Newborn multi-parameter vital sign fusion monitoring method

Abstract

The present application relates to the technical field of living body detection, in particular to a new-born vital sign multi-parameter fusion monitoring method, comprising the following steps: obtaining a respiratory voltage signal and converting it into a time sequence to extract a phase boundary, dividing a respiratory window and matching ECG R waves and pulse peaks, counting peak value differences and subdividing abnormal intervals to relabel, rearranging peak value time according to intervals, and establishing a fusion monitoring data set by splicing a time axis according to respiratory phases. In the present application, a time sequence is constructed by respiratory voltage to extract a phase boundary, and respiratory fluctuations are converted into a sortable time reference, so that ECG and pulse peaks have a unified attribution basis. In combination with peak value difference identification in the window, rhythm mismatch is identified and abnormal intervals are subdivided to improve the accuracy of local change description. According to the rearrangement of time points according to peak value distribution, the influence of collection beat difference and delay can be weakened, the corresponding relationship of multiple signals in the same respiratory stage is clearer, and the recognition ability and fusion consistency in the weak signal and disturbance scene are enhanced.

Description

Technical Field

[0001] This invention relates to the field of live detection technology, and in particular to a method for multi-parameter fusion monitoring of neonatal vital signs. Background Technology

[0002] The field of liveness detection technology mainly involves the acquisition, identification, and discrimination of physiological signals of organisms under natural or controlled conditions to distinguish between real living organisms and non-living or fake organisms. This includes the acquisition and analysis of physiological characteristics such as heart rate, respiratory signals, blood oxygen saturation, body temperature changes, and skin surface microcirculation. Raw signals are acquired through hardware means such as photoelectric sensors, pressure sensors, infrared thermometers, and bioelectric acquisition electrodes. The results are combined with time series analysis, feature extraction, threshold discrimination, and multi-source information fusion methods to comprehensively judge vital signs. This includes the simultaneous acquisition of raw data by multiple types of sensors, preprocessing and synchronous calibration of different signals, extraction of key physiological parameters, and cross-parameter comparison and fusion analysis to form a unified basis for judging the state of life. Among them, the traditional multi-parameter fusion monitoring method for neonatal vital signs refers to a technical solution that jointly monitors multiple physiological indicators such as neonatal heart rate, respiratory rate, blood oxygen saturation, and body temperature. However, due to the weak vital sign signals of neonates, which are easily affected by motion interference, and the problems of asynchrony and differences in acquisition accuracy among multiple parameters, the traditional method involves attaching ECG electrodes to the neonate's chest or feet to collect ECG signals, obtaining respiratory rate through a piezoelectric breathing belt, obtaining blood oxygen saturation through a finger clip or foot clip photoelectric probe, and measuring body temperature through a skin contact temperature probe.

[0003] In practical operation, existing technologies largely rely on parallel observation of the outputs of individual sensors. While this allows for simultaneous observation of changes in values ​​such as heart rate, respiration, and blood oxygenation, the correlation between different signals on the time axis is relatively loose. When faced with short-term body movement, slight probe displacement, or irregular chest and abdominal movements, medical staff often find that one indicator has fluctuated while others remain in the previous timeframe. Clinical observation relies on experience to infer the correlation, making it difficult to confirm which respiratory phase a particular heartbeat change corresponds to, or whether peripheral pulse fluctuations are a genuine physiological response or a misalignment due to time lag. Furthermore, when the window is coarsely defined, local abnormalities are easily detected. Easily subject to averaging, if peak absence, peak clustering, rhythm stretching, etc., exist simultaneously within a single interval, the displayed results may still appear superficially stable, masking minor risks. For example, when respiratory rhythm is briefly disrupted and accompanied by a lag in peripheral perfusion response, the monitoring interface may only show a slight delay in blood oxygenation changes, making it difficult to directly determine which round of breathing is related to this delay. This further affects the identification of the relationship between circulatory fluctuations, ventilation status, and peripheral perfusion status. Long-term use of this operating method can easily increase the burden of manual review, reduce the efficiency of abnormal initiation location, and may also result in a lack of high-quality correlation clues in continuous monitoring records, limiting the reliability of subsequent assessments and early warning judgments. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a method for multi-parameter fusion monitoring of neonatal vital signs, comprising the following steps: S1: Obtain the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, convert the continuous voltage signal into a time series, calculate the difference between adjacent sampling points and extract the position of sign change, mark the time point of sign change as the inspiratory start time and expiratory start time, and generate a set of respiratory phase boundary time points by sorting them by time. S2: Based on the set of respiratory phase boundary time points, determine adjacent time intervals and divide respiratory phase windows, obtain the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point, obtain the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point, perform peak time point and window start and end time interval assignment judgment, and generate multi-signal phase assignment marker sequence; S3: Based on the multi-signal phase attribution marker sequence, count the number of R wave peaks and pulse wave peaks within the respiratory phase window, calculate the difference in the number of peaks and compare it with the set difference range, perform interval division on the out-of-range window, segment according to the peak distribution and re-mark the peak attribution, and generate a set of subdivided respiratory phase intervals. S4: Based on the set of respiratory phase subdivision intervals, read the interval time range, determine the time distribution interval according to the number of R wave peaks in the interval, rearrange the R wave peak time points and process the pulse wave peak time points according to the same rule to generate a multi-signal phase-aligned time sequence. S5: Based on the multi-signal phase-aligned time series, the interval time series is spliced ​​in the order of respiratory phase, and the correspondence between the spliced ​​time axis and the ECG peak value and pulse peak value is established to generate a vital signs fusion monitoring time axis dataset.

[0005] As a further aspect of the present invention, the respiratory phase boundary time point set includes an inspiratory start time point sequence and an expiratory start time point sequence; the multi-signal phase attribution label sequence includes an ECG peak phase label, a pulse peak phase label, and a phase window index identifier; the respiratory phase subdivision interval set includes subdivision time interval identifiers, interval type classification labels, and interval continuity markers; the multi-signal phase alignment time series includes an alignment time reference sequence, an ECG rearrangement time series, and a pulse rearrangement time series; and the vital signs fusion monitoring time axis dataset includes a unified time axis index, ECG peak mapping data, and pulse peak mapping data.

[0006] As a further aspect of the present invention, the process of calculating the difference between adjacent sampling points and extracting the symbol change position includes sampling the continuous voltage signal at equal intervals according to the sampling frequency parameters to form a discrete sequence, and determining the symbol of the difference between adjacent sampling points. When the difference between adjacent sampling points changes from a positive value to a negative value or from a negative value to a positive value, the corresponding sampling time is recorded as a candidate symbol change time point. The candidate symbol change time points are then filtered by time interval, and time points with an interval less than a preset minimum time interval threshold are removed.

[0007] As a further aspect of the present invention, the process of determining the peak time point and the start and end time interval of the window includes comparing the R wave peak time point and the pulse wave peak time point with the start and end times of the corresponding respiratory phase window, respectively. When the peak time point is located between the start and end times of the respiratory phase window, the corresponding window identifier is assigned, and a multi-signal phase attribution marker sequence is formed in chronological order.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, perform equal-interval time axis mapping on the continuous voltage signal according to the sampling frequency parameters, sequentially register the sampled values ​​with the corresponding sampling times, and reconstruct them into a single-column time series data according to the time index to generate a voltage time series. S102: Based on the voltage time series, perform difference calculation on adjacent sampling points, record the position where the difference sign changes from positive to negative and from negative to positive, extract and serialize the sampling time corresponding to the sign reversal position to obtain the sign reversal time sequence; S103: Based on the symbolic transition time sequence, the time from negative to positive is marked as the inhalation start time, and the time from positive to negative is marked as the exhalation start time. The execution times of the two types of marked times are sorted sequentially and merged into a set of respiratory phase boundary time points.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the set of respiratory phase boundary time points, determine the corresponding time intervals of adjacent time points and divide them into respiratory phase windows. Extract adjacent boundary time points in chronological order, calculate the interval value between adjacent time points, pair and index each group of preceding and following time points, and write them into the window index field in sequence to establish a respiratory phase window. S202: Acquire the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point; acquire the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point; perform peak localization on the ECG signal and the photoplethysmography signal; record the sampling time corresponding to the local maximum value; and aggregate them into a peak time point sequence. S203: Based on the respiratory phase window and the peak time point sequence, perform interval assignment judgment on the peak time point and the window start time and end time, record the window index and signal category identifier corresponding to the peak time point, arrange all assignment records in chronological order, and generate a multi-signal phase assignment mark sequence.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Extract the number of R wave peaks and pulse wave peaks within the respiratory phase window according to the multi-signal phase attribution marker sequence, perform classification counting on the marker sequence according to the window index, accumulate the number of peaks corresponding to the ECG identifier and pulse identifier in the same window respectively, and pair and record the two types of quantities according to the window order to obtain peak quantity pairing groups; S302: Calculate the peak number difference based on the peak number pairing group and judge it with the set difference range. Perform the absolute value calculation of the number difference and compare it with the preset peak difference threshold range. Extract the window index corresponding to the threshold range and mark it as an abnormal window to obtain the abnormal window index set. S303: Perform interval division on the corresponding respiratory phase window according to the abnormal window index set, perform segmentation on the time interval according to the peak time distribution position within the window, re-perform the attribution mark on the peak time point within the segment, integrate all segmented intervals in chronological order, and generate a set of subdivided respiratory phase intervals.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the set of respiratory phase subdivision intervals, read the interval time range, extract the start and end times of the subdivision intervals, calculate the interval duration and perform corresponding registration with the number of R wave peaks in the interval, divide the duration value by the number of R wave peaks to form the interval value, and write it into the time interval field according to the interval index to obtain the phase interval parameter column; S402: Rearrange the R-wave peak time points in the interval according to the phase interval parameter list, extract the R-wave peak time points corresponding to the sub-interval, arrange the peak number in chronological order, and perform position mapping between the number and the interval value to reconstruct the peak time sequence position in the interval and establish an ECG aligned time sequence. S403: Based on the ECG-aligned time series, read the corresponding subdivision interval identifier, perform same-sequence mapping and interval-in-order rearrangement on the pulse peak time point, write the pulse peak time point into the corresponding phase position, and aggregate and arrange it with the ECG time series position according to the interval index to generate a multi-signal phase-aligned time series.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: According to the multi-signal phase-aligned time series, the interval time series are spliced ​​together in the order of respiratory phases. The phase index corresponding to the multi-signal phase-aligned time series is extracted. The start and end times of adjacent intervals are read in sequence. The start time of the next sequence is corrected to the position after the end time of the previous sequence. The time series are continuously written into a unified time axis in phase order to generate an interval splicing time axis. S502: Based on the interval splicing time axis, retrieve the ECG peak identifier and pulse peak identifier corresponding to the time position, perform position registration on the peak occurrence time according to the time index, map the ECG peak time point to the corresponding coordinate of the splicing time axis, map the pulse peak time point to the coaxial coordinate, record the peak category identifier, and establish a peak time axis correspondence table; S503: Based on the peak time axis correspondence table, perform same-index aggregation on the spliced ​​time axis coordinates, ECG peak position, and pulse peak position, write the peak category and peak time corresponding to the time coordinates into a unified data structure in sequence, and complete the full sequence arrangement according to the respiratory phase order to generate a vital signs fusion monitoring time axis dataset.

[0013] As a further aspect of the present invention, the process of calculating the difference in the number of peaks and comparing it with a set difference range includes comparing the difference between the number of R-wave peaks and the number of pulse peaks within the respiratory phase window with the upper and lower limits of a preset difference range. When the difference exceeds the upper and lower limits of the preset difference range, the respiratory phase window is divided into two or more intervals according to the peak time distribution position, and the peak assignment marking is re-executed on the divided intervals.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a time series is constructed by respiratory voltage and the phase boundary is extracted, transforming respiratory fluctuations into a sortable time reference. This provides a unified basis for attribution of ECG and pulse peaks. By combining peak differences within a window to identify rhythm mismatches and subdivide abnormal intervals, the accuracy of local change characterization is improved. Rearranging time points based on peak distribution can reduce the impact of acquisition beat differences and delays, making the correspondence between multiple signals in the same respiratory stage clearer and enhancing the identification ability and fusion consistency in weak signal and disturbance scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a method for multi-parameter fusion monitoring of neonatal vital signs, comprising the following steps: S1: Obtain the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, convert the continuous voltage signal into a time series, perform difference calculation on adjacent sampling points and extract the position of sign change, mark the time points corresponding to the sign change as the inspiratory start time and expiratory start time respectively, and arrange them in chronological order to generate a set of respiratory phase boundary time points; S2: Based on the set of respiratory phase boundary time points, determine the time intervals corresponding to adjacent time points and divide them into respiratory phase windows. At the same time, acquire the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point. Acquire the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point. Determine the interval assignment between the peak time point and the start and end time of the respiratory phase window to generate a multi-signal phase assignment marker sequence. S3: Extract the number of R wave peaks and pulse wave peaks within the respiratory phase window based on the multi-signal phase attribution marker sequence, calculate the difference in the number of peaks and judge it against the set difference range, perform interval division for respiratory phase windows that exceed the range, segment the window time interval according to the peak distribution, and re-attribute the peak time points of the segments to generate a set of subdivided respiratory phase intervals. S4: Based on the respiratory phase subdivision interval set, read the interval time range, determine the time distribution interval according to the number of R wave peaks in the interval, rearrange the R wave peak time points in the interval, and apply the same rules to the pulse wave peak time points to generate a multi-signal phase aligned time series. S5: Based on the multi-signal phase-aligned time series, the interval time series are spliced ​​together in the order of respiratory phase, and the spliced ​​time axis is established with the corresponding ECG peak value and pulse peak value to generate a vital signs fusion monitoring time axis dataset.

[0020] The set of respiratory phase boundary time points includes the inspiratory start time point sequence and the expiratory start time point sequence; the multi-signal phase attribution label sequence includes ECG peak phase labels, pulse peak phase labels, and phase window index identifiers; the set of respiratory phase subdivision intervals includes subdivision time interval identifiers, interval type classification labels, and interval continuity markers; the multi-signal phase aligned time series includes aligned time reference sequences, ECG rearranged time series, and pulse rearranged time series; the vital signs fusion monitoring time axis dataset includes a unified time axis index, ECG peak mapping data, and pulse peak mapping data.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, perform equal-interval time axis mapping on the continuous voltage signal according to the sampling frequency parameters, sequentially register the sampled values ​​with the corresponding sampling times, and reconstruct them into a single-column time series data according to the time index to generate a voltage time series. Taking 10 seconds of continuous data recorded by the monitoring bed's side acquisition plate as an example, the sampling frequency of the respiratory detection band was set to 50 Hz. This sampling frequency is within the common low-frequency continuous acquisition range for clinical respiratory effort monitoring, covering the respiratory rhythm of 30 to 60 breaths per minute in newborns. After conversion to 50 Hz, the time interval between each adjacent sampling is fixed at 0.02 seconds. Therefore, the first sampling value is registered to 0.00 seconds, the second sampling value to 0.02 seconds, the third sampling value to 0.04 seconds, and so on, until the 500th sampling value corresponds to 9.98 seconds, forming a sequential correspondence between a time index and a voltage value. Subsequently, the original records were cleaned, deleting dropout points, saturation points, and null points. If the voltage at a certain point exceeds 2.5 times the average of the five points before and after it, the point is identified as a motion glitch and rewritten as the median value of the four adjacent points. Baseline drift is checked using a 0.5-second window. When the difference between the first and last volts of the window exceeds 0.08 volts, the mean of the window is taken as the local baseline and subtracted point by point. In the example, the first 10 samples of the original sequence are 1.21, 1.24, 1.28, 1.33, 1.37, 1.40, 1.43, 1.45, 1.44, and 1.41 volts. After cleaning, the 8th point remains at 1.45 volts, the 9th point at 1.44 volts, and the 10th point at 1.41 volts, indicating that there is no dropout or saturation in this segment. The data is then reconstructed into a single-column time series based on the time index. Each row in the column retains only the sampling time and the corresponding voltage, facilitating subsequent point-by-point interpolation. To connect with subsequent examples, this embodiment selects the segment from 0.00 seconds to 2.00 seconds for further explanation. From 0.00 seconds to 0.90 seconds, the voltage rises overall; from 0.92 seconds to 1.84 seconds, the voltage falls overall; and after 1.86 seconds, it rises again. This arrangement is directly written into the voltage time series. Table 1 shows a portion of the mapping data. Subsequent segments are based on this sequence for transition identification and phase boundary correction. The respiratory rate of a newborn at rest is between 30 and 60 breaths per minute. Therefore, when recording at 50 Hz, at least 50 sampling points can be obtained within a single respiratory cycle, which is sufficient to support the localization of fluctuations and transitions.

[0022] Table 1 Voltage Time Series Mapping Table Sampling sequence number Sampling time (seconds) Voltage value (volts) after cleaning 1 0.00 1.21 2 0.02 1.24 3 0.04 1.28 4 0.06 1.33 5 0.08 1.37 6 0.10 1.40 7 0.12 1.43 8 0.14 1.45 9 0.16 1.44 10 0.18 1.41 46 0.90 1.62 47 0.92 1.61 92 1.82 1.09 93 1.84 1.10 94 1.86 1.12 As shown in Table 1, the table lists the key rows from the original sampling to the time mapping. The voltage rises continuously in the first stage, falls back in the middle stage, and rises again in the last stage. These can be directly used as input data for the next step of difference sign reversal judgment.

[0023] S102: Based on the voltage time series, perform difference calculation on adjacent sampling points, record the position where the difference sign changes from positive to negative and from negative to positive, extract and serialize the sampling time corresponding to the sign flip position to obtain the sign inflection time sequence; Starting with the second sample value minus the first, then the third sample value minus the second, and so on, until the end of the sequence. If the difference is greater than 0, it is recorded as an ascending sign; if the difference is less than 0, it is recorded as a descending sign; if the absolute value of the difference is not higher than 0.005 volts, it continues to merge into the same trend and does not generate a separate reversal. Taking the aforementioned 0.00 to 2.00 second segment as an example, 0.02 seconds corresponds to a difference of 0.03 volts, 0.04 seconds corresponds to a difference of 0.04 volts, and 0.06 seconds corresponds to a difference of 0.05 volts, all of which are recorded as ascending signs; 0.92 seconds corresponds to a negative difference, indicating the end of the ascending segment and the beginning of the descending segment, thus forming a sign reversal from positive to negative near 0.92 seconds; at 1.86 seconds, the previous difference is negative, and the next difference turns positive, forming a sign reversal from negative to positive. To avoid false inflection points introduced by slight vibrations in the respiration detection band, three differences are taken before and after each candidate inflection point to form a local judgment segment. A valid inflection point is confirmed only when at least two of the three differences before the candidate point maintain the same sign, and at least two of the three differences after the candidate point maintain the opposite sign. In the example, the differences of the six sampling points before 0.92 seconds are 0.02, 0.02, 0.01, 0.01, 0.01, and 0.00 volts, which are still considered an increase after threshold merging. The differences of the six sampling points after 0.92 seconds are -0.01, -0.02, -0.02, -0.03, -0.02, and -0.02 volts, satisfying the inflection confirmation condition. Therefore, 0.92 seconds enters the sign inflection time sequence. The same condition is met before and after 1.86 seconds, so 1.86 seconds is then written. If 11 flip points are detected within a 10-second recording period, the sequence is arranged chronologically as follows: 0.92, 1.86, 2.78, 3.70, 4.64, 5.58, 6.50, 7.42, 8.34, 9.28, and 9.92 seconds. This sequence only saves the sampling time and flip direction markers, and no longer retains the original difference values. It is then directly used to distinguish between the start of inspiration and the start of expiration.

[0024] S103: Based on the sequence of sign transition moments, mark the moment from negative to positive as the inspiratory start time and the moment from positive to negative as the expiratory start time. Sort the execution times of the two types of marked moments in order and merge them into a set of respiratory phase boundary time points. First, a direction field is added next to the transition record, and then a phase field is generated from the direction field. Taking the aforementioned data as an example, 0.92 seconds corresponds to the transition from positive to negative, recorded as the exhalation start time; 1.86 seconds corresponds to the transition from negative to positive, recorded as the inhalation start time; 2.78 seconds is recorded as the exhalation start time again; and 3.70 seconds is recorded as the inhalation start time again. Then, the two types of marked times are placed in the same time column and reordered. The sorting rule is based solely on the size of the sampled time, not on the phase category. To prevent overlap of two types of marks at the same instant, an adjacent time difference check is performed. When the time difference between two adjacent marks is less than 0.10 seconds, only the record with the higher continuity of the three consecutive differences is retained. The 0.10-second screening interval is set based on the minimum stable span of the five sampling points under the aforementioned 50 Hz sampling condition. In actual testing, after substituting the four intervals of 0.06 seconds, 0.08 seconds, 0.10 seconds, and 0.12 seconds into 30 sets of monitoring records, the false deletion rate corresponding to 0.10 seconds was 3.3%, and the false transition residue rate was 6.7%, which was better than the other three intervals. Therefore, it was fixed at 0.10 seconds. In the example, 0.92 seconds, 1.86 seconds, 2.78 seconds, 3.70 seconds, and 4.64 seconds appeared alternately without triggering overlapping deletion. After processing, a set of respiratory phase boundary time points is obtained, in which each element carries both the boundary time and boundary type. Based on the aforementioned respiratory rate range, a respiratory rate of 30 to 60 breaths per minute corresponds to a single respiratory cycle of approximately 1 to 2 seconds. In the example, the time span between adjacent similar boundaries is approximately 1.84 seconds, which is within an acceptable range. If similar boundary intervals of less than 0.60 seconds or greater than 2.50 seconds occur, a verification mark is added to the end of the current record, and the record is rechecked before proceeding to the next window stage. The respiratory rate of a newborn in a resting state falls within the range of 30 to 60 breaths per minute; therefore, similar boundary intervals falling around 1 to 2 seconds are considered normal for monitoring rhythm. The heart rate of a healthy newborn is commonly between 120 and 160 beats per minute, providing a quantitative reference for subsequent judgment of the number of ECG peak values.

[0025] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the set of respiratory phase boundary time points, determine the corresponding time intervals of adjacent time points and divide them into respiratory phase windows. Extract adjacent boundary time points in chronological order, calculate the interval value between adjacent time points, pair and index the preceding and following time points of each group, and write them into the window index field in sequence to establish a respiratory phase window. First, the first and second boundaries are read to generate the first window. Then, the second and third boundaries are read to generate the second window, and so on, recursively. Each window contains four items: the preceding time point, the following time point, the window duration, and the window sequence number. Taking the aforementioned sequence as an example, the first window lasts 0.94 seconds, from 0.92 seconds to 1.86 seconds; the second window lasts 0.92 seconds, from 1.86 seconds to 2.78 seconds; the third window lasts 0.92 seconds, from 2.78 seconds to 3.70 seconds; and the fourth window lasts 0.94 seconds, from 3.70 seconds to 4.64 seconds. After sorting, a phase label is attached to the window based on the type of its starting boundary. If the starting boundary is the beginning of exhalation, the window is recorded as an expiratory window; if the starting boundary is the beginning of inhalation, the window is recorded as an inhalation window. To ensure uninterrupted window continuity, an adjacent window connection check is performed, requiring the end time of the previous window to be exactly the same as the start time of the next window. If a misalignment of less than 0.02 seconds occurs, the end time of the previous window is corrected according to the start time of the next window. In the example, after the interval from 0.92 seconds to 1.86 seconds, the start time of the next window is also 1.86 seconds, indicating that no correction is needed. For windows with extremely short durations, a minimum window limit of 0.30 seconds is used for verification. This limit is set based on the fact that a single inspiratory or expiratory segment generally does not compress to less than 0.30 seconds at a rate of 60 breaths per minute. When the window duration is less than 0.30 seconds, the window is merged with the adjacent shorter-duration side, and the preceding phase label is retained. In the example, the durations of 0.92 seconds and 0.94 seconds are both greater than 0.30 seconds, and are all directly retained. After establishment, the window index field is recorded as 1, 2, 3, 4, and 5 respectively. All subsequent peak attribution, anomaly detection, and subdivision only refer to the window index and do not return to the original transition sequence.

[0026] S202: Acquire the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point; acquire the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point; perform peak localization on the ECG signal and photoplethysmography signal; record the sampling time corresponding to the local maximum value; and aggregate them into a peak time point sequence. The sampling frequency for the ECG signal was set to 250 Hz, and the sampling frequency for the photoplethysmography (PPG) signal was set to 100 Hz. 250 Hz falls within the acceptable sampling range for heart rate variability analysis, and 100 Hz falls within the common sampling range for PPG, sufficient to support pulse peak identification. First, baseline correction was performed on the ECG signal. The median level was calculated and subtracted using a 0.8-second sliding segment. Then, spikes exceeding three times the local mean were replaced with the average of two adjacent points. Next, a local peak was searched within each 0.20-second scan window. If this point was 0.15 mV higher than the window mean and 10 points higher than either the preceding or following point, it was recorded as a candidate ECG peak. If the time interval between candidate peaks was less than 0.20 seconds, only the peak with the higher amplitude was retained. In the example, ECG peaks were detected at 2.14, 2.54, 2.92, 3.30, and 3.68 seconds between 2.00 and 4.00 seconds, a total of five. The photoplethysmography (PPG) signal is then smoothed and artifact removed. A moving average of five adjacent points is used to suppress high-frequency fluctuations. Local peaks are identified within a 0.30-second scan window. Candidate points must simultaneously exceed the window mean by 0.8% transmittance change and have an interval of at least 0.25 seconds between them and the previous pulse peak. In the example, pulse peaks were detected at 2.22, 2.62, 3.00, 3.38, and 3.76 seconds, a total of five. To ensure that both types of peaks correspond to the actual physiological waveform, the intervals between peaks are checked again. If the heart rate calculated from adjacent ECG peak intervals falls between 100 and 180 beats per minute, it is retained; similarly, if the pulse rate calculated from adjacent PPG peak intervals falls between 100 and 180 beats per minute, it is retained. In the example, the average interval of the five peaks in the ECG was 0.385 seconds, which translates to approximately 156 beats per minute. The average interval of the five peaks in the photoplethysmography (PPG) was also 0.385 seconds, translating to approximately 156 beats per minute. Both fall within the normal neonatal heart rate monitoring range. Table 2 lists the peak location results for this segment. Subsequent assignments will directly compare the time points in the table with the start and end times of the window. The heart rate of healthy newborns is typically between 120 and 160 beats per minute. A sampling frequency of 250 Hz is suitable for more precise peak location. Related studies also indicate that ECG sampling frequencies between 250 Hz and 500 Hz are suitable for heart rate variability analysis, while PPG devices typically have a sampling range of approximately 20 Hz to 100 Hz.

[0027] Table 2. Peak Time Points for Multiple Signals Signal Category Sampling frequency (Hertz) Peak number Peak time (seconds) Interval between adjacent peaks (seconds) electrocardiogram signal 250 1 2.14 - electrocardiogram signal 250 2 2.54 0.40 electrocardiogram signal 250 3 2.92 0.38 electrocardiogram signal 250 4 3.30 0.38 electrocardiogram signal 250 5 3.68 0.38 Photoplethysmography signal 100 1 2.22 - Photoplethysmography signal 100 2 2.62 0.40 Photoplethysmography signal 100 3 3.00 0.38 Photoplethysmography signal 100 4 3.38 0.38 Photoplethysmography signal 100 5 3.76 0.38 As shown in Table 2, both types of peaks show a stable increasing distribution within the same time period, and the interval between adjacent peaks is concentrated between 0.38 seconds and 0.40 seconds, which can be directly used to determine the phase window assignment.

[0028] S203: Based on the respiratory phase window and peak time point sequence, perform interval assignment judgment on the peak time point and the window start time and end time, record the window index and signal category identifier corresponding to the peak time point, arrange all assignment records in chronological order, and generate a multi-signal phase assignment mark sequence; Read the start and end times of window 1, then sequentially read all ECG peak times and PEP peak times. If a peak time is greater than or equal to the window start time and less than the window end time, write that peak into the window 1's assignment record. Then switch to window 2 and continue the same process. Taking window 2 as an example, the range of window 2 is 1.86 seconds to 2.78 seconds. ECG peaks at 2.14 seconds and 2.54 seconds fall into this range, as do PEP peaks at 2.22 seconds and 2.62 seconds. Therefore, write window index 2 into the assignment table, with signal categories of ECG and PEP, and peak times recorded as 2.14, 2.22, 2.54, and 2.62 seconds respectively. Window 3 ranges from 2.78 seconds to 3.70 seconds. ECG peaks at 2.92, 3.30, and 3.68 seconds are assigned to window 3, as are PEP peaks at 3.00 and 3.38 seconds. If the peak value is exactly equal to the end time of the window, it is assigned to the next window, ensuring that one peak value corresponds to only one window. To prevent assignment drift near the boundary due to clock asynchrony, both ECG peaks and photoplethysmography (PPG) peaks undergo unified time base correction before assignment. In this example, the measured starting trigger error of the ECG acquisition reference and blood oxygen acquisition reference is 0.03 seconds. Therefore, the PPG time series is corrected forward by 0.03 seconds before comparing intervals. After correction, the original 3.00-second pulse peak is rewritten as 2.97 seconds, still falling into window 3. After assignment, the signals are reordered according to time sequence to form a multi-signal phase assignment marker sequence. Each record in the sequence contains four items: peak time, window index, phase category, and signal category. In the example, the sorted segments can be written as: 2.14 seconds Window 2 ECG, 2.19 seconds Window 2 PPG, 2.54 seconds Window 2 ECG, 2.59 seconds Window 2 PPG, 2.92 seconds Window 3 ECG, 2.97 seconds Window 3 PPG. This sequence retains both peak time information and phase assignment, allowing for direct classification and counting of the number of peaks of the two types within the same window without repeatedly traversing the original waveform.

[0029] Please see Figure 4 The specific steps of S3 are as follows: S301: Extract the number of R wave peaks and pulse wave peaks within the respiratory phase window based on the multi-signal phase attribution marker sequence, perform classification counting on the marker sequence according to the window index, accumulate the number of peaks corresponding to the ECG and pulse markers in the same window respectively, and pair and record the two types of quantities according to the window order to obtain peak quantity pairing groups; First, all records are grouped by window index. Then, within each group, they are split into two columns based on signal type: ECG records and PSE records, and the counts are accumulated separately. Taking window 2 as an example, there are 4 records, 2 of which are labeled ECG and 2 are labeled PSE. Therefore, the peak count pairings in window 2 are recorded as 2 to 2. Window 3 contains 3 ECG records and 2 PSE records, so window 3 is recorded as 3 to 2. Continuing processing, if window 4 produces 2 to 2 and window 5 produces 3 to 3, then peak count pairing groups arranged in window order can be formed, i.e., window 2 is 2 to 2, window 3 is 3 to 2, window 4 is 2 to 2, and window 5 is 3 to 3. To ensure that the count is not affected by duplicate markings, deduplication of similar peaks is performed before accumulation. If two records of the same signal type exist in the same window with a time difference of less than 0.08 seconds, only the one with the higher amplitude is retained. The 0.08-second deduplication interval was derived from the joint test results of 250 Hz ECG and 100 Hz photoplethysmography (PPG) data. Tests were conducted at four intervals: 0.04 seconds, 0.06 seconds, 0.08 seconds, and 0.10 seconds. 0.08 seconds showed the most stable compression of repetitive peaks. In the example, the three ECG peaks in window 3 were located at 2.92, 3.30, and 3.68 seconds, respectively. The interval between any two points was much greater than 0.08 seconds, so all were retained. After counting, the peak count pairing group not only recorded the number of peaks but also simultaneously retained the start and end times of the window, facilitating direct comparison of the number differences in the next segment. Since the heart rate of healthy newborns typically falls between 120 and 160 beats per minute, the presence of 1 to 3 ECG peaks within a single window is within a reasonable range when the window duration is close to 0.9 seconds. PPG peaks are affected by pulse propagation lag, but under normal conditions, the number of peaks is consistent with or differs from the number of ECG peaks by only one. Therefore, a 3-to-2 pairing in window 3 constitutes a key area for verification in the next segment.

[0030] S302: Calculate the peak number difference based on the peak number pairing group and judge it with the set difference range. Perform the absolute value calculation of the number difference and compare it with the preset peak difference threshold range. Extract the window index corresponding to the threshold range and mark it as an abnormal window to obtain the abnormal window index set. The difference in quantity is obtained by subtracting the number of photoelectrolysis peaks from the number of intraocular peaks within the same window and taking the absolute value. In this embodiment, 0 to 1 is used as the preset peak difference threshold range, determined based on actual measurements from 60 monitoring samples across three batches. During the statistical analysis, the upper limit of the threshold was set to four levels: 0, 1, 2, and 3. Samples were replayed, and the percentage of false alarms and false negatives was recorded. The results showed that when the upper limit was 0, the false alarm rate was 21.7%; when the upper limit was 1, the false alarm rate was 6.7% and the false negative rate was 5.0%; when the upper limit was 2, the false negative rate was 18.3%; and when the upper limit was 3, the false negative rate was 31.7%. Therefore, the range of 0 to 1 was fixed. Taking the 2-to-2 pair in window 2 as an example, the difference in quantity is 0, falling within the range, and is not marked as abnormal. The 3-to-2 pair in window 3 has a difference in quantity of 1, still falling within the range, and is retained as a normal window. If window 6 has a 4-to-2 pair with a difference of 2, it exceeds the upper limit of the range and is written into the abnormal window index set. To provide a more stable screening basis for subsequent segmentation, a window duration consistency check is introduced. If the duration of an abnormal window is simultaneously greater than 1.35 times the average duration of the two preceding and following windows, the abnormality marker is retained; if this multiple is not reached, a manual review marker is added. This 1.35 multiple is determined by comparing three levels: 1.20, 1.35, and 1.50. The 1.35 level can filter out most of the false anomalies caused by slight boundary drift in the sample. Table 3 lists the window difference calculation results in the example, where the number difference of window 6 is 2 and the duration is 1.28 seconds, the average duration of the two preceding and following windows is 0.93 seconds, and the multiple is approximately 1.38, which meets the anomaly confirmation condition. Therefore, the abnormal window index set is written as 6. This comparison result does not terminate the data chain, but instead directly sends window 6 to the next segment for segmentation and reallocation.

[0031] Table 3. Peak Quantity Difference Judgment Table Window Index Start time (seconds) End time (seconds) Number of ECG peaks (number) Number of photoplethysmography peaks (individual) Quantity difference (pieces) Judgment Result 2 1.86 2.78 2 2 0 conventional 3 2.78 3.70 3 2 1 conventional 4 3.70 4.64 2 2 0 conventional 5 4.64 5.58 3 3 0 conventional 6 5.58 6.86 4 2 2 abnormal Referring to Table 3, window 6 exhibits both a large difference in the number of peaks and a long duration, and is therefore extracted into the abnormal window index set for subsequent re-segmentation.

[0032] S303: Perform interval division on the corresponding respiratory phase window according to the abnormal window index set, perform segmentation on the time interval according to the peak time distribution position within the window, re-perform the attribution label on the peak time point within the segment, integrate all segmented intervals in chronological order, and generate a set of subdivided respiratory phase intervals; Taking window 6 as an example, its original range is 5.58 seconds to 6.86 seconds. The ECG peaks within it are located at 5.74, 6.02, 6.34, and 6.66 seconds, while the photoplethysmography peaks are located at 5.82 and 6.40 seconds. First, the positions where the intervals significantly increase are found according to the peak time distribution within the window. The time differences between adjacent peaks are compared. If the time difference between two adjacent peaks is more than 1.6 times the median peak interval within this window, then that interval position is used as the segment boundary. After mixing and sorting all peak times within window 6, the intervals are 5.74, 5.82, 6.02, 6.34, 6.40, and 6.66 seconds, with adjacent intervals of 0.08, 0.20, 0.32, 0.06, and 0.26 seconds. The 0.32-second interval is significantly higher than the lower threshold of 0.32 seconds (1.6 times the median interval of 0.20 seconds). Therefore, a segment boundary is inserted between 6.02 seconds and 6.34 seconds. Window 6 is then divided into intervals 6 to 1 and 6 to 2. Interval 6 to 1 ranges from 5.58 seconds to 6.18 seconds, and interval 6 to 2 ranges from 6.18 seconds to 6.86 seconds. Peak assignment is then redone for each interval: interval 6 to 1 has 2 ECG peaks and 1 photoplethysmography (PPG) peak; interval 6 to 2 has 2 ECG peaks and 1 PPG peak. If the difference in the number of peaks still exceeds the limit, segmentation continues until the difference in the number of peaks in each sub-interval falls back to the 0-1 range. In this example, the difference in the number of peaks reached 1 after both sub-segments, so further segmentation was stopped. The segment boundary of 6.18 seconds is not directly taken as the midpoint, but rather as the midpoint of a significant interval of 6.18 seconds, ensuring that the time between the preceding and following intervals is continuous and does not overlap. All sub-segmentation results are integrated chronologically to generate a set of respiratory phase sub-intervals, recorded in the form of interval index, start time, end time, and original window index. After this processing, the original window 6 no longer participates in subsequent direct alignment; instead, intervals 6 to 1 and 6 to 2 participate in the phase interval calculation separately. For regular windows that are not included in the abnormal window index set, they are rewritten as a single range and written into the same set to ensure data structure consistency during subsequent readings.

[0033] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the set of respiratory phase subdivision intervals, read the interval time range, extract the start and end times of the subdivision interval, calculate the interval duration and perform corresponding registration with the number of R wave peaks in the interval, divide the duration value by the number of R wave peaks to form the interval value, and write it into the time interval field according to the interval index to obtain the phase interval parameter column; Taking interval 6 to 1 as an example, the start time is 5.58 seconds, the end time is 6.18 seconds, the interval duration is 0.60 seconds, and the number of ECG peaks in the interval is 2. Therefore, the time interval between each adjacent alignment position is 0.30 seconds. For interval 6 to 2, the start time is 6.18 seconds, the end time is 6.86 seconds, the interval duration is 0.68 seconds, and the number of ECG peaks in the interval is 2. Therefore, the time interval is 0.34 seconds. For the regular window 3, the start time is 2.78 seconds, the end time is 3.70 seconds, the duration is 0.92 seconds, the number of ECG peaks is 3, and the corresponding time interval is approximately 0.31 seconds. To prevent the time interval from being artificially large due to too few ECG peaks, the minimum number of ECG peaks is set to 1. If no ECG peak is detected in a certain interval, the interval will not directly enter the alignment, but will return to the previous segment for re-subdivision. In this embodiment, the number of ECG peaks in all intervals is not less than 1. The obtained time interval parameters were then checked for range. Based on the common range of 120 to 160 beats per minute for healthy newborns, the interval between a single heartbeat was calculated to be approximately 0.375 to 0.50 seconds. Since the interval uses phase-based time interval allocation, a time interval between 0.20 and 0.50 seconds was considered valid. A value below 0.20 seconds indicated overly dense peaks, while a value above 0.50 seconds indicated sparse peaks, both requiring further review. The values ​​of 0.30 seconds for interval 6 to 1, 0.34 seconds for interval 6 to 2, and 0.31 seconds for window 3 were all within the valid range. These values, along with the interval indices, were written into the phase interval parameter column. This parameter column directly determines the unified timing position of each ECG peak within the interval and serves as the baseline data for peak time rearrangement.

[0034] S402: Rearrange the R-wave peak time points within the interval according to the phase interval parameter column, extract the corresponding R-wave peak time points of the sub-interval, arrange the peak number in chronological order, and perform position mapping between the number and the interval value to reconstruct the peak time sequence position within the interval and establish an ECG aligned time sequence. Instead of directly using the absolute times of the original ECG peaks, the starting time of the interval is used as the zero reference. The ECG peaks within the interval are numbered chronologically, and these numbers are then matched one-to-one with the aforementioned time intervals to form the aligned timing positions within the interval. Taking interval 6 to 1 as an example, the interval starts at 5.58 seconds, with a time interval of 0.30 seconds. The original times of the ECG peaks within this interval are 5.74 seconds and 6.02 seconds, which, after being sorted by time, are numbered as peak 1 and peak 2. Therefore, the aligned timing positions are written as 0.30 seconds and 0.60 seconds, respectively. For interval 6 to 2, the starting time is 6.18 seconds, with a time interval of 0.34 seconds. The original times of the ECG peaks are 6.34 seconds and 6.66 seconds, which, after sorting, are written as 0.34 seconds and 0.68 seconds. Window 3 corresponds to an interval time interval of 0.31 seconds, and the three ECG peaks are written as 0.31 seconds, 0.62 seconds, and 0.93 seconds. To prevent order misalignment caused by forced straightening of locally crowded original peaks, a sequence consistency check is performed, requiring that the temporal order between adjacent original peaks be completely consistent with the aligned numbering order; if cross-ordering occurs, the original order is used for renumbering. In this embodiment, the original peak values ​​in each interval are monotonically increasing, without triggering rearrangement correction. Subsequently, the interval index, the original time of the ECG peak, and the alignment position of the ECG peak are combined into an ECG alignment time sequence. The alignment position in this sequence uses relative time within the interval, not global absolute time. This normalizes different interval lengths to a unified phase expansion scale, making it easier to map pulse peaks to corresponding positions according to the same sequence number in the next segment. For example, the original ECG peaks in window 3 are not equidistant at 2.92, 3.30, and 3.68 seconds, but after being written as 0.31, 0.62, and 0.93 seconds, subsequent segments can form a one-to-one positional relationship with the pulse peaks without having to repeatedly calculate global drift.

[0035] S403: Based on the ECG-aligned time series, read the corresponding subdivision interval identifier, perform same-sequence mapping and intra-interval order rearrangement on the pulse wave peak time point, write the pulse wave peak time point into the corresponding phase position, and aggregate and arrange it with the ECG time series position according to the interval index to generate a multi-signal phase-aligned time series. Taking interval 6 to 1 as an example, there is only one photoplethysmography (PPE) peak at 5.82 seconds, while there are already two ECG alignment positions at 0.30 seconds and 0.60 seconds respectively. In this case, the nearest sequence number mapping rule is used, mapping 5.82 seconds to position 1 (0.30 seconds), while marking position 2 as PPE missing. Interval 6 to 2 also has only one PPE peak at 6.40 seconds, which is also mapped to position 1 (0.34 seconds), with position 2 missing. For the interval corresponding to window 3, there are originally two PPE peaks and three ECG alignment positions. Therefore, 2.97 seconds is mapped to position 1 (0.31 seconds), 3.38 seconds is mapped to position 2 (0.62 seconds), and position 3 is marked as missing. If the number of PPE peaks in a certain interval exceeds the number of ECG peaks, the first few peaks with the same ECG sequence number are retained, and the remaining peaks are sent back to the abnormal interval for verification. After mapping, the ECG and PEP alignment positions are aggregated and arranged according to the interval index to form a multi-signal phase-aligned time series. Each interval in this series contains several phase positions, and each position can be associated with the original time of both the ECG peak and the PEP peak. If no corresponding value is found, it is indicated by a missing field. After this arrangement, subsequent splicing does not require cross-signal peak matching; it only requires moving the phase positions within the interval to a unified time axis. In the example, the multi-signal phase-aligned record in window 3 can be written as follows: position 1, 0.31 seconds, corresponds to ECG 2.92 seconds and PEP 2.97 seconds; position 2, 0.62 seconds, corresponds to ECG 3.30 seconds and PEP 3.38 seconds; position 3, 0.93 seconds, corresponds to ECG 3.68 seconds and PEP is missing. This structure has completed the time-series alignment and interval aggregation of the two types of peaks, meeting the requirements for the next continuous splicing segment.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the multi-signal phase-aligned time series, the interval time series are spliced ​​together according to the respiratory phase order. The corresponding phase index of the multi-signal phase-aligned time series is extracted. The start and end times of adjacent intervals are read sequentially. The start time of the next sequence is corrected to the position after the end time of the previous sequence. The time series are continuously written into a unified time axis according to the phase order to generate the interval splicing time axis. First, take the relative time column of the first interval as the starting segment. Then, shift the starting position of the second interval to after the ending position of the first interval. The third interval is then appended after the second interval. Taking window 3 as an example, its alignment positions are 0.31, 0.62, and 0.93 seconds. If the alignment positions of the next interval from 6 to 1 are 0.30 and 0.60 seconds, then during splicing, the first position of interval 6 to 1 is rewritten to 1.23 seconds, and the second position is rewritten to 1.53 seconds. This is because the previous segment ends at 0.93 seconds, and adding 0.30 and 0.60 seconds results in consecutive positions of 1.23 and 1.53 seconds. Then, append the 0.34 and 0.68 seconds of interval 6 to 2 after 1.53 seconds, resulting in rewritten times of 1.87 and 2.21 seconds. To avoid overlap or regression between adjacent intervals, the start time of the next sequence is corrected before splicing. The correction rule is to maintain at least a 0.01-second interval between the first position of the next sequence and the end position of the previous sequence. If the interval between the two is less than 0.01 seconds after splicing, the entire next sequence is shifted backward to a position that meets the interval requirement. In the example, 0.93 seconds is followed by 1.23 seconds, with an interval of 0.30 seconds, requiring no additional correction. After splicing, an interval splicing timeline is generated. The timeline only saves uniform coordinates and no longer distinguishes between differences in the original interval lengths. After this processing, multiple intervals that were originally scattered in different respiratory phase windows are connected into a single continuous timeline, and all subsequent peak positions can be projected onto this timeline for full sequence arrangement. The continuous coordinate segments obtained in the example are 0.31, 0.62, 0.93, 1.23, 1.53, 1.87, and 2.21 seconds, which are subsequently linked with ECG and photoplethysmography peak markers, respectively.

[0037] S502: Based on the interval splicing time axis, retrieve the ECG peak identifier and pulse peak identifier corresponding to the time position, perform position registration on the peak occurrence time according to the time index, map the ECG peak time point to the corresponding coordinate of the splicing time axis, map the pulse peak time point to the coaxial coordinate, record the peak category identifier, and establish a peak time axis correspondence table; First, iterate through each position record in the multi-signal phase-aligned time series, extract the unified time axis coordinates to which that position belongs, and then write the original time of the ECG peak or the original time of the PEP peak into the corresponding field. Taking the aforementioned spliced ​​coordinate 0.31 seconds as an example, it corresponds to the first phase position of window 3, so the ECG peak is written at 2.92 seconds and the PEP peak at 2.97 seconds; the 0.62-second position corresponds to ECG at 3.30 seconds and PEP at 3.38 seconds; the 0.93-second position corresponds to ECG at 3.68 seconds while the PEP is empty; the 1.23-second position corresponds to the first position of interval 6 to 1, so ECG at 5.74 seconds and PEP at 5.82 seconds are written simultaneously; the 1.53-second position corresponds to ECG at 6.02 seconds and the PEP is empty. Each unified coordinate also needs to record the peak category identifier. If two types of peaks exist simultaneously under the same coordinate, the category field is written as dual signal; if there is only one type of peak, it is written as single signal. To reduce the burden of subsequent retrieval, original interval indexes and original window indexes are added to form a traceable structure. If the original times of the ECG peak and photoplethysmography peak corresponding to a certain unified coordinate differ by more than 0.25 seconds, a time difference exceeding the limit is added to that row. The 0.25-second setting is based on the empirical range that the peripheral pulse propagation lag in newborns is generally significantly less than the median cardiac cycle. In this embodiment, the time difference at the 0.31-second position is 0.05 seconds, and the time difference at the 0.62-second position is 0.08 seconds, both within the limit. After completing all mappings, a peak time axis correspondence table is established. The unified coordinates in this table are responsible for reflecting the full sequence sorting position, and the original peak times are responsible for preserving the actual monitoring traces. Together, they provide an index framework for the fused dataset.

[0038] S503: Based on the peak time axis correspondence table, perform same-index aggregation on the spliced ​​time axis coordinates, ECG peak position, and pulse peak position, write the peak category and peak time corresponding to the time coordinates into a unified data structure in sequence, and complete the full sequence arrangement according to the respiratory phase order to generate a vital signs fusion monitoring time axis dataset. Each row should contain at least seven items: unified time axis coordinates, phase index, original window index, original interval index, original ECG peak time, original photoplethysmography peak time, and peak category identifier. Taking the aforementioned segment as an example, the row at coordinate 0.31 seconds contains phase index 3, original window index 3, original interval index 3, ECG peak time 2.92 seconds, photoplethysmography peak time 2.97 seconds, and the category is written as dual signal. The row at coordinate 0.93 seconds also contains phase index 3, but the photoplethysmography peak field is empty, and the category is written as ECG single signal. The row at coordinate 1.23 seconds contains original window index 6, original interval indexes 6 to 1, ECG peak time 5.74 seconds, photoplethysmography peak time 5.82 seconds, and the category is written as dual signal. After completing the same aggregation for all coordinates, the vital signs fusion monitoring time axis dataset is generated. This dataset can be used for rhythm verification or disease playback, but in this embodiment, data storage is completed up to this point. To verify the integrity of the dataset, a final entry count check is performed, requiring that the number of time axis coordinates matches the total number of ECG alignment positions. If any rows with missing photoplethysmography (PPG) values ​​exist, their corresponding coordinates must be retained and cannot be deleted. In the example, the splicing axis has 7 coordinates, and there are also 7 ECG alignment positions, therefore the consistency check passes. The PPG has only 4 valid peaks, but the 3 missing rows are still fully preserved. The resulting dataset simultaneously possesses full sequence order, respiratory phase correlation, and dual-signal correspondence. Thus, the vital signs fusion monitoring time axis formed by the respiratory detection band, ECG electrode patch, and blood oxygen saturation probe is now complete.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for multi-parameter fusion monitoring of neonatal vital signs, characterized in that, Includes the following steps: S1: Obtain the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, convert the continuous voltage signal into a time series, calculate the difference between adjacent sampling points and extract the position of sign change, mark the time point of sign change as the inspiratory start time and expiratory start time, and generate a set of respiratory phase boundary time points by sorting them by time. S2: Based on the set of respiratory phase boundary time points, determine adjacent time intervals and divide respiratory phase windows, obtain the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point, obtain the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point, perform peak time point and window start and end time interval assignment judgment, and generate multi-signal phase assignment marker sequence; S3: Based on the multi-signal phase attribution marker sequence, count the number of R wave peaks and pulse wave peaks within the respiratory phase window, calculate the difference in the number of peaks and compare it with the set difference range, perform interval division on the out-of-range window, segment according to the peak distribution and re-mark the peak attribution, and generate a set of subdivided respiratory phase intervals. S4: Based on the set of respiratory phase subdivision intervals, read the interval time range, determine the time distribution interval according to the number of R wave peaks in the interval, rearrange the R wave peak time points and process the pulse wave peak time points according to the same rule to generate a multi-signal phase-aligned time sequence. S5: Based on the multi-signal phase-aligned time series, the interval time series is spliced ​​in the order of respiratory phase, and the correspondence between the spliced ​​time axis and the ECG peak value and pulse peak value is established to generate a vital signs fusion monitoring time axis dataset.

2. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that: The set of respiratory phase boundary time points includes the inspiratory start time point sequence and the expiratory start time point sequence; the multi-signal phase attribution label sequence includes ECG peak phase labels, pulse peak phase labels, and phase window index identifiers; the set of respiratory phase subdivision intervals includes subdivision time interval identifiers, interval type classification labels, and interval continuity markers; the multi-signal phase alignment time series includes alignment time reference sequences, ECG rearrangement time series, and pulse rearrangement time series; the vital signs fusion monitoring time axis dataset includes a unified time axis index, ECG peak mapping data, and pulse peak mapping data.

3. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that: The process of calculating the difference between adjacent sampling points and extracting the symbol change position includes sampling the continuous voltage signal at equal intervals according to the sampling frequency parameters to form a discrete sequence, and determining the symbol of the difference between adjacent sampling points. When the difference between adjacent sampling points changes from a positive value to a negative value or from a negative value to a positive value, the corresponding sampling time is recorded as a candidate symbol change time point. The candidate symbol change time points are then filtered by time interval, and time points with an interval less than a preset minimum time interval threshold are removed.

4. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that: The process of determining the peak time point and the start and end time interval of the window includes comparing the R wave peak time point and the pulse wave peak time point with the start and end times of the corresponding respiratory phase window, respectively. When the peak time point is located between the start and end times of the respiratory phase window, the corresponding window identifier is assigned, and a multi-signal phase attribution marker sequence is formed in chronological order.

5. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire the output voltage signal and sampling frequency parameters of the respiratory detection belt of the neonatal monitoring bed, perform equal-interval time axis mapping on the continuous voltage signal according to the sampling frequency parameters, sequentially register the sampled values ​​with the corresponding sampling times, and reconstruct them into a single-column time series data according to the time index to generate a voltage time series. S102: Based on the voltage time series, perform difference calculation on adjacent sampling points, record the position where the difference sign changes from positive to negative and from negative to positive, extract and serialize the sampling time corresponding to the sign reversal position to obtain the sign reversal time sequence; S103: Based on the symbolic transition time sequence, the time from negative to positive is marked as the inhalation start time, and the time from positive to negative is marked as the exhalation start time. The execution times of the two types of marked times are sorted sequentially and merged into a set of respiratory phase boundary time points.

6. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the set of respiratory phase boundary time points, determine the corresponding time intervals of adjacent time points and divide them into respiratory phase windows. Extract adjacent boundary time points in chronological order, calculate the interval value between adjacent time points, pair and index each group of preceding and following time points, and write them into the window index field in sequence to establish a respiratory phase window. S202: Acquire the ECG signal output from the neonatal ECG electrode patch and extract the R wave peak time point; acquire the photoplethysmography signal output from the neonatal blood oxygen saturation probe and extract the pulse wave peak time point; perform peak localization on the ECG signal and the photoplethysmography signal; record the sampling time corresponding to the local maximum value; and aggregate them into a peak time point sequence. S203: Based on the respiratory phase window and the peak time point sequence, perform interval assignment judgment on the peak time point and the window start time and end time, record the window index and signal category identifier corresponding to the peak time point, arrange all assignment records in chronological order, and generate a multi-signal phase assignment mark sequence.

7. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Extract the number of R wave peaks and pulse wave peaks within the respiratory phase window according to the multi-signal phase attribution marker sequence, perform classification counting on the marker sequence according to the window index, accumulate the number of peaks corresponding to the ECG identifier and pulse identifier in the same window respectively, and pair and record the two types of quantities according to the window order to obtain peak quantity pairing groups; S302: Calculate the peak number difference based on the peak number pairing group and judge it with the set difference range. Perform the absolute value calculation of the number difference and compare it with the preset peak difference threshold range. Extract the window index corresponding to the threshold range and mark it as an abnormal window to obtain the abnormal window index set. S303: Perform interval division on the corresponding respiratory phase window according to the abnormal window index set, perform segmentation on the time interval according to the peak time distribution position within the window, re-perform the attribution mark on the peak time point within the segment, integrate all segmented intervals in chronological order, and generate a set of subdivided respiratory phase intervals.

8. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the set of respiratory phase subdivision intervals, read the interval time range, extract the start and end times of the subdivision intervals, calculate the interval duration and perform corresponding registration with the number of R wave peaks in the interval, divide the duration value by the number of R wave peaks to form the interval value, and write it into the time interval field according to the interval index to obtain the phase interval parameter column; S402: Rearrange the R-wave peak time points in the interval according to the phase interval parameter list, extract the R-wave peak time points corresponding to the sub-interval, arrange the peak number in chronological order, and perform position mapping between the number and the interval value to reconstruct the peak time sequence position in the interval and establish an ECG aligned time sequence. S403: Based on the ECG-aligned time series, read the corresponding subdivision interval identifier, perform same-sequence mapping and interval-in-order rearrangement on the pulse peak time point, write the pulse peak time point into the corresponding phase position, and aggregate and arrange it with the ECG time series position according to the interval index to generate a multi-signal phase-aligned time series.

9. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: According to the multi-signal phase-aligned time series, the interval time series are spliced ​​together in the order of respiratory phases. The phase index corresponding to the multi-signal phase-aligned time series is extracted. The start and end times of adjacent intervals are read in sequence. The start time of the next sequence is corrected to the position after the end time of the previous sequence. The time series are continuously written into a unified time axis in phase order to generate an interval splicing time axis. S502: Based on the interval splicing time axis, retrieve the ECG peak identifier and pulse peak identifier corresponding to the time position, perform position registration on the peak occurrence time according to the time index, map the ECG peak time point to the corresponding coordinate of the splicing time axis, map the pulse peak time point to the coaxial coordinate, record the peak category identifier, and establish a peak time axis correspondence table; S503: Based on the peak time axis correspondence table, perform same-index aggregation on the spliced ​​time axis coordinates, ECG peak position, and pulse peak position, write the peak category and peak time corresponding to the time coordinates into a unified data structure in sequence, and complete the full sequence arrangement according to the respiratory phase order to generate a vital signs fusion monitoring time axis dataset.

10. The method for multi-parameter fusion monitoring of neonatal vital signs according to claim 1, characterized in that: The process of calculating the difference in the number of peaks and comparing it with a set difference range includes comparing the difference between the number of R-wave peaks and the number of pulse peaks in the respiratory phase window with the upper and lower limits of a preset difference range. When the difference exceeds the upper and lower limits of the preset difference range, the respiratory phase window is divided into two or more intervals according to the peak time distribution position, and the peak assignment label is re-executed for the divided intervals.

Images