Method for constructing low-oxygen load variability index of sleep breathing disorder

Abstract

The invention discloses a method for constructing a low-oxygen load variability index of sleep breathing disorder. The method comprises the following steps of: 1, preprocessing data; step 2, extracting a specific fragment of generalized apnea; step 3, nonlinear feature extraction; step 4, calculation of statistical features of specific fragments of generalized apnea; and step 5, combined calculation to obtain a low-oxygen load variability index. According to the method, a new evaluation index for quantifying intermittent hypoxia is constructed, the instantaneous change of SPO2 can be quantified, the area under the corresponding oxygen desaturation curve is calculated, and the severity of intermittent hypoxia is evaluated more accurately.

Description

Technical Field

[0001] The present invention relates to evaluating the severity of intermittent hypoxia associated with sleep-disordered breathing, and specifically to a method for constructing a sleep-disordered breathing hypoxia load variability index, and belongs to the field of clinical medicine. Background Art

[0002] Intermittent hypoxia plays an important role in the pathological mechanism of obstructive sleep apnea (OSA) and can lead to a variety of adverse clinical outcomes. The traditional indicator, apnoea-hypopnoea index (AHI), only records the frequency of respiratory events and cannot fully characterize physiological disorders. Therefore, many new indicators have emerged for the assessment of the severity of hypoxemia, such as obstruction severity (ObsSev), hypoxic burden (HB), hypoxia burden index (HBI), sleep breathing impairment index (SBII), and respiratory event oxygen desaturation transient area (REDTA). These indicators cannot accurately calculate the hypoxic burden associated with OSA and cannot reflect the irregularity of blood oxygen saturation signals. Therefore, the present invention constructs a new evaluation index for quantifying intermittent hypoxia, which can quantify the instantaneous change of SP0, and calculate the corresponding area under the oxygen desaturation curve, so as to more accurately evaluate the severity of intermittent hypoxia.

[0003] Since the technical solutions of the above indicators are similar, and the introduction of HB is relatively detailed, this example is used to introduce the existing technical methods.

[0004] Hypoxic Burden (HB): defined as the total area under the respiratory-related desaturation curve divided by the total sleep time, it is a measure of the depth, duration, and frequency of respiratory events. Baseline saturation is defined as the maximum blood oxygen saturation within 100 seconds before the end of the event (apnea or hypopnea). The key to calculating hypoxic burden is to calculate the total area under the desaturation curve. The specific calculation steps of hypoxic burden are as follows:

[0005] 1) Obtaining the mean desaturation curve: The mean desaturation curve for each participant was obtained by superimposing the blood oxygen saturation (SpO) relative to the end of the event. 2 ) signal to determine.

[0006] 2) Determine the relevant event: For each individually identified apnea or hypopnea (collectively referred to as a respiratory event), the pre-event baseline saturation was defined as the maximum SpO within 100 seconds before the end of the event 2 , below which oxygen desaturation events are detected. The area under this baseline value is calculated over the search window obtained from the average desaturation curve.

[0007] 3) Determine the start and end points of desaturation events: If there is a respiratory event related to blood oxygen saturation lower than the baseline saturation, the time when the blood oxygen saturation is lower than the baseline value is taken as the start point, and the time when the blood oxygen saturation returns to the baseline value is taken as the end point; for events in which the blood oxygen saturation does not return to the baseline value, the average oxygen saturation is used as the baseline to determine the start and end points.

[0008] 4) Determine the desaturation area: The area enclosed by the oxygen reduction curve, the reoxygenation curve and the baseline saturation is the area under the desaturation curve. The sum of all individual desaturation areas is the total area under the overnight desaturation curve (e.g. Figure 1 shaded area).

[0009] 5) Calculate the total area under the desaturation curve: The figure under the desaturation curve is regarded as a triangle, whose height and low are the depth of blood oxygen saturation decrease and its duration, respectively, that is, the area under a single desaturation curve = (desaturation depth × desaturation duration) / 2, and the sum of all single desaturation areas is the total area.

[0010] 6) Calculation of HB: HB can be approximated by multiplying AHI by the area under the average desaturation curve: HB = AHI x (desaturation depth x desaturation duration) / 2. Summary of the invention

[0011] The purpose of the present invention is to provide a method for constructing a sleep apnea hypoxia load variability index, to introduce in detail the quantification process of hypoxia load variability, and to construct an index that can more accurately assess the severity of intermittent hypoxia.

[0012] To achieve the above object, the technical solution adopted by the present invention is: a method for constructing a sleep apnea hypoxia load variability index, which comprises the following steps:

[0013] Step 1: Data preprocessing;

[0014] Step 2, generalized apnea-specific fragment extraction;

[0015] Step 3: Non-linear feature extraction:

[0016] Step 4: Calculate the statistical characteristics of the generalized apnea-specific segments:

[0017] Step 5: Combine and calculate to obtain the hypoxic load variability index.

[0018] Furthermore, in step 1, the blood oxygen saturation time series is recorded as x[n], and is preprocessed by using two steps of Delta filtering and linear interpolation:

[0019] First, Delta filtering was used to remove nonphysiological data points caused by pulse oximeter movement or poor contact between the finger and the probe;

[0020] Then, linear interpolation was used to replace all non-physiological values ​​to ensure smooth and continuous time series.

[0021] Furthermore, for a given input sequence x[n], a first-order Delta filter detects outliers by comparing the differences between adjacent sample points; specifically, if Then x[n] is marked as a non-physiological value that needs to be processed.

[0022] Furthermore, the linear interpolation formula is used to find the previous valid value x[i] and the next valid value x[j] of each non-physiological interval. to estimate the missing value x[k].

[0023] Furthermore, in step 2, a window with a length of N hours is used to slide the original respiratory signal at intervals of M minutes, and the respiratory segment with the highest number of pauses for each subject and its corresponding blood oxygen saturation segment are taken for subsequent analysis, wherein the blood oxygen saturation segment is a generalized apnea-specific segment.

[0024] Furthermore, in step 3, the permutation entropy and sample entropy of the blood oxygen saturation segment are calculated to characterize the nonlinear characteristics of the signal.

[0025] Furthermore, the permutation entropy is used to measure the degree of random change of the signal time series. The larger the value of the permutation entropy, the more random the blood oxygen saturation time series is and the more complex the signal is; conversely, the more regular the blood oxygen saturation time series is and the less complex it is.

[0026] Define the blood oxygen saturation time series as u(i), i = 1, 2, ..., N, N is the total length of blood oxygen saturation;

[0027] After phase space reconstruction, the following matrix is ​​obtained:

[0028]

[0029] Where m is the embedding dimension, τ is the delay factor, k = n-(m-1)τ, j = 1, 2, ..., k;

[0030] There are k reconstruction components in the matrix, each of which has m-dimensional embedded elements. Arrange the j-th component x(j), x(j+τ), …, x(j+(m-1)τ) in the matrix in ascending order to obtain:

[0031] x(i+(j 1 -1)τ)≤x(i+(j 2 -1)τ)≤…≤x(i+(j m -1)τ)

[0032] where j 1 , j 2 , …, j m Represents the subscript index value of each element in the reconstruction component; if two or more reconstruction components are equal, then according to j 1 、j 2 Sort in ascending order of size;

[0033] Each reconstructed component can obtain a reconstructed symbol sequence:

[0034] S(l)=(j 1 ,j 2 ,…,j m )

[0035] Where l = 1, 2, ..., k, satisfying k ≤ m!; each reconstruction component is an m-dimensional space, mapped to an m-dimensional symbol sequence, with a total of m! permutations; the probability of calculating each symbol sequence is expressed as P 1 , P 2 ,…,P K , then according to the definition of Shannon entropy, the permutation entropy of the blood oxygen saturation time series is defined as:

[0036]

[0037] Furthermore, the specific steps of calculating the sample entropy are as follows:

[0038] (a) The blood oxygen saturation time series Y 2 (n)=(Y 2 (1),Y 2 (2),Y 2 (3),...,Y 2 (i),...,Y 2 (k)) are composed into a v-dimensional vector in order:

[0039] V i =[Y 2 (i),Y 2 (i+1),…,Y 2 (i+v-1)],i=1,2,…,k

[0040] Among them, k is the blood oxygen saturation time series Y 2 (n) length;

[0041] (b) Calculate the phase vector V for each value of i i and the remaining vector V j The distance, that is:

[0042] d ij =max|Y 2 (i+I)-Y 2 (j+I)|,I=0,1,…,v-1

[0043] Given a threshold r, for each i, for Y 2 (i) and the remaining v-dimensional vector Y 2 (j) (j = 1, 2, ..., kv; j ≠ i) calculate the distance d between ij ;

[0044] (c) Statistics ij The number of numbers less than r and this number N_d ij The ratio of the total distance kv-1 is expressed as Right now:

[0045]

[0046] (d) Ask again The average value is:

[0047]

[0048] (e) The sample entropy of the blood oxygen saturation time series is shown as follows:

[0049]

[0050] Furthermore, in step 4, the average value of the blood oxygen saturation signal related to the respiratory event is taken to determine the search window; and the following is calculated for each search window:

[0051] (1) The area S1 enclosed by the blood oxygen saturation curve and the blood oxygen saturation baseline before the event;

[0052] (2) area under the blood oxygen saturation curve S2;

[0053] (3) The amplitude of change of the blood oxygen saturation curve H1;

[0054] (4) Slope of the line connecting the maximum and minimum values ​​of blood oxygen saturation, K1;

[0055] (5) The ratio of S1 to the search window duration;

[0056] (6) The ratio of S2 to the search window duration;

[0057] The calculation results are arranged in time to obtain a 6-dimensional event-level indicator sequence within the generalized apnea-specific segment, with the time step being the number of apneas;

[0058] For all 6-dimensional event-level indicator sequences within the generalized apnea-specific segments, the following statistics were calculated: (1) maximum value; (2) minimum value; (3) range; (4) standard deviation; (5) mean; (6) skewness; (7) kurtosis; (8) slope;

[0059] Finally, the statistical characteristics of each generalized apnea-specific segment were obtained.

[0060] Furthermore, in step 5, the segment-level entropy features and the statistical features extracted from the event-level indicators are combined to obtain a 50-dimensional generalized apnea-specific hypoxic load fluctuation index that describes the variability of hypoxic load during apnea, namely, the hypoxic load variability index.

[0061] The beneficial effect of the present invention is that the present invention constructs a new evaluation index for quantifying intermittent hypoxia, which can quantify the instantaneous change of SPO2 and calculate the corresponding area under the oxygen desaturation curve, so as to more accurately evaluate the severity of intermittent hypoxia. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the total area under the overnight desaturation curve in the hypoxic load calculation.

[0063] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0064] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] like Figure 2 This embodiment discloses a method for constructing a sleep apnea hypoxia load variability index, which comprises the following steps:

[0066] 1. Data preprocessing:

[0067] The blood oxygen saturation time series is recorded as x[n] and is preprocessed by two steps: Delta filtering and linear interpolation.

[0068] First, Delta filtering is used to remove non-physiological data points caused by pulse oximeter movement or poor contact between the finger and the probe. For a given input sequence x[n], a first-order Delta filter detects outliers by comparing the differences between adjacent sample points. Specifically, if Then x[n] is marked as a non-physiological value that needs to be processed.

[0069] Then, linear interpolation was used to replace all non-physiological values ​​to ensure smooth and continuous time series.

[0070] For all data points marked as non-physiological values, linear interpolation is performed using the nearest valid values ​​before and after them.

[0071] Find the previous valid value x[i] and the next valid value x[j] of each non-physiological interval using the linear interpolation formula to estimate the missing value x[k].

[0072] By combining the above-mentioned Delta filtering and linear interpolation method, the noise in the blood oxygen saturation data can be effectively removed, and the missing values ​​caused by equipment problems or improper operation can be filled, so as to obtain a more accurate and reliable blood oxygen saturation time series.

[0073] 2. Generalized apnea-specific fragment extraction:

[0074] A window with a length of 1 hour was used to slide the original respiratory signal at intervals of 20 minutes. The respiratory segment with the highest number of pauses and its corresponding blood oxygen saturation segment were taken for subsequent analysis for each subject, where the blood oxygen saturation segment was the generalized apnea-specific segment.

[0075] 3. Nonlinear feature extraction:

[0076] The permutation entropy and sample entropy of the blood oxygen saturation segments were calculated to characterize the nonlinear characteristics of the signal.

[0077] Permutation entropy is used to measure the degree of random change in the signal time series. The larger the value of permutation entropy, the more random the blood oxygen saturation time series is and the more complex the signal is; conversely, the more regular the blood oxygen saturation time series is and the less complex it is. Define the blood oxygen saturation time series as u(i), i = 1, 2, ..., N, N is the total length of blood oxygen saturation. After reconstructing it in phase space, the following matrix is ​​obtained:

[0078]

[0079] Where m is the embedding dimension, τ is the delay factor, k = n-(m-1)τ, j = 1, 2,…, k.

[0080] The False Nearest Neighbors (FNN) method [6] is used to determine the embedding dimension, and the number of false neighbors decreases as the embedding dimension increases. When the number of false neighbors no longer decreases significantly, it can be considered that a suitable embedding dimension has been found. When determining the delay factor, the Short-Time Autocorrelation (STA) [7] method is used, which selects a suitable delay factor based on the correlation of the signal with its delayed version over a short period of time.

[0081] There are k reconstruction components in the matrix, each with m-dimensional embedded elements. Arrange the j-th component x(j), x(j+τ), …, x(j+(m-1)τ) in the matrix in ascending order, and we get:

[0082] x(i+(j 1 -1)τ)≤x(i+(j 2 -1)τ)≤…≤x(i+(j m -1)τ)

[0083] where j 1 , j 2 , …, j m Represents the subscript index value of each element in the reconstruction component. If two or more reconstruction components are equal, such as x(i+(j 1 -1)τ)=x(i+(j 2 -1)τ), then we need to calculate the 1 、j 2 Sort in ascending order of size.

[0084] Each reconstructed component can obtain a reconstructed symbol sequence:

[0085] S(l)=(j 1 ,j 2 ,…,j m )

[0086] Where l = 1, 2, ..., k, and satisfies k ≤ m!. Each reconstruction component is an m-dimensional space, mapped to an m-dimensional symbol sequence, with a total of m! permutations. The probability of calculating each symbol sequence is expressed as P 1 , P 2 ,…,P K , then according to the definition of Shannon entropy, the permutation entropy of the blood oxygen saturation time series is defined as:

[0087]

[0088] The specific steps for calculating sample entropy are as follows:

[0089] The blood oxygen saturation time series Y 2 (n)=(Y 2 (1),Y 2 (2),Y 2 (3),...,Y 2 (i),...,Y 2 (k)) are composed into a v-dimensional vector in order:

[0090] V i =[Y 2 (i),Y 2 (i+1),…,Y 2 (i+v-1)],i=1,2,…,k

[0091] Among them, k is the blood oxygen saturation time series Y 2 (n) length.

[0092] For each value of i calculate the phase vector V i and the remaining vector V j The distance, that is:

[0093] d ij =max|Y 2 (i+I)-Y 2 (j+I)|,I=0,1,…,v-1

[0094] Given a threshold r, for each i, for Y 2 (i) and the remaining v-dimensional vector Y 2 (j) (j = 1, 2, ..., kv; j ≠ i) calculate the distance d between ij .

[0095] Statistics ij The number of numbers less than r and this number N_d ij The ratio of the total distance kv-1 is expressed as Right now:

[0096]

[0097] Ask again The average value is:

[0098]

[0099] The sample entropy of the blood oxygen saturation time series is shown as follows:

[0100]

[0101] 4. Calculate the statistical characteristics of generalized apnea-specific segments:

[0102] The search window (time between two peaks) is determined by taking the average of the blood oxygen saturation signals related to respiratory events. Under each search window, the following are calculated: ① the area S1 enclosed by the blood oxygen saturation curve and the blood oxygen saturation baseline before the event; ② the area S2 under the blood oxygen saturation curve; ③ the amplitude of change of the blood oxygen saturation curve H1; ④ the slope K1 of the line connecting the maximum and minimum blood oxygen saturation values; ⑤ the ratio of S1 to the search window duration; ⑥ the ratio of S2 to the search window duration. The calculation results are arranged by time to obtain a 6-dimensional event-level indicator sequence within the generalized apnea-specific segment, with the time step being the number of apneas.

[0103] For all 6-dimensional event-level indicator sequences in generalized apnea-specific segments, the following statistics are calculated: ① maximum value; ② minimum value; ③ range; ④ standard deviation; ⑤ mean; ⑥ skewness; ⑦ kurtosis; ⑧ slope. Finally, a total of 48 statistical features can be obtained for each generalized apnea-specific segment.

[0104] 5. Hypoxic Load Variability Index:

[0105] The entropy features at the segment level and the statistical features extracted from the event-level indicators were combined to obtain a 50-dimensional generalized apnea-specific hypoxic load fluctuation index, namely the hypoxic load variability index, which describes the variability of hypoxic load during apnea.

[0106] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the protection scope of the present invention in any form, and all technical solutions obtained by equivalent replacement and the like fall within the protection scope of the present invention.

[0107] The parts not involved in the present invention are the same as the prior art or can be implemented by using the prior art.

Claims

1. A method for constructing a sleep apnea hypoxia load variability index, characterized in that: The steps include: Step 1: Data preprocessing; Step 2, generalized apnea-specific fragment extraction; Step 3: Nonlinear feature extraction: step 4. Calculate the statistical characteristics of generalized apnea-specific segments: Step 5: Combine and calculate to obtain the hypoxic load variability index.

2. The method for constructing a sleep apnea hypoxia load variability index according to claim 1, characterized in that: In step 1, the blood oxygen saturation time series is recorded as x[n], and is preprocessed by using two steps of Delta filtering and linear interpolation: First, Delta filtering was used to remove nonphysiological data points caused by pulse oximeter movement or poor contact between the finger and the probe; Then, linear interpolation was used to replace all non-physiological values ​​to ensure smooth and continuous time series.

3. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 2, characterized in that: For a given input sequence x[n], a first-order Delta filter detects outliers by comparing the differences between adjacent sample points; specifically, if Then x[n] is marked as a non-physiological value that needs to be processed.

4. The method for constructing a sleep apnea hypoxia load variability index according to claim 2, characterized in that: Find the previous valid value x[i] and the next valid value x[j] of each non-physiological interval using the linear interpolation formula to estimate the missing value x[k].

5. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 1, characterized in that: In step 2, a window with a length of N hours is used to slide the original respiratory signal at intervals of M minutes, and the respiratory segment with the highest number of pauses for each subject and its corresponding blood oxygen saturation segment are taken for subsequent analysis, wherein the blood oxygen saturation segment is a generalized apnea-specific segment.

6. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 1, characterized in that: In step 3, the permutation entropy and sample entropy of the blood oxygen saturation segment are calculated to characterize the nonlinear characteristics of the signal.

7. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 6, characterized in that: The permutation entropy is used to measure the degree of random change of the signal time series. The larger the value of the permutation entropy, the more random the blood oxygen saturation time series is and the more complex the signal is. On the contrary, the more regular the blood oxygen saturation time series is and the less complex it is. Define the blood oxygen saturation time series as u(i), i = 1, 2, ..., N, N is the total length of blood oxygen saturation; After phase space reconstruction, the following matrix is ​​obtained: Where m is the embedding dimension, τ is the delay factor, k = n-(m-1)τ, j = 1, 2, ..., k; There are k reconstruction components in the matrix, each of which has m-dimensional embedded elements. Arrange the j-th component x(j), x(j+τ), …, x(j+(m-1)τ) in the matrix in ascending order to obtain: x(i+(j1-1)τ)≤x(i+(j2-1)τ)≤…≤x(i+(j m -1)t) where j1, j2, …, j m Represents the subscript index value of each element in the reconstructed component; if two or more reconstructed components are equal, they are sorted in ascending order according to the size of j1 and j2; Each reconstructed component can obtain a reconstructed symbol sequence: S(l)=(j1,j2,…,j m ) Where l = 1, 2, ..., k, k ≤ m!; each reconstruction component is an m-dimensional space, mapped to an m-dimensional symbol sequence, with a total of m! permutations; the probability of calculating each symbol sequence is expressed as P1, P2, ..., P K , then according to the definition of Shannon entropy, the permutation entropy of the blood oxygen saturation time series is defined as:

8. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 6, characterized in that: The specific steps of calculating the sample entropy are as follows: (a) The blood oxygen saturation time series Y2(n) = (Y2(1), Y2(2), Y2(3), ..., Y2(i), ..., Y2(k)) is sequentially formed into a v-dimensional vector: V i =[Y2(i),Y2(i+1),…,Y2(i+v-1)],i=1,2,…,k Wherein, k is the length of the blood oxygen saturation time series Y2(n); (b) Calculate the phase vector V for each value of i i and the remaining vector V j The distance, that is: d ij =max|Y2(i+I)-Y2(j+I)|,I=0,1,…,v-1 Given a threshold r, for each i, the distance di between Y2(i) and the remaining v-dimensional vectors Y2(j) (j = 1, 2, ..., kv; j ≠ i) is calculated j ; (c) Statistics ij The number of numbers less than r and this number N_d ij The ratio of the total distance kv-1 is expressed as Right now: i=1,…,k-v+1.i≠j; (d) Find B again i v The average value of (r) is: (e) The sample entropy of the blood oxygen saturation time series is shown as follows:

9. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 1, characterized in that: In step 4, the average value of the blood oxygen saturation signal related to the respiratory event is taken to determine the search window; and the following is calculated for each search window: (1) The area S1 enclosed by the blood oxygen saturation curve and the blood oxygen saturation baseline before the event; (2) area under the blood oxygen saturation curve S2; (3) The amplitude of change of the blood oxygen saturation curve H1; (4) Slope of the line connecting the maximum and minimum values ​​of blood oxygen saturation, K1; (5) The ratio of S1 to the search window duration; (6) The ratio of S2 to the search window duration; The calculation results are arranged in time to obtain a 6-dimensional event-level indicator sequence within the generalized apnea-specific segment, with the time step being the number of apneas; For all 6-dimensional event-level indicator sequences within the generalized apnea-specific segments, the following statistics were calculated: (1) maximum value; (2) minimum value; (3) range; (4) standard deviation; (5) mean; (6) skewness; (7) kurtosis; (8) slope; Finally, the statistical characteristics of each generalized apnea-specific segment were obtained.

10. The method for constructing a sleep-disordered breathing hypoxia load variability index according to claim 1, characterized in that: In step 5, the segment-level entropy features and the statistical features extracted from the event-level indicators are combined to obtain a 50-dimensional generalized apnea-specific hypoxic load fluctuation index that describes the variability of hypoxic load during apnea, namely, a hypoxic load variability index.

Images