Sleep respiration regulation parameter estimation method, device, equipment, medium and product

Abstract

The application relates to the technical field of sleep monitoring, and discloses a sleep respiration regulation parameter estimation method, device, equipment, medium and product. The method comprises the following steps: acquiring a multi-dimensional space-time spectrum diagram in a human sleep process; processing the multi-dimensional space-time spectrum diagram to obtain a target evaluation window; determining a respiration regulation parameter fitting value by using a pre-trained neural network model based on the target evaluation window; and determining a sleep respiration regulation parameter based on a plurality of respiration regulation parameter fitting values. According to the application, the digital representation of physiological signals can be realized without the traditional contact type collection equipment. Moreover, the target evaluation window can be obtained by processing the multi-dimensional space-time spectrum diagram, effective data screening can be automatically completed based on signal characteristics, manual labeling, manual staging or manual operation is not needed, and the evaluation process is simplified.

Description

Technical Field

[0001] This application relates to the field of sleep monitoring technology, and more specifically, to a method, device, equipment, medium, and product for estimating sleep breathing regulation parameters. Background Technology

[0002] Obstructive sleep apnea (OSA) is a chronic respiratory disease caused by repeated upper airway collapse during sleep. It easily induces cardiovascular complications and has a high incidence rate. The pathogenesis of obstructive sleep apnea is closely related to respiratory regulation ability.

[0003] In related technologies, the assessment of sleep breathing regulation parameters mainly relies on two types of methods: one is the measurement method based on continuous positive airway pressure intervention experiments; the other is the measurement method based on mathematical modeling of polysomnography.

[0004] However, the relevant technologies rely on manual annotation and specialized facilities, which cannot achieve the convenience of assessing respiratory regulation parameters. Summary of the Invention

[0005] In view of this, this application provides a method, device, equipment, medium and product for estimating sleep breathing regulation parameters, in order to solve the problem that related technologies rely on manual annotation and professional venues, which cannot meet the convenience requirements for assessing breathing regulation parameters.

[0006] In a first aspect, this application provides a method for estimating sleep breathing regulation parameters, the method comprising: acquiring a multidimensional spatiotemporal spectrum of a human body during sleep; processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window; determining the fitting values ​​of breathing regulation parameters based on the target evaluation window using a pre-trained neural network model; and determining sleep breathing regulation parameters based on multiple fitting values ​​of breathing regulation parameters.

[0007] The sleep breathing regulation parameter estimation method provided in this application obtains a multidimensional spatiotemporal spectrum during human sleep, enabling digital representation of physiological signals without traditional contact-based acquisition devices. Furthermore, by processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window, effective data can be automatically filtered based on signal characteristics, eliminating the need for manual annotation, manual staging, or manual monitoring, thus simplifying the evaluation process.

[0008] Furthermore, by determining the fitted values ​​of respiratory regulation parameters based on a pre-trained neural network model, the parameters can be automatically calculated and output directly without the need for professional medical facilities or personnel intervention. Finally, sleep respiratory regulation parameters are determined based on multiple fitted values, achieving fully automated processing. This solves the problem that related technologies rely on manual annotation and professional facilities, making it difficult to conveniently evaluate respiratory regulation parameters.

[0009] In one possible implementation, obtaining a multidimensional spatiotemporal spectrum during human dormancy includes: collecting physiological signals during human dormancy using lightweight sensing devices; processing the physiological signals to determine a multidimensional spatiotemporal spectrum containing body movement intensity, respiratory intensity, and respiratory Doppler.

[0010] In one possible implementation, the multidimensional spatiotemporal spectrum is processed to obtain a target evaluation window, including: dividing the multidimensional spatiotemporal spectrum into multiple sliding windows using a preset step size; performing signal discrimination on each sliding window to detect whether the sliding window contains respiratory rhythm abnormality events and micro-arousal events; discarding sliding windows that do not contain respiratory rhythm abnormality events and micro-arousal events as invalid segments, and determining the sliding windows that contain respiratory rhythm abnormality events and micro-arousal events as the target evaluation window.

[0011] In one possible implementation, signal discrimination is performed on each sliding window to detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events. This includes: determining the variance of the respiratory channel intensity characteristics and the peak value of the sudden energy characteristics of the body movement channel within the sliding window; based on the variance and peak value, detecting whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events; wherein, when the variance is greater than a preset respiratory variance threshold and the peak value is greater than a preset body movement energy peak value threshold, it is determined that the sliding window contains abnormal respiratory rhythm events and micro-arousal events.

[0012] In one possible implementation, the pre-trained neural network model includes: a spatial feature extraction module, a temporal modeling module, an attention mechanism module, and a multi-branch fully connected module. Specifically, based on the target evaluation window, the pre-trained neural network model is used to determine the fitted values ​​of respiratory regulation parameters, including: using the spatial feature extraction module to convolve the target evaluation window to obtain spatial local texture features; using the temporal modeling module to serialize the spatial local texture features along the time axis to obtain temporal respiratory dynamics features; using the attention mechanism module to calculate the time step weights of the temporal respiratory dynamics features and perform weighted summation to obtain key abnormal physiological focusing features; and using the multi-branch fully connected module to map the key abnormal physiological focusing features to determine the fitted values ​​of respiratory regulation parameters.

[0013] In one possible implementation, sleep breathing regulation parameters are determined based on fitted values ​​of multiple breathing regulation parameters, including: summarizing the fitted values ​​of breathing regulation parameters; processing them based on statistical distribution characteristics; and outputting sleep breathing regulation parameters. Among these, sleep breathing regulation parameters include pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response capacity.

[0014] Secondly, this application provides a sleep breathing regulation parameter estimation device, which includes: an acquisition module for acquiring a multidimensional spatiotemporal spectrum during human sleep; a processing module for processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window; a first determination module for determining the fitting value of the breathing regulation parameter based on the target evaluation window and using a pre-trained neural network model; and a second determination module for determining the sleep breathing regulation parameter based on multiple fitting values ​​of the breathing regulation parameter.

[0015] Thirdly, this application provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the sleep breathing regulation parameter estimation method of the first aspect or any corresponding embodiment described above.

[0016] Fourthly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the sleep breathing regulation parameter estimation method of the first aspect or any corresponding embodiment described above.

[0017] Fifthly, this application provides a computer program product, including computer instructions for causing a computer to execute the sleep breathing regulation parameter estimation method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or 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 this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the sleep breathing regulation parameter estimation method according to an embodiment of this application; Figure 2 This is a flowchart illustrating another method for estimating sleep breathing regulation parameters according to an embodiment of this application; Figure 3 This is a flowchart illustrating another method for estimating sleep breathing regulation parameters according to an embodiment of this application; Figure 4 This is a schematic diagram of a neural network model architecture according to an embodiment of this application; Figure 5 This is a structural block diagram of a sleep breathing regulation parameter estimation device according to an embodiment of this application; Figure 6This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Based on relevant technologies, obstructive sleep apnea (OSA) is a chronic sleep-disordered breathing disease characterized by repeated upper airway collapse during sleep. It is characterized by high incidence and a tendency to cause cardiovascular and other systemic complications. In traditional medicine, the diagnosis of OSA mainly relies on polysomnography (PSG), which comprehensively assesses sleep structure and respiratory events by simultaneously recording multiple physiological signals such as electroencephalography (EEG), electrooculography (EOG), electromyography (EMG), respiratory airflow, and blood oxygenation. However, this method suffers from problems such as complex equipment, high testing costs, and discomfort for patients wearing sensors, making it unsuitable for long-term monitoring and home application. Recent studies have found that the pathogenesis of OSA is not only related to the anatomy of the upper airway but also closely related to various respiratory regulation factors. The field of sleep medicine has proposed a pathogenesis model based on respiratory regulation parameters, namely the PALM model, which includes: critical closing pressure (P), arousal threshold (A), loop gain (L), and upper airway muscle responsiveness (M). These factors reflect the feedback regulation capacity of the respiratory control system and the neuromuscular function of the upper airway, which are of great significance for individualized treatment of OSA.

[0022] Currently, the assessment of these indicators typically relies on complex experimental equipment or intricate modeling and analysis of polysomnography data, making the process cumbersome and difficult to implement. To address the issues of existing technologies that rely on complex contact-based equipment and involve tedious procedures for detecting OSA respiratory regulation parameters, this invention proposes a method for estimating sleep respiratory regulation parameters. This method can automatically estimate OSA-related respiratory regulation indicators directly from nighttime sleep signals acquired by low-load sensors (such as non-contact radar, or radar combined with a portable pulse oximeter), providing a novel solution for the assisted diagnosis and home assessment of OSA.

[0023] Related technical solutions: The main methods for assessing respiratory regulation parameters include: measurement methods based on continuous positive airway pressure intervention experiments; and measurement methods based on mathematical modeling of polysomnography.

[0024] Problems and defects in related technologies: 1) Measurement methods based on continuous positive airway pressure intervention experiments: Highly invasive: It requires wearing a mask during sleep and the airway pressure is repeatedly and instantaneously changed by external devices, which can easily induce micro-arousals in patients and severely disrupt the natural sleep structure.

[0025] Limited by the setting: It must be done in a professional medical sleep center and requires experienced sleep technicians to be on duty and operate it manually all night. It is completely impossible to achieve daily monitoring in a home environment.

[0026] Low compliance: Frequent changes in ventilation pressure can cause patients significant respiratory discomfort and psychological stress, making it difficult for most patients to endure and successfully complete the overnight test.

[0027] 2) Measurement method based on mathematical modeling of polysomnography: Highly contact-intensive: The test subject needs to be covered with more than ten kinds of sensors such as EEG, ECG, and breathing belt, which brings a serious "first night effect" and causes the measured data to often deviate from the patient's true daily physiological state.

[0028] Poor stability: Traditional calculation models rely heavily on extremely fragile airflow signals from the mouth and nose. During sleep, the nasal cannula is very prone to displacement or dislodgement due to turning over. Once the core signal is distorted, the calculation of indicators will fail throughout the night.

[0029] The process is cumbersome: before calculation, professional doctors usually need to manually perform sleep staging and respiratory event annotation on the massive PSG data. The automation level is low, the processing efficiency is extremely poor, and it does not meet the conditions for large-scale screening.

[0030] Based on this, the sleep breathing regulation parameter estimation method provided in this application can achieve digital representation of physiological signals without traditional contact acquisition equipment by acquiring a multidimensional spatiotemporal spectrum of the human body during sleep. Furthermore, by processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window, effective data can be automatically filtered based on signal characteristics, eliminating the need for manual annotation, manual staging, or manual monitoring, thus simplifying the evaluation process.

[0031] Furthermore, by determining the fitted values ​​of respiratory regulation parameters based on a pre-trained neural network model, the parameters can be automatically calculated and output directly without the need for professional medical facilities or personnel intervention. Finally, sleep respiratory regulation parameters are determined based on multiple fitted values, achieving fully automated processing. This solves the problem that related technologies rely on manual annotation and professional facilities, making it difficult to conveniently evaluate respiratory regulation parameters.

[0032] According to an embodiment of this application, a method for estimating sleep breathing regulation parameters is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] This embodiment provides a method for estimating sleep breathing regulation parameters, which can be used in computer devices such as computers and servers. Figure 1 This is a flowchart of a sleep breathing regulation parameter estimation method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain a multidimensional spatiotemporal spectrum during human dormancy.

[0034] The human body's dormancy process can be understood as the complete cycle of natural sleep. A multidimensional spatiotemporal spectrum can be understood as a multi-channel spectrum containing spatial and temporal features, generated after preprocessing by collecting physiological micro-movements and respiratory signals during human sleep using lightweight sensing devices. Specifically, it can include dimensions such as body movement intensity, respiratory intensity, and respiratory Doppler features.

[0035] Specifically, lightweight sensing devices are used to continuously collect physiological micro-movement and respiratory-related signals during the natural dormancy process of the human body. The raw collected signals are preprocessed to finally generate a multi-dimensional spatiotemporal spectrum containing spatial and temporal features.

[0036] Lightweight sensing devices can be non-contact radar, or a combination of non-contact radar and a portable pulse oximeter, or other low-load sensor devices capable of sleep monitoring, such as wristwatches and portable ECG patches.

[0037] In a feasible scenario, a non-contact millimeter-wave radar is placed at a suitable height at the head of the patient's bed (e.g., 50-80cm from the head of the bed, 1.2-1.5m high). The radar continuously monitors the human body during sleep throughout the night, collecting raw radar echo signals generated by chest and abdominal micro-movements and respiratory movements. These signals are then transmitted wirelessly to the cloud or a local terminal for storage via Wi-Fi or Bluetooth. The collected radar echo signals are then processed for denoising, filtering, and signal enhancement to improve the signal-to-noise ratio. Time-frequency analysis is performed on the preprocessed signals to extract three core dimensions: body movement intensity, respiratory intensity, and respiratory Doppler characteristics, generating a three-channel multidimensional spatiotemporal spectrum.

[0038] Step S102: Process the multidimensional spatiotemporal spectrum to obtain the target evaluation window.

[0039] The target assessment window can be understood as a sliding window segment selected from a multidimensional spatiotemporal spectrum that contains typical respiratory rhythm abnormalities and micro-arousal events.

[0040] Specifically, a combination of sliding window segmentation and validity discrimination is used to process the multidimensional spatiotemporal spectrum, automatically removing invalid data such as stable breathing segments and awake segments, and screening out target evaluation windows containing typical respiratory rhythm abnormalities and micro-arousal events.

[0041] As an example, a fixed-duration, continuous sliding window segmentation is performed on the multidimensional spatiotemporal spectrum. The window duration is set to 5-10 seconds, and the step size is set to 2-3 seconds, dividing the entire night's multidimensional spatiotemporal spectrum into multiple sliding windows. A dual-discrimination rule is built-in: first, respiratory abnormality detection (calculating the variance of the respiratory intensity channel within the window; if the variance > a preset threshold, it indicates unstable respiratory rhythm and an abnormality); second, micro-arousal event detection (calculating the peak value of the body movement intensity channel within the window; if the peak value > a preset threshold, it indicates the presence of a micro-arousal event). For each sliding window, the above dual-discrimination rule is executed simultaneously. If a window simultaneously satisfies both respiratory rhythm abnormality and the presence of a micro-arousal event, it is determined to be a valid window and retained as a target evaluation window; if only one of these conditions is met or neither is met, it is determined to be an invalid window and discarded. The selected target evaluation windows undergo continuity verification, and isolated valid segments with a duration < 3 consecutive windows are removed, ultimately resulting in a complete set of target evaluation windows.

[0042] Step S103: Based on the target evaluation window, determine the fitted values ​​of the respiratory regulation parameters using a pre-trained neural network model.

[0043] The fitted value of respiratory regulation parameters can refer to the predicted value (continuous value) of the four respiratory regulation parameters of the corresponding model after the neural network model infers on a single target evaluation window, namely the preliminary prediction results of pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response capacity.

[0044] The selected target evaluation window is input into a pre-trained deep neural network model. Through spatial feature extraction, temporal modeling, attention focusing, and multi-target fitting, the respiratory regulation features within the window are automatically extracted, and the fitting values ​​of four respiratory regulation parameters corresponding to a single target evaluation window are output in parallel.

[0045] Step S104: Determine sleep breathing regulation parameters based on the fitted values ​​of multiple breathing regulation parameters.

[0046] Sleep breathing regulation parameters can be understood as the final quantitative assessment result representing an individual patient's sleep breathing regulation ability, obtained by summarizing the fitted values ​​of multiple breathing regulation parameters and performing statistical optimization.

[0047] The fitted values ​​of respiratory regulation parameters corresponding to all target assessment windows output in step S103 are summarized. Through statistical distribution characteristic analysis, the random fluctuations of single window fitting are eliminated to obtain the sleep respiratory regulation parameters, which represent the final quantitative assessment results of the patient's individual sleep respiratory regulation ability.

[0048] The sleep breathing regulation parameter estimation method provided in this application obtains a multidimensional spatiotemporal spectrum during human sleep, enabling digital representation of physiological signals without traditional contact-based acquisition devices. Furthermore, by processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window, effective data can be automatically filtered based on signal characteristics, eliminating the need for manual annotation, manual staging, or manual monitoring, thus simplifying the evaluation process.

[0049] Furthermore, by determining the fitted values ​​of respiratory regulation parameters based on a pre-trained neural network model, the parameters can be automatically calculated and output directly without the need for professional medical facilities or personnel intervention. Finally, sleep respiratory regulation parameters are determined based on multiple fitted values, achieving fully automated processing. This solves the problem that related technologies rely on manual annotation and professional facilities, making it difficult to conveniently evaluate respiratory regulation parameters.

[0050] This embodiment provides a method for estimating sleep breathing regulation parameters, which can be used in computer devices such as computers and servers. Figure 2 This is a flowchart of a sleep breathing regulation parameter estimation method according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain a multidimensional spatiotemporal spectrum during human dormancy.

[0051] Specifically, step S201 includes: Step S2021: Collect physiological signals during the human body's dormancy process using lightweight sensing devices.

[0052] Lightweight sensing devices can be understood as those that do not require full-body contact, are small in size, lightweight, and do not interfere with the body's natural sleep. Examples of lightweight sensing devices include non-contact millimeter-wave radar, portable pulse oximeters, and miniature respiratory sensors.

[0053] Physiological signals refer to signals that can be captured by sensing devices during human dormancy, generated by respiratory movements, chest and abdominal micro-movements, and cardiovascular activity. Among these, physiological signals can be those related to respiratory regulation, specifically including chest and abdominal micro-movement signals, respiratory airflow signals, and blood oxygen saturation signals.

[0054] Specifically, a suitable lightweight sensing device is selected to continuously and stably collect physiological signals related to respiratory regulation (such as chest and abdominal micro-movement signals and respiratory airflow signals) during the body's natural sleep process (without disturbing sleep or requiring full-body contact).

[0055] As an example, combined Figure 2 As shown, a 24GHz or 60GHz non-contact millimeter-wave radar (lightweight design, weighing <100g, palm-sized) was selected and placed at a certain height above the patient's bedside, ensuring that the radar beam covered the main areas of the patient's chest and abdomen. Sleep monitoring was performed throughout the night, and the radar echo data was transmitted wirelessly (e.g., via Wi-Fi or Bluetooth) to the cloud for storage and real-time processing. After the acquisition process, the acquired echo signals were preprocessed to obtain three-channel radar spectral data representing the subject's body movement, respiratory energy intensity, and respiratory Doppler characteristics during nighttime sleep.

[0056] Step S2022: Process the physiological signals to determine a multidimensional spatiotemporal spectrum containing body movement intensity, respiratory intensity, and respiratory Doppler.

[0057] The raw physiological signals collected in step S2021 are subjected to a series of preprocessing (such as noise reduction, filtering, and signal enhancement) to remove interference signals and repair missing signals. Then, three core features, namely body movement intensity, respiratory intensity, and respiratory Doppler, are extracted from the preprocessed signals to finally generate a multidimensional spatiotemporal spectrum containing the above three core features.

[0058] As an example, a wavelet denoising algorithm is used to denoise the original radar echo signal, eliminating environmental noise (such as indoor air conditioning noise and outside noise) and equipment noise, while retaining effective signals related to chest and abdominal micro-movements and respiratory airflow, improving the signal-to-noise ratio to over 35dB. A low-pass filtering algorithm is used to retain the 0.1-10Hz respiratory-related signal band (which contains core information on respiratory intensity and body movement intensity), eliminating high-frequency interference signals (such as electromagnetic interference) and low-frequency redundant signals (such as large-amplitude body movement interference caused by the patient turning over). For signal loss during the acquisition process (such as signal interruption caused by brief radar obstruction), a linear interpolation algorithm is used to repair it, ensuring signal continuity and avoiding gaps in feature extraction. The gain of the preprocessed signal is adjusted to amplify the amplitude of respiratory airflow and chest and abdominal micro-movement related signals.

[0059] By calculating the peak amplitude of the preprocessed signal, the intensity of chest and abdominal micromovements is quantified. The larger the peak amplitude, the higher the intensity of body movement, generating a time series sequence of body movement intensity. By extracting the periodic features of the signal, the mean amplitude of respiratory airflow is calculated to quantify respiratory intensity. The larger the mean amplitude, the higher the intensity of respiratory movement, generating a time series sequence of respiratory intensity. The focus is on extracting changes in respiratory Doppler features. By calculating the frequency offset of the signal, the velocity changes of respiratory airflow are quantified to generate a respiratory Doppler time series sequence, reflecting the dynamic evolution of respiratory rhythm.

[0060] Step S202 involves processing the multidimensional spatiotemporal spectrum to obtain the target evaluation window. Please refer to [link / reference needed] for details. Figure 1 Step S102 in the process will not be elaborated on here.

[0061] Step S203: Based on the target evaluation window, determine the fitted values ​​of the respiratory regulation parameters using a pre-trained neural network model. Please refer to [link / reference needed] for details. Figure 1 Step S103 in the process will not be elaborated on here.

[0062] Step S204: Determine sleep breathing regulation parameters based on the fitted values ​​of multiple breathing regulation parameters.

[0063] Specifically, step S204 includes: Step S2041: Summarize the fitted values ​​of respiratory regulation parameters.

[0064] The fitted values ​​are categorized and summarized according to the type of sleep breathing regulation parameters (pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response); the target evaluation window number corresponding to each fitted value is labeled; abnormal fitted values ​​that exceed the physiologically reasonable range are removed; missing fitted values ​​are supplemented using appropriate methods; and finally, a standardized fitted value summary table is formed.

[0065] Step S2042: Process based on statistical distribution characteristics and output sleep breathing regulation parameters; among which, sleep breathing regulation parameters include pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response capacity.

[0066] Analyze the statistical distribution characteristics of the fitted values ​​for each parameter to determine whether they conform to a normal distribution; calculate the mean, variance, and standard deviation of the fitted values ​​for each parameter, and remove outliers with excessive dispersion (secondary removal); calibrate the fitted values ​​based on the statistical results to correct systematic errors; finally, output the final value of each parameter and label the confidence interval of the parameter.

[0067] The sleep breathing regulation parameter estimation method provided in this application process the collected physiological signals and determines a multidimensional spatiotemporal spectrum including body movement intensity, breathing intensity, and respiratory Doppler. It can effectively remove environmental interference, equipment noise and abnormal signals, repair missing signals, and transform the abstract original signal into an intuitive and analyzable multi-channel feature map. It can accurately extract the core features related to breathing regulation and avoid interference from irrelevant features.

[0068] This embodiment provides a method for estimating sleep breathing regulation parameters, which can be used in computer devices such as computers and servers. Figure 3 This is a flowchart of a sleep breathing regulation parameter estimation method according to an embodiment of this application, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain a multidimensional spatiotemporal spectrum of the human body during hibernation. Please refer to [reference needed] for details. Figure 1 Step S201 in the process will not be elaborated on here.

[0069] Step S302: Process the multidimensional spatiotemporal spectrum to obtain the target evaluation window.

[0070] Specifically, step S302 includes: Step 3021: The multidimensional spatiotemporal spectrum is divided into multiple sliding windows using a preset step size.

[0071] A sliding window can be understood as an independent window obtained after sliding window segmentation. Each window contains multidimensional spatiotemporal spectrogram features of a fixed duration (such as 30s, 60s), which is the basic unit for event detection. The window size needs to match the duration of abnormal respiratory rhythm events and micro-arousal events to ensure that the target event can be captured completely.

[0072] Specifically, determine the size of the sliding window (based on the respiratory rhythm cycle, specifically 30s-60s, to adapt to the duration of abnormal respiratory events); set a preset step size (preferably 1 / 2 of the window size); continuously slide the window on the multidimensional spatiotemporal spectrum with the preset step size to complete the sliding window segmentation, obtaining multiple non-repeating sliding windows that cover the entire spectrum; number each sliding window and label the corresponding time interval.

[0073] Step 3022: Perform signal discrimination for each sliding window to detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events.

[0074] Abnormal respiratory rhythm events can refer to the phenomenon that the respiratory rate, respiratory depth, and respiratory airflow are significantly abnormal during the human body's rest. The core includes apnea, shallow and slow breathing, and respiratory rhythm disorder, which are manifested in the multidimensional spatiotemporal spectrum as a sudden drop in respiratory intensity and respiratory Doppler abnormalities.

[0075] Micro-arousal events refer to physiological events during the human body's sleep process in which the person is not fully awake but shows signs of brief awakening. They are usually accompanied by a sudden increase in body movement intensity and fluctuations in respiratory rhythm. In the multidimensional spatiotemporal spectrum, they are manifested as an abnormally high peak in body movement intensity. They are key events reflecting the sleep-wake threshold and the responsiveness of the upper airway dilator muscles.

[0076] Extract core features such as peak body movement intensity, mean respiratory intensity and fluctuation amplitude, and respiratory Doppler trend within each sliding window; determine whether there are abnormal respiratory rhythm events (such as a sudden drop in respiratory intensity ≥50% or abnormal respiratory Doppler) based on a clear event discrimination threshold; determine whether there are micro-arousal events (such as peak body movement intensity ≥ a preset threshold) based on a body movement intensity threshold; record the detection results of each window and label the event type and event duration.

[0077] Specifically, step S3022 includes: Step a1: Determine the variance of the respiratory channel intensity characteristics and the peak value of the burst energy characteristics of the body movement channel within the sliding window.

[0078] The sudden energy characteristics of the body movement channel can be understood as the core characteristics of the body movement intensity channel in the multidimensional spatiotemporal spectrum. It refers to the signal energy generated by sudden movements of the chest, abdomen, limbs and other parts during human dormancy, which directly reflects the occurrence of micro-awakening events.

[0079] The peak value can be understood as the quantified extreme value of signal energy, specifically the maximum value of the sudden energy of the body movement channel within the sliding window. The larger the peak value, the higher the intensity of the sudden body movement and the greater the possibility of micro-arousal events.

[0080] In this embodiment of the application, for each sliding window obtained by segmentation in step 3021, two types of key quantitative indicators are extracted from the respiratory intensity channel and the body movement intensity channel of the multidimensional spatiotemporal spectrum, namely the variance of the respiratory channel intensity characteristics and the peak value of the burst energy characteristics of the body movement channel, so as to transform the abstract spectrum features into quantifiable and discriminable numerical indicators.

[0081] As an example, a neural network model can be used to determine the variance of the respiratory channel intensity characteristics and the peak value of the burst energy characteristics of the body movement channel. Specifically, 450 feature values ​​of the respiratory intensity channel and 450 feature values ​​of the body movement intensity channel are extracted from each sliding window (30s, 15Hz, 450 frames) as inputs to the neural network model, and the variance of the respiratory channel intensity characteristics and the peak value of the burst energy characteristics of the body movement channel can be used as the outputs of the neural network model.

[0082] Step a2: Based on variance and peak value, detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events; wherein, when the variance is greater than a preset respiratory variance threshold and the peak value is greater than a preset body energy peak value threshold, it is determined that the sliding window contains abnormal respiratory rhythm events and micro-arousal events.

[0083] Specifically, retrieve the preset respiratory variance threshold (e.g., 0.001) and preset peak body energy threshold (e.g., 0.15dB) based on statistical calibration of clinical sample data; compare the variance and peak value of each sliding window extracted in step a1 with the preset thresholds respectively. Only when the variance is greater than the preset respiratory variance threshold and the peak value is greater than the preset peak body energy threshold can it be determined that the sliding window contains both respiratory rhythm abnormality events and micro-arousal events; if only one condition is met, or both conditions are not met, it is determined that it does not contain both types of events (or only contains a single event).

[0084] Step 3023: Sliding windows that do not contain abnormal respiratory rhythm events and micro-arousal events are discarded as invalid segments, and sliding windows that contain abnormal respiratory rhythm events and micro-arousal events are determined as target evaluation windows.

[0085] If a sliding window containing abnormal respiratory rhythm events and micro-arousal events is determined, then sliding windows that do not contain abnormal respiratory rhythm events and micro-arousal events can be discarded as invalid segments, and sliding windows that contain abnormal respiratory rhythm events and micro-arousal events can be determined as target evaluation windows.

[0086] Based on the signal discrimination results of step 3022, all sliding windows are filtered out, and invalid segments that do not contain abnormal respiratory rhythm events and micro-arousal events (no target events, only normal sleep characteristics) are removed. Sliding windows that contain both abnormal respiratory rhythm events and micro-arousal events are determined as target evaluation windows.

[0087] Specifically, the judgment results of all sliding windows are summarized; windows that contain both abnormal respiratory rhythm events and micro-arousal events are selected as target assessment windows; windows that "contain only a single event" and / or "contain neither" are removed as invalid segments; and the target assessment windows are numbered and labeled with event information.

[0088] Step S303: Based on the target evaluation window, determine the fitted values ​​of the respiratory regulation parameters using a pre-trained neural network model. Please refer to [link / reference needed] for details. Figure 2 Step S203 in the process will not be elaborated on here.

[0089] Step S304: Determine the sleep breathing regulation parameters based on the fitted values ​​of multiple breathing regulation parameters. Please refer to [link / reference] for details. Figure 2 Step S204 in the process will not be elaborated on here.

[0090] The sleep breathing regulation parameter estimation method provided in this application process the collected physiological signals and determines a multidimensional spatiotemporal spectrum including body movement intensity, breathing intensity, and respiratory Doppler. This method can effectively eliminate environmental interference, equipment noise, and abnormal signals, repair missing signals, and transform the abstract original signals into intuitive and analyzable multi-channel feature maps. It can accurately extract the core features related to breathing regulation and avoid interference from irrelevant features.

[0091] In one possible implementation, the pre-trained neural network model includes: a spatial feature extraction module, a temporal modeling module, an attention mechanism module, and a multi-branch fully connected module; wherein, step S303 above includes the following steps: Step b1: Use the spatial feature extraction module to perform convolution processing on the target evaluation window to obtain spatial local texture features.

[0092] The spatial feature extraction module is used to extract the spatial features of the target evaluation window. The core uses convolution operation, which can capture the spatial distribution, local correlation and texture details of body motion intensity, breathing intensity, respiratory Doppler within the window, filter redundant information in the spatial dimension, and output spatial local texture features, adapting to the feature extraction requirements of multi-channel multi-dimensional spatiotemporal spectra.

[0093] Spatial local texture features refer to the local texture, distribution patterns, and related features of body movement and breathing features in the spatial dimension within the target evaluation window. They are represented in the form of a feature matrix, which includes the intensity differences and spatial distribution patterns of features at different locations within the window.

[0094] Obtain the target evaluation window after screening in step 3023, and transform the feature data of each window into a standardized feature matrix (dimension is window frame number × feature channel number × spatial size); configure the convolution parameters of the spatial feature extraction module (convolution kernel size, stride, padding, number of output channels), and initialize based on the parameters; perform multiple rounds of convolution processing on the feature matrix, and perform nonlinear transformation through activation function after each round of convolution to enhance feature representation ability; perform pooling processing (such as max pooling) on ​​the convolutional features to further reduce dimensionality and suppress overfitting; integrate the features after multiple rounds of convolution and pooling processing to obtain spatial local texture features.

[0095] Step b2: Use the temporal modeling module to serialize the spatial local texture features along the time axis to obtain temporal respiratory dynamics features.

[0096] Obtain the spatial local texture features output in step b1, arrange them in chronological order to form an initial temporal feature sequence; configure the core parameters of the temporal modeling module (hidden layer dimension, number of iterations, dropout ratio), and initialize based on the parameters; input the initial temporal feature sequence into the temporal modeling module, capture the temporal correlation of features through the temporal modeling algorithm, and explore the dynamic change patterns of respiratory and body movement features; normalize the modeled temporal features to eliminate feature bias in the temporal dimension; output standardized temporal respiratory dynamic features.

[0097] Step b3: Calculate the time step weights of the temporal respiratory dynamics features using the attention mechanism module and perform a weighted summation to obtain key abnormal physiological focusing features.

[0098] Obtain the temporal feature sequence output in step b2; configure the core parameters of the attention mechanism module (number of attention heads, hidden layer dimension), and initialize based on the parameters; calculate the correlation between the feature at each time step and the entire temporal sequence using the attention mechanism algorithm, and obtain the weight value of each time step; multiply the temporal respiratory dynamics feature of each time step with the corresponding weight value, and then perform a weighted summation of all results to obtain the aggregated feature vector; standardize the aggregated features to eliminate feature bias after weight allocation; output key abnormal physiological focusing features.

[0099] Step b4: Use a multi-branch fully connected module to map key abnormal physiological focusing features and determine the fitted values ​​of respiratory regulation parameters.

[0100] Obtain the key abnormal physiological focusing features output from step b3; configure the core parameters of the multi-branch fully connected module (number of fully connected layers, number of neurons, and activation function for each branch), and initialize based on the parameters; simultaneously input the key abnormal physiological focusing features into multiple parallel fully connected sub-networks (each sub-network corresponds to respiratory regulation parameters); through linear transformation of the fully connected layers and nonlinear activation functions, map the high-dimensional aggregated features into low-dimensional parameter fitting values; verify the fitting values ​​output by each sub-network and remove abnormal fitting values; output standardized respiratory regulation parameter fitting values.

[0101] Figure 4 This is a schematic diagram of a neural network architecture provided according to an embodiment of this application. (In conjunction with...) Figure 4As shown, by fusing radar multi-channel spectrograms through multi-layered cascaded two-dimensional convolutional layers (e.g., the first layer has 3 input channels and 16 output channels; the second layer has 32 output channels), spatial local texture features containing respiratory limitation and body movement compensation are extracted. These spatial features are then serialized along the time axis and input into a long short-term memory network (e.g., the hidden layer dimension is set to 64) to capture the dynamic evolution and rhythmic patterns of respiratory dynamics within a sliding window. A temporal attention mechanism is used to calculate time-step weights and perform weighted summation, allowing the network to adaptively focus on key abnormal physiological moment feature vectors such as airway collapse and struggle. The high-dimensional features are then input into a multi-branch fully connected network. The network's terminals contain multiple independent linear mapping branches, which output in parallel continuous fitted values ​​representing the respiratory regulation capacity within that window, including pharyngeal critical pressure, arousal threshold, loop gain, and upper airway dilator muscle response capacity.

[0102] It should be noted that, Figure 4 P, A, L, and M in the equation can correspond to the pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response, respectively.

[0103] The sleep breathing regulation parameter estimation method provided in this application uses a spatial feature extraction module to accurately mine local spatial texture features within the target evaluation window through convolution operations, effectively filtering redundant noise and achieving dimensionality reduction of high-dimensional features, laying a solid foundation for subsequent processing; by using a temporal modeling module to serialize spatial features along the time axis, it can deeply capture the dynamic evolution of abnormal breathing rhythms and micro-arousal events, making up for the lack of temporal correlation in discrete features and making feature expression more complete.

[0104] By introducing an attention mechanism module, which automatically calculates time step weights and performs weighted summation, it can accurately focus on the key physiological features that contribute most to parameter fitting and suppress interference from irrelevant information, significantly improving the targeting and efficiency of the fitting process. A multi-branch fully connected module is used for nonlinear mapping, with each branch independently adapting to different respiratory regulation parameters. It can simultaneously output multiple key parameters such as pharyngeal critical pressure and sleep-wake threshold with high precision, ensuring the fitting accuracy of each parameter and improving the parallel processing efficiency of the overall system.

[0105] This embodiment also provides a sleep breathing regulation parameter estimation device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0106] This embodiment provides a sleep breathing regulation parameter estimation device, such as... Figure 5As shown, it includes: an acquisition module 501 for acquiring a multidimensional spatiotemporal spectrum during human sleep; a processing module 502 for processing the multidimensional spatiotemporal spectrum to obtain a target evaluation window; a first determination module 503 for determining the fitting value of respiratory regulation parameters based on the target evaluation window and using a pre-trained neural network model; and a second determination module 504 for determining sleep respiratory regulation parameters based on the fitting values ​​of multiple respiratory regulation parameters.

[0107] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0108] In this embodiment, the sleep breathing regulation parameter estimation device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0109] This application also provides a computer device having the above-described features. Figure 5 The device shown is for estimating sleep breathing regulation parameters.

[0110] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of this application, such as... Figure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 6 Take a processor 10 as an example.

[0111] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0112] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0113] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0114] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0115] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0116] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0117] The computer device also includes a communication interface for communicating with other devices or communication networks.

[0118] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0119] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0120] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for estimating sleep breathing regulation parameters, characterized in that, The method includes: Obtain multidimensional spatiotemporal spectra during human dormancy; The multidimensional spatiotemporal spectrum is processed to obtain the target evaluation window; Based on the target evaluation window, the fitted values ​​of the respiratory regulation parameters are determined using a pre-trained neural network model; Based on the fitted values ​​of multiple breathing regulation parameters, the sleep breathing regulation parameters are determined. The multidimensional spatiotemporal spectrum is processed to obtain a target evaluation window, including: The multidimensional spatiotemporal spectrum is divided into multiple sliding windows by using a preset step size; For each sliding window, signal discrimination is performed to detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events; Sliding windows that do not contain abnormal respiratory rhythm events and micro-arousal events are discarded as invalid segments, and sliding windows that contain abnormal respiratory rhythm events and micro-arousal events are identified as target evaluation windows. The pre-trained neural network model includes: a spatial feature extraction module, a temporal modeling module, an attention mechanism module, and a multi-branch fully connected module; wherein, based on the target evaluation window, the pre-trained neural network model is used to determine the fitted values ​​of the respiratory regulation parameters, including: The spatial feature extraction module is used to perform convolution processing on the target evaluation window to obtain spatial local texture features; The spatial local texture features are serialized along the time axis using the temporal modeling module to obtain temporal breathing dynamics features; The attention mechanism module is used to calculate the time step weights of the temporal respiratory dynamics features and perform weighted summation to obtain key abnormal physiological focusing features; The key abnormal physiological focusing features are mapped using a multi-branch fully connected module to determine the fitted values ​​of respiratory regulation parameters; Based on the fitted values ​​of multiple breathing regulation parameters, the sleep breathing regulation parameters are determined, including: Summarize the fitted values ​​of the respiratory regulation parameters; Based on statistical distribution characteristics, the sleep breathing regulation parameters are output; wherein, the sleep breathing regulation parameters include pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response capacity.

2. The method for estimating sleep breathing regulation parameters according to claim 1, characterized in that, The acquisition of the multidimensional spatiotemporal spectrum during human dormancy includes: Physiological signals during human sleep are collected using lightweight sensing devices; The physiological signals are processed to determine a multidimensional spatiotemporal spectrum including body movement intensity, respiratory intensity, and respiratory Doppler.

3. The method for estimating sleep breathing regulation parameters according to claim 1, characterized in that, For each of the sliding windows, signal discrimination is performed to detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events, including: Determine the variance of the respiratory channel intensity characteristics and the peak value of the burst energy characteristics of the body movement channel within the sliding window; Based on the variance and the peak value, it is detected whether the sliding window contains respiratory rhythm abnormality events and micro-arousal events; wherein, when the variance is greater than a preset respiratory variance threshold and the peak value is greater than a preset body kinetic energy peak value threshold, it is determined that the sliding window contains respiratory rhythm abnormality events and micro-arousal events.

4. A sleep breathing regulation parameter estimation device, characterized in that, The device includes: The acquisition module is used to acquire multidimensional spatiotemporal spectra during human dormancy. The processing module is used to process the multidimensional spatiotemporal spectrum to obtain the target evaluation window; The first determining module is used to determine the fitted values ​​of respiratory regulation parameters based on the target evaluation window and using a pre-trained neural network model. The pre-trained neural network model includes: a spatial feature extraction module, a temporal modeling module, an attention mechanism module, and a multi-branch fully connected module; the first determination module is further used to perform convolution processing on the target evaluation window using the spatial feature extraction module to obtain spatial local texture features; The spatial local texture features are serialized along the time axis using the temporal modeling module to obtain temporal breathing dynamics features; The attention mechanism module is used to calculate the time step weights of the temporal respiratory dynamics features and perform weighted summation to obtain key abnormal physiological focusing features; The key abnormal physiological focusing features are mapped using a multi-branch fully connected module to determine the fitted values ​​of respiratory regulation parameters; The second determining module is used to determine the sleep breathing regulation parameters based on the fitted values ​​of multiple breathing regulation parameters; The processing module is further configured to perform sliding window segmentation on the multidimensional spatiotemporal spectrum using a preset step size to obtain multiple sliding windows; For each sliding window, signal discrimination is performed to detect whether the sliding window contains abnormal respiratory rhythm events and micro-arousal events; Sliding windows that do not contain abnormal respiratory rhythm events and micro-arousal events are discarded as invalid segments, and sliding windows that contain abnormal respiratory rhythm events and micro-arousal events are identified as target evaluation windows. The second determining module is further used to summarize the fitted values ​​of the respiratory regulation parameters; Based on statistical distribution characteristics, the sleep breathing regulation parameters are output; wherein, the sleep breathing regulation parameters include pharyngeal critical pressure, sleep-wake threshold, loop gain, and upper airway dilator muscle response capacity.

5. A computer device, characterized in that, include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to cause the processor to perform the sleep breathing regulation parameter estimation method as described in any one of claims 1-3.

6. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the sleep breathing regulation parameter estimation method according to any one of claims 1-3.

7. A computer program product comprising a computer program that, when executed by a processor, implements the sleep breathing regulation parameter estimation method according to any one of claims 1-3.

Images