Abnormal breathing recognition method and system based on AI

Abstract

The present invention relates to the field of pattern recognition technology, specifically to an AI-based respiratory abnormality recognition method and system, comprising the following steps: based on respiratory frequency monitoring data and respiratory depth analysis data, extracting time series analysis trend values, local amplitude change values ​​and periodic change values ​​in each group of data, comparing the time series trend values ​​with the local amplitude change values, obtaining a respiratory rhythm feature parameter set, and classifying and partitioning the periodic change values ​​to generate a preliminary respiratory rhythm partition result. In the present invention, by carefully analyzing the respiratory frequency and depth, extracting trend values, local amplitude and periodic changes, deeply analyzing the respiratory rhythm, using partitioning and classification methods to accurately capture respiratory dynamics, cross-matching fluctuation amplitudes and trend rates, improving comprehensiveness and recognition accuracy in respiratory monitoring, associating abnormal interval offset values ​​with respiratory frequency and depth differences, scientifically marking abnormal points, and effectively improving the detection efficiency and accuracy of respiratory abnormalities.

Description

Technical Field

[0001] The present invention relates to the field of pattern recognition technology, and in particular to an AI-based breathing abnormality recognition method and system. Background Art

[0002] The field of pattern recognition technology is an interdisciplinary subject involving computer science, statistics and artificial intelligence. It aims to identify and interpret patterns and regularities in data through algorithms and statistical models. This technical field mainly focuses on the automated data interpretation process, analyzing input data (such as images, videos, texts, sounds, etc.) through the design and development of algorithms, and classifying, predicting or making decisions based on the features in the data. In practical applications, pattern recognition technology is widely used in image recognition, speech recognition, biometric recognition and other fields, and improves the accuracy and efficiency of decision-making by extracting and utilizing data features.

[0003] Among them, the abnormal breathing recognition method refers to the use of pattern recognition technology to monitor and identify abnormalities in breathing patterns. This method uses sound, motion sensors or biosensor data as input and analyzes the regularity, depth and frequency of breathing through specific algorithms. The purpose is to monitor the individual's breathing status in real time and detect potential breathing abnormalities in time so that corresponding non-therapeutic response measures can be taken. For example, in sports training, sleep monitoring or elderly care, the guardian's attention can be prompted by reminders or alarm systems. The method focuses on providing auxiliary information to enhance the safety and responsiveness of individuals or monitoring environments.

[0004] Existing technologies mainly rely on relatively simple sensor data analysis, focusing only on basic respiratory rate and depth monitoring, and failing to delve into the dynamic changes in respiratory rhythm and its subtle fluctuations. This method shows obvious limitations when dealing with complex breathing patterns, such as the respiratory pattern recognition of patients with sleep apnea or chronic obstructive pulmonary disease (COPD). The lack of analysis of subtle changes in respiratory rhythm makes early warnings insufficient in sensitivity and specificity, resulting in the inability to timely detect potential serious respiratory abnormalities in practical applications. Existing technologies fail to effectively utilize advanced data processing technology and complex algorithm optimization, making the identification of respiratory abnormalities too dependent on surface data and failing to deeply explore potential health risks. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an AI-based breathing abnormality recognition method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an AI-based breathing abnormality recognition method, comprising the following steps:

[0007] S1: Based on the respiratory rate monitoring data and respiratory depth analysis data, extract the time series analysis trend value, local amplitude change value and period change value in each set of data, compare the time series trend value with the local amplitude change value, obtain the respiratory rhythm characteristic parameter set, and classify and partition the period change value to generate the preliminary partition result of the respiratory rhythm;

[0008] S2: based on the preliminary partitioning result of the respiratory rhythm, extract the trend value, fluctuation amplitude and difference value between adjacent partitions in each partition, sort the trend value and fluctuation amplitude by difference value, obtain the dynamic change result of the respiratory rhythm, and cross-match the fluctuation amplitude distribution with the trend change rate to generate a respiratory rhythm fluctuation feature matrix;

[0009] S3: Based on the respiratory rhythm fluctuation feature matrix, analyzing the offset values ​​of the frequency change and the depth change within the time period, synchronously comparing the offset values ​​with adjacent features, and generating a respiratory rhythm offset matching result;

[0010] S4: Based on the respiratory rhythm offset matching result, extract the interval index of the abnormal offset value, analyze the cumulative value of the respiratory frequency and depth difference change of each index point, mark the index point with a cumulative value greater than the threshold as an abnormal point, and sort the time series index distribution to identify respiratory abnormalities.

[0011] As a further solution of the present invention, the steps of acquiring the respiratory rhythm characteristic parameter set are specifically:

[0012] S111: Based on the respiratory rate monitoring data and the respiratory depth analysis data, the data is smoothed by Gaussian filtering technology, trend components are extracted from the smoothed data, the trend slope at each time point is analyzed, and a time series trend slope is generated;

[0013] S112: applying a peak detection algorithm to the respiratory depth analysis data to identify local amplitude changes according to the time series trend slope, identifying amplitude differences for each breath, and smoothing the amplitude differences to generate smoothed local amplitude change values;

[0014] S113: Extract local eigenvalues ​​and analyze local change trends based on the smoothed local amplitude change values, and apply the formula: ;

[0015] Generate a respiratory rhythm feature parameter set, where represents the trend slope of the time series, Represents the local amplitude change value after smoothing, and Represent the adjustment weights of trend slope and amplitude change values, Represents the adjusted respiratory rhythm characteristic parameters.

[0016] As a further solution of the present invention, the steps for obtaining the preliminary partitioning results of the respiratory rhythm are specifically as follows:

[0017] S121: extracting respiratory rhythm data from the respiratory rhythm characteristic parameter set, including the start and end time points of each breath, marking the extracted data, recording the depth and frequency of each breath, and constructing a respiratory characteristic data set;

[0018] S122: Analyze the breathing interval and duration according to the breathing feature data set, set a threshold, divide the breathing cycle into three types: fast, regular and slow according to the time length, classify the breathing events using the threshold, and generate a breathing periodicity analysis table;

[0019] S123: extracting classified data from the respiratory periodicity analysis table, comparing each category of data, determining key respiratory rhythm patterns in the classification, and subdividing the respiratory patterns into normal and abnormal categories based on the frequency and characteristics of the key respiratory rhythm patterns, thereby forming a preliminary respiratory rhythm zoning result.

[0020] As a further solution of the present invention, the step of obtaining the dynamic change result of the respiratory rhythm is specifically:

[0021] S211: Based on the preliminary respiratory rhythm zoning result, analyzing the respiratory rhythm change rate and standard deviation in each time window to obtain trend and fluctuation characteristic data;

[0022] S212: Extract key parameters of each partition from the trend and fluctuation characteristic data, analyze the normalized fluctuation amplitude, and use the formula: ;

[0023] Calculate the difference value of each partition and obtain the difference value feature data, where: represents the difference value of the i-th partition, is the average rate of change of the ith partition, is the average rate of change of the partition, is the standard deviation of the ith partition, and To adjust the weight parameters of fluctuation range and average rate of change;

[0024] S213: Utilizing the difference value feature data, combined with the trend difference values ​​between adjacent partitions, sorting them according to the size of the difference, and generating a result of dynamic change of respiratory rhythm.

[0025] As a further solution of the present invention, the steps of acquiring the respiratory rhythm fluctuation characteristic matrix are specifically as follows:

[0026] S221: Based on the dynamic change result of the respiratory rhythm, the amplitude value of the respiratory fluctuation is measured at each time point, the amplitude value is classified according to a preset numerical range, and the classified fluctuation data is arranged in chronological order to obtain a respiratory fluctuation amplitude interval matrix;

[0027] S222: Based on the respiratory fluctuation amplitude interval matrix, count the occurrence frequency of each fluctuation interval, arrange the frequency data in chronological order, analyze the frequency changes between adjacent time points, and obtain the fluctuation trend change rate matrix by analyzing the change rate of the frequency at each time point;

[0028] S223: Based on the fluctuation trend change rate matrix, cross-match the fluctuation amplitude interval and change rate of the time node, integrate the cross-matched data, and organize the current data structure to obtain the respiratory rhythm fluctuation feature matrix.

[0029] As a further solution of the present invention, the steps of obtaining the respiratory rhythm deviation matching result are specifically as follows:

[0030] S311: capturing the respiratory frequency and depth in the time series data according to the respiratory rhythm fluctuation characteristic matrix, calculating the real-time values ​​of the respiratory frequency and depth for each time point, and generating a respiratory characteristic matrix within the time period;

[0031] S312: Analyze the frequency and depth deviation of the baseline corresponding to each time point through the respiratory characteristic matrix in the time period, and use the formula: ;

[0032] Determine the respiratory change at a time point and generate an offset value for the time point, where Indicates the offset value of a time point. and is the weighting factor for the sensitivity to frequency and depth offsets, is the respiratory rate at the current time point, is the average respiratory rate, is the current depth, is the average breathing depth;

[0033] S313: Using the offset value of the time point, the offset values ​​of adjacent time periods are compared, the offset value is judged by a set threshold, the abnormality of the breathing pattern is identified by comparing the changes of consecutive time points, and a respiratory rhythm offset matching result is generated.

[0034] As a further solution of the present invention, the acquisition step for identifying breathing abnormalities is specifically:

[0035] S411: using the respiratory rhythm offset matching result, screening the interval index with abnormal offset value, analyzing the offset value of each index point and the set abnormal threshold, verifying the potential abnormal point in the index, and generating the abnormal interval index;

[0036] S412: For the abnormal interval index, calculate the cumulative value of the respiratory frequency and depth difference at the corresponding time point, by accumulating the absolute value of the frequency difference and the depth difference at each index point, and applying the formula: ;

[0037] Generates a list of cumulative difference values ​​for an index where is the index point The cumulative difference value of and The time points The frequency and depth values ​​of and is the corresponding average value;

[0038] S413: Filter the index points whose cumulative values ​​exceed the set threshold value from the cumulative difference value list of the index, mark the index points as abnormal points, sort the time series index distribution of the index points, integrate the time series data of the marked abnormal points, and identify respiratory abnormalities.

[0039] An AI-based breathing abnormality recognition system, the AI-based breathing abnormality recognition system is used to execute the above-mentioned AI-based breathing abnormality recognition method, the system comprising:

[0040] The data analysis startup module uses time series decomposition to extract the trend, amplitude change and periodicity indicators of each data point based on the respiratory rate monitoring data and respiratory depth analysis data, and analyzes the fluctuation characteristics from the data points to obtain the basic rhythm parameter set;

[0041] The rhythm analysis module uses the basic rhythm parameter set to analyze the data at the differentiated time points, identifies the partition characteristics of the respiratory rhythm by calculating the periodic fluctuations in the region, and marks the key change regions to obtain the partition characteristic division results;

[0042] The feature integration module summarizes the trend and fluctuation data of each area based on the partition feature division results, sorts and fuses the data, and analyzes the relationship between the features to construct a dynamic rhythm matrix;

[0043] The deviation analysis module analyzes the frequency and deviation in a time period based on the dynamic rhythm matrix, compares the characteristic deviations of adjacent time periods, identifies key deviation events and records them, and generates deviation analysis results;

[0044] The anomaly detection module uses the offset analysis results to check the time interval of the abnormal offset, analyze the data changes in the interval, identify and mark the abnormal points that exceed the threshold, organize the abnormal distribution, and identify the breathing abnormality.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, the method of real-time monitoring of respiratory status is used to improve the early warning ability of respiratory abnormalities. Through the careful analysis of respiratory frequency and depth, key features such as trend values, local amplitudes and periodic changes in time series are extracted to achieve in-depth analysis of respiratory rhythm. By comparing trend values ​​and local amplitude changes, supplemented by partitioning and classification methods, the dynamic changes of respiratory rhythm can be captured more accurately. In the process of generating a respiratory rhythm fluctuation feature matrix, the monitoring of respiratory status is made more comprehensive by cross-matching the fluctuation amplitude distribution and the trend change rate. The detailed analysis of the offset value further improves the recognition accuracy, and the abnormal interval of the offset value is associated with the difference in respiratory frequency and depth, providing a more scientific basis for the method of marking abnormal points, thereby effectively improving the recognition efficiency and accuracy of respiratory abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0048] Figure 2 is a flow chart of the respiratory rhythm characteristic parameter set in the present invention;

[0049] Figure 3 A flow chart showing the preliminary partitioning results of respiratory rhythm in the present invention;

[0050] Figure 4 It is a flow chart of the dynamic change results of respiratory rhythm in the present invention;

[0051] Figure 5 It is a flow chart of the respiratory rhythm fluctuation characteristic matrix in the present invention;

[0052] Figure 6 A flow chart showing the respiratory rhythm deviation matching results in the present invention;

[0053] Figure 7 This is a flow chart for identifying respiratory abnormalities in the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0056] Example

[0057] See also Figure 1 The present invention provides a technical solution: an AI-based breathing abnormality recognition method, comprising the following steps:

[0058] S1: Based on the respiratory rate monitoring data and respiratory depth analysis data, extract the time series analysis trend value, local amplitude change value and period change value in each set of data, compare the time series trend value with the local amplitude change value, obtain the respiratory rhythm characteristic parameter set, and classify and partition the period change value to generate the preliminary partition result of the respiratory rhythm;

[0059] S2: Based on the preliminary partitioning results of respiratory rhythm, the trend value, fluctuation amplitude and difference value between adjacent partitions in each partition are extracted, the trend value and fluctuation amplitude are sorted by difference value, the dynamic change results of respiratory rhythm are obtained, and the fluctuation amplitude distribution and trend change rate are cross-matched to generate the respiratory rhythm fluctuation feature matrix;

[0060] S3: Based on the respiratory rhythm fluctuation feature matrix, analyze the offset values ​​of frequency change and depth change within the time period, compare the offset values ​​with adjacent features synchronously, and generate respiratory rhythm offset matching results;

[0061] S4: Based on the respiratory rhythm offset matching results, extract the interval index of the abnormal offset value, analyze the cumulative value of the respiratory frequency and depth difference changes of each index point, mark the index points with cumulative values ​​greater than the threshold as abnormal points, and organize the time series index distribution to identify respiratory abnormalities.

[0062] The respiratory rhythm feature parameter set includes time identifier, amplitude index, and periodicity measurement. The preliminary partitioning results of the respiratory rhythm include trend partitioning, interval statistical data, and partition effect marking. The dynamic change results of the respiratory rhythm include trend deviation, fluctuation frequency, and dynamic change rate. The respiratory rhythm fluctuation feature matrix includes feature correlation matrix, stability score, and matching efficiency. The respiratory rhythm offset matching results include offset measurement value, feature synchronization index, and offset comparison analysis. The identification of respiratory abnormalities includes abnormality identification index, cumulative difference evaluation, and timing anomaly label.

[0063] See also Figure 2 , the specific steps for obtaining the respiratory rhythm characteristic parameter set are:

[0064] S111: Based on the respiratory rate monitoring data and the respiratory depth analysis data, the data is smoothed by Gaussian filtering technology, trend components are extracted from the smoothed data, the trend slope at each time point is analyzed, and a time series trend slope is generated;

[0065] In the processing of respiratory rate monitoring data, Gaussian filtering technology is first used to reduce the noise of the original signal. The Gaussian filter parameters used in this process are determined by the signal's bandwidth and expected noise level. Then, the filtered data is trend analyzed by a sliding window. The size of the sliding window is set according to the time resolution of the data and the rate of change of the respiratory frequency to ensure that changes in the respiratory trend can be effectively captured. The linear trend of the data in each window is calculated to obtain the time series trend slope. The slope reflects the change of the respiratory frequency over time, providing basic data for subsequent analysis.

[0066] S112: applying a peak detection algorithm to the respiratory depth analysis data to identify local amplitude changes according to the time series trend slope, identifying the amplitude difference for each breath, and smoothing the amplitude difference to generate a smoothed local amplitude change value;

[0067] The trend slope of the time series is further analyzed, and the peak detection algorithm is used to process the respiratory depth analysis data. This algorithm needs to adjust the detection threshold according to the physiological characteristics of breathing. The threshold setting is based on the previous monitoring of the breathing cycle of a small sample. The algorithm identifies the highest and lowest points of each breath, thereby calculating the maximum and minimum amplitude difference of the breath. The difference is smoothed by the weighted moving average method. The smoothing parameter is set according to the volatility of the data and the expected smoothing effect to reduce the impact of accidental errors and generate a smoothed local amplitude change value.

[0068] S113: According to the smoothed local amplitude change value, extract the local eigenvalue and analyze the local change trend, and apply the formula: ;

[0069] Generate a respiratory rhythm feature parameter set, where represents the trend slope of the time series, Represents the local amplitude change value after smoothing, and Represent the adjustment weights of trend slope and amplitude change values, represents the adjusted respiratory rhythm characteristic parameters;

[0070] The usefulness of the formula is that it allows us to adjust the weight parameter and To balance the time series trend slope And the smoothed local amplitude change value The final result The influence of makes the results closer to the actual physiological state;

[0071] Set the time series trend slope is 0.3, the local amplitude change value after smoothing is 0.7, weight and If they are 0.5 respectively, the calculation process is: ;

[0072] The results show that the adjusted respiratory rhythm characteristic parameters It is 0.5, which is the optimal solution under the specified parameter settings and weights. It represents a respiratory state that balances the time series trend and local amplitude changes, and effectively reflects the patient's respiratory rhythm characteristics.

[0073] See also Figure 3 ,The specific steps for obtaining the preliminary partition results of respiratory rhythm are:

[0074] S121: extracting respiratory rhythm data from the respiratory rhythm feature parameter set, including the start and end time points of each breath, marking the extracted data, recording the depth and frequency of each breath, and constructing a respiratory feature data set;

[0075] Respiratory rhythm data is extracted from the respiratory rhythm characteristic parameter set. The patient's respiratory activity data within 24 hours is received and recorded through sensors. Each data point records the specific timestamp, breathing depth and respiratory frequency. The data processing part converts the information into a digital format for subsequent processing and analysis. The respiratory activity data is subjected to time series analysis to extract each complete respiratory cycle, that is, the interval from the start of one breath to the start of the next breath. The data is marked and stored in categories to lay the foundation for the next detailed analysis. Environmental factors such as temperature and humidity are also recorded. The data will be analyzed together with the respiratory data to explore the potential impact of external environmental changes on breathing patterns and construct a respiratory feature data set.

[0076] S122: Analyze the breathing interval and duration according to the breathing feature data set, set a threshold, divide the breathing cycle into three types: fast, regular and slow according to the time length, classify the breathing events using the threshold, and generate a breathing periodicity analysis table;

[0077] According to the respiratory characteristic data set, during the data analysis process, the average duration and standard deviation of all respiratory cycles are first calculated, and three thresholds are set based on this information: the upper limit of fast breathing, the normal range of regular breathing, and the lower limit of slow breathing. By comparing with the threshold, each respiratory cycle is classified as fast, regular, or slow. For example, fast breathing is defined as a duration lower than the average time of regular breathing minus a certain standard deviation, while slow breathing is higher than the average time plus the standard deviation. Each type of respiratory data is recorded in the respiratory periodicity analysis table, which lists in detail the specific number of each type of breathing and the percentage of the total respiratory cycle, providing a quantitative basis for further analysis and generating a respiratory periodicity analysis table.

[0078] S123: extracting classified data from the respiratory periodicity analysis table, comparing each type of data, determining a key respiratory rhythm pattern in the classification, and subdividing the respiratory pattern into normal and abnormal categories according to the frequency and characteristics of the key respiratory rhythm pattern, thereby forming a preliminary respiratory rhythm zoning result;

[0079] Extract classified data from the respiratory periodicity analysis table, process the respiratory periodicity analysis table obtained in the previous step through statistical analysis software, identify breathing patterns with abnormally high or low frequencies, and further qualitatively analyze the patterns. Slight changes in regular breathing indicate potential physiological changes. Combined with the patient's historical health records and real-time monitoring data, it is possible to identify which breathing patterns are normal physiological phenomena and which indicate health problems. Respiratory rhythm patterns are further subdivided into normal or abnormal categories. Each category is based on specific physiological and environmental parameters to ensure the accuracy and reliability of the analysis results, forming a preliminary zoning result for respiratory rhythm.

[0080] See also Figure 4 , the specific steps for obtaining the dynamic change results of respiratory rhythm are:

[0081] S211: Based on the preliminary respiratory rhythm zoning results, analyzing the respiratory rhythm change rate and standard deviation in each time window to obtain trend and fluctuation characteristic data;

[0082] Based on the continuously monitored respiratory data, the respiratory rhythm change rate in each time window is first calculated. The change rate is obtained by adding and averaging the number of breaths per minute. For example, during continuous monitoring, the number of breaths recorded in each five-minute window is accumulated and divided by the monitoring time to calculate the average respiratory frequency. For the calculation of the standard deviation, the fluctuation of the respiratory frequency in each time window is taken into account, that is, the sum of the squares of the difference between each respiratory frequency and the average value is divided by the number of measurements, and the square root is taken. The standard deviation obtained in this way reflects the instability of the respiratory frequency. These two parameters together constitute the trend and fluctuation characteristic data. The data is obtained by detailed analysis of the statistical characteristics of the respiratory pattern in different time periods. The data provides the necessary basic information for the subsequent steps and ensures the accuracy and relevance of data processing. This lays the foundation for further data analysis and obtains trend and fluctuation characteristic data. The data will be used for further processing and analysis in the next step.

[0083] S212: Extract key parameters of each partition from trend and fluctuation characteristic data, analyze normalized fluctuation amplitude, and use formula: ;

[0084] Calculate the difference value of each partition and obtain the difference value feature data, where: represents the difference value of the i-th partition, is the average rate of change of the ith partition, is the average rate of change of the partition, is the standard deviation of the ith partition, and To adjust the weight parameters of fluctuation range and average rate of change;

[0085] The benefit of the formula is that by considering the weight adjustment of the average rate of change and the standard deviation, it can more flexibly reflect the actual differences between the partitions and adapt to the needs of different scenarios;

[0086] Set the average change rate of a partition during the monitoring period is 0.2, the overall average rate of change of all partitions is 0.15, the standard deviation of this partition is 0.05, the weight parameter and Set them to 1.5 and 0.5 respectively, and substitute them into the formula:

[0087] ;

[0088] . ;

[0089] ;

[0090] ;

[0091] The results show that according to the set weights, the partition presents a smaller difference value under the comprehensive consideration of standard deviation and rate of change, reflecting its relative stability. The design of this comprehensive weight allows a more accurate assessment of the magnitude of change in different partitions, thus providing an important quantitative basis for further data analysis and processing.

[0092] S213: using the difference value feature data, combined with the trend difference values ​​between adjacent partitions, sorting them according to the difference size, and generating a dynamic change result of the respiratory rhythm;

[0093] Using the difference value feature data, the data has been calculated by weighted processing in the previous step by combining the standard deviation and average change rate of each partition. Now, combined with the trend difference value between adjacent partitions, sorting is performed by calculating the difference size. The sorting operation compares the The sorted data can intuitively show which partitions have more significant changes in respiratory rhythm and which are more stable. Such analysis helps medical workers or researchers quickly identify data anomalies or special patterns that need attention, and can identify those partitions whose respiratory patterns show abnormalities or significant changes within a specific time period, providing data support for further clinical or research applications and generating dynamic changes in respiratory rhythm. The results will be used to evaluate the overall stability and variability of the respiratory pattern, providing a scientific basis for subsequent treatment decisions.

[0094] See also Figure 5 , the specific steps for obtaining the respiratory rhythm fluctuation feature matrix are:

[0095] S221: Based on the dynamic change result of respiratory rhythm, the amplitude value of respiratory fluctuation is measured at each time point, the amplitude value is classified according to a preset numerical range, and the classified fluctuation data is arranged in chronological order to obtain a respiratory fluctuation amplitude interval matrix;

[0096] Based on the results of dynamic changes in respiratory rhythm, the amplitude value of respiratory fluctuations is measured at each time point. When recording the respiratory amplitude value at each time point, noise interference needs to be eliminated during the continuous acquisition process, and the respiratory signal is divided into multiple independent segments. For each segment, data is sorted according to the amplitude value partition, and the amplitude range of each interval is used as the boundary to count the number of fluctuation data points in different intervals. After the amplitude interval classification, all fluctuation data are sorted in chronological order to ensure the continuity of the classified data, and a respiratory fluctuation amplitude interval matrix is ​​generated, in which each column of the matrix represents the amplitude distribution in a specific time period, and each row of the matrix is ​​the classification result of different amplitude intervals, forming a two-dimensional matrix data structure of the time series, which is convenient for subsequent analysis.

[0097] S222: Based on the respiratory fluctuation amplitude interval matrix, count the occurrence frequency of each fluctuation interval, arrange the frequency data in chronological order, analyze the frequency changes between adjacent time points, and obtain the fluctuation trend change rate matrix by analyzing the change rate of the frequency at each time point;

[0098] Based on the respiratory fluctuation amplitude interval matrix, the frequency of occurrence of each fluctuation interval is counted, and the frequency data is arranged in chronological order to analyze the frequency changes between adjacent time points. In the process of frequency change analysis, it is necessary to calculate the fluctuation frequency difference between two adjacent time points point by point and store it as the current variable matrix. When calculating the frequency change rate, it is necessary to verify the continuity of the frequency change results to avoid fluctuation trend errors caused by accidental abnormal points. At each time node, the change rate between adjacent time points is extracted to establish a fluctuation trend change rate matrix. Each element of the matrix represents the fluctuation trend rate in a specific interval on the time axis. By distinguishing the positive and negative values ​​of the rate value, the growth or attenuation trend of the respiratory fluctuation can be intuitively displayed.

[0099] S223: Based on the fluctuation trend change rate matrix, cross-match the fluctuation amplitude interval and change rate of the time node, integrate the cross-matched data, and organize the current data structure to obtain the respiratory rhythm fluctuation feature matrix;

[0100] Based on the fluctuation trend change rate matrix, the fluctuation amplitude interval and change rate of the time node are cross-matched. During the matching process, the matching points that meet both the amplitude and rate conditions are screened according to the distribution characteristics of the amplitude interval and the range of the rate value in the rate matrix. The cross-matching results are used as the basis to integrate the amplitude data and rate data of the corresponding time nodes into the current data structure. Through logical judgment of the matching points, the matching items that do not meet the conditions are eliminated to ensure the accuracy and consistency of the integrated data. The integrated matching data are organized into the current feature matrix to obtain the respiratory rhythm fluctuation feature matrix, providing an accurate data basis for the analysis of dynamic changes in breathing.

[0101] See also Figure 6 , the specific steps for obtaining the respiratory rhythm deviation matching result are:

[0102] S311: capturing the respiratory frequency and depth in the time series data according to the respiratory rhythm fluctuation characteristic matrix, calculating the real-time values ​​of the respiratory frequency and depth for each time point, and generating a respiratory characteristic matrix within the time period;

[0103] The respiratory rate and depth data collected from each monitoring point are processed. Each data point reflects the respiratory status within a specific time. The data first goes through basic data cleaning steps to remove outliers and noise to ensure the reliability and accuracy of the data. The cleaned data is then standardized to eliminate deviations caused by different devices or environmental factors. The standardized data is aggregated through an algorithm and summarized into a respiratory characteristic matrix within the time period. This matrix records the average respiratory rate and depth within each time period in detail, providing basic data for subsequent offset analysis and generating a respiratory characteristic matrix within the time period.

[0104] S312: Analyze the frequency and depth deviation of the baseline corresponding to each time point through the respiratory characteristic matrix within the time period, and use the formula: ;

[0105] Determine the respiratory change at a time point and generate an offset value for the time point, where Indicates the offset value of a time point. and is the weighting factor for the sensitivity to frequency and depth offsets, is the respiratory rate at the current time point, is the average respiratory rate, is the current depth, is the average breathing depth;

[0106] The benefit of the formula is that, through weighted processing, the formula can comprehensively consider the changes in frequency and depth, more accurately measure the severity of the deviation, and provide a quantitative method for abnormal detection of breathing patterns;

[0107] Set the actual breathing rate at a certain point in time The baseline frequency is 16 times / minute. 14 times / minute, actual depth 500ml, baseline depth is 450ml, weight coefficient is 0.7, is 0.3;

[0108] Substitute numerical calculations :

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] The results show that after taking the weight factor into account, the respiratory state at the current time point has a deviation of 15.06 compared with the baseline, which reflects a significant change from the baseline state and can be regarded as a potential respiratory abnormality.

[0115] S313: using the offset value of the time point, comparing the offset values ​​of adjacent time periods, judging the offset value by a set threshold, identifying abnormalities in the breathing pattern by comparing the changes at consecutive time points, and generating a respiratory rhythm offset matching result;

[0116] Through the offset value of the time point, adjacent time periods are compared, and the pattern and trend of the offset are analyzed, including the comparison of the offset of consecutive time points, and the identification of continuous or frequent large offsets. This analysis helps to identify irregular breathing or pathological states. By setting thresholds, it is determined which offset patterns are significant and the significant offset patterns are marked. The marking is based on the comparison of the offset with the average offset. The offset exceeding the threshold is considered abnormal, which provides a basis for further clinical evaluation and generates respiratory rhythm offset matching results, providing a comprehensive view showing the dynamic changes of the patient's respiratory state over time.

[0117] See also Figure 7 , the specific steps for identifying abnormal breathing are:

[0118] S411: using the respiratory rhythm offset matching result, screening the interval index with abnormal offset value, analyzing the offset value of each index point and the set abnormal threshold, verifying the potential abnormal points in the index, and generating the abnormal interval index;

[0119] From the previous respiratory rhythm deviation matching results, the interval index points marked as abnormal are screened out. The index points represent the moments when the deviation values ​​exceed the normal range during the observation period. Data verification is performed to ensure that the data collected from the monitoring equipment are accurate. By comparing the frequency and depth data of each time point with the average value of the overall monitoring period, the deviation degree of each index point is calculated. This can not only identify the specific abnormal time points, but also accurately locate the sudden changes in the respiratory pattern. The data collation at this stage is to prepare for more accurate abnormal pattern analysis, including generating abnormal interval indexes through aggregating data to provide reliable basic data for subsequent steps.

[0120] S412: For the abnormal interval index, calculate the cumulative value of the respiratory frequency and depth difference at the corresponding time point by accumulating the absolute value of the frequency difference and depth difference at each index point, and apply the formula: ;

[0121] Generates a list of cumulative difference values ​​for an index where is the index point The cumulative difference value of and The time points The frequency and depth values ​​of and is the corresponding average value;

[0122] The benefit of the formula is that, by accumulating the frequency and depth deviations at each time point, it provides a quantitative method to assess the stability of the respiratory rhythm throughout the monitoring period, which can allow medical personnel to quickly identify potential respiratory abnormalities;

[0123] Considering the time period The cumulative deviation value of all points in and At the time point The observed frequency and depth values ​​of and is the average value. For the sake of demonstration, if at a certain index point , the observed frequency BPM and Depth ml, compared to the average frequency Beats / minute and average depth ml, the difference is calculated as:

[0124] ;

[0125] ;

[0126] ;

[0127] Set the abnormality detected within ten minutes , calculate the cumulative deviation value: ;

[0128] This indicates that the index point The cumulative deviation value for the ten minutes under consideration was 540, indicating that the changes in respiratory rate and depth during this time period were significantly higher than the average, thus validating the outliers. This analysis provides a clear quantitative method for identifying anomalies, which helps in further diagnosis and treatment planning.

[0129] S413: Filtering index points whose cumulative values ​​exceed a set threshold from the cumulative difference value list of the indexes, marking the index points as abnormal points, sorting out the time series index distribution of the index points, integrating the time series data of the marked abnormal points, and identifying respiratory abnormalities;

[0130] Using the cumulative difference value list of the index, we further screen the index points whose cumulative values ​​exceed the set threshold (for example, 1000) and mark them as abnormal points. This process not only includes recording the timestamps of all abnormal points, but also requires a detailed analysis of the distribution and frequency of the abnormal points throughout the monitoring cycle. This sorting method helps to identify the occurrence patterns and severity of health problems, especially for patients with chronic diseases who require long-term monitoring. For the identification of respiratory abnormalities, the time distribution of abnormal respiratory status will be described in detail, providing medical personnel with key information for further analysis and evaluation.

[0131] The AI-based breathing abnormality recognition system is used to execute the above-mentioned AI-based breathing abnormality recognition method, and the system includes:

[0132] The data analysis startup module uses time series decomposition to extract the trend, amplitude change and periodicity indicators of each data point based on the respiratory rate monitoring data and respiratory depth analysis data, and analyzes the fluctuation characteristics from the data points to obtain the basic rhythm parameter set;

[0133] The rhythm analysis module uses the basic rhythm parameter set to analyze the data at different time points, identifies the partition characteristics of the respiratory rhythm by calculating the periodic fluctuations in the region, and marks the key change areas to obtain the partition feature division results;

[0134] The feature integration module summarizes the trend and fluctuation data of each area based on the partition feature division results, sorts and fuses the data, and analyzes the relationship between the features to construct a dynamic rhythm matrix;

[0135] The deviation analysis module analyzes the frequency and deviation in a time period based on the dynamic rhythm matrix, compares the characteristic deviations of adjacent time periods, identifies key deviation events and records them, and generates deviation analysis results;

[0136] The anomaly detection module uses the results of the offset analysis to check the time interval of the abnormal offset, analyze the data changes in the interval, identify and mark the abnormal points that exceed the threshold, organize the abnormal distribution, and identify breathing abnormalities.

[0137] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The AI-based breathing abnormality recognition method is characterized by: The following steps are involved: Based on the respiratory rate monitoring data and respiratory depth analysis data, the time series analysis trend value, local amplitude change value and period change value in each set of data are extracted, the time series trend value is compared with the local amplitude change value, the respiratory rhythm characteristic parameter set is obtained, and the period change value is classified and partitioned to generate the preliminary respiratory rhythm partition result; Based on the preliminary partitioning results of the respiratory rhythm, the trend value, the fluctuation amplitude and the difference value between the adjacent partitions in each partition are extracted, the trend value and the fluctuation amplitude are sorted by the difference value, the dynamic change result of the respiratory rhythm is obtained, and the fluctuation amplitude distribution and the trend change rate are cross-matched to generate a respiratory rhythm fluctuation feature matrix; Based on the respiratory rhythm fluctuation feature matrix, analyzing the offset values ​​of frequency change and depth change within a time period, synchronously comparing the offset values ​​with adjacent features, and generating a respiratory rhythm offset matching result; Based on the respiratory rhythm offset matching result, extract the interval index of the abnormal offset value, analyze the cumulative value of the respiratory frequency and depth difference change of each index point, mark the index point with a cumulative value greater than a threshold as an abnormal point, and perform time series index distribution sorting to identify respiratory abnormalities; The steps for obtaining the dynamic change result of the respiratory rhythm are specifically as follows: Based on the preliminary respiratory rhythm zoning results, analyzing the respiratory rhythm change rate and standard deviation in each time window to obtain trend and fluctuation characteristic data; Extract the key parameters of each partition from the trend and fluctuation characteristic data, analyze the normalized fluctuation amplitude, and use the formula: ； Calculate the difference value of each partition and obtain the difference value feature data, where: represents the difference value of the i-th partition, is the average rate of change of the ith partition, is the average rate of change of the partition, is the standard deviation of the ith partition, and To adjust the weight parameters of fluctuation range and average rate of change; The difference value feature data is used in combination with the trend difference values ​​between adjacent partitions, and the partitions are sorted according to the size of the difference to generate a dynamic change result of the respiratory rhythm.

2. The AI-based breathing abnormality recognition method according to claim 1, characterized in that: The steps of acquiring the respiratory rhythm characteristic parameter set are specifically as follows: Based on the respiratory rate monitoring data and respiratory depth analysis data, the data is smoothed by Gaussian filtering technology, the trend component is extracted from the smoothed data, the trend slope at each time point is analyzed, and the time series trend slope is generated; According to the time series trend slope, applying a peak detection algorithm to the respiratory depth analysis data to identify local amplitude changes, identifying amplitude differences for each breath, and smoothing the amplitude differences to generate smoothed local amplitude change values; According to the smoothed local amplitude change value, extract the local eigenvalue and analyze the local change trend, and apply the formula: ； Generate a respiratory rhythm feature parameter set, where represents the trend slope of the time series, Represents the local amplitude change value after smoothing, and Represent the adjustment weights of trend slope and amplitude change values, Represents the adjusted respiratory rhythm characteristic parameters.

3. The AI-based breathing abnormality recognition method according to claim 2, characterized in that: The steps for obtaining the preliminary partitioning results of the respiratory rhythm are specifically as follows: Extracting respiratory rhythm data from the respiratory rhythm characteristic parameter set, including the start and end time points of each breath, marking the extracted data, recording the depth and frequency of each breath, and constructing a respiratory characteristic data set; According to the respiratory feature data set, the respiratory interval and duration are analyzed, a threshold is set, and the respiratory cycle is divided into three types: fast, regular and slow by time length, and the threshold is used to classify respiratory events to generate a respiratory periodicity analysis table; The classified data are extracted from the respiratory periodicity analysis table, and each category of data is compared to determine the key respiratory rhythm pattern in the classification. The respiratory pattern is subdivided into normal and abnormal categories based on the frequency and characteristics of the key respiratory rhythm pattern, thus forming a preliminary respiratory rhythm zoning result.

4. The AI-based breathing abnormality recognition method according to claim 3, characterized in that: The steps of obtaining the respiratory rhythm fluctuation characteristic matrix are specifically as follows: Based on the dynamic change result of the respiratory rhythm, the amplitude value of the respiratory fluctuation is measured at each time point, the amplitude value is classified according to a preset numerical range, and the classified fluctuation data is arranged in chronological order to obtain a respiratory fluctuation amplitude interval matrix; Based on the respiratory fluctuation amplitude interval matrix, the occurrence frequency of each fluctuation interval is counted, and the frequency data is arranged in chronological order, and the frequency changes between adjacent time points are analyzed. By analyzing the change rate of the frequency at each time point, a fluctuation trend change rate matrix is ​​obtained; Based on the fluctuation trend change rate matrix, the fluctuation amplitude interval and change rate of the time node are cross-matched, the cross-matched data are integrated, and the current data structure is organized to obtain the respiratory rhythm fluctuation feature matrix.

5. The AI-based breathing abnormality recognition method according to claim 4, characterized in that: The steps for obtaining the respiratory rhythm deviation matching result are specifically as follows: According to the respiratory rhythm fluctuation characteristic matrix, the respiratory frequency and depth in the time series data are captured, the real-time values ​​of the respiratory frequency and depth are calculated for each time point, and the respiratory characteristic matrix within the time period is generated; Through the respiratory characteristic matrix in the time period, the frequency and depth deviation of the baseline corresponding to each time point are analyzed, and the formula is: ； Determine the respiratory change at a time point and generate an offset value for the time point, where Indicates the offset value of a time point. and is the weighting factor for the sensitivity to frequency and depth offsets, is the respiratory rate at the current time point, is the average respiratory rate, is the current depth, is the average breathing depth; The offset values ​​of adjacent time periods are compared using the offset values ​​at the time points, the offset values ​​are judged by a set threshold, the abnormality of the breathing pattern is identified by comparing the changes at consecutive time points, and a respiratory rhythm offset matching result is generated.

6. The AI-based breathing abnormality recognition method according to claim 5, characterized in that: The steps for obtaining the abnormal breathing detection and recognition result are specifically as follows: Using the respiratory rhythm offset matching result, screening the interval index with abnormal offset value, analyzing the offset value of each index point and the set abnormal threshold, verifying the potential abnormal points in the index, and generating the abnormal interval index; For the abnormal interval index, calculate the cumulative value of the respiratory frequency and depth difference at the corresponding time point, by accumulating the absolute value of the frequency difference and depth difference at each index point, and apply the formula: ； Generates a list of cumulative difference values ​​for an index where is the index point The cumulative difference value of and The time points The frequency and depth values ​​of and is the corresponding average value; The index points whose cumulative values ​​exceed the set threshold are screened from the cumulative difference value list of the index, and the index points are marked as abnormal points. The time series index distribution of the index points is sorted, and the time series data of the marked abnormal points are integrated to identify respiratory abnormalities.

7. The AI-based breathing abnormality recognition system is characterized by: According to any one of claims 1 to 6, the method for identifying abnormal breathing based on AI comprises: The data analysis startup module uses time series decomposition to extract the trend, amplitude change and periodicity indicators of each data point based on the respiratory rate monitoring data and respiratory depth analysis data, and analyzes the fluctuation characteristics from the data points to obtain the basic rhythm parameter set; The rhythm analysis module uses the basic rhythm parameter set to analyze the data at the differentiated time points, identifies the partition characteristics of the respiratory rhythm by calculating the periodic fluctuations in the region, and marks the key change regions to obtain the partition characteristic division results; The feature integration module summarizes the trend and fluctuation data of each area based on the partition feature division results, sorts and fuses the data, and analyzes the relationship between the features to construct a dynamic rhythm matrix; The deviation analysis module analyzes the frequency and deviation in a time period based on the dynamic rhythm matrix, compares the characteristic deviations of adjacent time periods, identifies key deviation events and records them, and generates deviation analysis results; The anomaly detection module uses the offset analysis results to check the time interval of the abnormal offset, analyze the data changes in the interval, identify and mark the abnormal points that exceed the threshold, organize the abnormal distribution, and identify the breathing abnormality.

Images