A single-channel maternal-fetal hybrid ECG R peak detection method and device, and storage medium

Abstract

The present invention discloses a single-channel unsupervised maternal-fetal mixed ECG signal R-peak detection method and device, storage medium, including: step S1, obtaining the maternal-fetal mixed ECG signal; step S2, grading multiple signal segments of the maternal-fetal mixed ECG signal, and determining the R-peak detection strategy of each signal segment; step S3, according to the R-peak detection strategy of each signal segment, determining the acquisition method of the reference R-peak sequence, and detecting the remaining R-peaks in each signal segment in a dynamic cycle extension manner based on the quasi-periodicity and morphological characteristics of the ECG signal; step S4, sequentially splicing the R-peak detection results of each signal segment as the final output. The technical solution of the present invention can simultaneously detect the mother's R peak and the fetal R peak in a single-channel unlabeled maternal-fetal mixed ECG signal, which is suitable for single-channel signal detection and does not need to rely on information interaction between multi-channel signals.

Description

Technical Field

[0001] The present invention belongs to the technical field of fetal electrocardiogram detection, and in particular relates to a single-channel maternal-fetal mixed electrocardiogram R-peak detection method and device, and a storage medium. Background Art

[0002] Non-invasive fetal ECG detection technology has the advantages of being non-invasive, easy to wear, and capable of dynamic monitoring. It can assist doctors in assessing fetal heart development and promptly detect sudden abnormalities such as fetal intrauterine hypoxia and arrhythmia, thereby reducing perinatal morbidity and mortality. However, it is extremely challenging to accurately detect the fetal R peak from the mixed maternal and fetal ECG signals. The main reasons are: (1) The fetal ECG signals collected in the abdomen of pregnant women are very weak and easily interfered by artifacts such as electromyography due to the obstruction of vernix caseosa, amniotic fluid, and uterine wall, resulting in a low signal-to-noise ratio; (2) The fetal R peak is similar in morphology to the maternal R peak, and there is aliasing in both the time domain and the frequency domain, which is easy to misjudge; (3) The morphology and amplitude of the maternal and fetal R peaks are affected by many factors such as electrode layout, gestational age, uterine wall thickness, and fetal position, and there are significant individual differences.

[0003] The existing technology has major deficiencies in processing the fetal R peak detection task in the maternal-fetal mixed ECG signal:

[0004] 1. Traditional machine learning or deep learning algorithms are difficult to directly process fetal ECG signals with significant individual differences and time-varying properties. Such methods are sensitive to network parameters, and improper parameter selection may lead to overfitting. A large amount of labeled data is required to learn complex nonlinear relationships. However, it is quite difficult to obtain large-scale labeled data in practical applications, which limits the practical application of such algorithms. This problem is particularly prominent in the detection of fetal R peaks in the task of processing maternal-fetal mixed ECG signals.

[0005] 2. Algorithms based on blind source separation usually require multi-channel synchronous acquisition of mixed abdominal wall ECG signals to make up for the lack of prior knowledge such as source signals and mixing matrices when processing mixed signals from the maternal abdomen. Single-channel mixed abdominal signals are often composed of multi-source nonlinear signals, and the presence of various interference sources will seriously affect the blind source separation effect, thereby having a negative impact on subsequent R peak detection.

[0006] 3. Adaptive filtering technology usually requires a reference signal (usually a chest or shoulder signal) for effective filtering processing. When facing a single-channel abdominal signal, there is no reference signal, and it is difficult to filter out the nonlinear noise components in the mixed signal, which affects the detection of the fetal R peak.

[0007] 4. The morphology-based R-peak detection method relies on the morphological template. However, due to the variability of ECG signals and individual differences, the same signal template cannot be applied to all R-peaks, thus affecting the detection effect. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a single-channel unsupervised maternal-fetal mixed ECG signal R-peak detection method and device, and storage medium, which can simultaneously detect the mother's R peak and the fetal R peak in a single-channel unlabeled maternal-fetal mixed ECG signal, and is suitable for single-channel signal detection without relying on information interaction between multi-channel signals.

[0009] To achieve the above object, the present invention adopts the following technical solution:

[0010] A single-channel unsupervised R-peak detection method for maternal-fetal mixed electrocardiogram signal, comprising:

[0011] Step S1, obtaining maternal-fetal mixed ECG signals;

[0012] Step S2, classifying multiple signal segments of the maternal-fetal mixed ECG signal, and determining an R peak detection strategy for each signal segment;

[0013] Step S3, according to the R peak detection strategy of each signal segment, determine the method for obtaining the reference R peak sequence, and detect the remaining R peaks in each signal segment in a dynamic cycle extension manner according to the quasi-periodicity and morphological characteristics of the ECG signal;

[0014] Step S4, sequentially splicing the R peak detection results of each signal segment as the final output.

[0015] Preferably, in step S2, the maternal-fetal mixed ECG signal is subjected to preprocessing, peak detection, data segmentation and grading processing in sequence to determine the R-peak detection strategy for each signal segment; wherein the detection strategy includes: an R-peak detection strategy for high-quality segments, an R-peak detection strategy for noisy long segments and an R-peak detection strategy for noisy short segments.

[0016] Preferably, in step S2, data segmentation is performed on the entire signal according to the peak detection result of the preprocessed maternal-fetal mixed ECG signal, specifically including:

[0017] Calculate the average amplitude A_arg of all peak points of the entire signal;

[0018] Create a full zero vector X_noise with the same length as the entire signal to save the noise calibration result;

[0019] Traverse the peak detection results. If the amplitude of a peak point is greater than A_arg+α, where α is a constant parameter, the position of the peak point is taken as the center point of the noise interval. With the center point as the center and the one-side extension parameter extension_length as the radius, the interval after left and right extension is taken as the noise interval. All corresponding points in the noise interval are marked as 1 in the all-zero vector X_noise.

[0020] According to the labeling results in X_noise, the all-1 fragments and the all-0 fragments are segmented and divided into three categories: high-quality fragments, noisy long fragments and noisy short fragments.

[0021] Preferably, step S3 comprises:

[0022] Perform peak detection on each signal segment;

[0023] The peak detection results are screened by amplitude to obtain two types of peak sequences, namely high amplitude sequence and low amplitude sequence;

[0024] Scoring high-amplitude sequences for periodicity;

[0025] According to the scoring results, the mother's baseline R peak sequence and the fetus' baseline R peak sequence are extracted;

[0026] Centered on the maternal baseline R peak sequence and the fetal baseline R peak sequence, dynamic cycle extension is performed step by step, and the morphological feature scores of the peak points near the cycle extension coordinates are performed to detect the R peaks that have not been detected in the entire signal.

[0027] The present invention also provides a single-channel unsupervised R-peak detection device for maternal-fetal mixed electrocardiogram signals, comprising:

[0028] An acquisition module, used for acquiring maternal-fetal mixed ECG signals;

[0029] A determination module, used to classify multiple signal segments of the maternal-fetal mixed ECG signal and determine an R peak detection strategy for each signal segment;

[0030] The detection module is used to determine the acquisition method of the reference R peak sequence according to the R peak detection strategy of each signal segment, and detect the remaining R peaks in each signal segment in a dynamic cycle extension manner according to the quasi-periodicity and morphological characteristics of the ECG signal;

[0031] The splicing module is used to sequentially splice the R peak detection results of each signal segment as the final output.

[0032] Preferably, the determination module is used to perform preprocessing, peak detection, data segmentation and grading processing on the maternal-fetal mixed ECG signal in sequence to determine the R-peak detection strategy for each signal segment; wherein the detection strategy includes: R-peak detection strategy for high-quality segments, R-peak detection strategy for noisy long segments and R-peak detection strategy for noisy short segments.

[0033] Preferably, the detection module comprises:

[0034] A first detection unit, used to perform peak detection on each signal segment;

[0035] A screening unit is used to perform amplitude screening on the peak detection results to obtain two types of peak sequences, namely, a high amplitude sequence and a low amplitude sequence;

[0036] A scoring unit, used for periodically scoring high-amplitude sequences;

[0037] An extraction unit, used for extracting a maternal baseline R peak sequence and a fetal baseline R peak sequence according to the scoring result;

[0038] The second detection unit is used to perform dynamic cycle extension step by step with the mother's baseline R peak sequence and the fetus' baseline R peak sequence as the center, and perform morphological feature scoring on the peak points near the cycle extension coordinates to detect R peaks that have not been detected in the entire signal.

[0039] The present invention also provides a storage medium storing machine executable instructions. When the machine executable instructions are called and executed by a processor, the machine executable instructions prompt the processor to implement a single-channel unsupervised R-peak detection method for maternal-fetal mixed electrocardiogram signals.

[0040] The present invention has the following technical effects:

[0041] Compared with the prior art (application: CN201811202868.4, a method for realizing single-channel fetal heart rate monitoring based on a search tree), the advantages of the present invention are as follows:

[0042] 1. The detection efficiency is higher by first detecting the benchmark R peak sequence and then extending the detection

[0043] The R peaks of the electrocardiographic signals of the mother and the fetus in the mixed signal are similar in shape, and there is overlap between the two in the time domain and the frequency domain, which can easily lead to confusion and false detection during the detection process. Correction requires a lot of analytical calculations, which is inefficient. The present invention classifies the reference sequence of R peaks into mothers and fetuses at the beginning of the detection, which to a certain extent avoids the misdetection of the mother's R peak as the fetal R peak caused by the similarity principle, and reduces the unnecessary redundant calculations caused by category correction. In contrast, the prior art only searches for potential fetal R peaks based on the R peak spacing during the initial detection, which can easily misdetect the mother's R peak as the fetal R peak, and uses a large number of unnecessary correction calculation links, which reduces the detection efficiency.

[0044] 2. Use Gaussian function family to extract QRS waveform features, which is more robust

[0045] The morphological features used in the intermediate process of the present invention are function coefficients based on the waveform fitting of the Gaussian function family, which are not affected by the integrity of the QRS waveform. Even if the QRS waveform is missing, the features can still be fully extracted for subsequent calculations. In contrast, the features selected in the process of screening the R peak in the prior art are strongly correlated with the amplitude difference or distance difference between the three peaks of the Q peak, R peak and S peak in the QRS wave, and can only be fully acquired and accurately calculated when the QRS waveform is complete. However, in practice, the ECG signals extracted from the mixed signal, especially the fetal ECG, have many Q peaks or S peaks missing, which limits the feature extraction of the prior art.

[0046] 3. Use feature operations based on adaptive strategies to save more memory

[0047] In the detection process, the present invention adopts an adaptive strategy and selects some peaks for morphological feature extraction operation only when the position deviation discrimination condition is met according to the property of local cyclic stationarity, and only saves the feature-based scoring results for subsequent processing after calculation, saving memory. In contrast, the prior art requires a large amount of computing and storage resources. In the process of constructing and sorting the fetal R peak sequence based on the search tree, in order to ensure the accuracy of the sorting, the method saves the intermediate calculation data of all paths in the path construction process for comparison and screening, which requires a large amount of data storage space. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0049] Figure 1 It is a flow chart of a method for detecting R peak of a single-channel unsupervised maternal-fetal mixed ECG signal according to an embodiment of the present invention;

[0050] Figure 2 It is a flow chart of another single-channel unsupervised R-peak detection method of maternal-fetal mixed ECG signal according to an embodiment of the present invention;

[0051] Figure 3 Flowchart for scoring periodicity;

[0052] Figure 4 Flowchart for extracting two sets of benchmark R peak sequences

[0053] Figure 5 is a schematic diagram of an example of the first group of R peak reference sequences;

[0054] Figure 6Schematic diagram of two groups of R peak benchmark sequence examples;

[0055] Figure 7 Flowchart of edge extension detection based on benchmark sequence. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Embodiment 1:

[0059] like Figure 1 As shown, an embodiment of the present invention provides a single-channel unsupervised R peak detection method for maternal-fetal mixed ECG signal, comprising:

[0060] Step S1, obtaining maternal-fetal mixed ECG signals;

[0061] Step S2, classifying multiple signal segments of the maternal-fetal mixed ECG signal and determining a detection strategy for each signal segment;

[0062] Step S3, according to the detection strategy of each signal segment and based on the quasi-periodicity of the ECG signal, the R peak in each signal segment is detected in a dynamic cycle extension manner;

[0063] Step S4, sequentially splicing the R peak detection results of each signal segment as the final output.

[0064] As an implementation method of an embodiment of the present invention, in step S2, the maternal-fetal mixed ECG signal is preprocessed, peak detected, data segmented and graded in sequence to determine the detection strategy for each signal segment; wherein the detection strategy includes: R-peak detection strategy for high-quality segments, R-peak detection strategy for noisy long segments and R-peak detection strategy for noisy short segments.

[0065] As an implementation of the embodiment of the present invention, in step S2, data segmentation of the entire signal is performed according to the peak detection result of the preprocessed maternal-fetal mixed ECG signal, specifically including:

[0066] Calculate the average amplitude A_arg of all peak points of the entire signal;

[0067] Traverse the peak detection results. If the amplitude of a peak point is greater than A_arg+α, where α is a constant parameter, the position of the peak point is taken as the center point of the noise interval. With the center point as the center and the one-sided extension parameter extension_length as the radius, the interval after left and right extension is taken as the noise interval, and the corresponding position of the noise interval is marked as 1 in X_noise.

[0068] Preferably, step S3 comprises:

[0069] Perform peak detection on each signal segment;

[0070] The peak detection results are screened by amplitude to obtain two types of peak sequences, namely high amplitude sequence and low amplitude sequence;

[0071] Scoring high-amplitude sequences for periodicity;

[0072] According to the scoring results, the mother's baseline R peak sequence and the fetus' baseline R peak sequence are extracted;

[0073] Centered on the mother's reference R peak sequence and the fetus's reference R peak sequence, the cycle is extended step by step to detect the R peaks that have not been detected in the entire signal.

[0074] Embodiment 2:

[0075] An embodiment of the present invention provides a single-channel unsupervised R-peak detection method for maternal-fetal mixed ECG signals, which includes 10 steps. First, in steps 1-4, the maternal-fetal mixed ECG signals are sequentially preprocessed, noise interval calibrated, signal segmented, and fragment graded, and the processing order and strategy of each fragment are determined. Then, in steps 5-9, according to the detection strategies of different fragments, based on the quasi-periodicity of the ECG signal, the R peak in each signal fragment is detected by dynamic periodic extension. Finally, in step 10, the R-peak detection results of each signal fragment are spliced ​​in sequence, thereby obtaining the complete maternal-fetal R-peak detection result of the input signal.

[0076] As shown in 2, the specific steps are as follows:

[0077] Step 1: Preprocessing

[0078] The input maternal-fetal mixed ECG signal is preprocessed, including using wavelet transform to remove the interference of myoelectric noise, using low-pass filter to remove baseline drift, and using notch filter to remove the noise of power frequency interference.

[0079] Step 2: Peak Detection

[0080] Perform peak detection on the ECG signal preprocessed in step 1. The essence of this is to find the local maximum point in the signal. Save the corresponding coordinates and amplitudes of all peak points in the signal.

[0081] Step 3: Data segmentation

[0082] The entire signal is segmented according to the peak detection result obtained in step 2.

[0083] Step 3.1: Initialization

[0084] Create an all-zero vector X_noise of the same length as the preprocessed ECG signal

[0085] Step 3.2: Calculate the average amplitude of the peak point

[0086] Calculate the average amplitude A_arg of all peak points of the entire signal obtained in step 2.

[0087] Step 3:3: Mark the high amplitude noise interval

[0088] Traverse the peak detection results of step 2. If the amplitude of a peak point is greater than A_arg+α (α is a constant parameter set manually), the position of the peak point is used as the center point of the noise interval. With the center point as the center and the one-side extension parameter extension_length as the radius, the interval after left and right extension is used as the noise interval, and the corresponding position of the noise interval is marked as 1 in X_noise.

[0089] Step 3.4: Data segmentation

[0090] According to the results marked in X_noise, the first and last coordinates of the all-1 segment are used as segmentation points to segment the data preprocessed in step 1 and divide the entire signal into several segments.

[0091] Step 4: Grading

[0092] The multiple signal fragments obtained in step 3 are graded, the purpose of which is to determine a specific detection strategy for the signal fragments.

[0093] Step 4.1 Signal fragment classification method

[0094] The segmented signal segments are divided into three levels: high-quality segments, noisy long segments, and noisy short segments.

[0095] In the segmented fragment signal results obtained in step 3, the fragments whose actual positions correspond to fragments whose corresponding marks in X_noise are all 0s are high-quality fragments; the fragments whose corresponding marks in X_noise are all 1s and whose fragment lengths are greater than L are noisy long fragments; and the remaining signal fragments are noisy short fragments.

[0096] Step 4.2 R-peak detection strategy for signal segments at different levels

[0097] Different detection strategies are used for signal fragments at different levels.

[0098] In terms of processing order, all high-quality segments are processed first, followed by noisy long segments, and finally noisy short segments.

[0099] Step 4.2.1: R-peak detection strategy for high-quality fragments

[0100] Perform R peak detection according to steps 5-9 and output the maternal-fetal R peak detection sequence of the entire segment.

[0101] Step 4.2.2: R-peak detection strategy for noisy long fragments

[0102] Two sets of peak sequences are obtained according to step 5 and step 6. The R peak detection result of the high-quality segment adjacent to the noisy long segment is used as the reference sequence, and the R peak is detected in step 9 to output the maternal-fetal R peak detection sequence of the entire segment.

[0103] Step 4.2.3: R-peak detection strategy for noisy short fragments

[0104] According to step 5 and step 6, two sets of peak sequences are obtained, and the R peak detection results of the high-quality segments adjacent to the noisy long segment are used as the reference sequence. The R peak is detected in step 9 and the maternal-fetal R peak detection sequence of the entire segment is output. The difference between step 9 and the R peak detection strategy of the noisy long segment is that when the position detection abnormality of step 9.3 is triggered, it is not processed according to step 9.3, but the abnormal position is directly marked as vacant in step 9.4.

[0105] Step 5: Peak detection for each segment

[0106] Perform peak detection on the segmented signal segments, similar to step 2, and save the position coordinates and corresponding amplitude of the peak point.

[0107] Step 6: Screening Sequences

[0108] Through the low threshold A L With high threshold A H The peak detection result of step 5 is subjected to amplitude screening to obtain two types of peak sequences, namely, high amplitude sequence and low amplitude sequence.

[0109] Step 7: Figure 3 As shown, periodic scoring

[0110] Step 7.1d(i) Screening

[0111] Traverse the high amplitude sequence obtained in step 6, and for each peak point in it, record it as t. Calculate its surrounding peak point Peak_pos j The distance is denoted as d(i), through d(i) min and d(i) maxScreen d(i). The calculation formula is as follows:

[0112] d(i)=|t-Peak_pos j |,d(i)∈(d(i) min , d(i) max ),

[0113] Among them, d(i) min and d(i) max A manually set threshold.

[0114] Step 7.2 Extension

[0115] Use d(i) in step 7.1 to perform bilateral extension with the peak point t as the center to obtain the extension sequence extension_pos(t). The calculation formula is as follows:

[0116] extention_pos(t)=t+qd(i), q∈(-m,m),

[0117] The length of the extention_pos(t) sequence output in this step is 2q+1.

[0118] Step 7.3: Perform differential operation on the extended signal

[0119] The distance difference between the extended coordinate sequence extension_pos and Peak_pos in step 7.2 is used to obtain the peak sequence Diff(t) closest to each point in the extended coordinates. The calculation formula is as follows:

[0120] Diff(t)=argmin(|extention_pos(t)-Peak_pos|)

[0121] Step 7.4 Scoring

[0122] Sum the peak sequence Diff(t) in step 7.3 to obtain the score S(i) of the peak point t and its corresponding d(i), denoted as d * (i). S(i) can reflect the periodicity strength of the extended sequence of the peak point t under the period d(i). The lower the score, the more significant the periodicity of the point. The formula is as follows:

[0123]

[0124] Step 8: Figure 4 As shown, extract the benchmark R peak sequence

[0125] Find the maternal baseline R-peak sequence and the fetal baseline R-peak sequence.

[0126] Step 8.1 Screening the central peak point of the benchmark sequence

[0127] Using S(i) obtained in step 7, sort the score sequences Score(t) of all peak points, and find the peak point t_best with the lowest score as the central peak point of the required benchmark sequence.

[0128] Step 8.2 Retrieve the reference peak sequence

[0129] The method is the same as step 7.3, and the peak sequence Diff(t_best) centered at t_best is obtained as the benchmark sequence.

[0130] Step 8.3 Extract two sets of benchmark sequences

[0131] Step 8.3.1 Extract the first set of benchmark sequences

[0132] In the detection process of the same signal segment, the peak sequence Diff(t_best) centered at t_best is first obtained in step 8.2 as the first set of reference R peak sequences. Figure 5 As shown in the figure, the peak points marked in red are the first set of reference R peak sequences extracted from the entire signal.

[0133] Step 8.3.2 Extract the second set of benchmark sequences

[0134] According to the coordinates of the first and last R peaks of the first set of benchmark sequences obtained in 8.3.1, retrieve all peak points within their coverage range, and screen out peak points that do not overlap with the first set of benchmark peaks to form a set peak_pre. With these peak points as the center, perform periodic scoring according to step 7. The difference from the first set of benchmark R peak detection is that the high amplitude sequence in step 7.1 is replaced by the peak point set sequence peak_pre, and the remaining processing steps are the same as step 7. Finally, the benchmark sequence obtained through steps 8.1 to 8.2 is the second set of benchmark R peak benchmark sequences. Figure 6 In the figure, the green dashed line represents the second set of R-peak reference sequences, while the red line represents the first set of R-peak reference sequences.

[0135] Step 8.4 Determine the categories of the two groups of sequences

[0136] The average RR intervals of the two groups of reference sequences are calculated and compared. The reference sequence with the larger RR interval is selected as the mother's reference R peak sequence, and the other reference sequence is selected as the fetus's reference R peak sequence.

[0137] Step 9: Figure 7 As shown, the edge extension detection based on the benchmark sequence

[0138] Centered around the two sets of reference sequences (the maternal R-peak position sequence m_loc and the fetal R-peak position sequence f_loc) obtained in step 9, gradually perform periodic extension to detect R-peaks that have not been detected in the entire signal.

[0139] Step 9.1: Edge extension detection

[0140] Step 9.1.1: Calculate the estimated position of adjacent R-peaks

[0141] Select m consecutive R-peaks (2 < m < 2q + 1) at the edge from the existing R-peak sequence, calculate the average RR interval of the selected m R-peaks. If the last R-peak at the current edge position is the k-th R-peak, then estimate the position of the (k + 1)-th R-peak:

[0142]

[0143] Step 9.1.2: Search for the target R-peak

[0144] In the sequence interval retrieval, find the peak closest to the target detection position as R k+1 , record the detected R k+1 peak and the predicted position The distance between them is calculated as follows:

[0145]

[0146]

[0147] Step 9.1.3: Update the reference R-peak sequence

[0148] Add the R k+1 detected in step 9.1.2 to the reference peak sequence, and repeat steps 9.1.1 and 9.1.2 until the predicted position of the next R-peak exceeds the signal length.

[0149] Step 9.2: Morphological evaluation

[0150] Step 9.2.1 Morphological feature extraction

[0151] Fit the peak points through a third-order Gaussian family function (Hermite function can also be used), and use the parameters in the Gaussian function family as the morphological feature Opt_all of the peak point. The calculation formula is as follows:

[0152]

[0153] Opt_all = [a 1i , a 2i , a 3i , a4i ,b 1i ,μ 1i ] i＝1,2,3

[0154] Step 9.2.2 Morphological feature scoring

[0155] Select three consecutive R peaks at the edge of the R peak sequence and extract the morphological features according to step 9.2.1. Average the morphological features of the three R peaks to obtain the average R peak template feature matrix Opt_all ref The Euclidean distance between the R peak feature and the R peak template feature is calculated as the morphological comparison score Shape_score. The calculation process is as follows:

[0156]

[0157] Step 9.3: Recheck position deviation abnormality

[0158] Step 9.3.1 Position deviation abnormal warning

[0159] If the D predicted in step 9.1.2 is greater than the deviation threshold σ, the deviation abnormality warning is triggered. If the deviation abnormality triggers the warning twice in a row, the deviation abnormality recheck procedure is started. If D is greater than the deviation threshold σ twice in a row, it is considered that there may be a deviation in the previous R peak position, and the R peak that triggered the abnormal warning is rechecked.

[0160] Step 9.3.2 Position deviation recheck procedure

[0161] After the recheck is triggered, the R peak that triggered the warning in step 9.2.1 is rechecked. First, the interval near the estimated position of the R peak in the low threshold peak sequence is rechecked. Detect the peaks within the interval, extract the morphological features of each peak point in the interval according to step 9.2 and calculate the morphological evaluation score Shape_score. Calculate the evaluation score of the peak point according to the morphological evaluation score Shape_score and the periodic evaluation score S(i), select the peak point with the lowest score as the R peak after the early warning R peak re-inspection, and use this R peak to replace the original R peak. The calculation formula is as follows:

[0162] Score_all=C1·Shape_scoe+C2·Score

[0163] Step 9.4: Empty space marking

[0164] Vacant position virtual marking is a special processing method for dealing with local peaks without peaks at the predicted position. When searching for the target R peak in step 9.1.2, such as If an R peak that meets the requirements is not detected in the interval, the predicted position of the R peak is directly saved in the benchmark sequence.

[0165] Step 10: Result stitching

[0166] The detection results of all the segments in step 9 are spliced ​​together to obtain the mother's R peak detection result and the fetus's R peak detection result of the entire signal and output them, and the detection is completed.

[0167] In the detection process, the embodiment of the present invention performs data cutting and grading on the pre-processed maternal-fetal mixed ECG signal. By adaptively adopting different detection strategies to process different signal segments, the difficulty of the R-peak detection task of the non-stationary maternal-fetal mixed signal is reduced. At the same time, the present application suppresses the negative impact of R-peak morphological differences on R-peak detection and classification tasks by first detecting the most periodically stable R-peak reference sequence in the overall signal and then extending the strategy of detecting the remaining R peaks. This makes the method have significant advantages in processing fetal ECG signals with obvious individual differences, time-varying and morphological differences.

[0168] The innovative point of the method of the embodiment of the present invention is:

[0169] 1. Signal quality grading based on high-amplitude noise detection (steps 3 and 4): In step 3, the pre-processed complete data is divided into three types of segments (high-quality segments, noisy long segments, and noisy short segments) through noise detection, and different strategies are used to process the signal segments. The specific strategies include: (1) Different processing order: all high-quality segments are processed first, followed by all noisy long segments, and finally all noisy sub-segments. (2) Different reference R peak sequences: The reference sequence of the high-quality segment is a sequence segment of length 2q+1 found in itself through the reference R peak sequence of the reference periodicity score, and the reference R peak of the noisy segment is the 2q+1 sequence segment closest to the noisy segment in the adjacent high-quality segment detection results; (3) When the position deviation condition warning is triggered, the noisy short segment is not re-checked, but the predicted position is directly used as the detection R peak position. .

[0170] 2. Find the maternal-fetal benchmark R-peak sequence based on local cyclic stationarity (step 7-step 9): When searching for the benchmark R-peak sequence, construct a period evaluation function through the RR interval constraint, the period robustness constraint, and the minimum number of periods, and score all R-peak sequence fragments of length 2q+1. Then sort the scoring results, and use the sequence corresponding to the lowest score as the first group of benchmark R-peak sequences, and use the same method to find the R-peak sequence of length 2q+1 that intersects but does not overlap with the first group of benchmark R-peak sequences in the entire signal as the second group of benchmark R-peak sequences to improve the robustness of R-peak detection. Key point three of the present invention: A rapid extraction method for mentality features based on a Gaussian function family, which uses a third-order Gaussian function family to fit the peak morphology, and represents the morphological characteristics of the peak point through the corresponding coefficients of the function.

[0171] Embodiment 3:

[0172] The embodiment of the present invention also provides a single-channel unsupervised maternal-fetal mixed electrocardiogram signal R peak detection device, comprising:

[0173] An acquisition module, used for acquiring maternal-fetal mixed ECG signals;

[0174] A determination module, used to classify multiple signal segments of the maternal-fetal mixed ECG signal and determine an R peak detection strategy for each signal segment;

[0175] The detection module is used to determine the acquisition method of the reference R peak sequence according to the R peak detection strategy of each signal segment, and detect the remaining R peaks in each signal segment in a dynamic cycle extension manner according to the quasi-periodicity and morphological characteristics of the ECG signal;

[0176] The splicing module is used to sequentially splice the R peak detection results of each signal segment as the final output.

[0177] As an implementation mode of an embodiment of the present invention, a determination module is used to perform preprocessing, peak detection, data segmentation and grading processing on the maternal-fetal mixed ECG signal in sequence to determine the R-peak detection strategy for each signal segment; wherein the detection strategy includes: R-peak detection strategy for high-quality segments, R-peak detection strategy for noisy long segments and R-peak detection strategy for noisy short segments.

[0178] As an implementation of an embodiment of the present invention, the detection module includes:

[0179] A first detection unit, used to perform peak detection on each signal segment;

[0180] A screening unit is used to perform amplitude screening on the peak detection results to obtain two types of peak sequences, namely, a high amplitude sequence and a low amplitude sequence;

[0181] A scoring unit, used for periodically scoring high-amplitude sequences;

[0182] An extraction unit, used for extracting a maternal baseline R peak sequence and a fetal baseline R peak sequence according to the scoring result;

[0183] The second detection unit is used to perform dynamic cycle extension step by step with the mother's baseline R peak sequence and the fetus' baseline R peak sequence as the center, and perform morphological feature scoring on the peak points near the cycle extension coordinates to detect R peaks that have not been detected in the entire signal.

[0184] Embodiment 4:

[0185] An embodiment of the present invention also provides a storage medium, which stores machine executable instructions. When the machine executable instructions are called and executed by a processor, the machine executable instructions prompt the processor to implement a single-channel unsupervised R-peak detection method for maternal-fetal mixed ECG signals.

[0186] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A single-channel unsupervised R-peak detection method for maternal-fetal mixed ECG signal, characterized in that: include: Step S1, obtaining maternal-fetal mixed ECG signals; Step S2, classifying multiple signal segments of the maternal-fetal mixed ECG signal, and determining an R peak detection strategy for each signal segment; Step S3, according to the R peak detection strategy of each signal segment, determine the method for obtaining the reference R peak sequence, and detect the remaining R peaks in each signal segment in a dynamic cycle extension manner according to the quasi-periodicity and morphological characteristics of the ECG signal; Step S4, sequentially splicing the R peak detection results of each signal segment as the final output; In step S2, the maternal-fetal mixed ECG signal is sequentially preprocessed, peak detected, data segmented and graded to determine the R peak detection strategy for each signal segment; wherein the detection strategy includes: the R peak detection strategy for high-quality segments, the R peak detection strategy for noisy long segments and the R peak detection strategy for noisy short segments; In step S2, data segmentation is performed on the entire signal according to the peak detection result of the pre-processed maternal-fetal mixed ECG signal, specifically including: Calculate the average amplitude A_arg of all peak points of the entire signal; Create a full zero vector X_noise with the same length as the entire signal to save the noise calibration result; Traverse the peak detection results. If the amplitude of a peak point is greater than A_arg+α, where α is a constant parameter, the position of the peak point is taken as the center point of the noise interval. With the center point as the center and the one-side extension parameter extension_length as the radius, the interval after left and right extension is taken as the noise interval. All corresponding points in the noise interval are marked as 1 in the all-zero vector X_noise. According to the labeling results in X_noise, the all-1 fragments and the all-0 fragments are segmented and divided into three categories: high-quality fragments, noisy long fragments, and noisy short fragments; Step S3 includes: Perform peak detection on each signal segment; The peak detection results are screened by amplitude to obtain two types of peak sequences, namely high amplitude sequence and low amplitude sequence; According to the quasi-periodicity and morphological characteristics of the ECG signal, the maternal and fetal R peak detection sequence in the high-quality segments is detected by dynamic cycle extension; According to the high amplitude sequence and the low amplitude sequence, the R peak detection results of the high-quality segments adjacent to the noisy long segment are used as the reference sequence, and the R peak is detected based on the edge extension of the reference sequence and the maternal and fetal R peak detection sequence of the noisy long segment is output; According to the high amplitude sequence and the low amplitude sequence, the R peak detection results of the high-quality fragments adjacent to the noisy short fragments are used as the benchmark sequence. The R peak is detected based on the edge extension of the benchmark sequence and the maternal and fetal R peak detection sequence of the noisy short fragment is output.

2. A single-channel unsupervised maternal-fetal mixed ECG signal R-peak detection device that implements the single-channel unsupervised maternal-fetal mixed ECG signal R-peak detection method of claim 1, characterized in that: include: An acquisition module, used for acquiring maternal-fetal mixed ECG signals; A determination module, used to classify multiple signal segments of the maternal-fetal mixed ECG signal and determine an R peak detection strategy for each signal segment; The detection module is used to determine the acquisition method of the reference R peak sequence according to the R peak detection strategy of each signal segment, and detect the remaining R peaks in each signal segment in a dynamic cycle extension manner according to the quasi-periodicity and morphological characteristics of the ECG signal; The splicing module is used to sequentially splice the R peak detection results of each signal segment as the final output.

3. A storage medium, characterized in that: The storage medium stores machine executable instructions. When the machine executable instructions are called and executed by the processor, the machine executable instructions prompt the processor to implement the R peak detection method of the single-channel unsupervised maternal-fetal mixed electrocardiogram signal according to claim 1.

Images