Multi-parameter dynamic monitoring method for critical patients

Abstract

This invention relates to the field of physiological signal monitoring technology, specifically a method for multi-parameter dynamic monitoring of critically ill patients, comprising the following steps: acquiring multi-source waveforms and constructing a cardiac cycle sequence; segmenting the time axis according to the cycle to extract frequency domain and phase features to form a feature set; normalizing and resampling the phase of the arterial pressure segment and calculating the phase difference to construct a differential sequence; determining the coupling state based on phase polarity and combining it with cycle energy to generate a change sequence; screening abnormal time windows and fusing blood flow variability to form a monitoring index; and combining gradient changes and state transition decoding to correct anomalies and obtain multi-parameter monitoring results. In this invention, by uniformly segmenting multi-source waveforms of the cardiac cycle and extracting frequency domain and phase features to form a correlation expression, determining the coupling state based on phase difference and screening abnormal time windows based on energy changes, anomaly identification is achieved and blood flow variability is fused to construct an index. Correction and estimation are completed through gradient changes and state transition decoding, thereby improving the stability and consistency of monitoring.

Description

Technical Field

[0001] This invention relates to the field of physiological signal monitoring technology, and in particular to a multi-parameter dynamic monitoring method for critically ill patients. Background Technology

[0002] The field of physiological signal monitoring technology involves the acquisition, transmission, and analysis of human vital signs information. It mainly includes the continuous detection of various parameters such as electrocardiogram signals, blood pressure values, blood oxygen saturation, respiratory rate, and body temperature. Typically, sensors are attached to or implanted on the surface or inside the human body to convert the collected electrical or physical signals into processable voltage or digital signals. These signals are then transmitted to monitoring equipment via leads or wireless means. The processing unit then records the data, performs numerical calculations, and interprets the status, thereby forming a continuous observation system for an individual's physiological state. This technology is widely used in scenarios such as intensive care, postoperative recovery, and chronic disease management.

[0003] The traditional multi-parameter dynamic monitoring method for critically ill patients refers to the process of collecting electrocardiogram waveforms through electrocardiogram electrodes, measuring blood pressure using cuff or invasive catheters, detecting blood oxygen saturation through finger clip sensors, and obtaining respiratory rate by combining respiratory sensors in an intensive care environment. The data collected from multiple independent sources are input into the monitor for time-series recording. At the same time, multiple parameters are compared one by one according to preset thresholds, and the waveforms and numerical changes are displayed on the screen to achieve continuous observation of multiple physiological indicators of the patient.

[0004] Current critical care multi-parameter monitoring mainly relies on the independent acquisition and recording of multiple physiological signals in a time-series manner. Data processing focuses on threshold comparison and waveform display, lacking the ability to analyze the temporal correlation between signals. This makes it difficult to reveal the dynamic coupling relationship between different physiological parameters. Furthermore, the monitoring results depend on fixed threshold judgments, which are easily affected by individual differences and short-term fluctuations, resulting in insufficient sensitivity in identifying abnormal states. At the same time, there is a lack of effective suppression mechanisms for noise interference and abnormal waveforms, which can easily lead to misjudgment or missed judgment, further affecting the continuity and reliability of clinical decision-making. Summary of the Invention

[0005] To achieve the above objectives, the present invention employs the following technical solution: a multi-parameter dynamic monitoring method for critically ill patients, comprising the following steps: S1: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform sequences through intensive care equipment, collect electrocardiogram waveform signals, extract the time points of electrocardiogram peak extreme values ​​to construct cardiac cycle sequences, segment the continuous arterial pressure and central venous pressure according to the cardiac cycle sequences, extract frequency domain and phase features, and generate multi-parameter feature sets; S2: Extract the phase components of the arterial pressure rise and fall segments from the multi-parameter feature set and perform time-normalized resampling. Calculate the phase difference between adjacent cardiac cycle time points based on the resampled data and splice them together to construct a phase difference sequence. S3: Extract the phase difference distribution polarity from the phase difference sequence, calculate the polarity ratio and determine the phase coupling state, extract the single-cycle arterial pressure and time interval and multiply and accumulate them to generate an energy change sequence; S4: Based on the energy change sequence and the overlapping time interval of the phase coupling state, abnormal time windows are screened, the duration of abnormal time windows is calculated and the abnormal duration ratio is extracted, the peak extreme points and trough extreme points are extracted from the central venous pressure and blood oxygen waveform sequences and the blood flow variability is analyzed, and the blood flow variability and abnormal duration ratio are weighted and combined to construct a multidimensional monitoring index sequence. S5: Based on the multidimensional monitoring index sequence, extract the monitoring index at adjacent time points, calculate the gradient change, and perform fixed boundary comparison to extract outlier coordinates. Perform state transition decoding on the multidimensional monitoring index sequence and outlier coordinates to obtain the implicit state estimation sequence. Then, replace the corresponding values ​​of the outlier coordinates to generate the multidimensional monitoring result.

[0006] As a further embodiment of the present invention, the multi-parameter feature set includes frequency domain amplitude features, phase offset features, and period segmentation identifiers; the phase difference sequence includes a phase difference vector, a period splicing index, and a time normalization identifier; the energy change sequence includes a periodic energy accumulation value, an energy fluctuation trend value, and a periodic energy distribution identifier; the multidimensional monitoring index sequence includes an abnormal duration ratio index, a blood flow variability index, and a weighted mapping coefficient; and the multidimensional monitoring result includes an abnormality correction index, a state estimation mapping, and a sequence reconstruction result.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform data sequences by outputting electrical signals from intensive care equipment, collect electrocardiogram waveform electrical signals, extract the time points of electrocardiogram peak extreme values ​​and record the interval values ​​between adjacent extreme values, determine the interval values ​​with the preset cardiac cycle duration threshold, and generate a cardiac cycle sequence. S102: Locate the continuous arterial pressure time axis sampling index position according to the cardiac cycle sequence and simultaneously mark the central venous pressure sampling index position. Determine the cycle boundary of the corresponding sampling index segment and perform time axis segmentation to obtain a single cycle waveform dataset. S103: Extract the amplitude sequence of multi-cycle sampling points based on the single-cycle waveform dataset and calculate the amplitude of discrete frequency components. At the same time, record the phase angle values ​​of the corresponding frequency components and complete the frequency index sequence registration operation to generate a multi-parameter feature set.

[0008] As a further aspect of the present invention, the cardiac cycle duration threshold is determined by extracting the initial cardiac electrode value interval time set, sorting the interval time set in ascending order and extracting the median value of the interval at the center index position, calculating the discrete variance of the multiple values ​​in the interval time set and the median value of the interval, and adding the median value of the interval and the discrete standard deviation value.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Extract the phase components of the arterial pressure rise and arterial pressure decay segments from the multi-parameter feature set, reconstruct the phase components by linear interpolation according to a unified time scale, map the multi-segment phase data to a fixed number of sampling points sequence, and perform amplitude normalization processing on the multi-sampling points to obtain a phase-normalized sampling sequence. S202: Based on the phase-normalized sampling sequence, perform point-by-point difference calculation on the phase values ​​at the corresponding index positions of adjacent cardiac cycles, arrange the multiple difference results in position alignment according to the cycle order, and impose consistency constraints on the difference amplitude to obtain a phase difference vector group; S203: Based on the phase difference vector group, perform head-to-tail splicing on the multi-period difference sequence, concatenate the corresponding position differences of consecutive periods at the sequence level, and reorder the splicing result by index to obtain the phase difference sequence.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Extract the polarity of the phase difference distribution between adjacent cardiac cycles from the phase difference sequence, perform sign discrimination along the sequence index, compare the phase difference values ​​of multiple sampling points with the zero value benchmark and mark the positive and negative attributes, and serialize the marking results according to the cardiac cycle order to obtain the phase polarity distribution set; S302: Based on the phase polarity distribution set, the proportion of polarity markers in adjacent cardiac cycles is statistically analyzed. The number of positive polarity markers and the number of negative polarity markers are counted along the cycle range and the proportions are calculated respectively. The proportions are compared with a preset polarity offset threshold to determine the direction and obtain a polarity offset determination sequence. S303: Based on the polarity offset determination sequence, call the continuous arterial pressure data after segmenting the time axis and obtain the single-cycle arterial pressure and time interval, perform corresponding parameter multiplication operations along the cycle order and accumulate the multiplication results of multiple cycles, rearrange the accumulated results according to the time index, and obtain the energy change sequence.

[0011] As a further aspect of the present invention, the polarity offset threshold is determined by extracting cardiac cycle polarity distribution samples under baseline physiological conditions, statistically analyzing the polarity baseline proportion of multiple sample sequences, calculating the distribution mean and deviation standard deviation of all baseline proportion values, and weighting and summing the distribution mean and deviation standard deviation according to a preset tolerance weight coefficient.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the energy change sequence and the phase coupling state, perform an intersection comparison of overlapping time windows, match the corresponding indices of the energy change sequence and the phase coupling state within multiple time windows point by point, calculate the ratio of the intersection coverage length to the total length of the time window, and determine it with a preset anomaly identification threshold to generate an abnormal time window set. S402: Extract the timestamps corresponding to the start and end indices of multiple time windows from the abnormal time window set, calculate the duration period and perform duty cycle calculation to obtain the abnormal duration ratio, call the central venous pressure and blood oxygen waveform sequences and extract the peak and trough extreme points, calculate the adjacent extreme value difference sequence and normalize it to obtain the blood flow variability sequence. S403: Call the blood flow variability sequence and the abnormal duration ratio to perform synchronous index alignment, perform weighted summation on the corresponding position values ​​and map them to a unified value range, and serialize the mapping results in chronological order to obtain a multidimensional monitoring index sequence.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Extract the monitoring index at adjacent time points based on the multidimensional monitoring index sequence, perform differential operation on the monitoring index at adjacent time points to obtain the gradient change, filter the position index corresponding to the gradient change exceeding the preset fixed boundary value limit interval and integrate it with spatial coordinate encoding to obtain the outlier noise coordinate set. S502: Call the outlier coordinate set and the multidimensional monitoring index sequence to construct a state sequence mapping structure, determine the transition relationship of the state identifiers corresponding to multiple time points and recursively calculate the location of the outlier coordinate set, encode and reorganize the state transition path, and obtain the implicit state estimation sequence. S503: Extract the state prediction value at the corresponding time point based on the implicit state estimation sequence, perform numerical traversal and comparison on the corresponding position of the outlier noise coordinate set inside the multidimensional monitoring index sequence, replace the abnormal index value at the corresponding position with the state prediction value and re-sort it to generate the multidimensional monitoring result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a unified time segmentation of multi-source waveforms is performed by constructing a cardiac cycle, and frequency domain and phase features are extracted to form a cross-parameter correlation expression, which enables the characterization of the intrinsic coupling relationship between different physiological signals. At the same time, coupling state determination is realized based on phase difference sequence and polarity distribution, and dynamic identification of abnormal states is realized by combining energy change and abnormal time window screening mechanism. On this basis, a comprehensive index is constructed by integrating blood flow variability and abnormal duration ratio, and abnormal correction and state estimation are completed through gradient change and state transition decoding, which effectively improves the stability and interpretation consistency of monitoring results. 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 2 This 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 dynamic monitoring of critically ill patients, comprising the following steps: S1: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform sequences through intensive care equipment, collect electrocardiogram waveform signals, extract the time points of electrocardiogram peak extreme values ​​to construct cardiac cycle sequences, segment the continuous arterial pressure and central venous pressure according to the cardiac cycle sequences, extract frequency domain and phase features, and generate multi-parameter feature sets; S2: Extract the phase components of the arterial pressure rise and fall segments from the multi-parameter feature set and perform time-normalized resampling. Calculate the phase difference between adjacent cardiac cycle time points based on the resampled data and splice them together to construct a phase difference sequence. S3: Extract the phase difference distribution polarity from the phase difference sequence, calculate the polarity ratio and determine the phase coupling state, extract the single-cycle arterial pressure and time interval and multiply and accumulate them to generate an energy change sequence; S4: Based on the energy change sequence and the overlapping time interval of the phase coupling state, abnormal time windows are screened, the duration of abnormal time windows is calculated and the abnormal duration ratio is extracted, the peak and trough extreme points of the central venous pressure and blood oxygen waveform sequences are extracted and the blood flow variability is analyzed, and the blood flow variability and abnormal duration ratio are weighted and mapped to construct a multidimensional monitoring index sequence. S5: Extract monitoring indices at adjacent time points based on the multidimensional monitoring index sequence, calculate the gradient change and perform fixed boundary comparison to extract outlier coordinates, perform state transition decoding on the multidimensional monitoring index sequence and outlier coordinates to obtain the implicit state estimation sequence, and perform replacement on the corresponding values ​​of the outlier coordinates to generate multidimensional monitoring results.

[0020] The multi-parameter feature set includes frequency domain amplitude features, phase shift features, and period segmentation identifiers; the phase difference sequence includes phase difference vectors, period splicing indexes, and time normalization identifiers; the energy change sequence includes periodic energy accumulation values, energy fluctuation trend values, and periodic energy distribution identifiers; the multidimensional monitoring index sequence includes abnormal duration ratio indicators, blood flow variability indicators, and weighted mapping coefficients; and the multidimensional monitoring results include abnormality correction indicators, state estimation mappings, and sequence reconstruction results.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform data sequences by outputting electrical signals from intensive care equipment, collect electrocardiogram waveform electrical signals, extract the time points of electrocardiogram peak extreme values ​​and record the interval values ​​between adjacent extreme values, determine the interval values ​​with the preset cardiac cycle duration threshold, and generate a cardiac cycle sequence. The patient's vital signs are collected via the data interface of the intensive care unit (ICU) equipment. Continuous arterial pressure, central venous pressure, and blood oxygen saturation waveforms are read at a sampling frequency of 500 times per second. Mean filtering and noise reduction are performed on each waveform sequence. The sliding filter window length is set to 10 sampling points. The amplitude data within each window is read sequentially, and the summation is performed. The summation is then divided by the window length of 10 to obtain the corrected smooth amplitude for the current central sampling point. Subsequently, ECG waveform signals for the corresponding time period were acquired via ECG leads. The real-time amplitude of each ECG waveform signal was read sequentially over time. When the real-time amplitude of the current sampling point was greater than the amplitudes of the 15 adjacent sampling points before and after it, and this real-time amplitude was greater than the preset ECG peak amplitude benchmark of 1.5 mV, the timestamp corresponding to the current sampling point was recorded as the ECG peak extreme value time point. The aforementioned ECG peak amplitude benchmark of 1.5 mV was determined by continuously collecting resting ECG data from 50 healthy volunteers and calculating the arithmetic mean of their peak amplitudes. The timestamp values ​​of two adjacent ECG peak extreme value time points were extracted sequentially. The interval between adjacent extreme values ​​was calculated by subtracting the previous ECG peak extreme value from the latter ECG peak extreme value. The lower limit of the cardiac cycle duration threshold interval was set to 0.6 seconds, and the upper limit to 1.2 seconds. This threshold interval was set based on the cardiac cycle distribution range of normal sinus rhythm in a clinical ECG statistical database. The calculated interval between adjacent extreme values ​​is compared with a preset cardiac cycle duration threshold range. If the interval is greater than or equal to 0.6 seconds and less than or equal to 1.2 seconds, the corresponding two extreme timestamps are used as the start and end boundaries of one valid cardiac cycle, and stored sequentially to generate a cardiac cycle sequence. In the actual data processing, the first ECG peak extreme time point was read as 2.5 seconds, and the second ECG peak extreme time point was 3.4 seconds. By subtracting 2.5 seconds from 3.4 seconds, the interval between adjacent extreme values ​​was calculated to be 0.9 seconds. Substituting 0.9 seconds into the threshold range, it was confirmed to fall within the valid range of 0.6 seconds to 1.2 seconds, and the corresponding time period was then included in the cardiac cycle sequence.

[0022] S102: Locate the continuous arterial pressure time axis sampling index position based on the cardiac cycle sequence and simultaneously mark the central venous pressure sampling index position. Determine the cycle boundary of the corresponding sampling index segment and perform time axis segmentation to obtain a single-cycle waveform dataset. The generated cardiac cycle sequence is retrieved, and the start and end timestamps recorded are read one by one. The acquired start and end timestamps are multiplied by the baseline sampling frequency of 500 times per second to calculate the start and end sampling index values ​​corresponding to the continuous arterial pressure waveform data sequence, thus locating the sampling index position on the continuous arterial pressure time axis. Using the same time reference, the same start and end sampling index values ​​are used to perform a synchronization marking operation in the central venous pressure waveform data sequence, generating a strict correspondence between multi-source physiological signals in the time dimension. For the marked sampling index segments, cycle boundary determination is performed. The arithmetic mean of the continuous arterial pressure amplitudes of the first 10 sampling points of each segment is extracted as the left boundary baseline pressure value, and the arithmetic mean of the continuous arterial pressure amplitudes of the last 10 sampling points is extracted as the right boundary baseline pressure value. The difference between the left and right boundary baseline pressure values ​​is calculated, and the absolute value is taken to obtain the boundary baseline drift value. A preset waveform segmentation error compensation threshold of 2.5 mmHg is introduced. This threshold is a limit tolerance value obtained by continuously monitoring the baseline fluctuation of arterial pressure in 200 critically ill patients. When the boundary baseline drift value is less than or equal to this waveform segmentation error compensation threshold, the current segment's period boundary consistency is deemed to meet the standard. Subsequently, the original continuous time series is truncated and segmented according to the starting sampling index value and the ending sampling index value to generate independent single-period waveform datasets, as shown in Table 1, the waveform truncation index table below.

[0023] Table 1 Waveform Extraction Index Table 101 1.5 seconds 750th point 1.2 mmHg 102 2.4 seconds 1200th point 1.8 mmHg As shown in Table 1, the starting timestamp value is extracted as 1.5 seconds. Multiplying this by the sampling frequency of 500 times per second, the starting sampling index value is calculated as the 750th sampling point. If the ending timestamp value is 2.3 seconds, the ending sampling index value is calculated using the same multiplication method as the 1150th sampling point. The corresponding left boundary base pressure value is extracted as 75.8 mmHg, and the right boundary base pressure value is 74.6 mmHg. The absolute difference between the two is calculated to obtain a boundary baseline drift value of 1.2 mmHg. Substituting this 1.2 mmHg value into the judgment condition, since it is less than the waveform segmentation error compensation threshold of 2.5 mmHg, time axis segmentation is performed along the 750th and 1150th sampling points, producing the corresponding single-cycle waveform data.

[0024] S103: Extract the amplitude sequence of multi-cycle sampling points based on the single-cycle waveform dataset and perform discrete frequency component amplitude calculation. At the same time, record the phase angle values ​​of the corresponding frequency components and complete the frequency index sequence registration operation to generate a multi-parameter feature set. The pressure amplitude of each continuous arterial pressure waveform sampling point is retrieved sequentially from the generated single-cycle waveform dataset and combined according to the time acquisition order to form a complete multi-cycle sampling point amplitude sequence. A discrete frequency conversion calculation process is initiated on this multi-cycle sampling point amplitude sequence. For the selected target frequency value, the pressure amplitude of each sampling point is multiplied by a time conversion weighting factor. This time conversion weighting factor is obtained by multiplying the current sampling time value by the target angular frequency value to obtain the phase angle, which is then calculated using cosine and negative sine functions. The cosine function result is used as the real part weighting factor, and the negative sine function result is used as the imaginary part weighting factor. Next, the pressure amplitude of all sampling points within the multi-cycle is multiplied by the corresponding real and imaginary part weighting factors, and the products are summed to generate a complex intermediate result for the corresponding target frequency. The real and imaginary parts of the complex intermediate result are extracted, and the squares of the real and imaginary parts are calculated respectively. These two squares are then summed, and the square root of the sum is taken to obtain the precise discrete frequency component amplitude. In the same calculation process, the extracted imaginary part is divided by the real part to obtain the tangent ratio. This tangent ratio is then solved using the arctangent trigonometric function to obtain and record the corresponding frequency component phase angle. Following a frequency range from 0 Hz to 20 Hz, the discrete frequency component amplitudes and phase angles at corresponding frequency nodes are associated and bound, and a frequency index sequence registration operation is performed. The 20 Hz cutoff frequency is a threshold set according to the definition of the effective frequency band for cardiovascular pressure signals by the Medical Device Association. The data after the above registration and filtering is combined and output as a multi-parameter feature set. Based on specific operating parameters, for the calculation node with a target frequency of 2.0 Hz, the real part of the complex intermediate result is analyzed to be 8.0, and the imaginary part is 6.0. Multiplying the real part 8.0 by itself yields 64.0, and multiplying the imaginary part 6.0 by itself yields 36.0. Adding these two values ​​together gives a total of 100.0. Taking the square root of 100.0 gives the discrete frequency component amplitude at that frequency as 10.0. Then, dividing the imaginary part 6.0 by the real part 8.0 yields the tangent ratio 0.75. The arctangent value is then calculated to determine the phase angle of the frequency component. This discrete frequency component amplitude of 10.0 is then associated with the corresponding phase angle value and appended to the 2.0 Hz index position, incorporating it into the final multi-parameter feature set.

[0025] Please see Figure 3 The specific steps of S2 are as follows: S201: Extract the phase components of the arterial pressure rise and arterial pressure decay segments from the multi-parameter feature set, reconstruct the phase components by linear interpolation according to a unified time scale, map the multi-segment phase data to a fixed number of sampling points sequence, and perform amplitude normalization processing on the multi-sampling points to obtain a phase-normalized sampling sequence. The continuous time series of arterial pressure and the corresponding frequency component phase angle values ​​for each cardiac cycle were retrieved one by one from the multi-parameter feature set. First, the minimum and maximum points of the pressure amplitude within each cardiac cycle were located. The time interval between the minimum and maximum points was defined as the arterial pressure rising segment, and the time interval between the maximum point and the minimum point of the next cycle was defined as the arterial pressure decay segment. Then, the frequency component phase angle values ​​within these two segments were extracted and used as the phase components of the arterial pressure rising segment and the arterial pressure decay segment, respectively. To eliminate the problem of inconsistent sampling point numbers due to differences in the duration of different cardiac cycles, a fixed number of sampling points of 100 was set, with the first 40 points allocated to the arterial pressure rising segment and the last 60 points allocated to the arterial pressure decay segment. Based on a unified time scale, linear interpolation is performed on the extracted phase components to calculate the ratio of the current timestamp to the timestamps of the known sampling points before and after it. This ratio is multiplied by the phase component values ​​of the known sampling points before and after it and then summed, thereby accurately mapping the irregular multi-segment phase data to a fixed sequence of 100 sampling points. Next, the maximum and minimum phase values ​​in this sequence are extracted. The numerator is obtained by subtracting the minimum phase value from the current phase value of each sampling point in the sequence, and the denominator is obtained by subtracting the minimum phase value from the maximum phase value. Amplitude normalization is completed by dividing the numerator by the denominator, ultimately yielding a phase-normalized sampling sequence with values ​​ranging from 0 to 1. For example, within a certain arterial pressure attenuation range, if the phase value of the preceding sampling point is 1.2 radians and the phase value of the following sampling point is 1.6 radians, and the interpolation time point is exactly in the middle of the two, then the ratio is 0.5 for each. Multiplying 1.2 radians by 0.5 gives 0.6 radians, and multiplying 1.6 radians by 0.5 gives 0.8 radians. Summing these two gives the interpolated phase value of 1.4 radians. Subsequently, during the normalization process, if the maximum phase value of the sequence is 2.0 radians and the minimum phase value is 0.0 radians, subtracting 0.0 radians from the previously obtained 1.4 radians and dividing by 2.0 radians yields a normalized phase value of 0.7. This 0.7 is stored in the sequence as the final result, representing the normalized relative phase.

[0026] S202: Based on the phase-normalized sampling sequence, perform point-by-point difference calculation on the phase values ​​at corresponding index positions of adjacent cardiac cycles, arrange the multiple difference results in position alignment according to the cycle order, and impose consistency constraints on the difference amplitude to obtain a phase difference vector group; The output phase-normalized sampling sequence is retrieved, and the first and second adjacent cardiac cycles are extracted based on the chronological order of cardiac cycles. For these two adjacent cycles, the corresponding normalized phase values ​​are extracted point-by-point at fixed sampling index positions from the first to the 100th. The normalized phase value of the second cardiac cycle at the current index position is used as the minuend, and the normalized phase value of the first cardiac cycle at the same index position is used as the subtrahend. Point-by-point subtraction is performed to obtain a series of phase fluctuation values ​​reflecting the changes in the cyclic rhythm. The calculated multiple difference results are aligned and arranged according to the cycle order from the first to the 100th to form the original difference sequence. To prevent transient extreme values ​​caused by occasional electromagnetic interference or severe coughing by the patient from intermittently interfering with subsequent analysis, a consistency constraint amplitude threshold of 0.15 is introduced. This threshold is the upper limit of the 95% confidence interval obtained by statistically analyzing the continuous phase difference fluctuation amplitudes of 30 healthy subjects under different exercise states. The absolute value of each phase fluctuation in the original difference sequence is checked to see if it exceeds the consistency constraint amplitude threshold. If the absolute value at any point exceeds 0.15, the sign of the value is retained, and its absolute value is forcibly corrected to 0.15. Through this consistency constraint operation, a smoothed and filtered phase difference vector group is obtained. In the specific calculation example, the previously obtained normalized phase value of 0.7 is substituted as the value at the 60th index position of the first cardiac cycle. Then, the normalized phase value at the 60th index position of the second cardiac cycle is extracted as 0.92. Subtracting 0.7 from 0.92, the phase fluctuation value is calculated to be 0.22. The absolute value of 0.22 is compared with the consistency constraint amplitude threshold of 0.15. Since 0.22 is greater than 0.15, the consistency constraint mechanism is triggered, maintaining its positive sign and forcibly correcting the difference at this point to 0.15. The corrected value of 0.15 is stored sequentially in the 60th index position to form part of the final phase difference vector group. This corrected value of 0.15 represents that the nonlinear mutations that exceed normal physiological fluctuations have been truncated and smoothed, and are directly transformed into stable and continuous effective difference features.

[0027] S203: Based on the phase difference vector group, perform first and last splicing on the multi-period difference sequence, concatenate the corresponding position differences of consecutive periods at the sequence level, and rearrange the splicing results by index to obtain the phase difference sequence; The phase difference vector sets of all consecutive cardiac cycle pairs are obtained. Following the natural extension direction of the patient's vital signs monitoring time axis, the first and last phase difference sequence is concatenated. The first set of phase difference vector sets generated from the first and second cardiac cycles, and the second set of phase difference vector sets generated from the second and third cardiac cycles are extracted. The tail data point of the first set of phase difference vector sets is directly connected to the head data point of the second set of phase difference vector sets, and so on, concatenating the 49 sets of vector data corresponding to 50 consecutive cardiac cycles at the sequence level. After concatenation, the original independent internal indexes of each cycle can no longer represent the precise global position on the time axis; therefore, a global index rearrangement operation is required for the concatenation results. The index of the starting data point of the sequence is set as 1. Following a fixed length of 100 sampling points per vector group, the indices are incremented sequentially. The indices of the first group are set to 1 to 100, and the indices of the second group are set to 101 to 200. By continuously accumulating the basic offset, a long-range time series of 4900 discrete data points is obtained, which is the final output phase difference sequence. For example, substituting the first phase difference vector group containing the correction value 0.15, its corresponding internal local indices are 1 to 100, where the correction value 0.15 is located at position 60. Then, the second phase difference vector group is extracted, and its local indices are also 1 to 100. After performing head-to-tail concatenation and sequence-level chaining, the absolute order of the data points in the original first group remains unchanged, and the global index is established as numbers 1 to 100. The data points in the original second group follow immediately after, and their original first position is updated to the 101st position in the global index after adding the preceding 100 basic offsets. This order continues sequentially. The correction value of 0.15 at the 60th position in the first group is precisely anchored at the 60th absolute position in the global sequence, while the value originally located at the 60th position in the second group is rearranged to the 160th position, thus obtaining a complete and ordered phase difference sequence. This continuously increasing global index represents the continuous reconstruction result of discrete features on a macroscopic time scale.

[0028] Please see Figure 4 The specific steps of S3 are as follows: S301: Extract the polarity of the phase difference distribution between adjacent cardiac cycles from the phase difference sequence, perform sign discrimination along the sequence index, compare the phase difference values ​​of multiple sampling points with the zero value benchmark and mark the positive and negative attributes, and serialize the marking results according to the cardiac cycle order to obtain the phase polarity distribution set; The generated phase difference sequence is retrieved, and the phase fluctuation values ​​corresponding to each discrete data point in the long program sequence are read one by one. The zero reference is strictly set to an absolute value of 0.0, which represents the ideal state where the waveform phase does not shift during propagation on the time axis. For each extracted phase fluctuation value, it is compared with the zero reference of 0.0 along the global index direction of the long program sequence. If the currently read phase fluctuation value is strictly greater than 0.0, the sampling point is determined to have a forward phase lag distribution and is assigned a positive polarity label value of 1. If the currently read phase fluctuation value is strictly less than 0.0, the sampling point is determined to have a backward phase lead distribution and is assigned a negative polarity label value of -1. If the currently read phase fluctuation value is exactly equal to 0.0, a neutral label value of 0 is assigned. Substituting the previously obtained correction value of 0.15 at the 60th position of the global index, this value is extracted and compared with the zero baseline of 0.0. Since 0.15 is greater than 0.0, the forward phase lag condition is met, and therefore a positive polarity marker value of 1 is assigned. Following the natural order of cardiac cycles and the increasing trajectory of the global absolute index, the discrete polarity marker values ​​generated by the sign discrimination and comparison operations performed on each of the 4900 discrete data points are sequentially concatenated and arranged to construct a phase polarity distribution set perfectly aligned with the original time-domain monitoring scale. The aforementioned marker value of 1 at the 60th position is accurately filled into the corresponding index of this set. The advantage of this operational logic is that by directly mapping continuous floating-point differences to discrete polarity signs, it effectively amplifies the rhythm deviation characteristics within the microscopic hemodynamic cycle.

[0029] S302: Based on the phase polarity distribution set, the proportion of polarity markers in adjacent cardiac cycles is statistically analyzed. The number of positive polarity markers and the number of negative polarity markers are counted along the cycle range and the proportions are calculated respectively. The proportions are compared with the preset polarity offset threshold to determine the direction and obtain the polarity offset judgment sequence. Discrete labeled numerical sequences are extracted from the constructed phase polarity distribution set. Using a fixed number of 100 sampling points corresponding to a single cardiac cycle as the basic statistical window span, statistical intervals are divided sequentially along the natural occurrence order of the cardiac cycle. Within any independent cycle interval containing 100 sampling points, a numerical accumulator is used to sum the occurrence counts of positive polarity label value 1 and negative polarity label value -1, respectively, to obtain the total number of positive and negative polarity labels in the current cycle. Subsequently, the total number of positive polarity labels is divided by the basic number of points in the statistical window (100) to obtain the positive polarity occurrence ratio, and the total number of negative polarity labels is divided by the basic number of points (100) to obtain the negative polarity occurrence ratio. A polarity offset threshold of 0.60 is introduced, which is the upper limit tolerance of physiological polarity fluctuations obtained by statistically analyzing dynamic monitoring data from 200 healthy volunteers. The calculated occurrence ratios are compared with the polarity offset threshold of 0.60 for magnitude judgment. Substituting the statistical interval containing the positive polarity marker value of 1 from the previous text into the calculation example, if, after cumulative statistics, the total number of positive polarity markers in this cycle is 75 and the total number of negative polarity markers is 20, dividing 75 by 100 yields a positive polarity occurrence ratio of 0.75, and dividing 20 by 100 yields a negative polarity occurrence ratio of 0.20. The positive polarity occurrence ratio of 0.75 and the polarity shift threshold of 0.60 are used to determine the direction. Since 0.75 is greater than 0.60, a positive polarity shift is established in the current cycle, and a positive shift status marker value of 2 is assigned. If the negative polarity occurrence ratio is greater than 0.60, a negative shift status marker value of -2 is assigned; if neither exceeds this value, a balance marker value of 0 is assigned. The status marker values ​​of each cycle are sequentially integrated according to the cardiac cycle order to generate a polarity shift determination sequence. The advantage of this operational logic is that, by introducing a fixed window span for proportional calculation and a threshold filtering mechanism, the interference of single discrete outliers is eliminated.

[0030] S303: Based on the polarity offset determination sequence, call the continuous arterial pressure data after the time axis is segmented and obtain the single-cycle arterial pressure and time interval. Perform the corresponding parameter multiplication operation along the cycle order and accumulate the multiplication results of multiple cycles. Rearrange the accumulated results according to the time index to obtain the energy change sequence. Based on the cycle index order recorded in the generated polarity offset determination sequence, the corresponding single-cycle waveform dataset and its matching start and end timestamps are synchronously retrieved from the database. For the retrieved single-cycle waveform data, all arterial pressure sampling amplitudes within each cycle are read, and the absolute values ​​of all sampling amplitudes are summed to obtain the single-cycle comprehensive pressure value. Then, the end and start timestamps are extracted, and the start timestamp is subtracted from the end timestamp to calculate the precise single-cycle time interval value. Along the cycle's continuation trajectory, the calculated comprehensive pressure value and the single-cycle time interval value are multiplied by the corresponding parameters to obtain the single-cycle baseline energy value. Subsequently, a sliding accumulation window is established, with the accumulation window spanning 5 consecutive cardiac cycles. The 5 single-cycle baseline energy values ​​within the window's coverage area are summed to derive the local total energy value. Substituting the data from the previous example, the start timestamp of 1.5 seconds and the end timestamp of 2.3 seconds are extracted, and 1.5 seconds is subtracted from 2.3 seconds to obtain a single-cycle time interval value of 0.8 seconds. The absolute values ​​of the continuous arterial pressure sampling amplitudes retrieved within this cycle are summed to obtain a comprehensive pressure value of 7500.0 mmHg. Multiplying this comprehensive pressure value of 7500.0 mmHg by the time interval of 0.8 seconds yields a single-cycle baseline energy value of 6000.0. The single-cycle baseline energy values ​​for the subsequent four cycles are obtained as 5800.0, 6200.0, 5900.0, and 6100.0, respectively. These five values ​​are then summed to calculate a local total energy value of 30000.0. This summation result is associated with the aforementioned state identifier value 2, and the local total energy values ​​are rearranged strictly according to the initial timestamp order on the global time axis to construct the final energy change sequence. The advantage of this operational logic is that by combining pressure amplitude with the time dimension to form an energy concept and supplementing it with windowed summation, it effectively smooths out the oscillations of cardiac energy over a very short period.

[0031] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the intersection comparison of overlapping time windows of energy change sequence and phase coupling state, the corresponding indices of energy change sequence and phase coupling state in multiple time windows are matched point by point, the ratio of intersection coverage length to total time window length is calculated, and the result is determined by a preset anomaly identification threshold to generate an abnormal time window set. Retrieve the generated energy change sequence and the polarity shift determination sequence generated in the previous step. Set the overlapping time window span to 60.0 seconds and the time window sliding step size to 30.0 seconds. Along the global time axis, sequentially overlay the overlapping time window onto the energy change sequence and the polarity shift determination sequence to obtain the data segments within the corresponding time window. For each data segment within the time window, data points in the energy change sequence whose local total energy value exceeds the preset high-energy warning standard (set to 28000.0 here) are identified as energy anomalies, and their corresponding index positions are obtained. Simultaneously, read the status flag value at the same index position in the polarity shift determination sequence. If the status flag value is equal to 2 or negative 2, it indicates the existence of a polarity shift anomaly, and the point is confirmed as a double anomaly matching point. Accumulate the number of all double anomaly matching points within the current time window to obtain the intersection coverage length. Then, extract the total number of data points contained in the current time window as the total time window length. Divide the intersection coverage length by the total time window length to obtain the intersection ratio. Here, a preset anomaly detection threshold of 0.40 is introduced. This threshold is selected by comparing continuous waveforms from 500 clinically diagnosed patients with heart failure using a sliding window method, choosing the optimal cutoff point that balances sensitivity and specificity. The resulting intersection ratio is compared with the preset anomaly detection threshold of 0.40. If the intersection ratio is greater than or equal to 0.40, the current time window is determined to be an abnormal time window. Substituting the previously calculated local total energy value of 30000.0, this value is greater than the high-energy warning standard of 28000.0, so the index where it is located is determined to be an energy anomaly point. The polarity offset status flag value at this index is read synchronously as 2, satisfying the double anomaly matching condition, and the intersection coverage length is counted as 1 point. If 45 double anomaly matching points are obtained within this time window containing 100 data points, dividing 45 by 100 yields an intersection ratio of 0.45. The intersection ratio of 0.45 is compared with the preset anomaly identification threshold of 0.40 for directional determination. Since 0.45 is greater than 0.40, the overlapping time window is identified as an abnormal time window. All overlapping time windows that meet the conditions are saved in time sliding order to construct an abnormal time window set, which covers all time segments with significant hemodynamic deterioration.

[0032] S402: Extract the timestamps corresponding to the start and end indices of multiple time windows from the abnormal time window set, calculate the duration period and perform duty cycle calculation to obtain the abnormal duration ratio, call the central venous pressure and blood oxygen waveform sequences and extract the peak and trough extreme points, calculate the adjacent extreme value difference sequence and normalize it to obtain the blood flow variability sequence. For the generated set of abnormal time windows, the start and end indices of each abnormal time window are extracted and associated with the vital signs monitoring time axis to obtain the corresponding start and end timestamps. The start timestamp is subtracted from the end timestamp to obtain the actual span of the abnormal time window. This is then divided by the average span of a single cardiac cycle (0.8 seconds) to calculate the duration of the cycle covered by the window. Subsequently, the total observation time of the entire monitoring cycle is extracted, and the aforementioned actual span time is divided by the total observation time, performing a duty cycle division operation to obtain the abnormal duration ratio. Simultaneously, continuous central venous pressure (CVP) sampling sequences and blood oxygen waveform sequences within the same time period as the current abnormal time window are synchronously retrieved from the intensive care unit (ICU) database. For the continuous CVP sampling sequences, a first-order difference calculation is performed along the time direction to find the zero-crossing points of the derivative, identifying local maxima as peak extremes and local minima as trough extremes. The pressure amplitude corresponding to each peak extreme is subtracted from the pressure amplitude corresponding to the adjacent trough extreme to obtain the difference between adjacent CVP extreme values. The same operation was performed on the blood oxygen waveform sequence to obtain the difference between adjacent extreme values ​​of blood oxygen. The difference between adjacent extreme values ​​of central venous pressure was selected as the dividend, and the difference between adjacent extreme values ​​of blood oxygen was selected as the divisor. A ratio calculation was performed to obtain the baseline blood flow variability. Then, using range standardization, the baseline variability was obtained by subtracting the historical minimum variability and dividing by the historical variability range. Substituting the data from the previous analysis, if the start time stamp of an abnormal time window is 10.0 seconds and the end time stamp is 42.0 seconds, subtracting 10.0 seconds from 42.0 seconds yields an actual span of 32.0 seconds. Dividing this by the average span of a single cardiac cycle (0.8 seconds) yields a duration of 40 cycles. If the total observation duration is 600.0 seconds, dividing 32.0 seconds by 600.0 seconds yields an abnormal duration ratio of 0.053. In waveform extreme value calculation, if the peak pressure amplitude of the central venous pressure waveform is 15.0 mmHg and the adjacent trough pressure amplitude is 8.0 mmHg, the difference between the two extreme values ​​is 7.0 mmHg, and the peak-trough extreme value difference of the blood oxygen waveform is 4.0 percentage points. Dividing 7.0 mmHg by 4.0 percentage points yields a baseline blood flow variability of 1.75. If the historical minimum variability is 0.50 and the historical variability range is 2.0, subtracting 0.50 from 1.75 and dividing by 2.0 yields a normalized blood flow variability of 0.625. These values ​​are then sequentially entered into the corresponding time indices to construct the final blood flow variability sequence.

[0033] S403: Call the blood flow variability sequence and the abnormal duration ratio for synchronous index alignment, perform weighted summation on the corresponding position values ​​and map them to a unified value range, and serialize the mapping results in chronological order to obtain the multidimensional monitoring index sequence. The generated blood flow variability sequence and the corresponding abnormal duration ratios for each time period were obtained. An alignment and matching mechanism based on a unified time scale was established, scanning second-by-second along the global time axis to synchronously align the variability values ​​at specific time scales in the blood flow variability sequence with the abnormal duration ratios within the same time scale range. A weighted summation operation was performed using variability weighting coefficients and duration ratio weighting coefficients. The variability weighting coefficient was set to 0.6, and the duration ratio weighting coefficient was set to 0.4. These two coefficients were determined through a multivariate linear regression analysis of multidimensional physiological indicators from 300 critically ill patients with complete prognostic labels to obtain the best-fit weighted distribution. The aligned blood flow variability values ​​were multiplied by the variability weighting coefficient of 0.6 to obtain the variability weighted component, and the abnormal duration ratios were multiplied by the duration ratio weighting coefficient of 0.4 to obtain the phenomenon persistence weighted component. These two weighted components were then summed to obtain the comprehensive monitoring characteristic value. To facilitate intuitive clinical interpretation, this comprehensive monitoring characteristic value was mapped to a unified numerical range of 0 to 100. The specific mapping operation involves multiplying the comprehensive monitoring characteristic value by 100 for amplification. If the amplified value exceeds 100, it is forcibly truncated to 100. Following the natural chronological order of occurrence, the mapping results are pieced together and serialized point by point to output the final multidimensional monitoring index sequence. Substituting the normalized blood flow variability value of 0.625 and the abnormal duration ratio of 0.053 obtained earlier, the actual calculation is performed. Multiplying the blood flow variability value of 0.625 by the variability weighting coefficient of 0.6 yields a variability weighted component of 0.375. Multiplying the abnormal duration ratio of 0.053 by the duration ratio weighting coefficient of 0.4 yields a phenomenon duration weighted component of 0.0212. Summing 0.375 and 0.0212 yields a comprehensive monitoring characteristic value of 0.3962. Multiplying 0.3962 by 100 yields a mapping result of 39.62. The value of 39.62 represents the patient's overall hemodynamic risk score at the current monitoring time; a higher score indicates a more severe risk of decompensation and worsening. This value of 39.62 is stored in the corresponding time node of the multidimensional monitoring index sequence, and new index nodes are continuously generated over time for continuous tracking.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S501: Extract monitoring indices at adjacent time points based on the multidimensional monitoring index sequence, perform differential operations on the monitoring indices at adjacent time points to obtain gradient changes, filter the position indices corresponding to the gradient changes exceeding the preset fixed boundary value limit interval and integrate them with spatial coordinate encoding to obtain the outlier noise coordinate set. Retrieve the generated multidimensional monitoring index sequence. Along the time scale of this sequence, read the monitoring index values ​​corresponding to the current time node and the previous adjacent time node point by point. Extract the monitoring index value of the current time node and subtract it from the monitoring index value of the previous adjacent time node, performing a first-order difference operation to obtain the gradient change within the corresponding time step. Through statistical analysis of the range of continuous monitoring index fluctuations in 500 historical stable-phase critically ill patients, a fixed boundary value limit interval is set from -10.0 to +10.0, representing a reasonable tolerance for index fluctuations under normal physiological conditions. Perform a numerical inclusion judgment between the previously calculated gradient change and this fixed boundary value limit interval. If the gradient change is strictly less than -10.0 or strictly greater than +10.0, the current time node is determined to have experienced a non-physiological jump and is considered an outlier. Substitute the previously obtained multidimensional monitoring index value of 39.62 as the value of the current time node, and simultaneously read the monitoring index value of the previous adjacent time node as 22.10. Subtracting 22.10 from 39.62 yields a gradient change of 17.52. This gradient change of 17.52 is extracted and compared to a fixed boundary value range of -10.0 to +10.0. Since 17.52 is greater than +10.0, the value corresponding to this node is determined to be outside the defined range. The time index value of the outlier is extracted, the sequence number on the time axis is used as the x-axis value, and the exponential magnitude is used as the y-axis value. The x-axis and y-axis values ​​are combined to construct a two-dimensional coordinate system. After traversing the entire sequence, the coordinates of all the filtered outliers are collected and reassembled to output the outlier coordinate set.

[0035] S502: Construct a state sequence mapping structure by calling the outlier coordinate set and the multidimensional monitoring index sequence, determine the transition relationship of the state identifiers corresponding to multiple time points, recursively calculate the location of the outlier coordinate set, encode and reorganize the state transition path, and obtain the implicit state estimation sequence. The outlier coordinate set and multidimensional monitoring index sequence were extracted, and continuous monitoring indices at non-noise locations were read as valid observation numerical sequences to construct a state sequence mapping structure. Pathophysiological states were quantified into three discrete levels, assigned the identifiers 1 (stable state), 2 (fluctuating state), and 3 (deteriorating state). 100,000 historical cardiovascular critical care monitoring records with clinical annotations were retrieved, and the frequency of state transitions between adjacent time points was statistically analyzed to establish state transition probability parameters. The probability parameter for maintaining a stable state was set to 0.8, the probability parameter for transitioning from a stable state to a fluctuating state to 0.2, the probability parameter for maintaining a fluctuating state to 0.7, and the probability parameter for transitioning from a fluctuating state to a deteriorating state to 0.3. Starting from the initial position of the sequence along consecutive time points, the state probability distribution parameter at the current time point was multiplied by the corresponding state transition probability parameter to calculate the prior state probability distribution for the next time point. When the time index advances to the outlier location recorded in the outlier coordinate set, the actual observation values ​​are stopped. The optimal state probability value of the previous time point is extracted and multiplied with the state transition probability parameter to complete the one-way recursion of the implicit state of the noise location. Substituting the 150th sampling point of the outlier time index from the previous analysis, the state determination result of the 149th sampling point is a fluctuating state identifier value of 2, and its corresponding state probability parameter is read as 0.9. The fluctuating state maintenance probability parameter of 0.7 is extracted, and 0.9 is multiplied by 0.7 to calculate the recursive probability value of the 150th sampling point being a fluctuating state, which is 0.63. The probability parameter of the fluctuating state transitioning to a deteriorating state of 0.3 is extracted, and 0.9 is multiplied by 0.3 to calculate the recursive probability value of 0.27. By comparing the recursive probability values, the state identifier corresponding to the highest probability is selected. Since 0.63 is greater than 0.27, the state identifier value of this noise location is confirmed as 2. All the derived state identifier values ​​are integrated and recombined in chronological order to generate a continuous implicit state estimation sequence.

[0036] S503: Extract the state prediction value at the corresponding time point based on the implicit state estimation sequence, perform numerical traversal and comparison on the corresponding position of the outlier noise coordinate set inside the multidimensional monitoring index sequence, replace the abnormal index value at the corresponding position with the state prediction value and re-sort it to generate the multidimensional monitoring result. The constructed implicit state estimation sequence is retrieved, and the state identifier values ​​under the corresponding time indices are read one by one. Based on the distribution pattern of the mean values ​​of the multidimensional monitoring indices corresponding to each pathological state in the historical database, a state value mapping benchmark is established. The state prediction benchmark value is set to 25.0 for stable state identifier value 1, 45.0 for fluctuating state identifier value 2, and 75.0 for deteriorating state identifier value 3. The state identifier values ​​in the implicit state estimation sequence are extracted, and the state prediction benchmark value corresponding to the current time point is obtained as the state prediction value, based on the aforementioned state value mapping benchmark. A point-by-point scan is performed within the original multidimensional monitoring index sequence to extract all abnormal index positions recorded in the outlier coordinate set. Abnormal index values ​​corresponding to the abnormal index positions in the original sequence are completely cleared, and the state prediction values ​​for the corresponding time points are extracted and used to overwrite the original positions. For normal index positions not recorded in the outlier coordinate set, their initially calculated monitoring index values ​​are retained. Substituting the outlier noise at the 150th sampling point indexed earlier, the original monitoring index value at this point is 39.62, and it is determined to be a non-physiological jump noise point. The state identifier value corresponding to this index in the implicit state estimation sequence is extracted as 2. According to the state value mapping benchmark, the state prediction benchmark value corresponding to the state identifier value 2 is 45.0. The outlier value of 39.62 at the 150th sampling point of the original sequence is cleared, and the state prediction value of 45.0 is filled in to complete the value replacement. According to the natural chronological order of global timestamps, the discrete node values ​​after replacement and repair operations and the retained original normal values ​​are rearranged and integrated to generate the final multidimensional monitoring result.

[0037] The above are merely specific embodiments 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 dynamic monitoring of critically ill patients, characterized in that, Includes the following steps: S1: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform sequences through intensive care equipment, collect electrocardiogram waveform signals, extract the time points of electrocardiogram peak extreme values ​​to construct cardiac cycle sequences, segment the continuous arterial pressure and central venous pressure according to the cardiac cycle sequences, extract frequency domain and phase features, and generate multi-parameter feature sets; S2: Extract the phase components of the arterial pressure rise and fall segments from the multi-parameter feature set and perform time-normalized resampling. Calculate the phase difference between adjacent cardiac cycle time points based on the resampled data and splice them together to construct a phase difference sequence. S3: Extract the phase difference distribution polarity from the phase difference sequence, calculate the polarity ratio and determine the phase coupling state, extract the single-cycle arterial pressure and time interval and multiply and accumulate them to generate an energy change sequence; S4: Based on the energy change sequence and the overlapping time interval of the phase coupling state, abnormal time windows are screened, the duration of abnormal time windows is calculated and the abnormal duration ratio is extracted, the peak and trough extreme points of the central venous pressure and blood oxygen waveform sequences are extracted and the blood flow variability is analyzed, and the blood flow variability and abnormal duration ratio are weighted and combined to construct a multidimensional monitoring index sequence.

2. The multi-parameter dynamic monitoring method for critically ill patients according to claim 1, characterized in that, The multi-parameter feature set includes frequency domain amplitude features, phase offset features, and period segmentation identifiers; the phase difference sequence includes a phase difference vector, a period splicing index, and a time normalization identifier; the energy change sequence includes a period energy accumulation value, an energy fluctuation trend value, and a period energy distribution identifier; and the multidimensional monitoring index sequence includes an abnormal duration ratio index, a blood flow variability index, and a weighted mapping coefficient.

3. The multi-parameter dynamic monitoring method for critically ill patients according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire continuous arterial pressure, central venous pressure and blood oxygen waveform data sequences by outputting electrical signals from intensive care equipment, collect electrocardiogram waveform electrical signals, extract the time points of electrocardiogram peak extreme values ​​and record the interval values ​​between adjacent extreme values, determine the interval values ​​with the preset cardiac cycle duration threshold, and generate a cardiac cycle sequence. S102: Locate the continuous arterial pressure time axis sampling index position according to the cardiac cycle sequence and simultaneously mark the central venous pressure sampling index position. Determine the cycle boundary of the corresponding sampling index segment and perform time axis segmentation to obtain a single cycle waveform dataset. S103: Extract the amplitude sequence of multi-cycle sampling points based on the single-cycle waveform dataset and calculate the amplitude of discrete frequency components. At the same time, record the phase angle values ​​of the corresponding frequency components and complete the frequency index sequence registration operation to generate a multi-parameter feature set.

4. The multi-parameter dynamic monitoring method for critically ill patients according to claim 3, characterized in that, The cardiac cycle duration threshold is determined by extracting the initial cardiac electrode value interval time set, sorting the interval time set in ascending order, extracting the median value of the interval at the center index position, calculating the discrete variance of the multiple values ​​in the interval time set and the median value of the interval, and adding the median value of the interval and the discrete standard deviation value.

5. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Extract the phase components of the arterial pressure rise and arterial pressure decay segments from the multi-parameter feature set, reconstruct the phase components by linear interpolation according to a unified time scale, map the multi-segment phase data to a fixed number of sampling points sequence, and perform amplitude normalization processing on the multi-sampling points to obtain a phase-normalized sampling sequence. S202: Based on the phase-normalized sampling sequence, perform point-by-point difference calculation on the phase values ​​at the corresponding index positions of adjacent cardiac cycles, arrange the multiple difference results in position alignment according to the cycle order, and impose consistency constraints on the difference amplitude to obtain a phase difference vector group; S203: Based on the phase difference vector group, perform head-to-tail splicing on the multi-period difference sequence, concatenate the corresponding position differences of consecutive periods at the sequence level, and reorder the splicing result by index to obtain the phase difference sequence.

6. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Extract the polarity of the phase difference distribution between adjacent cardiac cycles from the phase difference sequence, perform sign discrimination along the sequence index, compare the phase difference values ​​of multiple sampling points with the zero value benchmark and mark the positive and negative attributes, and serialize the marking results according to the cardiac cycle order to obtain the phase polarity distribution set; S302: Based on the phase polarity distribution set, the proportion of polarity markers in adjacent cardiac cycles is statistically analyzed. The number of positive polarity markers and the number of negative polarity markers are counted along the cycle range and the proportions are calculated respectively. The proportions are compared with a preset polarity offset threshold to determine the direction and obtain a polarity offset determination sequence. S303: Based on the polarity offset determination sequence, call the continuous arterial pressure data after segmenting the time axis and obtain the single-cycle arterial pressure and time interval, perform corresponding parameter multiplication operations along the cycle order and accumulate the multiplication results of multiple cycles, rearrange the accumulated results according to the time index, and obtain the energy change sequence.

7. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 6, characterized in that, The polarity offset threshold is determined by extracting cardiac cycle polarity distribution samples under baseline physiological conditions, statistically analyzing the polarity baseline proportion of multiple sample sequences, calculating the distribution mean and deviation standard deviation of all baseline proportion values, and then weighting and summing the distribution mean and deviation standard deviation according to a preset tolerance weight coefficient.

8. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the energy change sequence and the phase coupling state, perform an intersection comparison of overlapping time windows, match the corresponding indices of the energy change sequence and the phase coupling state within multiple time windows point by point, calculate the ratio of the intersection coverage length to the total length of the time window, and determine it with a preset anomaly identification threshold to generate an abnormal time window set. S402: Extract the timestamps corresponding to the start and end indices of multiple time windows from the abnormal time window set, calculate the duration period and perform duty cycle calculation to obtain the abnormal duration ratio, call the central venous pressure and blood oxygen waveform sequences and extract the peak and trough extreme points, calculate the adjacent extreme value difference sequence and normalize it to obtain the blood flow variability sequence. S403: Call the blood flow variability sequence and the abnormal duration ratio to perform synchronous index alignment, perform weighted summation on the corresponding position values ​​and map them to a unified value range, and serialize the mapping results in chronological order to obtain a multidimensional monitoring index sequence.

9. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 1, characterized in that, The method further includes: S5: Based on the multidimensional monitoring index sequence, extract the monitoring index at adjacent time points, calculate the gradient change and perform fixed boundary comparison to extract outlier coordinates, perform state transition decoding on the multidimensional monitoring index sequence and outlier coordinates to obtain the implicit state estimation sequence, and replace the corresponding values ​​of the outlier coordinates to generate multidimensional monitoring results. The multidimensional monitoring results include anomaly correction indices, state estimation mappings, and sequence reconstruction results.

10. The method for multi-parameter dynamic monitoring of critically ill patients according to claim 9, characterized in that, The specific steps of S5 are as follows: S501: Extract the monitoring index at adjacent time points based on the multidimensional monitoring index sequence, perform differential operation on the monitoring index at adjacent time points to obtain the gradient change, filter the position index corresponding to the gradient change exceeding the preset fixed boundary value limit interval and integrate it with spatial coordinate encoding to obtain the outlier noise coordinate set. S502: Call the outlier coordinate set and the multidimensional monitoring index sequence to construct a state sequence mapping structure, determine the transition relationship of the state identifiers corresponding to multiple time points and recursively calculate the location of the outlier coordinate set, encode and reorganize the state transition path, and obtain the implicit state estimation sequence. S503: Extract the state prediction value at the corresponding time point based on the implicit state estimation sequence, perform numerical traversal and comparison on the corresponding position of the outlier noise coordinate set inside the multidimensional monitoring index sequence, replace the abnormal index value at the corresponding position with the state prediction value and re-sort it to generate the multidimensional monitoring result.

Images