Adaptive calibration system and method for invasive blood pressure signal drift

Abstract

The invention belongs to the technical field of calibration, and discloses a self-adaptive calibration system and method for invasive blood pressure signal drift. The method comprises the steps that original blood pressure, posture and height signals are collected and preprocessed through an invasive sensor, initial blood pressure signals, posture angles and height changes are obtained, and first waveform features are extracted; calibrating the blood density of the patient, judging interference in combination with the attitude angle and height change, and extracting to obtain a second waveform feature; inputting into an attenuation classification model to obtain an attenuation probability to judge an attenuation type, and extracting corresponding blood pressure signals and waveform features according to the attenuation type; calculating a theoretical reference value by taking a calibration reference point, and calculating an actual drift distance calibration blood pressure signal by combining waveform characteristics; finally, non-invasive blood pressure measurement values are synchronized, attenuation classification model parameters are optimized, and accurate calibration is achieved. According to the method, the influence of macroscopic interference on signal judgment is thoroughly eliminated, and accurate distinguishing between microcosmic physical interference and electron drift is realized.

Description

Technical Field

[0001] This invention relates to the field of calibration technology, and more specifically, to an adaptive calibration system and method for invasive blood pressure signal drift. Background Technology

[0002] In clinical settings such as intensive care and cardiovascular surgery, invasive blood pressure monitors can acquire key physiological parameters such as arterial systolic blood pressure, diastolic blood pressure, and pulse waveform in real time and with high accuracy. The accuracy of these monitors is directly related to the accuracy of clinical decisions and patient safety.

[0003] Chinese patent application CN119366892A discloses an invasive blood pressure sensor, pressure correction, and sensitivity calibration method. The invasive blood pressure sensor includes components such as a position correction unit, a pressure sensor, a substrate, and a pre-filling structure. Specifically, the position correction unit uses an angle measurement unit and a height measurement unit to measure the offset angle of the sensor substrate relative to the vertical direction and the change in the vertical height of the substrate relative to a reference point. When performing pressure correction, the effects of sensor tilt and height changes are comprehensively considered. It uses the measured tilt angle, height change, density of the pre-filling medium, and gravitational acceleration to calculate the corresponding pressure compensation value using specific trigonometric functions and hydrostatic formulas. Finally, the original pressure value measured by the sensor is added to these compensation values ​​to output the corrected blood pressure value.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] The above method mainly relies on measuring the macroscopic angle and height changes of the sensor substrate and compensating for the hydrostatic pressure of the fluid through a fixed physical formula. It cannot identify and distinguish the microscopic displacement of the sensor catheter tip in the blood vessel, such as the local pressure signal attenuation and waveform distortion caused by touching the blood vessel wall. If this physical interference is misjudged as electronic drift and calibrated, it may lead to a dangerous false high blood pressure output after calibration.

[0006] In view of this, the present invention proposes an adaptive calibration system and method for invasive blood pressure signal drift to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an adaptive calibration method for invasive blood pressure signal drift, comprising:

[0008] The raw blood pressure signal, raw posture signal and raw altitude signal are acquired by invasive sensors and preprocessed to obtain the initial blood pressure signal, posture angle and altitude change.

[0009] Waveform features are extracted from the initial blood pressure signal to obtain the first waveform features;

[0010] The patient's blood density is calibrated to obtain calibrated blood density. Interference is judged based on attitude angle, height change and calibrated blood density. Based on the interference judgment result, the first waveform feature is extracted to obtain the second waveform feature.

[0011] The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability. The attenuation type is determined based on the attenuation probability. The corresponding blood pressure signal and waveform features are extracted based on the attenuation type.

[0012] Extract calibration reference points, calculate theoretical reference values, and calculate the actual drift amount of the current pulsation cycle based on the calibration reference points and theoretical reference values, combined with waveform characteristics. Then, perform drift calibration on the blood pressure signal based on the actual drift amount.

[0013] Non-invasive blood pressure measurements are collected simultaneously, and the parameters of the attenuation classification model are calibrated based on the blood pressure signal after drift calibration.

[0014] Furthermore, methods for determining the attenuation type based on the attenuation probability include:

[0015] If the attenuation probability is higher than the first threshold, it is determined that the catheter tip is in contact with the blood vessel wall, calibration is paused, and an alarm is triggered.

[0016] If the attenuation probability is lower than the second threshold, the signal is determined to have no physical interference; the corresponding initial blood pressure signal and second waveform features are extracted as the blood pressure signal and waveform features.

[0017] If the attenuation probability is between the first threshold and the second threshold, it is judged as suspected interference; the attenuation probability is recalculated after observing for L periods; where the first threshold is greater than the second threshold.

[0018] Furthermore, methods for obtaining the actual drift amount of the current pulsation cycle include:

[0019] The end-diastolic pressure of each pulse cycle is used as the calibration reference point; the average end-diastolic pressure at the initial calibration time of the system is calculated to obtain the theoretical reference value; the initial drift is set to 0; the end-diastolic pressure of the current pulse cycle is obtained, and the difference between the theoretical reference value and the end-diastolic pressure is calculated to obtain the instantaneous deviation of the current pulse cycle.

[0020] Calculate the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude of the previous S pulsation cycles. Based on the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude, calculate the waveform deviation.

[0021] The complete waveform segment of the current pulsation cycle is extracted and zero-mean processed to obtain the waveform segment. The db4 wavelet is selected as the wavelet basis, and the number of decomposition levels is calculated based on the sampling rate and noise frequency. The waveform segment is then decomposed into N-level wavelet coefficients to obtain approximation coefficients and detail coefficients. The noise standard deviation is calculated based on the Nth-level detail coefficients. The square of the noise standard deviation is calculated to obtain the noise power. Soft-threshold denoising is performed on the detail coefficients from the first level to the (N-1)th level to obtain the denoised detail coefficients. The reconstructed signal is obtained based on the denoised detail coefficients and approximation coefficients. The variance of the reconstructed signal is calculated to obtain the signal power. The ratio of signal power to noise power is calculated to obtain the signal-to-noise ratio.

[0022] The signal-to-noise ratio (SNR) and waveform deviation are normalized to obtain normalized SNR and morphological stability values, respectively. The signal quality index is calculated based on the normalized SNR and waveform deviation values, and the adaptive coefficient is calculated based on the signal quality index. The actual drift of the current pulse cycle is calculated using a first-order recursive filter based on the adaptive coefficient, the drift of the previous pulse cycle, and the instantaneous deviation of the current pulse cycle.

[0023] Furthermore, methods for obtaining the signal-to-noise ratio normalized value include:

[0024] When the signal-to-noise ratio (SNR) is not higher than the first signal threshold, the SNR normalization value is set to 0; when the SNR is between the first and second signal thresholds, the SNR is normalized by combining the first and second signal thresholds to obtain the SNR normalization value; when the SNR is not lower than the second signal threshold, the SNR normalization value is set to 1; wherein, the first signal threshold is lower than the second signal threshold.

[0025] Methods for obtaining morphological stability normalization values ​​include:

[0026] When the waveform deviation is not higher than the first waveform threshold, the normalized value of the waveform deviation is set to 0; when the waveform deviation is between the first waveform threshold and the second waveform threshold, the waveform deviation is standardized by combining the first waveform threshold and the second waveform threshold to obtain the normalized value of the waveform deviation; when the waveform deviation is not lower than the second waveform threshold, the normalized value of the waveform deviation is set to 1; wherein, the first waveform threshold is lower than the second waveform threshold.

[0027] Furthermore, methods for obtaining calibrated blood density include:

[0028] Keep the patient in a supine position and ensure the sensor is level with the heart;

[0029] Adjust the sensor to the first height, adjust the attitude angle to the first attitude angle, stabilize for the first time period, and record the end-diastolic pressure and non-invasive diastolic pressure at this time;

[0030] Calculate the difference between end-diastolic pressure and non-invasive diastolic pressure to obtain static pressure deviation;

[0031] The calibrated blood density is calculated based on the static pressure deviation, gravitational acceleration, first height, and first attitude angle.

[0032] Furthermore, methods for determining interference based on attitude angle, altitude changes, and calibrated blood density include:

[0033] The real-time static pressure deviation is calculated based on attitude angle, altitude change, calibrated blood density, and gravitational acceleration. If the real-time static pressure deviation exceeds the static pressure threshold, it is determined that there is positional interference. The difference between the blood pressure signal and the real-time static pressure deviation is used as the new blood pressure signal, and feature extraction is performed again to obtain the second waveform feature. Otherwise, it is determined that there is no positional interference, and the first waveform feature is directly used as the second waveform feature.

[0034] Furthermore, methods for obtaining the first waveform features include:

[0035] The initial blood pressure signal is periodically segmented by dividing a single pulsation cycle into two adjacent systolic peaks. The peak systolic blood pressure and the lowest diastolic blood pressure, as well as the corresponding time, are extracted from each pulsation cycle.

[0036] Calculate the difference between the peak systolic blood pressure and the lowest diastolic blood pressure to obtain the pulse wave amplitude;

[0037] Calculate the time difference between the peak systolic blood pressure and the lowest diastolic blood pressure, and calculate the ratio of the difference to the time difference to obtain the slope of the rising branch;

[0038] Calculate the second derivative of the initial blood pressure signal, and take the maximum value of the second derivative as the dicrotic notch;

[0039] Integrate the initial blood pressure signal to calculate the waveform area of ​​the corresponding pulsation cycle;

[0040] The pulse wave amplitude, the slope of the ascending limb, the dicrotic notch, and the waveform area are spliced ​​together as the first waveform feature.

[0041] Furthermore, methods for obtaining approximation coefficients and detail coefficients include:

[0042] Calculate the minimum length of the N-level decomposition, and pad the number of sampling points in the current pulsation cycle with zeros to extend it to an integer multiple of the minimum length to obtain the initial length;

[0043] The Mallat algorithm is used to achieve layer-by-layer decomposition. Its core is to extract approximate coefficients through a low-pass filter, extract detailed coefficients through a high-pass filter, and reduce the amount of data through downsampling.

[0044] Starting with the initial approximation coefficients set to the initial length, decompose layer by layer until the Nth layer;

[0045] The decomposition process is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting The approximation coefficient of the first layer is used as the first... The input of the layer is used to linearly convolve the low-pass filter with the initial approximation coefficients to obtain the initial approximation coefficients. Then, the high-pass filter is linearly convolved with the initial approximation coefficients to obtain the initial detail coefficients. Finally, the initial approximation coefficients and the initial detail coefficients are downsampled by a factor of P to obtain the first layer. Approximation coefficients and detail coefficients of the layers.

[0046] Furthermore, the method for calibrating the parameters of the attenuation classification model based on the drift-calibrated blood pressure signal includes:

[0047] The difference between the average end-diastolic pressure of the blood pressure signal after drift calibration and the average non-invasive diastolic pressure is calculated to obtain the detection bias.

[0048] If the detection deviation is greater than the first deviation threshold, adjust the parameters of the attenuation classification model so that the detection deviation is not greater than the first deviation threshold; otherwise, keep the parameters of the attenuation classification model unchanged and adjust the adaptive coefficients until the detection deviation is less than the second deviation threshold.

[0049] An adaptive calibration system for invasive blood pressure signal drift, implementing the aforementioned adaptive calibration method for invasive blood pressure signal drift, includes:

[0050] Data acquisition module: Acquires raw blood pressure signal, raw posture signal and raw height signal through invasive sensors and performs preprocessing to obtain initial blood pressure signal, posture angle and height change;

[0051] Feature extraction module: Extracts waveform features from the initial blood pressure signal to obtain the first waveform features;

[0052] Interference detection module: calibrates the patient's blood density to obtain calibrated blood density, performs interference detection based on attitude angle, height change and calibrated blood density, extracts the first waveform features based on the interference detection results, and obtains the second waveform features;

[0053] Attenuation judgment module: The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability, and the attenuation type is determined based on the attenuation probability; the corresponding blood pressure signal and waveform features are extracted based on the attenuation type.

[0054] Drift calibration module: Extracts calibration reference point, calculates theoretical reference value, calculates actual drift amount of the current pulsation cycle based on calibration reference point and theoretical reference value, combined with waveform characteristics, and performs drift calibration on blood pressure signal based on actual drift amount;

[0055] Calibration and detection module: Synchronously acquires non-invasive blood pressure measurements and calibrates the parameters of the attenuation classification model based on the blood pressure signal after drift calibration.

[0056] The technical effects and advantages of the adaptive calibration system and method for invasive blood pressure signal drift proposed in this invention are as follows:

[0057] This invention acquires raw blood pressure, posture, and altitude signals, and combines this with patient-specific blood density calibration to eliminate the influence of macroscopic positional interference on the signal. It then extracts waveform features such as pulse wave amplitude, ascending limb slope, dicrotic notch, and waveform area. An attenuation classification model is used to input nearly K cycles of these features to accurately identify physical interference caused by microscopic displacement and the absence of interference. Subsequently, using end-diastolic pressure as a benchmark, the adaptive coefficients are dynamically adjusted based on the signal-to-noise ratio and waveform stability after wavelet transform denoising. A first-order recursive filter is used to accurately calculate and calibrate the electronic drift. Finally, the attenuation classification model parameters and adaptive coefficients are optimized inversely using synchronously acquired non-invasive blood pressure as a benchmark. This approach eliminates the influence of macroscopic interference on signal judgment and achieves accurate differentiation between microscopic physical interference and electronic drift. Mechanistically, it avoids mistaking physical interference for electronic drift calibration, ultimately ensuring the safety and accuracy of adaptive calibration of invasive blood pressure signals. This addresses the shortcomings of existing technologies and effectively prevents the generation of dangerous hypertension false alarms. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the adaptive calibration method for invasive blood pressure signal drift according to the present invention;

[0059] Figure 2 This is a schematic flowchart of the method for obtaining the first waveform feature according to the present invention;

[0060] Figure 3 This is a schematic diagram of the method for obtaining the actual drift amount of the current pulsation cycle according to the present invention;

[0061] Figure 4 This is a schematic diagram of the adaptive calibration system for invasive blood pressure signal drift according to the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1

[0064] Please see Figure 1As shown in this embodiment, the adaptive calibration method for invasive blood pressure signal drift includes:

[0065] The raw blood pressure signal, raw posture signal and raw altitude signal are acquired by invasive sensors and preprocessed to obtain the initial blood pressure signal, posture angle and altitude change.

[0066] Methods for obtaining initial blood pressure signals, posture angles, and altitude changes include:

[0067] The patient's original blood pressure signal from the artery was subjected to analog-to-digital conversion and filtering at a sampling rate of not less than 1000Hz to obtain the initial blood pressure signal in the continuous time domain.

[0068] The original attitude signal and the original altitude signal are filtered to obtain the attitude angle and altitude changes.

[0069] Waveform features are extracted from the initial blood pressure signal to obtain the first waveform features;

[0070] Reference Figure 2 Methods for obtaining the first waveform features include:

[0071] The initial blood pressure signal is periodically segmented by dividing a single pulsation cycle into two adjacent systolic peaks. The peak systolic blood pressure and the lowest diastolic blood pressure, as well as the corresponding time, are extracted from each pulsation cycle.

[0072] Calculate the difference between the peak systolic blood pressure and the lowest diastolic blood pressure to obtain the pulse wave amplitude;

[0073] Calculate the time difference between the peak systolic blood pressure and the lowest diastolic blood pressure, and calculate the ratio of the difference to the time difference to obtain the slope of the rising branch;

[0074] Calculate the second derivative of the initial blood pressure signal, and take the maximum value of the second derivative as the dicrotic notch; as in the... Diponeme notch per beat cycle ,in, For the first The contraction peak time of the nth pulsation cycle is determined using an adaptive threshold method: Within each pulsation cycle, the moment when the initial blood pressure signal exceeds 90% of the average peak systolic blood pressure of the past 5 cycles is considered the [time value]. ; For the first The search time for the notch of each pulsation cycle. Represents the variable Take the maximum value; For the notch search interval, The search start time, This is the search end time; For the first A smoothed blood pressure signal per pulsation cycle. For a moment, , For the first The time of each sampling point Here, b is the index of the sampling point, and b+2 is the total number of sampling points. For the first The initial blood pressure signal of each pulse cycle; second derivative ,in, and For the first The and the first The time of each sampling point; For the first The amplitude of the pulse wave in one pulsating cycle;

[0075] Integrate the initial blood pressure signal to calculate the waveform area of ​​the corresponding pulsation cycle;

[0076] The pulse wave amplitude, the slope of the ascending limb, the dicrotic notch, and the waveform area are spliced ​​together as the first waveform feature.

[0077] By acquiring and preprocessing raw blood pressure, posture, and altitude signals at a sampling rate of no less than 1000Hz, the initial blood pressure signal, posture angle, and altitude changes are obtained. This ensures the integrity of the blood pressure signal details and lays the foundation for eliminating macroscopic position interference through filtering of posture and altitude signals. Periodic segmentation of the initial blood pressure signal and extraction of pulse wave amplitude, ascending limb slope, dicrotic notch based on second derivative and normalization, and waveform area as the first waveform features can accurately capture the local pressure signal attenuation and waveform distortion caused by the micro-displacement of the sensor catheter tip in the blood vessel. This effectively distinguishes such physical interference from real electronic drift, avoiding the dangerous hypertension spurious caused by the inability to identify the above-mentioned micro-displacement interference and mistaking it as electronic drift for calibration in the existing technology. This provides key feature basis for subsequent safe and accurate adaptive calibration.

[0078] The patient's blood density is calibrated to obtain calibrated blood density. Interference is judged based on attitude angle, height change and calibrated blood density. Based on the interference judgment result, the first waveform feature is extracted to obtain the second waveform feature.

[0079] Methods for obtaining calibrated blood density include:

[0080] Keep the patient in a supine position and ensure the sensor is level with the heart;

[0081] Adjust the sensor to the first height, adjust the attitude angle to the first attitude angle, stabilize for the first time period, and record the end-diastolic pressure and non-invasive diastolic pressure at this time;

[0082] Calculate the difference between end-diastolic pressure and non-invasive diastolic pressure to obtain static pressure deviation;

[0083] The calibrated blood density is calculated based on the static pressure deviation, gravitational acceleration, initial height, and initial attitude angle. (For example, blood density...) ,in, For static pressure deviation, unit conversion is required before calculating static pressure deviation, from mmHg to Pa, such as 1 mmHg ≈ 133.322 Pa; It is the acceleration due to gravity; The first height is the difference between the sensor height and the heart height; This is the first attitude angle, which is generally set to 0° to avoid angular interference.

[0084] Methods for determining interference based on attitude angle, altitude changes, and calibrated blood density include:

[0085] The real-time static pressure deviation is calculated based on attitude angle, altitude change, calibrated blood density, and gravitational acceleration. If the real-time static pressure deviation exceeds the static pressure threshold, it is determined that there is positional interference. The difference between the blood pressure signal and the real-time static pressure deviation is used as the new blood pressure signal, and feature extraction is performed again to obtain the second waveform feature. Otherwise, it is determined that there is no positional interference, and the first waveform feature is directly used as the second waveform feature.

[0086] By having the patient lie supine and adjusting the sensor to a specific height and posture angle, and calculating the calibrated blood density using the difference between end-diastolic pressure and non-invasive diastolic pressure, the influence of individual blood density differences on hydrostatic pressure calculation can be eliminated, improving the accuracy of hydrostatic pressure deviation calculation caused by macroscopic position changes. Based on this, real-time hydrostatic pressure deviation is calculated according to the calibrated blood density, real-time posture angle, and height changes. If the deviation exceeds a threshold, the blood pressure signal is corrected and features are re-extracted to obtain a second waveform feature; otherwise, the first waveform feature is directly used. This method can accurately identify and compensate for interference caused by changes in the macroscopic position of the sensor, avoiding misjudging macroscopic position interference as signal attenuation or electronic drift caused by microscopic displacement of the catheter tip. This ensures that the waveform features used to distinguish between microscopic physical interference and electronic drift have eliminated the influence of macroscopic interference, thus laying the foundation for accurately identifying signal distortion caused by microscopic displacement of the catheter tip. This effectively solves the problem that existing technologies, which rely on fixed physical formulas to compensate for hydrostatic pressure, cannot distinguish between macroscopic and microscopic interference, and may misjudge and miscalibrate, avoiding the generation of dangerous hypertension false alarms.

[0087] The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability. The attenuation type is determined based on the attenuation probability. The corresponding initial blood pressure signal and waveform features are extracted based on the attenuation type.

[0088] Methods for determining attenuation type based on attenuation probability include:

[0089] If the attenuation probability is higher than the first threshold, it is determined that the catheter tip is in contact with the blood vessel wall, calibration is paused, and an alarm is triggered.

[0090] If the attenuation probability is lower than the second threshold, the signal is determined to have no physical interference; the corresponding initial blood pressure signal and second waveform features are extracted as the blood pressure signal and waveform features.

[0091] If the attenuation probability is between the first threshold and the second threshold, it is judged as suspected interference; the attenuation probability is recalculated after observing for L periods; where the first threshold is greater than the second threshold.

[0092] The second waveform features of nearly K pulsation cycles are input into the attenuation classification model to obtain the attenuation probability. The attenuation type is then determined based on the attenuation probability. When the attenuation probability is higher than the first threshold, it is determined that the catheter tip is in contact with the blood vessel wall, and calibration is paused and an alarm is triggered. When it is lower than the second threshold, it is determined that there is no physical interference, and the corresponding signal and features are extracted. When it is between the two, the observation period is extended and re-judgment is performed. Through the classification analysis of waveform features, the local pressure signal attenuation and waveform distortion caused by the micro-displacement of the catheter tip in the blood vessel can be accurately identified. This effectively distinguishes such physical interference from real electronic drift, avoiding the problem of existing technologies misjudging the above-mentioned micro-interference as electronic drift for calibration. By pausing calibration when physical interference is confirmed and only processing signals without physical interference, the mechanism prevents dangerous hypertension spurious caused by incorrect calibration, and makes up for the shortcomings of existing technologies that only rely on macroscopic angles and highly compensated static pressure and cannot deal with micro-interference.

[0093] Extract calibration reference points, calculate theoretical reference values, and calculate the actual drift amount of the current pulsation cycle based on the calibration reference points and theoretical reference values, combined with waveform characteristics. Then, perform drift calibration on the blood pressure signal based on the actual drift amount.

[0094] Reference Figure 3 Methods for obtaining the actual drift amount of the current pulsation cycle include:

[0095] The end-diastolic pressure of each pulsation cycle is used as the calibration reference point;

[0096] Calculate the average end-diastolic pressure at the initial calibration time of the system to obtain the theoretical reference value;

[0097] Set the initial drift to 0;

[0098] Obtain the end-diastolic pressure of the current pulsation cycle, calculate the difference between the theoretical reference value and the end-diastolic pressure, and obtain the instantaneous deviation of the current pulsation cycle;

[0099] Calculate the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude of the previous S pulsation cycles. Based on the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude, calculate the waveform deviation. Determine the waveform integrity based on the waveform deviation. (For example, waveform deviation...) ,in, This represents the amplitude of the pulse wave during the current pulsation cycle. The average amplitude of the pulse wave over the previous S pulsation cycles is used. If the waveform deviation rate is not higher than the deviation threshold, the waveform is considered complete; otherwise, the waveform is considered incomplete.

[0100] Extract the complete waveform segment of the current pulsation cycle, and perform zero-mean processing on the complete waveform segment, that is, subtract the average value of the pulse wave amplitude of the current pulsation cycle to eliminate DC component interference;

[0101] The db4 wavelet is selected as the wavelet basis. The number of decomposition levels is calculated based on the sampling rate and noise frequency. The waveform segment is decomposed into N-level wavelet decomposition to obtain the approximation coefficients and detail coefficients.

[0102] Methods for obtaining approximation coefficients and detail coefficients include:

[0103] Calculate the minimum length of the N-level decomposition, and pad the number of sampling points in the current pulsation cycle with zeros to extend it to an integer multiple of the minimum length to obtain the initial length;

[0104] The Mallat algorithm is used to achieve layer-by-layer decomposition. Its core principle is to extract approximate coefficients using a low-pass filter, extract detailed coefficients using a high-pass filter, and reduce the amount of data through downsampling. ,in, For the first Approximation coefficients of the layer; For the first Approximation coefficients of the layer; For the first The detail factor of the layer; It is a low-pass filter; It is a high-pass filter; This is a convolution operation; This is a 2x downsampling used to preserve points with even indices in the convolution result;

[0105] Starting with the initial approximation coefficients set to the initial length, decompose layer by layer until the Nth layer;

[0106] The decomposition process is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting The approximation coefficient of the first layer is used as the first... The input of the layer is used to linearly convolve the low-pass filter with the initial approximation coefficients to obtain the initial approximation coefficients. Then, the high-pass filter is linearly convolved with the initial approximation coefficients to obtain the initial detail coefficients. Finally, the initial approximation coefficients and the initial detail coefficients are downsampled by a factor of P to obtain the first layer. Approximation coefficients and detail coefficients of the layers;

[0107] The noise standard deviation is calculated based on the detail coefficients of the Nth layer; the square of the noise standard deviation is calculated to obtain the noise power; if the noise standard deviation... ,in, is the detail coefficient of the Nth level; 0.6745 is the 0.75 quantile of the standard normal distribution. It is a median function;

[0108] Soft thresholding is performed on the detail coefficients from the first layer to the (N-1)th layer to obtain the denoised detail coefficients. The reconstructed signal is obtained by reconstructing the detail coefficients and the approximation coefficients.

[0109] Methods for obtaining the detail coefficients after denoising include:

[0110] A soft threshold is calculated based on the number of sampling points and the noise standard deviation. Detail coefficients with values ​​below the soft threshold are set to zero, while those with values ​​at or above the soft threshold are retained and subtracted from the soft threshold to obtain the denoised detail coefficients. The blood pressure signal is then reconstructed using these denoised detail coefficients and the Nth-level approximation coefficients to obtain the reconstructed signal. (The soft threshold is then used in conjunction with the calculation of the soft threshold.) ,in, This represents the number of sampling points;

[0111] Calculate the variance of the reconstructed signal to obtain the signal power, and calculate the ratio of signal power to noise power to obtain the signal-to-noise ratio.

[0112] The signal-to-noise ratio is normalized by combining the first signal threshold and the second signal threshold to obtain a normalized signal-to-noise ratio value; the waveform deviation is normalized by combining the first waveform threshold and the second waveform threshold to obtain a normalized morphological stability value.

[0113] Methods for obtaining the normalized signal-to-noise ratio include:

[0114] When the signal-to-noise ratio (SNR) is not higher than the first signal threshold, the SNR normalization value is set to 0; when the SNR is between the first and second signal thresholds, the SNR is normalized by combining the first and second signal thresholds to obtain the SNR normalization value; when the SNR is not lower than the second signal threshold, the SNR normalization value is set to 1; wherein, the first signal threshold is lower than the second signal threshold.

[0115] Methods for obtaining morphological stability normalization values ​​include:

[0116] When the waveform deviation is not higher than the first waveform threshold, the normalized value of the waveform deviation is set to 0; when the waveform deviation is between the first waveform threshold and the second waveform threshold, the waveform deviation is standardized by combining the first waveform threshold and the second waveform threshold to obtain the normalized value of the waveform deviation; when the waveform deviation is not lower than the second waveform threshold, the normalized value of the waveform deviation is set to 1; wherein, the first waveform threshold is lower than the second waveform threshold.

[0117] The signal quality composite index is calculated based on the normalized signal-to-noise ratio and waveform deviation, and the adaptive coefficient is calculated based on the signal quality composite index. Adaptive coefficient ,in, The weighting coefficients can be obtained through natural heuristic optimization algorithms. This is the normalized signal-to-noise ratio value; This is the normalized value for waveform deviation; This is the lower limit of the adaptive coefficient, set to match clinical needs; This is the upper limit of the adaptive coefficient, set to match clinical needs;

[0118] Based on the adaptive coefficients, the drift amount of the previous pulse cycle, and the instantaneous deviation of the current pulse cycle, the actual drift amount of the current pulse cycle is calculated using a first-order recursive filter; for example, the actual drift amount of the current pulse cycle... ,in, This represents the instantaneous deviation of the current pulsation cycle; This represents the actual drift amount in the previous pulse cycle.

[0119] By using end-diastolic pressure as the calibration reference point and combining it with the theoretical reference value of the initial system calibration, the instantaneous deviation is calculated. Waveform integrity is assessed through waveform deviation, and the signal-to-noise ratio (SNR) is obtained through db4 wavelet decomposition and soft threshold denoising. The SNR and waveform deviation are then normalized to obtain a comprehensive signal quality index, which is used to dynamically adjust the adaptive coefficients. Finally, a first-order recursive filter is used to calculate the actual drift based on the drift amount from the previous cycle and the current instantaneous deviation. This method accurately captures the true electronic drift after eliminating macroscopic position interference and confirming the absence of physical interference caused by microscopic displacement of the duct tip. Waveform integrity assessment and signal-to-noise ratio evaluation can effectively filter out abnormal signals affected by microscopic interference, preventing them from participating in drift calculation. The adaptive coefficient driven by the signal quality comprehensive index makes drift updates more sensitive when the signal is reliable and more robust when the signal is questionable, preventing the signal attenuation or distortion caused by microscopic displacement from being misjudged as electronic drift. Recursive filtering further ensures the stability of drift calibration, thereby fundamentally avoiding the dangerous spurious hypertension that may be caused by the inability of existing technologies to distinguish between microscopic physical interference and electronic drift, achieving safe and accurate calibration only for real electronic drift.

[0120] Non-invasive blood pressure measurements are collected simultaneously, and the parameters of the attenuation classification model are calibrated based on the blood pressure signal after drift calibration.

[0121] Methods for calibrating the parameters of the attenuation classification model based on the drift-calibrated blood pressure signal include:

[0122] The difference between the average end-diastolic pressure of the blood pressure signal after drift calibration and the average non-invasive diastolic pressure is calculated to obtain the detection bias.

[0123] If the detection deviation is greater than the first deviation threshold, adjust the parameters of the attenuation classification model so that the detection deviation is not greater than the first deviation threshold; otherwise, keep the parameters of the attenuation classification model unchanged and adjust the adaptive coefficients until the detection deviation is less than the second deviation threshold.

[0124] Simultaneously acquiring non-invasive blood pressure measurements as a reference standard, the detection deviation is obtained by calculating the difference between the average end-diastolic pressure of the blood pressure signal after drift calibration and the average non-invasive diastolic pressure. If the deviation is greater than the first deviation threshold, the attenuation classification model parameters are adjusted until the deviation meets the standard; otherwise, the model parameters are kept unchanged and the adaptive coefficient is adjusted until the deviation is less than the second deviation threshold. This allows the attenuation classification model to be optimized in reverse using non-invasive blood pressure as a reliable benchmark, enabling the model to more accurately distinguish between physical interference caused by microscopic displacement of the sensor catheter tip and real electronic drift, avoiding the model from misjudging interference as drift due to parameter deviation. At the same time, by adjusting the adaptive coefficient, it is further ensured that drift calibration only targets real electronic drift, preventing errors introduced by calibration logic deviation. With dual protection from both model recognition capability and calibration logic, this effectively makes up for the shortcomings of existing technologies that cannot distinguish between microscopic interference and electronic drift, and completely avoids the problem of creating dangerous hypertension false alarms by misjudging microscopic physical interference as electronic drift for calibration.

[0125] Example 2

[0126] Please see Figure 4 As shown, the adaptive calibration system for invasive blood pressure signal drift described in this embodiment includes:

[0127] Data acquisition module: Acquires raw blood pressure signal, raw posture signal and raw height signal through invasive sensors and performs preprocessing to obtain initial blood pressure signal, posture angle and height change;

[0128] Feature extraction module: Extracts waveform features from the initial blood pressure signal to obtain the first waveform features;

[0129] Interference detection module: calibrates the patient's blood density to obtain calibrated blood density, performs interference detection based on attitude angle, height change and calibrated blood density, extracts the first waveform features based on the interference detection results, and obtains the second waveform features;

[0130] Attenuation judgment module: The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability, and the attenuation type is determined based on the attenuation probability; the corresponding blood pressure signal and waveform features are extracted based on the attenuation type.

[0131] Drift calibration module: Extracts calibration reference point, calculates theoretical reference value, calculates actual drift amount of the current pulsation cycle based on calibration reference point and theoretical reference value, combined with waveform characteristics, and performs drift calibration on blood pressure signal based on actual drift amount;

[0132] Calibration and detection module: Synchronously acquires non-invasive blood pressure measurements and calibrates the parameters of the attenuation classification model based on the blood pressure signal after drift calibration.

[0133] 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.

[0134] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive calibration method for invasive blood pressure signal drift, characterized in that, include: The raw blood pressure signal, raw posture signal and raw altitude signal are acquired by invasive sensors and preprocessed to obtain the initial blood pressure signal, posture angle and altitude change. Waveform features are extracted from the initial blood pressure signal to obtain the first waveform features; The patient's blood density is calibrated to obtain calibrated blood density. Interference is judged based on attitude angle, height change and calibrated blood density. Based on the interference judgment result, the first waveform feature is extracted to obtain the second waveform feature. The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability. The attenuation type is determined based on the attenuation probability. The corresponding blood pressure signal and waveform features are extracted based on the attenuation type. Extract calibration reference points, calculate theoretical reference values, and calculate the actual drift amount of the current pulsation cycle based on the calibration reference points and theoretical reference values, combined with waveform characteristics. Then, perform drift calibration on the blood pressure signal based on the actual drift amount. Non-invasive blood pressure measurements are collected simultaneously, and the parameters of the attenuation classification model are calibrated based on the blood pressure signal after drift calibration.

2. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for determining attenuation type based on attenuation probability include: If the attenuation probability is higher than the first threshold, it is determined that the catheter tip is in contact with the blood vessel wall, calibration is paused, and an alarm is triggered. If the attenuation probability is lower than the second threshold, the signal is determined to have no physical interference; the corresponding initial blood pressure signal and second waveform features are extracted as the blood pressure signal and waveform features. If the attenuation probability is between the first threshold and the second threshold, it is judged as suspected interference; the attenuation probability is recalculated after observing for L periods; where the first threshold is greater than the second threshold.

3. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for obtaining the actual drift amount of the current pulsation cycle include: The end-diastolic pressure of each pulse cycle is used as the calibration reference point; the average end-diastolic pressure at the initial calibration time of the system is calculated to obtain the theoretical reference value; the initial drift is set to 0; the end-diastolic pressure of the current pulse cycle is obtained, and the difference between the theoretical reference value and the end-diastolic pressure is calculated to obtain the instantaneous deviation of the current pulse cycle. Calculate the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude of the previous S pulsation cycles. Based on the pulse wave amplitude of the current pulsation cycle and the average pulse wave amplitude, calculate the waveform deviation. The complete waveform segment of the current pulsation cycle is extracted and zero-mean processed to obtain the waveform segment. The db4 wavelet is selected as the wavelet basis, and the number of decomposition levels is calculated based on the sampling rate and noise frequency. The waveform segment is then decomposed into N-level wavelet coefficients to obtain approximation coefficients and detail coefficients. The noise standard deviation is calculated based on the Nth-level detail coefficients. The square of the noise standard deviation is calculated to obtain the noise power. Soft-threshold denoising is performed on the detail coefficients from the first level to the (N-1)th level to obtain the denoised detail coefficients. The reconstructed signal is obtained based on the denoised detail coefficients and approximation coefficients. The variance of the reconstructed signal is calculated to obtain the signal power. The ratio of signal power to noise power is calculated to obtain the signal-to-noise ratio. The signal-to-noise ratio (SNR) and waveform deviation are normalized to obtain normalized SNR and morphological stability values, respectively. The signal quality index is calculated based on the normalized SNR and waveform deviation values, and the adaptive coefficient is calculated based on the signal quality index. The actual drift of the current pulse cycle is calculated using a first-order recursive filter based on the adaptive coefficient, the drift of the previous pulse cycle, and the instantaneous deviation of the current pulse cycle.

4. The adaptive calibration method for invasive blood pressure signal drift according to claim 3, characterized in that, Methods for obtaining the normalized signal-to-noise ratio include: When the signal-to-noise ratio (SNR) is not higher than the first signal threshold, the SNR normalization value is set to 0; when the SNR is between the first and second signal thresholds, the SNR is normalized by combining the first and second signal thresholds to obtain the SNR normalization value; when the SNR is not lower than the second signal threshold, the SNR normalization value is set to 1; wherein, the first signal threshold is lower than the second signal threshold. Methods for obtaining morphological stability normalization values ​​include: When the waveform deviation is not higher than the first waveform threshold, the normalized value of the waveform deviation is set to 0; when the waveform deviation is between the first waveform threshold and the second waveform threshold, the waveform deviation is standardized by combining the first waveform threshold and the second waveform threshold to obtain the normalized value of the waveform deviation; when the waveform deviation is not lower than the second waveform threshold, the normalized value of the waveform deviation is set to 1; wherein, the first waveform threshold is lower than the second waveform threshold.

5. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for obtaining calibrated blood density include: Keep the patient in a supine position and ensure the sensor is level with the heart; Adjust the sensor to the first height, adjust the attitude angle to the first attitude angle, stabilize for the first time period, and record the end-diastolic pressure and non-invasive diastolic pressure at this time; Calculate the difference between end-diastolic pressure and non-invasive diastolic pressure to obtain static pressure deviation; The calibrated blood density is calculated based on the static pressure deviation, gravitational acceleration, first height, and first attitude angle.

6. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for determining interference based on attitude angle, altitude changes, and calibrated blood density include: The real-time static pressure deviation is calculated based on attitude angle, altitude change, calibrated blood density, and gravitational acceleration. If the real-time static pressure deviation exceeds the static pressure threshold, it is determined that there is positional interference. The difference between the blood pressure signal and the real-time static pressure deviation is used as the new blood pressure signal, and feature extraction is performed again to obtain the second waveform feature. Otherwise, it is determined that there is no positional interference, and the first waveform feature is directly used as the second waveform feature.

7. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for obtaining the first waveform features include: The initial blood pressure signal is periodically segmented by dividing a single pulsation cycle into two adjacent systolic peaks. The peak systolic blood pressure and the lowest diastolic blood pressure, as well as the corresponding time, are extracted from each pulsation cycle. Calculate the difference between the peak systolic blood pressure and the lowest diastolic blood pressure to obtain the pulse wave amplitude; Calculate the time difference between the peak systolic blood pressure and the lowest diastolic blood pressure, and calculate the ratio of the difference to the time difference to obtain the slope of the rising branch; Calculate the second derivative of the initial blood pressure signal, and take the maximum value of the second derivative as the dicrotic notch; Integrate the initial blood pressure signal to calculate the waveform area of ​​the corresponding pulsation cycle; The pulse wave amplitude, the slope of the ascending limb, the dicrotic notch, and the waveform area are spliced ​​together as the first waveform feature.

8. The adaptive calibration method for invasive blood pressure signal drift according to claim 3, characterized in that, Methods for obtaining approximation coefficients and detail coefficients include: Calculate the minimum length of the N-level decomposition, and pad the number of sampling points in the current pulsation cycle with zeros to extend it to an integer multiple of the minimum length to obtain the initial length; The Mallat algorithm is used to achieve layer-by-layer decomposition. Its core is to extract approximate coefficients through a low-pass filter, extract detailed coefficients through a high-pass filter, and reduce the amount of data through downsampling. Starting with the initial approximation coefficients set to the initial length, decompose layer by layer until the Nth layer; The decomposition process is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The approximation coefficient of the first layer is used as the first... The input of the layer is used to linearly convolve the low-pass filter with the initial approximation coefficients to obtain the initial approximation coefficients. Then, the high-pass filter is linearly convolved with the initial approximation coefficients to obtain the initial detail coefficients. Finally, the initial approximation coefficients and the initial detail coefficients are downsampled by a factor of P to obtain the first layer. Approximation coefficients and detail coefficients of the layers.

9. The adaptive calibration method for invasive blood pressure signal drift according to claim 1, characterized in that, Methods for calibrating the parameters of the attenuation classification model based on the drift-calibrated blood pressure signal include: The difference between the average end-diastolic pressure of the blood pressure signal after drift calibration and the average non-invasive diastolic pressure is calculated to obtain the detection bias. If the detection deviation is greater than the first deviation threshold, adjust the parameters of the attenuation classification model so that the detection deviation is not greater than the first deviation threshold; otherwise, keep the parameters of the attenuation classification model unchanged and adjust the adaptive coefficients until the detection deviation is less than the second deviation threshold.

10. An adaptive calibration system for invasive blood pressure signal drift, comprising the adaptive calibration method for invasive blood pressure signal drift according to any one of claims 1-9, characterized in that, include: Data acquisition module: Acquires raw blood pressure signal, raw posture signal and raw height signal through invasive sensors and performs preprocessing to obtain initial blood pressure signal, posture angle and height change; Feature extraction module: Extracts waveform features from the initial blood pressure signal to obtain the first waveform features; Interference detection module: calibrates the patient's blood density to obtain calibrated blood density, performs interference detection based on attitude angle, height change and calibrated blood density, extracts the first waveform features based on the interference detection results, and obtains the second waveform features; Attenuation judgment module: The second waveform features of nearly K pulsation cycles are used as input to the attenuation classification model to obtain the attenuation probability, and the attenuation type is determined based on the attenuation probability; the corresponding blood pressure signal and waveform features are extracted based on the attenuation type. Drift calibration module: Extracts calibration reference point, calculates theoretical reference value, calculates actual drift amount of the current pulsation cycle based on calibration reference point and theoretical reference value, combined with waveform characteristics, and performs drift calibration on blood pressure signal based on actual drift amount; Calibration and detection module: Synchronously acquires non-invasive blood pressure measurements and calibrates the parameters of the attenuation classification model based on the blood pressure signal after drift calibration.

Images