Blood oxygen and nitric oxide asthma recognition snore stopping system and method based on breath sound

Abstract

The invention relates to the technical field of respiratory function monitoring, in particular to a snore-ceasing system and method for identifying blood oxygen and nitric oxide asthma based on respiratory sound, which enables abnormal judgment not to stay in independent identification in a single time window by introducing abnormal time sequence correlation constraint formed by a time sequence convolutional network. The abnormal types of adjacent time positions are used as a continuous sequence to carry out overall constraint and consistency verification, and by introducing node propagation weights formed by a graph neural network, risk assessment is changed from simple time sequence judgment to propagation and consistency analysis based on a connection relationship between nodes; abnormal grade changes and time position information of adjacent turning nodes are gathered through multiple rounds of adjacent node information transmission, a stable link identifier is formed in a continuous risk segment under propagation weight consistency verification, and the matching degree of an intervention opportunity and a night ventilation risk evolution rhythm is improved. And quantitative description of risk accumulation and turning is realized in a sleep scene.

Description

Technical Field

[0001] This invention relates to the field of respiratory function monitoring technology, and in particular to an anti-snoring system and method for identifying asthma based on breath sounds, blood oxygen saturation, and nitric oxide. Background Technology

[0002] The field of respiratory function monitoring technology focuses on measurable physiological signals generated during human respiration. By continuously and periodically monitoring parameters related to respiratory activity, it aims to assess respiratory function status, identify abnormalities, and determine risks. Respiratory function monitoring typically focuses on airflow patency, gas exchange capacity, and respiratory rhythm stability, and covers application scenarios such as sleep-disordered breathing, chronic airway diseases, and respiratory attenuation. The field emphasizes the direct correlation between the monitored object and the respiratory physiological process, requiring monitoring results to reflect the degree of airway obstruction, changes in respiratory efficiency, and disease-inducing characteristics, providing quantitative evidence for subsequent intervention and treatment.

[0003] The breath sound-based blood oxygen and nitric oxide asthma identification anti-snoring system is a breathing state identification and intervention system for users in sleep and resting states. The system uses sound signals generated during breathing, combined with blood oxygen saturation and nitric oxide-related parameters, as the basis for discrimination to distinguish between normal breathing, snoring, and asthma-induced abnormal breathing. The system aims to identify breathing abnormalities that may be accompanied by airway constriction and airway inflammation without relying on complex operations. When the identification results meet preset conditions, it triggers the anti-snoring and intervention process, thereby reducing the amplitude of breathing resistance fluctuations, improving the continuity of nighttime ventilation, reducing the probability of repeated snoring and breathing interruptions, and improving breathing stability during sleep.

[0004] Existing technologies use breath sounds combined with blood oxygen saturation and nitric oxide-related parameters as the basis for distinguishing between normal breathing, snoring, and asthma-induced abnormal breathing. Their operation is biased towards immediate state identification and triggering based on preset conditions in a resting environment. They lack a mechanism for serializing and solidifying the direction of changes in the duration of the respiratory cycle and the stable driving level. This makes it difficult to establish a unified reference for the cyclical structural differences that occur within the same night due to changes in posture, short-term breath-holding, and fluctuations in snoring intensity. The identification criteria rely more on instantaneous fragments, easily leading to the same state being repeatedly classified into different types in adjacent time windows, resulting in discontinuous nighttime state assessments. While existing technologies emphasize anomaly identification and risk assessment, they lack clear constraints on the distribution of anomaly types on the sleep monitoring cycle timeline and the consistency between adjacent time periods. Risk assessment struggles to provide support for risk escalation turning points and phased links. The triggering process is more likely to remain at the level of single threshold fulfillment, making it difficult to distinguish between short-term, occasional anomalies and continuously accumulating anomalies, thus leading to the risk of both frequent triggering and missed interventions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and to propose an anti-snoring system and method based on blood oxygen and nitric oxide asthma identification using breath sounds.

[0006] To achieve the above objectives, the present invention employs the following technical solution: an anti-snoring system based on breath sound-based blood oxygen and nitric oxide asthma identification includes: Breath sound drive mapping module: Based on the respiratory cycle trigger state, locate the start and end nodes of oral and nasal ventilation and mark the ventilation period of a single cycle, compare the direction of change of duration of adjacent cycles, assign a stable cycle ventilation drive level and solidify it sequentially, and generate ventilation drive distribution markers. Respiratory abnormality joint discrimination module: Based on the ventilation drive distribution marker, read the change range of blood oxygen value in the corresponding time window and mark the continuous decreasing segment, read the change direction of nitric oxide value, combine the states and screen out conflicting combinations to obtain the respiratory abnormality type identifier. Sleep monitoring cycle state segmentation module: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the continuous time period of the sleep monitoring cycle is divided into the sleep monitoring cycle, and the number of abnormal occurrences in each segment is counted. The dominant abnormality type is selected and the consistency of abnormalities in adjacent time periods is compared to establish nighttime ventilation state segmentation. Risk evolution path module: Based on the nighttime ventilation state segmentation, sort the time position of the state segment, combine the node propagation weight formed by the graph neural network to screen the risk increasing turning point segment, simultaneously verify the corresponding ventilation drive distribution marker and blood oxygen and nitric oxide change segment, and output the risk change trajectory; Anti-snoring control trigger module: Based on the nighttime ventilation state segmentation and the risk change trajectory, align high-risk state segments and risk turning points, count the number of consecutive abnormal segments and calculate the time interval between segments, match the trigger segment and the observation segment, and obtain anti-snoring intervention instructions.

[0007] As a further aspect of the present invention, the ventilation drive distribution marker includes a ventilation drive level, a level arrangement order, and a corresponding time window identifier; the respiratory abnormality type identifier includes abnormal types associated with blood oxygenation changes, abnormal types associated with nitric oxide changes, and abnormal types combining blood oxygenation and nitric oxide; the nighttime ventilation status segmentation includes a time period number, a time period start and end range, and a dominant abnormal type for that time period; the risk change trajectory includes the location of risk inflection points, the connection relationship between adjacent inflection points, and the corresponding propagation weight results; and the anti-snoring intervention instruction includes a trigger segment identifier, an observation segment identifier, a start identifier, and an end identifier.

[0008] As a further aspect of the present invention, the breathing sound driving mapping module comprises: Ventilation cycle positioning submodule: Based on the respiratory cycle trigger state, identify continuous trigger time points and pair adjacent trigger point intervals, mark the start and end nodes of oral and nasal ventilation to form a single cycle boundary, record the duration of each cycle and remove abnormal offset cycles, and generate a valid identifier for the ventilation cycle. Drive level solidification submodule: Based on the effective identifier of the ventilation cycle, read the duration of each effective cycle and arrange them in chronological order, group and classify the continuous change direction of adjacent cycles, assign corresponding ventilation drive levels and lock the order of continuous levels, and generate ventilation drive distribution markers.

[0009] As a further aspect of the present invention, the respiratory abnormality joint discrimination module includes: Blood oxygen change localization submodule: Based on the ventilation drive distribution marker, read the blood oxygen values ​​within the corresponding time window and arrange them in chronological order, perform point-by-point difference calculation on adjacent values ​​and mark the direction of decrease, connect consecutive segments with the same direction and check the continuous span, exclude discontinuous segments, and generate blood oxygen change segment identifiers. Nitrogen and oxygen joint discrimination submodule: Based on the blood oxygen change segment identifier, read the nitric oxide value within the corresponding time window and mark the direction of change, connect time periods with the same direction and verify the duration, align the time of each blood oxygen segment with the nitric oxide time period, eliminate inconsistent segments and classify consistent segments to obtain the respiratory abnormality type identifier.

[0010] As a further aspect of the present invention, the sleep monitoring cycle state segmentation module includes: Time period segmentation submodule: Based on the respiratory abnormality type identifier, combined with the temporal correlation constraint results of the temporal convolutional network on the abnormal distribution of the sleep monitoring cycle, the time period segment is divided into equal time segments in sequence along the time axis of the sleep monitoring cycle. The abnormal type of each time point is written into the segment to which it belongs and the abnormal occurrence position in the segment is recorded one by one. After accumulating the number of occurrences of each abnormality, the segment content is arranged in chronological order to generate the abnormal distribution of the time period. Dominant anomaly determination submodule: Based on the anomaly distribution over the time period, read the cumulative anomaly results for each time period, compare the number of anomalies occurring in the same segment item by item, select the peak percentage of anomalies as the dominant item, and record the changes of dominant anomalies between adjacent segments to obtain the dominant anomaly ranking. State Boundary Solidification Submodule: Based on the dominant anomaly arrangement, the consistency of dominant anomalies in adjacent segments is compared segment by segment. Consistent segments are continuously merged and their duration is extended. The boundary position of inconsistent segments is retained, and the segment number and length are written for the merged segments to establish nighttime ventilation state segments.

[0011] As a further aspect of the present invention, the temporal convolutional network receives the time sequence of the respiratory abnormality type identifiers in the order of the sleep monitoring cycle time axis, selects adjacent continuous positions on the time axis to form a fixed-length window, performs convolutional kernel coverage and position displacement processing on the abnormality type arrangement within the window, accumulates and records the response values ​​generated at the same time position under different coverage windows, and outputs the corresponding temporal correlation constraint results at each position on the time axis. The temporal correlation constraint results are written into the abnormal distribution data structure in a time index aligned form.

[0012] As a further aspect of the present invention, the risk evolution path module includes: State sequence construction submodule: Based on the nighttime ventilation state segmentation, read the segment number of each state segment and sort them according to the occurrence time, connect adjacent state segments one by one and write the connection relationship, record the duration of each connection and verify the sequential continuity, remove missing connections, and generate the state segment sequence structure. Turning point filtering submodule: Based on the state segment sequence structure, compare the abnormal level change direction of adjacent state segments one by one, mark the position of change from low level to high level, and synchronously read the corresponding ventilation drive distribution mark and blood oxygen and nitric oxide change segment and align the time to obtain the risk turning point set; Evolutionary chain solidification submodule: Based on the risk turning point node set, it introduces the inter-node propagation weight and link consistency judgment result formed by graph neural network, connects adjacent turning points in chronological order, checks the continuity of state segments between nodes and excludes broken segments, writes continuous nodes into a unified link identifier, records the start and end positions of the link and the coverage segment number, and outputs the risk change trajectory.

[0013] As a further aspect of the present invention, the graph neural network constructs a node set through a risk inflection node set, writes the temporal adjacency relationship and state segment continuity relationship between adjacent inflection nodes into the node connection relationship, performs multiple rounds of adjacent node information transmission according to the connection relationship, summarizes the abnormality level change information and time position identifier of adjacent nodes in each round of transmission, and records the propagated weight value at the node level. At the same time, it performs consistency judgment on the continuous link composed of multiple nodes, writes the node connection that meets the continuity condition into a unified link identifier structure and outputs the corresponding propagation weight result.

[0014] As a further aspect of the present invention, the anti-snoring control trigger module includes: Risk Alignment Screening Submodule: Based on the nighttime ventilation status segmentation and the risk change trajectory, the time range of the status segment and the time position of the turning point are compared segment by segment. Alignment is completed through overlapping interval verification. The number of consecutive high-risk status segments is counted and the time interval between adjacent segments is calculated. Combinations with excessive intervals are screened out and intervention triggering segment identifiers are generated. Intervention instruction generation submodule: Based on the intervention trigger segment identifier, it matches the corresponding observation segment time range one by one and locates the risk fall-off node position, verifies the start and end of the trigger segment and the order of fall-off nodes, writes the start identifier and end identifier and summarizes them in time order to obtain the anti-snoring intervention instruction.

[0015] A method for stopping snoring based on blood oxygenation and nitric oxide asthma identification using breath sounds, wherein the method is executed based on the aforementioned system for stopping snoring based on blood oxygenation and nitric oxide asthma identification using breath sounds, includes the following steps: S1: Based on the respiratory cycle trigger state, determine the start and end points of oral and nasal ventilation and form a single-cycle ventilation period. Compare the direction of change of duration of adjacent cycles, assign a stable ventilation drive level and solidify it in time sequence to generate a ventilation drive distribution marker. S2: Based on the ventilation-driven distribution marker, read the changes in blood oxygen values ​​within the corresponding time window and mark the continuously decreasing segments. At the same time, read the direction of nitric oxide value changes. Combine the two types of changes and eliminate conflicting situations to obtain the respiratory abnormality type identifier. S3: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the sleep monitoring cycle time axis is divided into continuous time periods and the number of abnormal occurrences in each period is counted. The dominant abnormality type is selected and the consistency of adjacent periods is compared to establish nighttime ventilation status segmentation. S4: Based on the nighttime ventilation state segmentation, the state segments are arranged in chronological order. The node propagation weights formed by the graph neural network are used to filter the transition segments with increasing abnormal levels. The corresponding ventilation drive distribution markers and blood oxygen and nitric oxide change segments are checked simultaneously, and the risk change trajectory is output. S5: Based on the nighttime ventilation status segmentation and the risk change trajectory, align the high-risk status segment with the risk turning point, count the number of consecutive abnormal segments and calculate the time interval between segments, match the trigger segment with the observation segment, and obtain the anti-snoring intervention instruction.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. In this invention, by introducing anomaly temporal correlation constraints formed by temporal convolutional networks, the anomaly determination is no longer limited to independent identification within a single time window, but instead treats the anomaly types at adjacent time positions as a continuous sequence for overall constraint and consistency verification, thereby suppressing the discrete fluctuations caused by changes in body position, instantaneous breathing instability, and short-term snoring enhancement in the time dimension. 2. In this invention, by introducing node propagation weights formed by graph neural networks, risk assessment is shifted from simple time sequence judgment to propagation and consistency analysis based on the connection relationship between nodes. The risk increment position is screened by the joint constraint of the direction of change of the abnormal level between nodes and the temporal adjacency relationship, avoiding risk inference based solely on the intensity of a single abnormal segment. 3. In this invention, the abnormality level changes and time location information of adjacent turning points are summarized through multiple rounds of adjacent node information transmission. Continuous risk segments form stable link identifiers under the consistency verification of propagation weights, thereby reducing the probability of false triggering and missed triggering, improving the matching degree between intervention timing and the evolution rhythm of nighttime ventilation risks, and realizing a quantifiable description of risk accumulation and turning points in the sleep scenario. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages 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 and not intended to limit the invention.

[0019] Example 1 Please see Figure 1 The present invention provides a technical solution: an anti-snoring system based on breath sound-based blood oxygen and nitric oxide asthma identification, comprising: Breath sound drive mapping module: Based on the respiratory cycle trigger state, locate the start and end nodes of oral and nasal ventilation and mark the ventilation period of a single cycle, compare the direction of change of duration of adjacent cycles, assign a stable cycle ventilation drive level and solidify it sequentially, and generate ventilation drive distribution markers. Respiratory abnormality joint discrimination module: Based on ventilation-driven distribution markers, read the amplitude of blood oxygen value changes in the corresponding time window and mark the continuously decreasing segment, read the direction of nitric oxide value changes, combine the states and screen out conflicting combinations to obtain respiratory abnormality type identifiers. Sleep monitoring cycle state segmentation module: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the continuous time period of the sleep monitoring cycle is divided into the sleep monitoring cycle, and the number of abnormal occurrences in each segment is counted. The dominant abnormality type is selected and the consistency of abnormalities in adjacent time periods is compared to establish nighttime ventilation state segmentation. Risk evolution path module: Based on nighttime ventilation status segmentation, the time position of the status segment is sorted, and the node propagation weight formed by graph neural network is used to screen the risk increasing turning point segment. The corresponding ventilation drive distribution marker and blood oxygen and nitric oxide change segment are simultaneously verified, and the risk change trajectory is output. Anti-snoring control trigger module: Based on nighttime ventilation status segmentation and risk change trajectory, it aligns high-risk status segments and risk inflection points, counts the number of consecutive abnormal segments and calculates the time interval between segments, matches the trigger segment and the observation segment, and obtains anti-snoring intervention instructions.

[0020] The ventilation drive distribution markers include ventilation drive level, level order and corresponding time window identifier; the respiratory abnormality type identifiers include abnormal types associated with blood oxygenation changes, abnormal types associated with nitric oxide changes, and abnormal types of combined blood oxygenation and nitric oxide; the nighttime ventilation status segmentation includes time period number, time period start and end range and dominant abnormal type of time period; the risk change trajectory includes the location of risk inflection point, the connection relationship between adjacent inflection points and the corresponding propagation weight results; the anti-snoring intervention instructions include trigger segment identifier, observation segment identifier, start identifier and end identifier.

[0021] The breath sound drive mapping module includes: Ventilation cycle positioning submodule: Based on the respiratory cycle trigger state, identify continuous trigger time points and pair adjacent trigger point intervals, mark the start and end nodes of oral and nasal ventilation to form a single cycle boundary, record the duration of each cycle and remove abnormal offset cycles, and generate a valid identifier for the ventilation cycle. Drive level solidification submodule: Based on the valid identifier of the ventilation cycle, read the duration of each valid cycle and arrange them in chronological order, group and classify the continuous change direction of adjacent cycles, assign corresponding ventilation drive levels and lock the order of consecutive levels, and generate ventilation drive distribution markers. Ventilation cycle positioning submodule: Based on the respiratory cycle trigger state, it identifies event points in the trigger state time series. The preset sampling frequency is 1000 times per second, the preset trigger threshold is 1, and the time point when the trigger state changes from 0 to 1 is recorded as the trigger start point. The preset release threshold is 0, and the time point when the trigger state changes from 1 to 0 is recorded as the trigger end point. The preset minimum trigger duration is 200 milliseconds, and trigger segments shorter than 200 milliseconds are removed. The preset minimum interval between adjacent trigger segments is 800 milliseconds, and trigger segments with an interval less than 800 milliseconds are merged to obtain a continuous trigger time point sequence. Adjacent trigger start points and adjacent trigger end points are paired in chronological order to form paired boundaries. The time point before each pair of boundaries is marked as the start node of oral-nasal ventilation, and the time point after each pair of boundaries is marked as the end node of oral-nasal ventilation to form a single cycle boundary. The duration of each single cycle boundary is recorded and written into the cycle duration sequence. The preset effective range of cycle duration is 1 second to 8 seconds, and cycle records with durations shorter than 1 second and longer than 8 seconds are removed, generating a valid identifier for the ventilation cycle. Drive level solidification submodule: Based on the valid identifier of the ventilation cycle, it reads the duration of each valid cycle and arranges them in chronological order to form a cycle duration sequence. The preset direction determination tolerance is 0.05 seconds, and the difference between the durations of two adjacent cycles is used as the direction determination basis. A difference greater than 0.05 seconds is determined as an upward direction, a difference between -0.05 seconds and 0.05 seconds is determined as a stationary direction, and a difference less than -0.05 seconds is determined as a downward direction, resulting in a change direction sequence. The change direction sequence is grouped and classified according to the consistency of adjacent cycles, and a group number is written for each group. The ventilation drive level is represented by an integer encoding method, and the encoding value is fixed as 2 for the upward direction, 1 for the stationary direction, and 0 for the downward direction. The level code is mapped to the corresponding cycle number and written into the level sequence in chronological order. Sequence locking is performed on the continuous level sequence, and the preset locking rule is that the order of the level code cannot be exchanged when it first appears, and subsequent records can only be appended, not inserted or rewritten. A ventilation drive distribution mark is generated.

[0022] The respiratory abnormality joint discrimination module includes: Blood oxygen change localization submodule: Based on ventilation-driven distribution markers, read blood oxygen values ​​within the corresponding time window and arrange them in chronological order. Perform point-by-point difference calculation on adjacent values ​​and mark the direction of decrease. Continuously connect segments with consistent directions and check the continuous span. Exclude discontinuous segments and generate blood oxygen change segment identifiers. Nitrogen and oxygen joint discrimination submodule: Based on the blood oxygen change segment identifier, read the nitric oxide value in the corresponding time window and mark the change direction, connect the time periods with the same direction and verify the duration, align the time of each blood oxygen segment with the nitric oxide time period, remove inconsistent segments and classify consistent segments to obtain the respiratory abnormality type identifier. Blood oxygen change localization submodule: Based on ventilation-driven distribution markers, it reads the blood oxygen value sequence within the corresponding time window. The preset blood oxygen sampling period is 1 second and at least 30 sampling points are read continuously. The blood oxygen values ​​are arranged in the order of sampling time to form a time sequence. The difference between adjacent sampling points is subtracted point by point and recorded. The preset threshold for the decrease is -0.5 and adjacent points with a difference less than -0.5 are marked as the direction of decrease. The preset minimum number of consecutive decrease points is 5 and the decrease direction points that meet the continuity condition are connected into segments. The connection process is completed by traversing the decrease marking sequence point by point and recording the start and end time points. The preset minimum duration is 5 seconds and the duration of each connected segment is checked. Segments with a duration of less than 5 seconds and those with discontinuities of more than 2 seconds are removed. The remaining consecutive decrease segments are written with segment numbers and corresponding time ranges to generate blood oxygen change segment identifiers. The nitrogen oxide joint discrimination submodule reads the nitric oxide numerical sequence within the corresponding time window based on the blood oxygen change segment identifier. The preset nitric oxide sampling period is 1 second, and values ​​are read point-by-point within the blood oxygen change segment coverage time range. Numerical difference is performed on adjacent sampling points, and the direction of change is recorded. The preset threshold for determining the direction of change is +0.2 and -0.2. A difference greater than 0.2 is marked as an upward direction, a difference less than -0.2 is marked as a downward direction, and a difference between +0.2 and -0.2 is marked as a stable direction. For consecutive sampling points with the same direction, segment connection is performed, and the start and end times are recorded. The preset minimum duration is 5 seconds, and segments with insufficient duration are removed. The remaining nitric oxide segments and blood oxygen change segments are aligned segment by segment according to the time index. The alignment rule is that the time overlap length of the two types of segments is not less than 60% of their respective segment durations. Segments that do not meet the overlap condition are removed. Segments that meet the condition are classified according to the directional combination relationship and written with a classification number, generating a respiratory abnormality type identifier.

[0023] The sleep monitoring cycle state segmentation module includes: Time period segmentation submodule: Based on the respiratory abnormality type identifier, combined with the temporal correlation constraint results of the sleep monitoring cycle abnormal distribution of the temporal convolutional network, the time period segment is divided into equal time segments in sequence along the time axis of the sleep monitoring cycle. The abnormality type of each time point is written into the segment to which it belongs and the location of the abnormality in the segment is recorded one by one. After accumulating the number of occurrences of each abnormality, the segment content is arranged in chronological order to generate the time period abnormal distribution. Dominant Anomaly Determination Submodule: Based on the anomaly distribution over time periods, read the cumulative anomaly results for each time period, compare the number of anomalies occurring in the same segment item by item, select the anomaly with the highest percentage of occurrences as the dominant item, and record the changes of the dominant anomaly between adjacent segments to obtain the dominant anomaly ranking. State Boundary Solidification Submodule: Based on the dominant anomaly arrangement, the consistency of the dominant anomalies in adjacent segments is compared segment by segment. Consistent segments are continuously merged and their duration is extended. The boundary position of inconsistent segments is retained, and the segment number and length are written to the merged segments to establish nighttime ventilation state segments. The time-segmentation submodule: Based on the respiratory abnormality type identifier, a temporal convolutional network is used to perform temporal correlation processing on the abnormal distribution of sleep monitoring cycles. The start time of the sleep monitoring cycle time axis is preset to 0 seconds and the end time is recorded. The preset time axis sampling resolution is 1 second. The respiratory abnormality type identifiers are written into the abnormal sequence array according to the time index order, constructing an abnormal sequence with a length equal to the number of seconds in the sleep monitoring cycle. A multi-layer one-dimensional convolutional structure is used to perform coverage calculation on the abnormal sequence. The number of convolutional layers is set to 3. The kernel length of the first layer is set to 5 time points with a stride of 1. The kernel length of the second layer is set to 7 time points with a stride of 1. The third layer... The convolutional kernel length is set to 9, and the stride at each time point is set to 1. The convolutional kernel weights are pre-set and stored through offline training models. The output response values ​​from different convolutional coverage ranges are accumulated at each time point to generate a temporal correlation constraint label sequence. Then, the time segments are divided into equal-length segments along the sleep monitoring cycle time axis. The length of a single time segment is preset to 300 seconds and numbered in order of start time. The anomaly type and corresponding temporal correlation constraint label at each time point are written into the time segment. The time position and type number of the anomaly occurrence are recorded one by one. The anomaly type is counted and accumulated within the segment, and the segment content is organized in chronological order to generate the time period anomaly distribution. The dominant anomaly determination submodule reads the cumulative number of occurrences of each anomaly type in each time period based on the anomaly distribution over time. It constructs a segment count table by associating the anomaly type number with the corresponding occurrence count. It performs a step-by-step comparison operation on the counts of all anomaly types within the same segment. The preset dominant determination rule is that the anomaly type number with the maximum occurrence count corresponds to the anomaly type number. When multiple anomaly types have the same occurrence count and the maximum value, the earliest occurrence anomaly type number is selected as the dominant item according to the first-in-time principle. It establishes a mapping relationship between the dominant anomaly type number and the corresponding segment number. Then, it arranges the dominant anomaly type number sequence according to the segment number order and records whether the dominant anomaly type changes in adjacent segments. The changes are written into the adjacent segment relationship table to obtain the dominant anomaly arrangement. The State Boundary Solidification Submodule: Based on the dominant anomaly arrangement, it performs a segment-by-segment comparison operation on the segment sequence. Starting from the first segment, it reads the dominant anomaly type number and compares it with the dominant anomaly type number of the next adjacent segment. When the numbers match, the two segments are marked as mergingable and the starting segment number of the merge is recorded. When the numbers do not match, the current segment boundary position is retained and the merge record is ended. Then, the same judgment process is continued to be executed. For consecutive segments with the same number, the merge process is performed and the duration of the merged segments is accumulated. The duration is calculated by multiplying the number of merged segments by the duration of a single segment of 300 seconds. A new segment number is written to each merged segment and the start time position and accumulated duration are recorded. For unmerged segments, the original segment boundary is retained and an independent segment number and duration are written. Finally, a continuous segment structure record is formed, and the nighttime ventilation state segmentation is established.

[0024] The temporal convolutional network receives a time sequence of respiratory abnormality type identifiers in the order of the sleep monitoring cycle time axis. It selects adjacent continuous positions on the time axis to form a fixed-length window, performs convolutional kernel coverage and position displacement processing on the abnormality type arrangement within the window, accumulates and records the response values ​​generated at the same time position under different coverage windows, and outputs the corresponding temporal association constraint results at each position on the time axis. The temporal association constraint results are written into the abnormal distribution data structure in time index aligned form. Temporal convolutional networks, according to the formula: in: Indicates the time index is The calculated temporal correlation constraint result value, This represents the time index number on the sleep monitoring cycle timeline and increments according to a fixed time resolution. Indicates that at time index The number of coverage windows involved in the cumulative calculation. Indicates the coverage window index and from 0 to Take values ​​in sequence, This represents the coverage window attenuation coefficient and is used to assign time weights to different coverage windows. This represents the length of the convolution window and the number of time points covered by a single convolution kernel. Indicates the weight index inside the convolution kernel and ranges from 0 to... Take values ​​in sequence, Indicates the first The convolution weight coefficients and their corresponding convolution kernels. One location, Indicates the time index is The numerical value indicating the type of respiratory abnormality at that location. This represents the time shift step parameter and is used to control the convolution sampling time interval. This represents the bias term and is used to perform numerical translation correction on the overall convolution calculation result; Execution process: The respiratory abnormality type identifier is first mapped to a time index sequence according to the time axis of the sleep monitoring cycle. The system constructs a continuous-time index based on a fixed time resolution. Index at various times Select in sequence at each location. A covering window and indexed by window Each coverage level is labeled, and the content within each coverage window is arranged according to the length of the convolution window. With time displacement step Read the exception type identifier in the direction of the historical time index and the corresponding convolution weights After performing term-by-term multiplication, the values ​​are summed to form the convolution response values ​​within the window, which are then processed by a decay coefficient. The response values ​​of different coverage windows are weighted, summed, and then the result is determined by the number of windows. Perform normalization calculations and finally introduce a bias term. The results are corrected to improve the time index. The unique and definite temporal correlation constraint result value is obtained at this point. The results are written into the anomaly distribution data structure in chronological order, serving as the input basis for subsequent nighttime ventilation status segmentation and risk evolution path analysis.

[0025] The risk evolution path module includes: State sequence construction submodule: Based on nighttime ventilation state segmentation, read the segment number of each state segment and sort them according to the occurrence time, connect adjacent state segments one by one and write the connection relationship, record the corresponding duration of each connection and verify the sequential continuity, remove missing connections, and generate the state segment sequence structure. Turning point filtering submodule: Based on the state segment sequence structure, it compares the direction of abnormal level change of adjacent state segments one by one, marks the position of change from low level to high level, and synchronously reads the corresponding ventilation drive distribution marker and blood oxygen and nitric oxide change segment and aligns the time to obtain the risk turning point set. Evolutionary chain solidification submodule: Based on the risk inflection node set, it introduces the inter-node propagation weight and link consistency judgment result formed by graph neural network, connects adjacent inflection nodes in chronological order, checks the continuity of state segments between nodes and excludes broken segments, writes continuous nodes into a unified link identifier, records the link start and end positions and coverage segment number, and outputs the risk change trajectory. The state sequence construction submodule is based on nighttime ventilation state segmentation. The nighttime start time is preset to 0 seconds and the end time is recorded. Each state segment is read into a structure record containing four items: segment number, start time, end time, and duration. All state segments are sorted in ascending order based on the start time field. The sorting rule is preset to time values ​​from smallest to largest and no duplicate start times are allowed. After sorting, adjacent state segments are sequentially connected to each other. The connection record contains four fields: previous segment number, next segment number, connection start time, and connection end time. The duration is calculated for each connection. The duration is obtained by subtracting the previous segment's end time from the next segment's start time. The preset continuity check rule is that a duration of 0 seconds or less than or equal to 5 seconds is considered sequentially continuous, while a duration greater than 5 seconds is considered sequentially broken. Connection records that do not meet the continuity rule are removed. Only connection records that meet the continuity rule are retained. The retained connection records are written into the sequence table in chronological order to generate the state segment sequence structure. The inflection point filtering submodule, based on the state segment sequence structure, uses an adjacent anomaly level direction comparison method to perform pairwise judgment on the connected state segments. It reads the anomaly level of the preceding and following state segments corresponding to each connected record. The anomaly level is represented by a preset integer encoding method, where low level is encoded as 0, medium level as 1, and high level as 2. The preceding and following levels are compared numerically. When the value of the following level is greater than that of the preceding level, the connection position is marked as an incrementing position, and the corresponding preceding and following segment numbers are recorded as node indices. At the same time, the ventilation drive distribution marking time index, blood oxygen change segment time index, and nitric oxide change segment time index within the connection coverage time range are read. Time alignment is performed on the three types of time indices. The alignment rule is preset to be that the overlap length of the three types of time indices is not less than 60% of the connection duration. Records that do not meet the overlap rule are removed. Records that meet the rule are written with node numbers and the occurrence time position is recorded, generating a risk inflection point set. The evolution chain solidification submodule: Based on the risk inflection node set, a graph neural network is used to construct links for node relationships. The preset node set consists of all node numbers in the risk inflection node set. The preset node connection rule is that the nodes are adjacent in time order and their corresponding state segments are continuous. For nodes that meet the rule, a directed connection relationship is established and written into the adjacency list. A fixed 3-layer graph propagation structure is used to perform multiple rounds of node information transmission. In each round of propagation, the propagation weights of adjacent nodes are updated. The initial propagation weight is uniformly set to 1. The update rule is preset to take the average weight of adjacent nodes and retain two decimal places. After completing 3 rounds of propagation, the final weight value of each node is recorded. Then, adjacent nodes are connected according to the node time order to form candidate links. Consistency judgment is performed on the candidate links. The consistency judgment rule is preset to be that the difference in propagation weights of nodes in the link does not exceed 0.3 and the corresponding state segment numbers are continuous. Links that do not meet the consistency rule are eliminated. Links that meet the rule are written with a unified link number, and the start time, end time, and coverage of the state segment number range are recorded. The risk change trajectory is output.

[0026] The graph neural network constructs a node set through a risk inflection node set, writes the temporal adjacency and state segment continuity between adjacent inflection nodes into the node connection relationship, performs multiple rounds of adjacent node information transmission according to the connection relationship, summarizes the abnormality level change information and time position identifier of adjacent nodes in each round of transmission, and records the propagated weight value at the node level. At the same time, it performs consistency judgment on the continuous link composed of multiple nodes, writes the node connection that meets the continuity condition into a unified link identifier structure and outputs the corresponding propagation weight result. Graph neural networks, according to the formula: in: Represents a node In the transmission rounds The node propagation weight values ​​are calculated. Indicates the name of the propagation weight variable. Indicates the index of the risk inflection point number. This indicates the propagation round index and takes the value 3. Indicates the threshold mapping rule symbol. This represents the round decay coefficient and has a value of 0.8. This represents the self-feedback coefficient and has a value of 1.0. Represents a node In the round Input feature values, Indicates the name of the node feature variable. This indicates that the accumulation symbol is applied to the set of adjacent nodes. Represents nodes There are adjacent node indexes with connection relationships. Represents a node The corresponding set of adjacent nodes Represents a node With nodes The connection flag is set to 1 when the time intervals are adjacent and the state segments are consecutive, and 0 otherwise. Represents a node With nodes The time interval coefficient is set to 1 / (1 + time difference). Represents a node With nodes The continuous relation coefficient of the state segment takes a value of 1 when continuous and a value of 0 when otherwise. Represents a node The number of connections is obtained by counting adjacent nodes. Represents a node The number of connections is obtained by counting adjacent nodes. This represents the node degree normalization term. Represents a node In the round Input feature values, Indicates the offset item and sets its value to 0; Execution process: Based on the risk inflection point set, a node number index is constructed and a node sequence is formed according to the occurrence time. Based on the temporal adjacency of nodes and the segmented continuity of nighttime ventilation status, connection tags are written. And generate the adjacent set of each node. The size of the adjacent set of each node is obtained by statistically analyzing the values ​​of adjacent sets. And obtain adjacent nodes After reading the node's time position and calculating the time difference, write the time interval coefficient as 1 / (1+time difference). Then read the state segment number and write the state segment continuity coefficient according to the numbering continuity rule. Write the previous round input feature values ​​of each node into Write the previous round input feature values ​​of adjacent nodes into In the transmission rounds Calculations are performed sequentially for values ​​ranging from 1 to 3, and the node's own items are counted. Adjacent node items exist The sums are accumulated and combined into the summation value within the parentheses. The summation value is then multiplied by the round decay term. And add offset terms The result is obtained by inputting the threshold mapping rule. get And will be the final round of propagation correspond Write the propagation weight result record table according to the node number for use in link consistency determination and unified link identifier structure writing.

[0027] The anti-snoring control trigger module includes: Risk Alignment Screening Submodule: Based on nighttime ventilation status segmentation and risk change trajectory, it compares the time range of status segments and the time position of turning points segment by segment, completes alignment through overlapping interval verification, counts the number of consecutive high-risk status segments and calculates the time interval between adjacent segments, filters out combinations with excessive intervals, and generates intervention triggering segment identifiers. Intervention instruction generation submodule: Based on the intervention trigger segment identifier, it matches the corresponding observation segment time range one by one and locates the risk fallback node position, verifies the start and end of the trigger segment and the order of fallback nodes, writes the start and end identifiers and summarizes them in time order to obtain the anti-snoring intervention instruction; Risk Alignment Screening Submodule: Based on nighttime ventilation status segmentation and risk change trajectory, the module uses a time interval overlap determination method to read the start and end times of each status segment and construct time intervals. Simultaneously, it reads the time positions of risk turning points in the risk change trajectory and forms a node time sequence. It performs segment-by-segment inclusion determination on status segments and node times, and records node times that are within the time interval of a status segment as overlapping relationships. The preset alignment rule is that if there is at least one turning point in a status segment, it is written into the alignment record, and status segments that meet the condition are marked as high-risk status segments. Then, the high-risk status segments are arranged in chronological order and the number of consecutive segments is counted. The consecutive determination rule is preset to the difference between adjacent segment numbers being 1. At the same time, the time interval between adjacent high-risk status segments is calculated by subtracting the end time of the previous segment from the start time of the next segment. The preset upper limit of the interval is 60 seconds, and segments exceeding the upper limit are removed. For the retained segment groups, the segment number range and time range information are written, and an intervention trigger segment identifier is generated. The intervention instruction generation submodule: Based on the intervention trigger segment identifier, it reads the corresponding time range of each trigger segment through time window matching and sequential labeling methods, and retrieves the subsequent node time in the risk change trajectory. The preset observation segment length is 120 seconds, and the start time of the observation segment is set as the end time of the trigger segment. It performs interval retrieval on the node time series and locates the risk fall-off node within the observation segment. The fall-off node judgment rule is preset to the abnormality level changing from high-level code 2 to medium-level code 1 and low-level code 0. It checks the order of the start time of the trigger segment, the end time of the trigger segment, and the fall-off node time for the retrieved node records, and writes the start time of the trigger segment into the start identifier and the fall-off node time into the end identifier. Finally, it summarizes all records in chronological order to obtain the anti-snoring intervention instruction.

[0028] Please see Figure 1 A method for stopping snoring based on blood oxygen and nitric oxide levels in breath sounds to identify asthma includes the following steps: S1: Based on the respiratory cycle trigger state, determine the start and end points of oral and nasal ventilation and form a single-cycle ventilation period. Compare the direction of change of duration of adjacent cycles, assign a stable ventilation drive level and solidify it in time sequence to generate a ventilation drive distribution marker. S2: Based on the ventilation-driven distribution marker, read the changes in blood oxygen values ​​within the corresponding time window and mark the continuously decreasing segments. At the same time, read the direction of nitric oxide value changes. Combine the two types of changes and eliminate conflicting situations to obtain the respiratory abnormality type identifier. S3: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the sleep monitoring cycle time axis is divided into continuous time periods and the number of abnormal occurrences in each period is counted. The dominant abnormality type is selected and the consistency of adjacent periods is compared to establish the nighttime ventilation status segmentation. S4: Based on nighttime ventilation status segmentation, the status segments are arranged in chronological order. The node propagation weights formed by the graph neural network are used to filter the transition segments with increasing abnormal levels. The corresponding ventilation drive distribution markers and blood oxygen and nitric oxide change segments are verified simultaneously, and the risk change trajectory is output. S5: Based on the segmentation of nighttime ventilation status and risk change trajectory, align high-risk status segments and risk turning points, count the number of consecutive abnormal segments and calculate the time interval between segments, match the trigger segment and the observation segment, and obtain the anti-snoring intervention instruction.

[0029] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An anti-snoring system based on breath sound-based blood oxygenation and nitric oxide asthma identification, characterized in that, The system includes: Breath sound drive mapping module: Based on the respiratory cycle trigger state, locate the start and end nodes of oral and nasal ventilation and mark the ventilation period of a single cycle, compare the direction of change of duration of adjacent cycles, assign a stable cycle ventilation drive level and solidify it sequentially, and generate ventilation drive distribution markers. Respiratory abnormality joint discrimination module: Based on the ventilation drive distribution marker, read the change range of blood oxygen value in the corresponding time window and mark the continuous decreasing segment, read the change direction of nitric oxide value, combine the states and screen out conflicting combinations to obtain the respiratory abnormality type identifier. Sleep monitoring cycle state segmentation module: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the continuous time period of the sleep monitoring cycle is divided into the sleep monitoring cycle, and the number of abnormal occurrences in each segment is counted. The dominant abnormality type is selected and the consistency of abnormalities in adjacent time periods is compared to establish nighttime ventilation state segmentation. Risk evolution path module: Based on the nighttime ventilation state segmentation, sort the time position of the state segment, combine the node propagation weight formed by the graph neural network to screen the risk increasing turning point segment, simultaneously verify the corresponding ventilation drive distribution marker and blood oxygen and nitric oxide change segment, and output the risk change trajectory; Anti-snoring control trigger module: Based on the nighttime ventilation state segmentation and the risk change trajectory, align high-risk state segments and risk turning points, count the number of consecutive abnormal segments and calculate the time interval between segments, match the trigger segment and the observation segment, and obtain anti-snoring intervention instructions.

2. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The ventilation drive distribution markers include ventilation drive levels, level order, and corresponding time window identifiers. The respiratory abnormality type identifiers include abnormal types associated with blood oxygenation changes, abnormal types associated with nitric oxide changes, and abnormal types combining blood oxygenation and nitric oxide. The nighttime ventilation status segmentation includes time period numbers, time period start and end ranges, and dominant abnormal types for each time period. The risk change trajectory includes the location of risk inflection points, the connection relationship between adjacent inflection points, and the corresponding propagation weight results. The anti-snoring intervention instructions include trigger segment identifiers, observation segment identifiers, activation identifiers, and termination identifiers.

3. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The breath sound driving mapping module includes: Ventilation cycle positioning submodule: Based on the respiratory cycle trigger state, identify continuous trigger time points and pair adjacent trigger point intervals, mark the start and end nodes of oral and nasal ventilation to form a single cycle boundary, record the duration of each cycle and remove abnormal offset cycles, and generate a valid identifier for the ventilation cycle. Drive level solidification submodule: Based on the effective identifier of the ventilation cycle, read the duration of each effective cycle and arrange them in chronological order, group and classify the continuous change direction of adjacent cycles, assign corresponding ventilation drive levels and lock the order of continuous levels, and generate ventilation drive distribution markers.

4. The anti-snoring system based on breath sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The respiratory abnormality joint discrimination module includes: Blood oxygen change localization submodule: Based on the ventilation drive distribution marker, read the blood oxygen values ​​within the corresponding time window and arrange them in chronological order, perform point-by-point difference calculation on adjacent values ​​and mark the direction of decrease, connect consecutive segments with the same direction and check the continuous span, exclude discontinuous segments, and generate blood oxygen change segment identifiers. Nitrogen and oxygen joint discrimination submodule: Based on the blood oxygen change segment identifier, read the nitric oxide value within the corresponding time window and mark the direction of change, connect time periods with the same direction and verify the duration, align the time of each blood oxygen segment with the nitric oxide time period, eliminate inconsistent segments and classify consistent segments to obtain the respiratory abnormality type identifier.

5. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The sleep monitoring cycle state segmentation module includes: Time period segmentation submodule: Based on the respiratory abnormality type identifier, combined with the temporal correlation constraint results of the temporal convolutional network on the abnormal distribution of the sleep monitoring cycle, the time period segment is divided into equal time segments in sequence along the time axis of the sleep monitoring cycle. The abnormal type of each time point is written into the segment to which it belongs and the abnormal occurrence position in the segment is recorded one by one. After accumulating the number of occurrences of each abnormality, the segment content is arranged in chronological order to generate the abnormal distribution of the time period. Dominant anomaly determination submodule: Based on the anomaly distribution over the time period, read the cumulative anomaly results for each time period, compare the number of anomalies occurring in the same segment item by item, select the peak percentage of anomalies as the dominant item, and record the changes of dominant anomalies between adjacent segments to obtain the dominant anomaly ranking. State Boundary Solidification Submodule: Based on the dominant anomaly arrangement, the consistency of dominant anomalies in adjacent segments is compared segment by segment. Consistent segments are continuously merged and their duration is extended. The boundary position of inconsistent segments is retained, and the segment number and length are written for the merged segments to establish nighttime ventilation state segments.

6. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The temporal convolutional network receives the time sequence of the respiratory abnormality type identifiers according to the time axis of the sleep monitoring cycle. It selects adjacent continuous positions on the time axis to form a fixed-length window, performs convolutional kernel coverage and position displacement processing on the abnormality type arrangement within the window, accumulates and records the response values ​​generated at the same time position under different coverage windows, and outputs the corresponding temporal correlation constraint results at each position on the time axis. The temporal correlation constraint results are written into the abnormal distribution data structure in a time index aligned form.

7. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The risk evolution path module includes: State sequence construction submodule: Based on the nighttime ventilation state segmentation, read the segment number of each state segment and sort them according to the occurrence time, connect adjacent state segments one by one and write the connection relationship, record the duration of each connection and verify the sequential continuity, remove missing connections, and generate the state segment sequence structure. Turning point filtering submodule: Based on the state segment sequence structure, compare the abnormal level change direction of adjacent state segments one by one, mark the position of change from low level to high level, and synchronously read the corresponding ventilation drive distribution mark and blood oxygen and nitric oxide change segment and align the time to obtain the risk turning point set; Evolutionary chain solidification submodule: Based on the risk turning point node set, it introduces the inter-node propagation weight and link consistency judgment result formed by graph neural network, connects adjacent turning points in chronological order, checks the continuity of state segments between nodes and excludes broken segments, writes continuous nodes into a unified link identifier, records the start and end positions of the link and the coverage segment number, and outputs the risk change trajectory.

8. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The graph neural network constructs a node set through a risk inflection node set, writes the temporal adjacency relationship and state segment continuity relationship between adjacent inflection nodes into the node connection relationship, performs multiple rounds of adjacent node information transmission according to the connection relationship, summarizes the abnormality level change information and time position identifier of adjacent nodes in each round of transmission, and records the propagated weight value at the node level. At the same time, it performs consistency judgment on the continuous link composed of multiple nodes, writes the node connection that meets the continuity condition into a unified link identifier structure and outputs the corresponding propagation weight result.

9. The anti-snoring system based on respiratory sound-based blood oxygenation and nitric oxide asthma identification according to claim 1, characterized in that, The anti-snoring control trigger module includes: Risk Alignment Screening Submodule: Based on the nighttime ventilation status segmentation and the risk change trajectory, the time range of the status segment and the time position of the turning point are compared segment by segment. Alignment is completed through overlapping interval verification. The number of consecutive high-risk status segments is counted and the time interval between adjacent segments is calculated. Combinations with excessive intervals are screened out and intervention triggering segment identifiers are generated. Intervention instruction generation submodule: Based on the intervention trigger segment identifier, it matches the corresponding observation segment time range one by one and locates the risk fall-off node position, verifies the start and end of the trigger segment and the order of fall-off nodes, writes the start identifier and end identifier and summarizes them in time order to obtain the anti-snoring intervention instruction.

10. A method for stopping snoring based on blood oxygenation and nitric oxide asthma identification using breath sounds, characterized in that, The anti-snoring system based on breath sound-based blood oxygenation and nitric oxide asthma identification, according to any one of claims 1-9, comprises the following steps: S1: Based on the respiratory cycle trigger state, determine the start and end points of oral and nasal ventilation and form a single-cycle ventilation period. Compare the direction of change of duration of adjacent cycles, assign a stable ventilation drive level and solidify it in time sequence to generate a ventilation drive distribution marker. S2: Based on the ventilation-driven distribution marker, read the changes in blood oxygen values ​​within the corresponding time window and mark the continuously decreasing segments. At the same time, read the direction of nitric oxide value changes. Combine the two types of changes and eliminate conflicting situations to obtain the respiratory abnormality type identifier. S3: Based on the respiratory abnormality type identifier, combined with the abnormal temporal association constraint formed by the temporal convolutional network, the sleep monitoring cycle time axis is divided into continuous time periods and the number of abnormal occurrences in each period is counted. The dominant abnormality type is selected and the consistency of adjacent periods is compared to establish nighttime ventilation status segmentation. S4: Based on the nighttime ventilation state segmentation, the state segments are arranged in chronological order. The node propagation weights formed by the graph neural network are used to filter the transition segments with increasing abnormal levels. The corresponding ventilation drive distribution markers and blood oxygen and nitric oxide change segments are checked simultaneously, and the risk change trajectory is output. S5: Based on the nighttime ventilation status segmentation and the risk change trajectory, align the high-risk status segment with the risk turning point, count the number of consecutive abnormal segments and calculate the time interval between segments, match the trigger segment with the observation segment, and obtain the anti-snoring intervention instruction.

Images