AI decision support system for acute kidney injury dynamic target drug screening

Abstract

The present application relates to the technical field of medical data mining, in particular to an AI decision support system for dynamic target drug screening of acute kidney injury. The present application first divides a single beat time window in a preprocessing and decoupling module, obtains a diastolic tension coefficient, a total stress load integral of a blood vessel wall and a microcirculation driving load integral; further extracts an edema damping index and a perfusion conversion efficiency in an analysis module; finally, through a state marking module, the current diastolic tension coefficient and the perfusion conversion efficiency are mapped to a two-dimensional decision plane, and the overall characteristics of the central venous pressure in the current single beat time window are combined to mark the microcirculation fluid state. The present application solves the problem of disconnection between macroscopic blood pressure and microcirculation perfusion and the problem of difficult identification of the root cause of disorder by decoupling tension and energy characteristics, constructing a lagging circulation to quantify damping and correcting efficiency, mapping a decision plane to mark a fluid state, and providing decision support.

Description

Technical Field

[0001] This invention relates to the field of medical data mining technology, specifically to an AI decision support system for dynamic target drug screening in acute kidney injury. Background Technology

[0002] Acute kidney injury following cardiac surgery is a common complication in intensive care, with a complex pathological mechanism involving the dynamic coupling of systemic hemodynamic disturbances and renal microcirculatory dysfunction. Clinically, current treatment decisions heavily rely on macroscopic static indicators such as mean arterial pressure and central venous pressure, typically with pre-set fixed pressure management targets. However, after major surgeries such as cardiopulmonary bypass, patients often experience special fluid states such as vasomotor syndrome, tissue edema, or obstructed venous return. In these states, macroscopic blood pressure indicators often become disconnected from the actual perfusion level of the microcirculation. For example, in severe vasomotor syndrome, blood flows too rapidly through the dilated vascular bed; although the mean arterial pressure may barely remain at the lower limit of the normal range, the effective perfusion pressure is severely insufficient. In cases of severe tissue edema or venous hypertension, the elevated hydrostatic pressure creates significant adverse resistance, making it difficult for energy to be transferred from the arterial end to the microcirculation.

[0003] Existing linear-indicator-based monitoring methods cannot effectively distinguish between these distinct hydrodynamic states with vastly different underlying mechanisms. This makes it difficult for clinicians to make precise, pathophysiologically oriented choices among different treatment strategies, such as vasoconstrictors, diuretics, or microcirculation protectants. Deviations in medication use strategies not only fail to reverse kidney damage but may also exacerbate renal ischemia or volume overload, thus delaying optimal treatment. Summary of the Invention

[0004] To address the technical problem of the disconnect between existing macroscopic blood pressure monitoring and actual microcirculation perfusion levels, which makes it difficult to accurately identify the root causes of various complex hemodynamic disturbances and thus provide effective decision support, the present invention aims to provide an AI decision support system for dynamic target drug screening in acute kidney injury. The specific technical solution adopted is as follows: The preprocessing and decoupling module is used to divide the single pulsation time window based on the fluctuation of the arterial pressure signal. For the current single pulsation time window, the attenuation characteristics of arterial pressure during diastole are analyzed. Combined with the duration of the single pulsation time window, the diastolic tension coefficient is obtained. The square of the arterial pressure and the square of the difference between the arterial pressure and the central venous pressure are integrated to obtain the total stress load integral of the vessel wall and the microcirculation driving load integral. The analysis module is used to construct a two-dimensional hysteresis loop within the current preset historical neighborhood based on the integral of the total stress load of the blood vessel wall and the integral of the microcirculation driving load, and calculate the geometric area enclosed by the two-dimensional hysteresis loop as the edema damping index; based on the linear characteristics of the integral of the total stress load of the blood vessel wall and the integral of the microcirculation driving load, and combined with the result of correcting the edema damping index by the diastolic tension coefficient, the perfusion conversion efficiency is obtained. The status labeling module is used to map the current diastolic tension coefficient and the perfusion conversion efficiency to a two-dimensional decision plane. Based on the spatial distribution of the current data points in the two-dimensional decision plane, and combined with the overall characteristics of the central venous pressure within the current single pulsation window, the microcirculation fluid state is labeled.

[0005] Furthermore, the method for obtaining the single-pulse time window includes: The first derivative of the arterial pressure signal is obtained, and local maxima exceeding a preset noise threshold are identified as dividing points. Adjacent dividing points are then divided into a single beat time window.

[0006] Furthermore, the method for obtaining the diastolic tension coefficient includes: Based on the difference between the diastolic initiation pressure and the mean central venous pressure within the current single pulsation window, the arterial pressure signal is fitted using a two-element Windkessel model to obtain the absolute decay time constant; the diastolic tension coefficient is obtained by fusing the absolute decay time constant and the duration of the single pulsation window.

[0007] Furthermore, the method for obtaining the two-dimensional hysteresis loop includes: Within the current preset historical neighborhood, the data points corresponding to each single pulsation time window in the two-dimensional coordinate system are connected in chronological order, with the normalized total stress load integral of the blood vessel wall as the horizontal axis and the normalized microcirculation driving load integral as the vertical axis, to form a two-dimensional hysteresis loop.

[0008] Furthermore, the method for obtaining the perfusion conversion efficiency includes: The linear characteristics of the integral of the total stress load on the blood vessel wall and the integral of the microcirculation-driven load are analyzed to obtain the basic slope; The effective perfusion factor is obtained by integrating the preset damping sensitivity coefficient, the edema damping index, the diastolic tension coefficient, and the preset stability constant; the perfusion conversion efficiency is obtained by integrating the effective perfusion factor and the baseline slope.

[0009] Furthermore, the method for obtaining the basic slope includes: Using the integral of the total stress load on the blood vessel wall as the abscissa and the integral of the microcirculation driving load as the ordinate, the basic slope is obtained using the Theil-Sen estimator based on the data points corresponding to each time window within the current preset historical neighborhood.

[0010] Furthermore, the method for obtaining the effective perfusion factor includes: The product of the preset damping sensitivity coefficient and the edema damping index is used as the numerator, the sum of the diastolic tension coefficient and the preset stability constant is used as the denominator, and the difference between the constant 1 and the fractional ratio is used as the effective perfusion factor.

[0011] Furthermore, methods for mapping to a two-dimensional decision plane include: The current diastolic tension coefficient is divided by a preset benchmark diastolic tension coefficient as the horizontal axis, and the perfusion conversion efficiency is divided by a preset benchmark conversion efficiency as the vertical axis, and mapped onto a two-dimensional decision plane.

[0012] Furthermore, the method for labeling the microcirculation fluid state includes: In the two-dimensional decision plane, if the current data point is in a preset first state region, it is marked as a preset first state; if the ordinate of the current data point is less than a preset blocking threshold, and the duration of being less than the preset collapse threshold does not exceed a preset safety time limit, and the average central venous pressure in the current time window is greater than a preset warning value, it is marked as a preset second state; if the average central venous pressure in the current time window is less than or equal to the preset warning value, it is marked as a preset third state; if the ordinate of the current data point is less than the preset collapse threshold, and the duration exceeds a preset safety time limit, it is marked as a preset fourth state.

[0013] Furthermore, the geometric area enclosed by the two-dimensional hysteresis loop is calculated using the Discrete Green's Theorem.

[0014] The present invention has the following beneficial effects: This invention first divides the single-pulse time window in the preprocessing and decoupling module, segmenting the continuous hemodynamic signal into a series of independent physiological event units, obtaining the diastolic tension coefficient to characterize the physical contraction state of vascular smooth muscle, obtaining the total stress load integral of the vascular wall to reflect the total pressure potential energy exerted by cardiac ejection on the arterial wall, and obtaining the microcirculation driving load integral to characterize the net potential energy that can actually be used to generate tissue perfusion after overcoming venous back pressure. Further, in the analysis module, a two-dimensional hysteresis loop is constructed within the current preset historical neighborhood, and the geometric area enclosed by the two-dimensional hysteresis loop is calculated as the edema damping index, quantifying the irreversible energy loss caused by the fluid viscosity resistance on the microcirculation path, and then obtaining the perfusion conversion efficiency, realizing the physical correction and quantification of the microcirculation functional state. Finally, through the state labeling module, the current diastolic tension coefficient and perfusion conversion efficiency are mapped to the two-dimensional decision plane, and combined with the overall characteristics of the central venous pressure within the current single-pulse time window, the microcirculation fluid state is labeled. This invention decouples tension and energy characteristics, constructs a hysteresis loop to quantify damping and correct efficiency, maps the decision plane to mark fluid state, solves the problem of disconnect between macroscopic blood pressure and microcirculation perfusion and the difficulty in identifying the root causes of disorder, and provides decision support. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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 system block diagram of an AI decision support system for dynamic target drug screening in acute kidney injury, provided as an embodiment of the present invention. Figure 2 This is a flowchart illustrating a method for obtaining perfusion conversion efficiency according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an AI decision support system for dynamic target drug screening in acute kidney injury proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of an AI decision support system for dynamic target drug screening in acute kidney injury provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a system block diagram of an AI decision support system for dynamic target drug screening in acute kidney injury, provided by an embodiment of the present invention. The system includes: a preprocessing and decoupling module 101, an analysis module 102, and a status marking module 103.

[0021] The preprocessing and decoupling module 101 is used to divide the single pulsation time window based on the fluctuation of the arterial pressure signal. For the current single pulsation time window, it analyzes the attenuation characteristics of arterial pressure during diastole and obtains the diastolic tension coefficient by combining the duration of the single pulsation time window. It integrates the square of the arterial pressure and the square of the difference between the arterial pressure and the central venous pressure to obtain the total stress load integral of the vessel wall and the microcirculation driving load integral.

[0022] Since the heart's beating is a rhythmic pressure source, each systolic-diastolic cycle constitutes a complete energy input unit. In order to establish a discretized and periodically synchronized analytical basis, the time window of a single beating is first divided based on the fluctuation of the arterial pressure signal, thereby dividing the continuous hemodynamic signal into a series of independent physiological event units, providing a time reference for subsequent periodically aligned temporal analysis and feature extraction. Since high-risk patients with acute kidney injury are often in a state of stress-induced tachycardia, the shortened diastolic phase can lead to bias in the assessment of vascular resistance based on absolute time constant. In order to eliminate the interference of heart rate factors on the assessment of vascular tension, we analyze the attenuation characteristics of arterial pressure during diastole for the current single-pulse time window, and obtain the diastolic tension coefficient by combining the duration of the single-pulse time window. This coefficient characterizes the physical contraction state of vascular smooth muscle and provides a basis for accurately distinguishing vascular status in the future. Next, the integral of the total stress load on the vessel wall is obtained by integrating the square of the arterial pressure, which reflects the total pressure potential energy exerted by the heart's blood ejection on the arterial wall. The integral of the microcirculation driving load is obtained by integrating the square of the difference between the arterial pressure and the central venous pressure, which characterizes the net potential energy that can actually be used to generate tissue perfusion after overcoming the venous back pressure. This provides a direct physical quantity input for quantifying the energy transfer loss caused by tissue edema or venous congestion in subsequent steps.

[0023] Preferably, in one embodiment of the invention, arterial pressure is collected via a catheter placed in the radial or femoral artery. Central venous pressure is collected via a catheter placed in the internal jugular vein or subclavian vein. The sampling frequency is set to no less than 100Hz to preserve high-frequency features such as dicrotic notches.

[0024] Considering that the rate of pressure rise is a more sensitive indicator for identifying the onset of each cardiac ejection, obtaining the first derivative of the arterial pressure signal allows us to use the slope of pressure change over time as a recognition feature, thus more accurately pinpointing the moment of sharp pressure rise during each cardiac systole. Therefore, using local maxima exceeding a preset noise threshold as effective dividing points can effectively filter out misjudgments caused by high-frequency noise or small fluctuations. Defining the time segment divided by adjacent dividing points as a single-beat time window ensures that each analysis unit precisely corresponds to a complete cardiac cycle.

[0025] As an example, with a preset noise threshold of 100 mmHg / s, the duration interval corresponding to the k-th single-pulse time window is... , The k-th effective dividing point corresponds to the start time of the k-th single-beat time window. This is the (k+1)th valid dividing point, corresponding to the starting time of the (k+1)th single-pulse time window.

[0026] It should be noted that a single beat window is considered as one beat cycle. For each single beat window acquired, the system also includes a verification operation, and the system calculates the pulse pressure difference (the range of the arterial pressure signal) within the current single beat window. ,like If the signal is less than the preset validity threshold (e.g., 20 mmHg), it indicates that there is severe physical damping or sensor failure. The system will discard the data of the current single-pulse window and maintain the characteristic output of the previous window until a new valid single-pulse window appears.

[0027] Preferably, in one embodiment of the present invention, the diastolic phase is first located, and the system is positioned within a time interval. Internal positioning dicrotic notch. The dicrotic notch marks the closure of the aortic valve and the beginning of diastole. If the waveform characteristics are not obvious, known techniques can be used, such as taking the minimum point of the first derivative after the peak systolic blood pressure as a substitute.

[0028] Based on the difference between the diastolic initiation pressure and the mean central venous pressure within the current single pulsation window, the arterial pressure signal is fitted using a two-element Windkessel model to obtain the absolute decay time constant; the diastolic tension coefficient is obtained by fusing the absolute decay time constant and the duration of the single pulsation window.

[0029] This is because the physical process of diastolic pressure decay is primarily determined by the aortic compliance and peripheral vascular resistance, and can be approximated as a Windkessel model circuit composed of an elastic element (arterial compliance) and a resistance element (peripheral vascular resistance). By using the pressure at the diastolic initiation point (dicrotic notch) as the initial value and correcting the pressure decay asymptote at the end of diastole to the average central venous pressure of that cycle (rather than zero or a fixed value), the closed loop of systemic circulation can be simulated more accurately. By performing an exponential fit on the arterial pressure decay waveform within this interval, a key parameter—the absolute decay time constant "τ"—can be obtained. This time constant τ comprehensively reflects the elastic and resistance characteristics of the blood vessels during that cardiac cycle and is a direct measure of the vascular physical state.

[0030] As an example, the fitting formula for the two-element Windkessel model includes: ; In the formula, This represents the pressure value at the onset of diastole. This represents the average central vein within the current k-th single pulsation window; , This is the start time of the current time window's diastolic phase. This is the absolute decay time constant.

[0031] Based on least squares fitting, Substitute the discrete data points into the fitting formula to solve for the optimal absolute decay time constant; The physical meaning of is the average pressure level of the venous return system during the current cardiac cycle, which quantifies the reverse obstruction pressure of the venous system on arterial blood flow.

[0032] Using the absolute time constant τ directly to assess vascular tension can lead to misdiagnosis: when a patient is in stress-induced tachycardia, their entire cardiac cycle (including diastole) is significantly shortened. In this case, even if the physical properties of the blood vessel itself (elasticity and resistance) remain unchanged, a shorter diastolic period will cause the pressure decay curve to appear "steeper" on the absolute time scale, thus fitting a smaller τ value. If this τ value, compressed by heart rate, is used directly, it will be misdiagnosed as "decreased vascular tension" or "vasoparalysis," which is actually an illusion caused by an increased heart rate.

[0033] Therefore Divide by the duration of a single pulsation window to obtain the diastolic tension coefficient. This is equivalent to normalizing the "velocity" of diastolic decay to the full cycle scale of the current heartbeat, quantifying the ability of the blood vessel wall to maintain tension. This normalization process ensures that the assessment of vascular tension is no longer confused by the speed of heart rate, providing an independent and reliable assessment dimension for accurately distinguishing pathological states.

[0034] It should be noted that the installation and setup of the acquisition equipment, the first derivative of the time-series signal, the acquisition of local extrema, and the least squares fitting method are all technical means well known to those skilled in the art. Implementers can adjust them according to actual needs, and will not elaborate further.

[0035] Furthermore, in order to distinguish between the total load the heart puts on the vascular system and the effective work actually done to drive the flow of blood in the microcirculation, embodiments of the present invention calculate the energy integral characteristics of the pressure waveform.

[0036] Specifically, within the current single pulsation time window (hereinafter referred to as the time window), the square of the arterial pressure is integrated, and the integrated result is used as the total stress load of the blood vessel wall. It characterizes the total pressure potential energy density exerted by blood on the arterial wall, reflecting the total pressure load cost of the cardiovascular system.

[0037] Then, the square of the difference between arterial pressure and central venous pressure is integrated, and the result is used as the microcirculation driving load integral. Since microcirculation perfusion depends not only on the thrust at the arterial end but also on the resistance of the hydrostatic pressure at the venous end, this index quantifies the net potential energy remaining after overcoming the venous back pressure that can actually generate transmural flow, representing the driving force of the "effective work" or "output benefit" that can generate actual tissue perfusion.

[0038] The system combines the decoupled tension, load, drive, and back pressure information into a single-pulse fluid characteristic set. , .

[0039] It should be noted that the definite integral of time-series signals is a well-known technique in the art, and will not be elaborated upon here.

[0040] The analysis module 102 is used to construct a two-dimensional hysteresis loop within the current preset historical neighborhood based on the integral of total stress load of the blood vessel wall and the integral of microcirculation driving load, and calculate the geometric area enclosed by the two-dimensional hysteresis loop as the edema damping index; based on the linear characteristics of the integral of total stress load of the blood vessel wall and the integral of microcirculation driving load, and combined with the result of correcting the edema damping index by the diastolic tension coefficient, the perfusion conversion efficiency is obtained.

[0041] Because energy dissipation in the microcirculation (such as fluid viscous damping caused by tissue edema or venous congestion) is a dynamically evolving process, single-beat data cannot capture this hysteretic characteristic of energy transfer and is easily affected by random physiological noise. Therefore, analysis must be performed within a pre-defined historical neighborhood encompassing multiple consecutive cardiac cycles. The microcirculatory system exhibits significant viscoelastic memory characteristics under pathological conditions (such as tissue edema and venous congestion). This implies a dynamic hysteresis in the microcirculation's response to the cumulative load applied by continuous cardiac beats.

[0042] Therefore, within the current preset historical neighborhood, a two-dimensional hysteresis loop is constructed based on the integral of the total stress load of the blood vessel wall and the integral of the microcirculation driving load. The two-dimensional hysteresis loop reflects the dynamic response relationship between the output (driving energy) and the input (stress energy) of the microcirculation system as a viscoelastic damper. Its area represents the total energy that is consumed due to phase lag and fails to be converted into effective driving work within a complete fluctuation cycle. Therefore, the geometric area enclosed by the two-dimensional hysteresis loop is calculated as the edema damping index, which quantifies the irreversible energy loss caused by the fluid viscosity resistance on the microcirculation path. This transforms the state of tissue edema or venous congestion, which cannot be directly observed by traditional time-domain indicators, into a measurable physical quantity.

[0043] Considering that the area of ​​the hysteresis loop alone only reflects the total amount of energy dissipation but fails to reveal the actual impact of this dissipation on perfusion efficiency under the current vascular physical state, and that the tension level of the vessel wall significantly alters the sensitivity of microcirculation to the same edema damping, this study extracts the maximum efficiency that a unit cardiac load input can theoretically be converted into driving energy output under ideal conditions where damping is ignored, based on the linear characteristics of the integral of the total stress load of the vessel wall and the integral of the driving load of microcirculation. Then, by combining the results of correcting the edema damping index with the diastolic tension coefficient, damping correction is performed to obtain the perfusion conversion efficiency. This characterizes the actual proportion that each unit increase in cardiac load can be converted into effective microcirculation perfusion under the current specific coupling effect of vascular tension and edema damping, thus realizing the physical correction and quantification of the microcirculation functional state.

[0044] Preferably, in one embodiment of the present invention, the preset historical neighborhood ends with the current time window and covers the most recent W=60 single-pulse time windows, corresponding to about 3-5 respiratory cycles, in order to smooth the effect of breathing on intrathoracic pressure.

[0045] If the number of time windows within the current preset historical neighborhood is insufficient, it indicates that the system is in the initial startup phase, meaning that the data has not yet accumulated to a level sufficient to construct a stable phase space. At this time, the system remains silent, performing no further calculations and not outputting labeling results to avoid generating meaningless geometric fittings or slope estimations based on sparse data.

[0046] If the number of time windows within the current preset historical neighborhood is sufficient, then proceed with the step of constructing a two-dimensional hysteresis loop: extract two time series from the current preset historical neighborhood: This corresponds to the time sequence formed by the integral of the total stress load on the blood vessel wall; This corresponds to the time sequence formed by the integral of the micro-circulation driven load; To eliminate the influence of absolute numerical differences among different patients or at different time periods on geometric morphology analysis, normalization is performed before constructing the two-dimensional hysteresis loop. In this embodiment, max-min normalization is used, and the data dimensions of the total stress load integral of the blood vessel wall and the microcirculation-driven load integral are taken as the dimensions to be analyzed. The maximum and minimum values ​​of the dimensions to be analyzed are taken from the time series within the current preset historical neighborhood. If the range of the dimensions to be analyzed is less than the preset noise threshold, such as 2... If the normalization value is set to 0, the area of ​​the two-dimensional hysteresis loop is forced to be 0 to avoid oscillations in the normalization calculation caused by the range approaching zero due to signal stability.

[0047] For each time window within the current preset historical neighborhood, the data points corresponding to each single pulsation time window in the two-dimensional coordinate system (Cartesian coordinate system) are connected in chronological order, with the normalized total stress load integral of the blood vessel wall as the abscissa and the normalized microcirculation driving load integral as the ordinate, to form a two-dimensional hysteresis loop.

[0048] Due to the periodic fluctuations and damping effects of the organism, the trajectory approaches each other, forming a closed or nearly closed loop structure, namely a stress-driven two-dimensional hysteresis loop. In an ideal undamped elastic pipe, pressure and flow change in phase, the phase space trajectory tends to be a straight line, and the enclosed area approaches zero. However, in microcirculation with tissue edema or congestion, fluid viscosity increases, and the establishment of driving energy lags behind the application of stress to the blood vessel wall, causing the trajectory to open and form a hysteresis loop.

[0049] Therefore, the system uses the Discrete Green's Formula (also known as the Shoelace Formula) to calculate the geometric area enclosed by the two-dimensional hysteresis loop, which is then used as the edema damping index for the current single pulsation time window. . The larger the value, the stronger the "devouring" effect of microcirculation on energy transfer, that is, the more severe the pathological fluid damping.

[0050] It should be noted that the two-dimensional hysteresis loop constructed in this embodiment does not describe the instantaneous phase difference within a single heartbeat waveform, but rather depicts the dynamic evolution path of the energy conversion efficiency of the microcirculation system as the load changes within a preset historical neighborhood (such as 60 heartbeats).

[0051] It should be noted that the method of calculating geometric area using the Discrete Green's formula is a well-known technique. In other embodiments of the present invention, the implementer may use other existing area calculation algorithms to solve for the geometric area, or may adjust the length of the preset historical neighborhood and the value of the preset noise threshold according to actual needs, which will not be elaborated further.

[0052] Preferably, in one embodiment of the present invention, please refer to Figure 2 The diagram illustrates a flowchart of a method for obtaining perfusion conversion efficiency according to an embodiment of the present invention, specifically including: Step S201: Analyze the linear characteristics of the integral of total stress load on the blood vessel wall and the integral of microcirculation-driven load to obtain the baseline slope.

[0053] In clinical monitoring settings, the real-time acquired arterial pressure and central venous pressure signals may be affected by various factors, such as sensor momentary jitter, patient body movement or respiratory movements, which may cause the characteristic data points of a few time windows to deviate from their true physical change trend and become "outliers". This affects the conventional least squares fitting and causes slope distortion. Therefore, Theil-Sen estimator is used to obtain the baseline slope. Considering that the data has been normalized for the subsequent construction of hysteresis loops and the calculation of their areas, and that the basic slope reflects the inherent proportional relationship between the original physical quantities, it should not be affected by the normalization scale transformation. Therefore, the unnormalized original data must be used. The coordinates of the data points corresponding to each time window are constructed with the total stress load integral of the blood vessel wall as the x-axis and the microcirculation driving load integral as the y-axis. Based on the data points corresponding to each time window in the current preset historical neighborhood, the Theil-Sen estimator is used to obtain the basic slope.

[0054] Specifically, the data points corresponding to any two time windows within the current preset historical neighborhood are combined, the slope of all combinations is calculated, and the median of the slopes of all combinations is taken as the base slope. , representing the linear characteristic of the integral of total stress load on the blood vessel wall and the integral of microcirculation-driven load.

[0055] When calculating the slope of a combination, the two time windows within the combination naturally contain temporal information. Combinations (j,m) and (m,j) are considered as the same slope calculation sample. If j is less than m, the slope corresponds to... If the denominator is 0, the calculation of the corresponding combination is skipped and not included in the median calculation.

[0056] Step S202: Combine the preset damping sensitivity coefficient, edema damping index, diastolic tension coefficient and preset stability constant to obtain the effective perfusion factor; combine the effective perfusion factor and the baseline slope to obtain the perfusion conversion efficiency.

[0057] Considering that simple linear conduction (characterized by the baseline slope) only describes the average proportion of energy transfer, neglecting the phase loss caused by fluid resistance, and that the severity of this loss depends not only on the magnitude of the damping itself (characterized by the edema damping index) but also on the nonlinear modulation of the vessel wall tension state, an effective perfusion factor is obtained by integrating a preset damping sensitivity coefficient, edema damping index, diastolic tension coefficient, and preset stability constant.

[0058] Then, the effective perfusion factor and the base slope are combined to obtain the perfusion conversion efficiency.

[0059] As an example, because when the blood vessel wall tension is normal ( When the blood vessels are relatively large, they have good elastic recoil ability, which can resist the compression of interstitial edema to a certain extent, resulting in a smaller loss of perfusion efficiency due to the same damping; conversely, when the blood vessels are in a paralyzed and relaxed state ( When the fluid is relatively small, the microcirculation lumen is easily compressed by external edematous tissue, resulting in a catastrophic energy cutoff effect from the same fluid damping, i.e., "low tension amplifies the destructive force of damping".

[0060] Based on this, for the current time window, a preset damping sensitivity coefficient will be used. With edema damping index The product of these factors is used as the numerator, representing the diastolic tension coefficient. and preset stability constant The sum of the values ​​is used as the denominator, and the difference between the constant 1 and the fractional ratio is used as the effective perfusion factor. , .

[0061] In this example, focusing on the monitoring of renal microcirculation, the renal capsule has poor compliance and is extremely sensitive to fluid resistance caused by interstitial edema. The value is set to 1.0 (or higher, such as 1.2) to ensure the negative correction effect of the edema damping index on the perfusion conversion efficiency; Take 0.1g to prevent extreme vasoparalysis ( Approaching This leads to numerical divergence.

[0062] Since the integral of the total stress load on the blood vessel wall and the integral of the microcirculation driving load have the same unit, the basic slope is dimensionless. The annular area calculated from the normalized coordinate data is also dimensionless; It is a time constant The quotient divided by the window duration is also dimensionless.

[0063] The infusion conversion efficiency is further truncated for negative values; when the calculated infusion conversion efficiency is negative, it is forcibly set to 0.

[0064] The base slope represents the average energy conversion benchmark within the current time window, without considering phase lag; while the effective perfusion factor represents the "survival rate" (i.e., how much energy is not lost) due to damping and tension states. Therefore, the effective perfusion factor and the base slope are multiplied and combined to obtain the effective perfusion factor. .

[0065] It should be noted that, The value of is negatively correlated with the interstitial compliance of the target tissue: the worse the compliance (the more easily it is compressed). The larger the value, the better. Implementers can determine the optimal value through calibration experiments based on the specific clinical application scenario. Adjust the preset stability constant value using the parameters.

[0066] The system will calculate , and current mean central venous pressure Encapsulated as a microcirculation state feature set , This is for use by subsequent modules.

[0067] The status labeling module 103 is used to map the current diastolic tension coefficient and perfusion conversion efficiency to a two-dimensional decision plane. Based on the spatial distribution of the current data points in the two-dimensional decision plane, combined with the overall characteristics of the central venous pressure within the current single pulsation window, the microcirculation fluid state is labeled.

[0068] Considering that the diastolic tension coefficient independently characterizes the physical contraction state of vascular smooth muscle (i.e., "vascular support force"), while the perfusion conversion efficiency quantifies the proportion of actual microcirculation perfusion converted by a unit cardiac load after edema damping coupling correction (i.e., "energy transfer efficiency"), the two exhibit orthogonal decoupling characteristics in physiological mechanisms. Therefore, the current diastolic tension coefficient and perfusion conversion efficiency are mapped to a two-dimensional decision plane. The spatial distribution of current data points in the two-dimensional decision plane intuitively reflects the independent contribution and coupling state of the two pathological mechanisms of vascular tone loss and fluid damping blockage. The overall characteristics of central venous pressure within the current time window directly reflect the level of venous return resistance and volume load. Therefore, based on the spatial distribution of current data points in the two-dimensional decision plane, combined with the overall characteristics of central venous pressure within the current single pulsation time window, the microcirculation fluid state is marked, the root cause of hemodynamic disturbances that lead to the disconnect between macroscopic blood pressure and the actual microcirculation perfusion level is accurately identified, and decision support information for assisting clinical intervention is generated.

[0069] Preferably, in one embodiment of the present invention, since there are significant individual differences in baseline vascular tone and baseline cardiac efficiency among different patients (e.g., hypertensive patients and normotensive patients), the current diastolic tension coefficient can be divided by a preset benchmark diastolic tension coefficient as the abscissa. The perfusion conversion efficiency divided by the preset benchmark conversion efficiency is used as the ordinate. It is mapped to a two-dimensional decision plane and uses a preset benchmark value to eliminate individual baseline differences.

[0070] The system includes an initialization calibration process to obtain preset reference tensile coefficients and preset reference conversion efficiencies. Specifically, the system sets a calibration counter to capture the data before system startup. Microcirculation state characteristic group In this example When the number of microcirculation state feature groups is insufficient, the system accumulates data in real time without outputting labeling results. When the number of features reaches a certain threshold, the system continues to accumulate data. When calculating quantities, the system calculates beforehand. indivual The arithmetic mean of the diastolic tension coefficients is used as the initial tension baseline, and the arithmetic mean of the perfusion conversion efficiency is used as the initial efficiency baseline. Meanwhile, the system loads a preset safe physiological range. As an example, based on large-scale historical normal population data, the system retrieves arterial pressure and central venous pressure data from strictly screened healthy subjects (excluding those with cardiovascular disease, diabetes, renal insufficiency, and a history of hypertension) from the database. The diastolic tension coefficient and perfusion conversion efficiency data are extracted, and the range between the 5th percentile (P5) and the 95th percentile (P95) is used as the safe physiological range. The diastolic tension corresponds to [0.5, 2.0], and the perfusion conversion efficiency corresponds to [0.6, 1.0].

[0071] If the calculated initial tension baseline and initial efficiency baseline both fall within the corresponding safe physiological range, they are directly locked as the individual's baseline reference values. The initial tension baseline is used as the preset baseline relaxation coefficient, and the initial efficiency baseline is used as the preset baseline conversion efficiency.

[0072] If the calculated initial tension or initial efficiency benchmark exceeds the corresponding safe physiological range, it indicates that the patient may have been in a pathological state when monitoring was initiated (such as severe vascular paralysis due to shock or insufficient perfusion due to excessive diuresis). In this case, if the abnormal value is directly set as the benchmark, the system may mistakenly regard the state as "normal," leading to a sluggish response to further deterioration in subsequent monitoring (i.e., "false negative" misjudgment). Therefore, this embodiment of the invention uses a boundary truncation method to truncate the initial benchmark that exceeds the corresponding safe physiological range. For example, if the initial tension benchmark is 2.1, it is truncated to 2.0; if the initial efficiency benchmark is 0.5, it is truncated to 0.6.

[0073] It should be noted that the number of healthy subjects retrieved and the duration of the analyzed signals can be customized by the implementer based on the significance requirements of clinical statistics. To ensure the stability and representativeness of the statistical results, it is recommended that the number of subjects retrieved be no less than 100, and the signal analysis time for each subject be no less than 10 minutes.

[0074] Preferably, in one embodiment of the present invention, in a two-dimensional decision plane, if Significantly reduced (e.g.) )and Maintain normal (e.g.) This indicates that the loss of vascular wall tension is the primary problem, while microcirculatory permeability remains relatively good, therefore... and The region is set as the preset first state region. If the current data point is in the preset first state region, it is marked as the preset first state. For example, if the identifier is set to D1, it reflects that the current microcirculation fluid state is dominated by the significant lack of physical support capacity of vascular smooth muscle, while the microcirculation energy transfer efficiency is not significantly damped and blocked, which is a simple vascular tension deficiency (i.e., simple vascular paralysis state). If the ordinate of the current data point is ( If the duration of the blockade threshold (e.g., 0.7) and the collapse threshold (e.g., 0.3) is less than the preset blockade threshold (e.g., 0.3) does not exceed the preset safety time limit (e.g., 15 minutes), it indicates severe fluid damping leading to energy dissipation, but it has not yet evolved into irreversible loss of cellular metabolic function. It is still within the pathophysiological window period reversible by hemodynamic intervention. At this time, secondary discrimination should be performed in conjunction with mean central pulse pressure. When the average central venous pressure within the current time window is greater than the preset warning value (e.g., 12 mmHg), it indicates that the damping is mainly due to obstructed venous return, and it is marked as the preset second state, such as the identifier being set to D2-1. When the average central venous pressure within the current time window is less than or equal to the preset warning value, it indicates that the damping originates from fluid resistance caused by microcirculatory interstitial edema or endothelial swelling, and is marked as the preset third state, such as the identifier being set to D2-2; If the ordinate of the current data point is ( If the energy depletion of the microcirculation is less than the preset collapse threshold and the duration exceeds the preset safety time limit, it indicates that hemodynamic intervention can no longer reverse the energy depletion of the microcirculation, and is then marked as the preset fourth state, such as D3.

[0075] Among them, the mean central pulse pressure is the average value of the central pulse pressure within the time window, representing the overall characteristics of the central venous pressure.

[0076] It should be noted that the various judgment thresholds involved in this embodiment (such as 0.8, 0.9, 0.7, 0.3, and 12 mmHg) were determined based on preliminary clinical experiments and retrospective analysis of large-scale multi-center studies. Specifically, the system performs cluster analysis on hemodynamic data from hundreds of patients after cardiac surgery, and annotates them with clinical gold standards (such as renal function biochemical indicators and tissue oxygen saturation), selecting characteristic critical values ​​that can maximize the differentiation of different pathophysiological states as preset thresholds. Implementers can personalize and fine-tune the above thresholds according to the clinical management goals of different departments (such as ICU or operating room).

[0077] To enhance the stability of the decision-making system, the undefined empty area between each boundary threshold is set as a decision buffer area. When real-time data points drift into the buffer area due to physiological noise or short-term fluctuations, the system will maintain the labeling result from the previous moment. This mechanism effectively avoids frequent switching of labeling results (i.e., "jumping") caused by minor data jitter in critical states.

[0078] It should be noted that, in principle, the physical meanings of the horizontal and vertical axes in the two-dimensional decision plane can be arbitrarily chosen and interchanged. Choosing "tension ratio" as the horizontal axis and "efficiency ratio" as the vertical axis is more intuitive: the horizontal axis from left to right represents the change in vascular tension from "paralysis" to "tension," and the vertical axis from bottom to top represents the change in perfusion efficiency from "low" to "high." If the horizontal axis coordinates are interchanged, their respective boundary thresholds can be strictly reversed.

[0079] In summary, to address the technical problem of the disconnect between existing macroscopic blood pressure monitoring and the actual microcirculation perfusion level, making it difficult to accurately identify the root causes of various complex hemodynamic disturbances and thus provide effective decision support, this invention aims to provide an AI decision support system for dynamic target drug screening in acute kidney injury. This invention first divides the single-pulse time window in the preprocessing and decoupling module to obtain the diastolic tension coefficient, the integral of the total stress load on the vessel wall, and the integral of the microcirculation driving load. Further, in the analysis module, it extracts the edema damping index and perfusion conversion efficiency. Finally, through the state labeling module, it maps the current diastolic tension coefficient and perfusion conversion efficiency to a two-dimensional decision plane, and then combines this with the overall characteristics of the central venous pressure within the current single-pulse time window to label the microcirculation fluid state. This invention solves the problem of the disconnect between macroscopic blood pressure and microcirculation perfusion and the difficulty in identifying the root causes of disturbances by decoupling tension and energy characteristics, constructing a hysteresis loop to quantify damping and correcting efficiency, and mapping the decision plane to label the fluid state, thus providing decision support.

[0080] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0081] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An AI decision support system for dynamic target drug screening in acute kidney injury, characterized in that, The system includes: The preprocessing and decoupling module is used to divide the single pulsation time window based on the fluctuation of the arterial pressure signal. For the current single pulsation time window, the attenuation characteristics of arterial pressure during diastole are analyzed. Combined with the duration of the single pulsation time window, the diastolic tension coefficient is obtained. The square of the arterial pressure and the square of the difference between the arterial pressure and the central venous pressure are integrated to obtain the total stress load integral of the vessel wall and the microcirculation driving load integral. The analysis module is used to construct a two-dimensional hysteresis loop within the current preset historical neighborhood based on the integral of the total stress load of the blood vessel wall and the integral of the microcirculation driving load, and calculate the geometric area enclosed by the two-dimensional hysteresis loop as the edema damping index; based on the linear characteristics of the integral of the total stress load of the blood vessel wall and the integral of the microcirculation driving load, and combined with the result of correcting the edema damping index by the diastolic tension coefficient, the perfusion conversion efficiency is obtained. The status labeling module is used to map the current diastolic tension coefficient and the perfusion conversion efficiency to a two-dimensional decision plane. Based on the spatial distribution of the current data points in the two-dimensional decision plane, and combined with the overall characteristics of the central venous pressure within the current single pulsation window, the microcirculation fluid state is labeled.

2. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, The method for obtaining the single-pulse time window includes: The first derivative of the arterial pressure signal is obtained, and local maxima exceeding a preset noise threshold are identified as dividing points. Adjacent dividing points are then divided into a single beat time window.

3. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, The method for obtaining the diastolic tension coefficient includes: Based on the difference between the diastolic initiation pressure and the mean central venous pressure within the current single pulsation window, the arterial pressure signal is fitted using a two-element Windkessel model to obtain the absolute decay time constant; the diastolic tension coefficient is obtained by fusing the absolute decay time constant and the duration of the single pulsation window.

4. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, The method for obtaining the two-dimensional hysteresis loop includes: Within the current preset historical neighborhood, the data points corresponding to each single pulsation time window in the two-dimensional coordinate system are connected in chronological order, with the normalized total stress load integral of the blood vessel wall as the horizontal axis and the normalized microcirculation driving load integral as the vertical axis, to form a two-dimensional hysteresis loop.

5. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, The method for obtaining the infusion conversion efficiency includes: The linear characteristics of the integral of the total stress load on the blood vessel wall and the integral of the microcirculation-driven load are analyzed to obtain the basic slope; The effective perfusion factor is obtained by integrating the preset damping sensitivity coefficient, the edema damping index, the diastolic tension coefficient, and the preset stability constant; the perfusion conversion efficiency is obtained by integrating the effective perfusion factor and the baseline slope.

6. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 5, characterized in that, The method for obtaining the basic slope includes: Using the integral of the total stress load on the blood vessel wall as the abscissa and the integral of the microcirculation driving load as the ordinate, the basic slope is obtained using the Theil-Sen estimator based on the data points corresponding to each time window within the current preset historical neighborhood.

7. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 5, characterized in that, The method for obtaining the effective perfusion factor includes: The product of the preset damping sensitivity coefficient and the edema damping index is used as the numerator, the sum of the diastolic tension coefficient and the preset stability constant is used as the denominator, and the difference between the constant 1 and the fractional ratio is used as the effective perfusion factor.

8. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, Methods for mapping to a two-dimensional decision plane include: The current diastolic tension coefficient is divided by a preset benchmark diastolic tension coefficient as the horizontal axis, and the perfusion conversion efficiency is divided by a preset benchmark conversion efficiency as the vertical axis, and mapped onto a two-dimensional decision plane.

9. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 8, characterized in that, The method for labeling the state of microcirculation fluid includes: In the two-dimensional decision plane, if the current data point is in a preset first state region, it is marked as a preset first state; if the ordinate of the current data point is less than a preset blocking threshold, and the duration of being less than the preset collapse threshold does not exceed a preset safety time limit, and the average central venous pressure in the current time window is greater than a preset warning value, it is marked as a preset second state; if the average central venous pressure in the current time window is less than or equal to the preset warning value, it is marked as a preset third state; if the ordinate of the current data point is less than the preset collapse threshold, and the duration exceeds a preset safety time limit, it is marked as a preset fourth state.

10. The AI ​​decision support system for dynamic target drug screening in acute kidney injury according to claim 1, characterized in that, The geometric area enclosed by the two-dimensional hysteresis loop is calculated using the Discrete Green's Theorem.

Images