body motion noise source signal are extracted respectively. The cardiac source signal is templated and enhanced to obtain the heartbeat event time sequence. S103. Based on respiratory source signals, pulse wave signals, body motion noise source signals, blood oxygen saturation values, chest vibration signals, and abdominal vibration signals, calculate the respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body motion energy index, respiratory effort index, and chest-abdominal contradiction index. S104. Construct a multi-feature fuzzy reasoning system to determine the event type based on five features: respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body energy index, respiratory effort index, and chest-abdominal contradiction index. S105. Obtain the sleep period label for each time period based on the heartbeat event time sequence, respiratory source signal and body energy index, calculate the total sleep time based on the sleep period label, and count the total number of effective respiratory events that occurred within the total sleep time interval based on the event markers. Calculate the sleep apnea-hypopnea index based on the total number of effective respiratory events and the total sleep time.
2. The method as described in claim 1, characterized in that, The method further includes the following steps: A unified microsecond-level hardware timestamp is applied to chest vibration signals, abdominal vibration signals, and pulse wave signals.
3. The method as described in claim 1, characterized in that, S102 specifically includes: S1021. Preprocess and filter the chest vibration signal and abdominal vibration signal to remove power frequency interference and baseline drift, and then perform ensemble empirical mode decomposition (EEMD) to obtain a set of intrinsic mode functions (IMFs) for each. S1022. For both chest and abdominal vibration signals, calculate the average instantaneous frequency and normalized energy of each IMF component in each group of intrinsic mode functions. Select components that simultaneously satisfy the condition that the average instantaneous frequency is in the range of 0.1 Hz to 10 Hz and the normalized energy is greater than 1% to form the chest candidate component set. and abdominal candidate component set ; S1023. Constructing a variable-step-size recursive least squares VSS-RLS adaptive filter for the chest candidate component set. and abdominal candidate component set Perform filtering; S1024. The chest vibration signal is bandpass filtered to obtain a low-frequency respiratory reference signal, and the chest mixed signal is... Abdominal mixed signal Together with the low-frequency respiratory reference signal, they constitute the observation matrix X(t); X(t) uses the FastICA algorithm based on maximizing negative entropy to output several independent components with the same dimension as the observation matrix. The cross-correlation coefficient between each independent component and the low-frequency respiratory reference signal is calculated, and the component with the highest cross-correlation coefficient is identified as the respiratory source signal S_resp(t); the power of each component in the 0.8-3Hz frequency band is calculated, and the component with the highest power is identified as the heartbeat source signal S_heart(t); the remaining main components are regarded as the body motion noise source signal S_motion(t); S1025. Template-enhanced the heartbeat source signal, including: bandpass filtering the heartbeat source signal S_heart(t), and using an adaptive thresholding method to detect candidate heartbeat peaks; extracting signal segments of fixed duration centered on each peak point, and generating typical heartbeat waveform templates through cluster analysis; performing cross-correlation calculations on all signal segments and the templates, retaining highly correlated segments as qualified segments, and performing subsampling for precise positioning based on the cross-correlation peaks to obtain a high-precision heartbeat event time sequence; aligning all qualified segments according to the heartbeat event time sequence, performing coherent averaging, and finally outputting the enhanced heartbeat source signal S_heart_enhanced(t); the heartbeat event time sequence will be used to calculate heart rate variability characteristics.
4. The method as described in claim 3, characterized in that, S1023 specifically includes: chest candidate component set and abdominal candidate component set The sum of the absolute values of the amplitudes of all components at each moment is used as the reference noise signal. This is used to comprehensively characterize global high-frequency vibrations and instantaneous large-amplitude motions; with reference noise signals. As reference noise input, the chest candidate component set IMF_C and the abdominal candidate component set... The synthesized signal is used as the main input for parallel adaptive filtering, with the step size factor μ based on... Instantaneous power adaptive adjustment: ,in, The initial step size, The attenuation coefficient; Finally, the preliminarily denoised chest-shaped mixed signal is output. Mixed signals from the abdomen .
5. The method as described in claim 3, characterized in that, S103 specifically includes: S1031. Perform Hilbert transform on the respiratory source signal to obtain the respiratory signal envelope Env_resp(t) that represents the amplitude change. Calculate the preset percentile of Env_resp(t) within the sliding time window and update it dynamically as the current respiratory baseline. Calculate the moving average of the blood oxygen saturation value with the same sliding time window and update it dynamically as the blood oxygen saturation baseline. Calculate the respiratory amplitude drop ratio R_drop: R_drop = [B_resp(t) - Env_resp(t)] / B_resp(t), which is the percentage decrease of the current respiratory envelope value relative to its dynamic baseline, where B_resp(t) is the respiratory baseline and Env_resp(t) is the respiratory signal envelope; Blood oxygen saturation drop value S_drop: S_drop = B_spo2(t) - SpO2(t), which is the absolute value of the current blood oxygen value relative to its dynamic baseline, where B_spo2(t) is the blood oxygen saturation baseline and SpO2(t) is the blood oxygen saturation value; S1032. High-pass filtering is applied to the body motion noise source signal S_motion(t) and the chest vibration signal V_chest(t). The root mean square value of each signal within a short time window is calculated to obtain the body motion noise source signal energy E_motion(t) and the original high-frequency vibration energy E_hf(t), which are then fused. The kinetic energy of the original fusion body is E_raw(t) = sqrt(α*E_motion(t)). 2 +β* E_hf(t) 2 ), where α and β are weighting coefficients and satisfy α+β= 1, calculate the 99th percentile of E_raw(t) in the past time window as the normalization benchmark value, divide the current E_raw(t) by the normalization benchmark value to obtain the body energy index A(t), A(t) > 1 indicates that there is significant body motion; S1033. Perform bandpass filtering on the chest vibration signal V_chest(t) and the abdominal vibration signal V_abd(t) respectively, and calculate the Hilbert envelope to obtain the chest respiratory envelope Env_c(t) and the abdominal respiratory envelope Env_a(t); Calculate the instantaneous respiratory effort amplitude: Eff_inst(t) = sqrt(Env_c(t)) 2 + Env_a(t) 2 ); Calculate the median of Eff_inst(t) within the sliding time window as the dynamic baseline Eff_base(t); The final respiratory effort index is Eff_idx(t) = Eff_inst(t) / Eff_base(t), where Eff_idx(t) > 1.2 indicates increased respiratory effort; S1034. Perform Hilbert transform on the bandpass filtered signals of chest vibration signal V_chest(t) and abdominal vibration signal V_abd(t) to obtain instantaneous phases φ_c(t) and φ_a(t); Calculate the instantaneous phase difference Δφ(t) = φ_c(t) - φ_a(t), and normalize it to the interval [-π, π]. Within a decision event window, the percentage of sampling points with |Δφ(t)| > (2π / 3) radians is counted, and this percentage is the thoracoabdominal contradiction index P(t).
6. The method as described in claim 3, characterized in that, S104 specifically includes: S1041. Define the overall confidence level of an event. The range of the overall confidence level of an event is [0, 1]. Define a fuzzy set {very low, low, medium, high, very high} and the corresponding membership function for the overall confidence level of an event. S1042. Input the respiratory amplitude decrease ratio, blood oxygen saturation decrease value, body energy index, respiratory effort index and chest-abdominal contradiction index into the predefined membership function to obtain their membership degree as "extremely low", "low", "medium", "high" or "extremely high" fuzzy linguistic variables. S1043. Construct a fuzzy rule library containing multiple if-rule rules; S1044. For each rule in the fuzzy rule base, calculate the minimum membership degree of all its antecedent conditions, which is used as the trigger strength of the rule. Each rule activates the output fuzzy set corresponding to its consequent with its trigger strength. The output fuzzy sets of all activated rules are maximized and aggregated into a total output fuzzy set. S1045. The centroid of the total output fuzzy set is calculated using the centroid method and mapped to a scalar between 0 and 1, which is the comprehensive confidence score C of this event. An event is officially adopted by the system if and only if the duration of a potential event is greater than or equal to 10 seconds and its comprehensive confidence score C is greater than or equal to the threshold. The specific type of the adopted event is determined by the event type indicated by the rule with the highest trigger strength.
7. The method as described in claim 6, characterized in that, S105 specifically includes: S1051. The signals monitored throughout the night, including the heartbeat event time series {t_k}, respiratory source signal S_resp(t), and body energy index A(t), are divided into continuous preset time periods. The heart rate variability features, body movement features, respiratory features, and event markers of each time period are extracted, input into a hidden Markov model for automatic staging, and the sleep period label for each time period is output. Heart rate variability characteristics: Based on the heartbeat event time sequence obtained in step S1025, calculate the time domain and frequency domain characteristics within this time period; Body dynamics characteristics: Based on the body dynamics energy index A(t), calculate the average body dynamics energy during this period; Respiratory characteristics: Based on the respiratory source signal S_resp(t), the average respiratory rate and respiratory amplitude variability during this period are calculated; Event marker: Whether the time period includes a valid breathing event determined in step S104. A valid breathing event refers to an adopted event determined in step S1045, with a duration of ≥10 seconds and a comprehensive confidence score C ≥ the threshold, including central apnea, obstructive apnea, mixed apnea, and hypoventilation. S1052. In the sleep period tag sequence, search backward from the starting point and determine the start time of the first period in the first consecutive non-wake period period as the sleep time. Search backward from the ending point and determine the end time of the last sleep period period before the last consecutive wake period period as the wake time. S1053. Determine the total sleep time (TST) based on the determined sleep onset time and wake-up time; calculate the apnea-hypopnea index (AHI) based on the total number of effective breathing events occurring within the total sleep time (TST) interval, according to the event markers. AHI = Total number of effective breathing events / Total sleep time (TST).
8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the non-contact sleep breathing monitoring method as described in any one of claims 1 to 7.
9. A non-contact sleep apnea monitoring device, characterized in that, include: Communication port; A processor for executing computer instructions, the computer instructions including the steps of performing the non-contact sleep apnea monitoring method as described in any one of claims 1 to 7; Internal communication bus, used for communication between systems; The memory is configured to store data and instructions. Input / output components are configured to support data input / output.
10. A non-contact sleep apnea monitoring system, characterized in that, include: One or more first fiber optic sensors are used to acquire chest vibration data of the object; One or more second fiber optic sensors are used to acquire abdominal vibration data of the object; One or more PPG sensors are used to synchronously acquire pulse wave data of the object; and The non-contact sleep breathing monitoring device as described in claim 9.