Method and system for blood filter patency detection based on flush impulse response

Abstract

The application discloses a blood filter patency detection method and system based on a flushing impulse response, and relates to the field of blood filter state monitoring, wherein the method comprises the following steps: introducing a controllable flushing impulse as an active excitation source during the operation of a blood purification device, inducing transient fluid disturbance of the blood filter without changing the main working state of the loop; synchronously collecting pressure / flow response in a preset time window to form an analyzable flushing response signal sequence; constructing a target response segment by identifying the flushing start and intercepting the decay segment, systematically identifying the filter dynamic decay characteristics based on the segment, and estimating the dynamic response parameters represented by the equivalent time constant; further, calculating the patency index and comparing it with a threshold value. Thus, real-time online quantitative analysis and determination of the patency of the blood filter are realized, and the sensitivity and interpretability of early blockage identification are improved compared with traditional passive monitoring.

Description

Technical Field

[0001] This application relates to the field of blood filter condition monitoring technology, and in particular to a blood filter patency detection method and system based on flushing pulse response. Background Technology

[0002] During extracorporeal blood purification procedures such as hemodialysis (HD) and continuous renal replacement therapy (CRRT), the patency of the blood filter (artificial kidney) is a key factor in ensuring treatment effectiveness and patient safety. However, the hollow fibers inside the blood filter are highly susceptible to clotting or blockage due to fibrin deposition and blood cell aggregation during prolonged contact with blood.

[0003] Currently, clinical monitoring of hemofilter status mainly relies on passive observation of hydrodynamic parameters such as venous pressure and transmembrane pressure (TMP), or on periodic measurement of the clearance rate of solutes (such as urea and myoglobin) in the dialysate. However, these monitoring methods have significant lag and limitations.

[0004] Specifically, pressure indicators such as transmembrane pressure typically only show significant abnormal increases when filter clogging reaches a considerably severe level, making it difficult to identify minor local blockages or membrane fouling in the early stages. This can easily lead to healthcare professionals missing the optimal intervention window. Furthermore, assessment methods based on solute clearance require offline sampling and testing, resulting in a considerable time delay. Moreover, the results are affected by various variables such as patient metabolic rate and dialysate formulation, making it impossible to support real-time, standardized online monitoring. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for detecting blood filter patency based on flushing pulse response, which aims to at least solve the problem of the lack of timely, sensitive, and quantifiable online monitoring and early warning of blood filter patency status in current related technologies.

[0006] In a first aspect, embodiments of this application provide a method for detecting the patency of a blood filter based on flushing pulse response. The method includes: during the operation of a blood purification device, controlling the fluid circuit of the blood purification device to inject a preset volume of flushing pulse into the input side of the blood filter to induce transient fluid disturbance corresponding to the flushing pulse inside the blood filter; in response to the injection of the flushing pulse, using a pressure sensor and / or a flow sensor to synchronously acquire the response signal of the blood filter under the transient fluid disturbance within a preset acquisition time window to generate a flushing response signal sequence; identifying the flushing start time from the flushing response signal sequence, and extracting a signal segment from the flushing response signal sequence that includes the attenuation process after the flushing start time to obtain a target response segment for patency assessment; analyzing the attenuation characteristics of the blood filter to the flushing pulse based on the target response segment to estimate a dynamic response parameter reflecting the current fluid dynamic characteristics of the blood filter; the dynamic response parameter includes at least an equivalent time constant; calculating a patency index characterizing the patency of the blood filter based on the dynamic response parameter, and comparing the patency index with a preset threshold to output a blood filter patency detection result.

[0007] Secondly, embodiments of this application provide a blood filter patency detection system based on flushing pulse response. The system includes: a flushing pulse control unit, used to control the fluid circuit of the blood purification equipment to inject a preset volume of flushing pulse into the input side of the blood filter during operation, so as to induce transient fluid disturbances corresponding to the flushing pulse inside the blood filter; a signal synchronization acquisition unit, used to synchronously acquire the response signal of the blood filter under the transient fluid disturbance within a preset acquisition time window in response to the injection of the flushing pulse, using a pressure sensor and / or a flow sensor, to generate a flushing response signal sequence; and a target fragment extraction unit, used to extract from the... The flushing response signal sequence identifies the flushing start time, and a signal segment containing the attenuation process after the flushing start time is extracted from the flushing response signal sequence to obtain a target response segment for patency assessment; a dynamic parameter analysis unit is used to analyze the attenuation characteristics of the blood filter to the flushing pulse based on the target response segment to estimate dynamic response parameters reflecting the current hydrodynamic characteristics of the blood filter; the dynamic response parameters include at least an equivalent time constant; a patency detection and evaluation unit is used to calculate a patency index characterizing the patency of the blood filter based on the dynamic response parameters, and compare the patency index with a preset threshold to output the patency detection result of the blood filter.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the blood filter patency detection method based on flushing pulse response according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the blood filter patency detection method based on flushing pulse response according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the blood filter patency detection method based on flushing pulse response according to any embodiment of this application.

[0011] The blood filter patency detection method and system based on flushing pulse response provided in this application can achieve at least the following technical effects:

[0012] (1) By applying a preset flushing pulse during the operation of the blood purification device, repeatable transient fluid disturbances are generated in the blood filter-extracorporeal circulation pipeline system, and the pressure / flow response is acquired synchronously within a limited time window. Since this response is directly governed by the physical fluid dynamic state such as the current equivalent resistance, effective flow cross-sectional area and local energy dissipation, the gradual restriction of the filter channel will be primarily reflected in the change of transient response morphology, thereby enabling the unobstructed state to be captured at the dynamic level with higher sensitivity, and realizing the detectability and trend characterization of early and local flow abnormalities.

[0013] (2) By locating the start of the flushing process and extracting the subsequent attenuation segment, the analysis is focused on the signal range strongly correlated with the excitation, reducing the impact of slow drift and uncorrelated fluctuations on the consistency of interpretation. Furthermore, based on the attenuation characteristics, dynamic response parameters containing at least an equivalent time constant are obtained, transforming the filter state from instantaneous numerical fluctuations into quantitative indicators that are closer to the inherent dynamic properties of the system. Thus, the complex response waveform is reduced to a set of comparable and traceable parameters, providing a stable measurement basis for continuous evaluation and cross-time period comparison.

[0014] This technical solution uses the flushing pulse as a standardized input to perform online dynamic identification of the blood filter, and maps the identified dynamic response parameters to a patency index to complete the threshold determination output. Thus, by constructing a closed-loop quantitative framework, the patency status is transformed from empirical observation into a calculable and standardized detection result, thereby supporting real-time trend monitoring and timely intervention triggering, improving the risk controllability and decision-making consistency during the operation of blood purification equipment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an example of a blood filter patency detection method based on flushing pulse response according to an embodiment of this application is shown;

[0017] Figure 2 A schematic diagram illustrating the operational mechanism of an example of a blood filter patency detection method based on flushing pulse response according to an embodiment of this application is shown.

[0018] Figure 3 The simulation curve of the blood filter flushing pressure response versus time obtained based on the first-order system impulse response model is shown.

[0019] Figure 4 Simulation curves of the blood filter patency index under different blockage ratios are shown.

[0020] Figure 5 A schematic diagram comparing ROC curves of different monitoring indicators used to determine the clogging status of blood filters is shown.

[0021] Figure 6 A structural block diagram of an example of a blood filter patency detection system based on flushing pulse response according to an embodiment of this application is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To address the limitations of traditional fluid dynamics monitoring, data-driven approaches are being explored to improve assessment accuracy. For example, some studies attempt to use machine learning algorithms (such as gradient boosters) to perform multidimensional modeling of collected pressure and flow data, aiming to uncover latent features beyond transmembrane pressure to predict coagulation risk. However, because the input data is essentially passively observed static or quasi-static physical quantities, the amount of information it contains about changes in the filter's internal microstructure is limited. This results in a limited improvement in the accuracy of such models in predicting early coagulation, and the fundamental problem of data lag remains unresolved.

[0024] In other aspects of the extracorporeal circulation system, such as the monitoring of vascular access (arteriovenous fistula or artificial blood vessel), some intelligent monitoring solutions have emerged. Some researchers have proposed using acoustic sensors to collect vascular murmurs or collecting fingertip photoplethysmography (PPG) signals, combined with artificial intelligence algorithms such as convolutional neural networks, to identify the stenosis or blockage of the access. While these non-invasive methods have achieved some success in the long-term prognostic assessment of vascular access, their monitoring primarily focuses on the patient's own vascular access, and their signal characteristics are difficult to map to the hydrodynamic state of the blood filter itself. Therefore, they cannot be directly used to assess the degree of local blockage or membrane fouling of the hollow fibers inside the blood filter.

[0025] Furthermore, some pipeline monitoring technologies propose detecting leaks or blockages by applying pressure to the pipeline and using pipeline pressure response models. However, these technologies are primarily designed for pipeline systems with simple components and structures. Blood purification operates in an extremely complex environment, involving highly viscous whole blood with non-Newtonian fluid properties, a complex semi-permeable membrane structure in blood filters, and bidirectional fluid interaction between the blood and dialysate sides. In particular, simple pipeline pressure response models cannot adapt to the complex fluid dynamics environment of an artificial kidney, making it difficult to directly apply these technologies to the real-time monitoring of blood filter patency.

[0026] In summary, although current technologies have made progress in data analysis algorithms, peripheral pathway monitoring, and simple pipeline testing, there is still a lack of an effective detection method that can actively induce responses in the blood filter, a core component, during the operation of blood purification equipment, and can quantitatively assess its microscopic patency in real time.

[0027] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0028] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0029] Figure 1 A flowchart illustrating an example of a blood filter patency detection method based on flushing pulse response according to an embodiment of this application is shown.

[0030] Regarding the execution subject of the method in this application embodiment, it can be any controller or processor with computing or processing capabilities, such as a data processing controller. It can be integrated into a blood purification device, which may also include a blood filter, a flushing pulse generation module, and a signal acquisition module. By constructing standardized excitation conditions with controllable flushing pulses and extracting dynamic response parameters such as the equivalent time constant through system identification of the transient decay process, a thresholdable patency index output mechanism is formed. This transforms patency assessment from a passive observation of the natural operating state to an active quantitative characterization of the filter's equivalent hydrodynamic characteristics.

[0031] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0032] like Figure 1 As shown, in step S110, during the operation of the blood purification device, the fluid circuit of the blood purification device is controlled to inject a preset volume of flushing pulse into the input side of the blood filter, so as to induce transient fluid disturbance corresponding to the flushing pulse inside the blood filter.

[0033] Here, during the operation of the blood purification equipment, a preset volume of flushing pulse is injected into the blood filter input side via the control fluid circuit. For example, by applying a preset flushing pulse to the external fluid circuit during the operation of the blood purification equipment, repeatable transient fluid disturbances are generated in the blood filter-circuit assembly, where the external fluid circuit is the circuit assembly of the blood purification equipment. More specifically, without changing the basic operating mode (such as basic flow setting and loop closed-loop stability), the dynamic response of the blood filter is actively stimulated by a "small disturbance input" with controlled amplitude and duration, so that the fluid resistance, compliance, and fiber bundle channel state inside the blood filter are manifested in a "transient form".

[0034] In some implementations, a flow increment command can be superimposed on the existing basic flow control command. This flow increment command satisfies the volume constraint within a preset pulse duration, thereby ensuring that the injection volume of the flushing pulse is repeatable and measurable. At the same time, an upper limit can be set on the increment amplitude, and the triggering time of the pulse injection can be limited to the stable operating range of the equipment (e.g., avoiding periods of rapid pump speed adjustment, ultrafiltration strategy switching, or abnormal alarm handling) to reduce the impact of disturbances on the stability of the loop.

[0035] It is worth noting that the flushing pulse in this embodiment is a hydrodynamic excitation applied to the blood filter assembly in the extracorporeal circulation loop. The "preset volume" and "duration" of this pulse are strictly limited within the device's hydrodynamic safety range; for example, the flow rate increment is much smaller than the fluctuation tolerance of the baseline blood flow velocity, and the pulse volume is only a tiny fraction of the dead space volume of the tubing. Therefore, this transient fluid disturbance mainly generates a pressure difference response between the blood filter's input and output ends, and its energy is rapidly attenuated after passing through the filter and subsequent tubing, without substantially affecting the hemodynamic state of the object connected to the other end of the loop. Therefore, the execution logic of this method is always limited to the physical operation of the blood purification device and its tubing consumables, aiming to detect the mechanical patency of the extracorporeal device (blood filter), rather than performing direct medical treatment on the organism.

[0036] In this way, transient fluid disturbances corresponding to the flushing pulses are generated inside the blood filter. In healthy filters, these disturbances typically exhibit a "rapid response and smooth decay" characteristic. However, in the presence of membrane fouling, local blockage, or fibrous bundle coagulation, the decay process of the disturbances is significantly prolonged or exhibits a tailing effect. This transforms the "early, local, and progressive blockages," which are difficult to distinguish through traditional passive observation, into transient response differences that can be captured by sensors, providing identifiable input stimuli for quantifying patency. Simultaneously, because the pulse volume and duration are controlled, comparable detection can be achieved under different times, different patients, or different equipment operating conditions.

[0037] It should also be noted that the "operation process of the blood purification device" in the embodiments of this application can be understood as the working state of the blood purification device when performing its fluid loop drive and monitoring functions, and is not limited to clinical treatment scenarios where blood purification operations must be performed on the human body. Specifically, the operation process may include, but is not limited to: equipment factory testing, maintenance and repair, self-test calibration, pre-operation checks after consumable loading, and functional testing processes of the blood filter and tubing loop under simulated loop or alternative fluid media (e.g., physiological saline, calibration solution, or simulated solution of equivalent viscosity). In these scenarios, the blood filter, as the tested component in the loop, is connected to the fluid loop, and the injection of flushing pulses and the acquisition of response signals can be completed on the device side to assess the patency of the blood filter or the loop dynamics state, thereby being suitable for online detection and quality control at the device end and reducing dependence on specific clinical operation subjects.

[0038] In step S120, in response to the injection of the flushing pulse, the response signal of the blood filter under transient fluid disturbance is synchronously acquired within a preset acquisition time window using a pressure sensor and / or a flow sensor to generate a flushing response signal sequence.

[0039] It should be noted that the disturbance caused by the flushing pulse is a short-term process. If the sampling time window is too short, it will truncate the attenuation tail. If it is too long, it will introduce slow drift unrelated to blockage. Therefore, a balance needs to be struck between "covering the complete attenuation process" and "suppressing the influence of operating condition drift".

[0040] Specifically, a data acquisition time window centered on the pulse injection moment can be set. The first half of the window is used to acquire the steady-state background (facilitating subsequent baseline estimation), while the second half covers the peak value and its decay tail. Synchronous acquisition means not only sampling on the same time axis but also aligning multiple channel signals according to the same trigger reference to avoid peak value shifts caused by sampling delays, which would affect the accuracy of subsequent segment truncation and fitting. This results in a flushing response signal sequence that reflects the entire transient disturbance process. Therefore, compared to the slow trend relying solely on a single pressure index, this sequence includes transient amplitude changes, decay patterns, and possible tailing characteristics, allowing changes in filter internal resistance and compliance to be observed with higher temporal resolution, thereby improving the sensitivity and specificity of subsequent unobstructedness assessment.

[0041] In step S130, the flushing start time is identified from the flushing response signal sequence, and a signal segment containing the attenuation process after the flushing start time is extracted from the flushing response signal sequence to obtain the target response segment for smoothness assessment.

[0042] It should be noted that dynamic parameter estimation should be based as much as possible on the "main response segment triggered by the flushing pulse" rather than the segment affected by noise, drift or other operational events. Therefore, it is necessary to first accurately locate the time anchor point of the flushing event, and then extract the target segment containing the peak and decay main process accordingly.

[0043] More specifically, the start time can be identified by combining signal abrupt change characteristics. For example, the first-order difference or gradient sequence can be calculated for the sequence, and the moment when the gradient first exceeds a preset trigger threshold can be taken as the flushing start time. Then, the peak point of the response can be searched backward with the start time as a reference to determine the peak time, and the segment termination point can be determined further based on the attenuation characteristics, such as "attenuation to below a preset proportional coefficient of the peak" or "attenuation derivative recovering to the steady-state noise level" as the termination criterion. To reduce the comparability differences under different pulse amplitudes or different operating conditions, amplitude normalization or baseline correction can be performed on the truncated segments to make the fitting focus more on the attenuation pattern rather than the absolute dimension. Thus, in complex operating backgrounds, the effective information segment can be separated from the complete acquisition sequence, avoiding the introduction of slow drift or random noise unrelated to blockage into parameter estimation, thereby improving the stability and repeatability of dynamic parameters.

[0044] In step S140, the attenuation characteristics of the blood filter to the flushing pulse are analyzed based on the target response fragment to estimate the dynamic response parameters that reflect the current hydrodynamic characteristics of the blood filter. These dynamic response parameters include at least the equivalent time constant.

[0045] It should be noted that a blood filter can be equivalent to a dynamic system with "resistance-compliance" characteristics under transient disturbances. Blockage or membrane fouling will change the equivalent resistance and effective fluid channel, resulting in a slower decay rate, an increased time constant, or enhanced tailing.

[0046] In some implementations, the target response segment can be fitted to a first-order inertial decay model, using the equivalent time constant as the core dynamic response parameter. Specifically, before fitting, necessary bias removal processing (such as subtracting the steady-state baseline) is performed on the segment, and robust parameter solving strategies (such as least squares fitting or log-linearized regression) are used to estimate the time constant. When a significant long tail or multi-timescale characteristics are detected in the decay tail, a multi-segment fitting or multi-order model selection mechanism can be introduced to avoid distortion of the fitting of a single model to extreme clogging states. In addition, reasonable parameter boundaries and convergence criteria can be set during parameter estimation to prevent non-physical results caused by noise (such as negative time constants or abnormal divergence). Thus, waveform changes that originally relied on empirical interpretation are transformed into physically meaningful and directly comparable dynamic response parameters. In particular, the equivalent time constant can directly reflect the filter's recovery speed to disturbances, making it more sensitive to early local clogging. Compared with only observing slowly varying indicators such as transmembrane pressure, dynamic parameters derived from the system identification results after active excitation are more robust to operating condition drift.

[0047] In step S150, a patency index characterizing the patency of the blood filter is calculated based on the dynamic response parameters, and the patency index is compared with a preset threshold to output the patency test result of the blood filter.

[0048] It should be noted that although a single parameter can reflect changes in a certain dimension, it may still be affected by individual differences, sensor offset, or changes in operating conditions in actual operation. Therefore, it is necessary to map dynamic parameters into more engineering-usable metrics and achieve executable detection output through threshold determination.

[0049] More specifically, the equivalent time constant is oriented to be consistent (for example, a larger time constant indicates slower decay and more severe blockage), and a patency index is constructed accordingly to ensure that the patency index and the degree of patency remain monotonically correlated. The threshold can be a fixed threshold or an individualized benchmark threshold established based on the initial health state, and it is updated in combination with trend stability during long-term operation to reduce systematic bias caused by different patients or different filter batches.

[0050] In addition, the blood filter patency test results output by the test can include not only "patency / abnormal" results, but can also be expanded into graded warnings (such as subcritical warnings and severe blockage alarms) to support different levels of alert strategies or equipment protection strategies.

[0051] This application's embodiments transform the results of dynamic system identification into directly usable engineering judgment signals, achieving a closed loop of "real-time quantification - standardized comparison - result output." Through the patency index and threshold mechanism, early warnings can be provided before blockage leads to significant transmembrane pressure anomalies, reducing the risk of missing the intervention window and improving the interpretability and feasibility of real-time intervention operations.

[0052] Regarding the operational details of injecting flushing pulses in step S110, in some examples of embodiments of this application, the current basic flow rate command of the flow-driven pump in the fluid circuit is obtained in real time; the instantaneous flow rate increment is calculated based on the preset volume and preset pulse duration, and the instantaneous flow rate increment is superimposed on the current basic flow rate command to construct a composite flow control command; the composite flow control command is used to drive the flow-driven pump to perform an instantaneous acceleration action, thereby maintaining the basic operating flow rate while superimposing a square wave or step flow pulse on the input side of the blood filter.

[0053] Here, by superimposing a controlled increment on the existing closed-loop operation command, the device outputs a repeatable and measurable small disturbance input without changing the basic operation target, thereby applying a standardized excitation to the fluid network inside the blood filter.

[0054] In some implementations, the instantaneous flow rate increment can be configured as a square wave or a step, maintaining a stable increment amplitude within a preset pulse duration to ensure consistency between the injection volume and duration of the flushing pulse. Simultaneously, an upper limit for the increment amplitude and an acceleration constraint can be set to prevent pump drive saturation or the introduction of excessive loop pressure fluctuations. After the flow-driven pump performs an instantaneous acceleration action using this composite flow control command, a square wave or step flow pulse will be superimposed on the blood filter input side, generating transient fluid disturbances within the blood filter.

[0055] Therefore, compared to monitoring methods that rely on passive trends, this active pulse excitation can amplify filter dynamic differences and form an identifiable decay response when early local blockage or membrane fouling has not yet significantly increased transmembrane pressure. At the same time, since the injection volume is controlled, it is convenient to conduct comparability detection and subsequent parametric modeling across time periods and operating conditions.

[0056] Accordingly, regarding the implementation details of generating the flushing response signal sequence in step S120, in some examples of embodiments of this application, the original sampling signal is acquired using a pressure sensor and / or a flow sensor, and the original sampling signal before the flushing pulse injection is buffered based on a sliding time window; the statistical mean of the original sampling signal within the sliding time window is calculated as the steady-state baseline value; the steady-state baseline value is subtracted from the original sampling signal within the preset acquisition time window to eliminate the DC bias component and obtain the debiased dynamic waveform.

[0057] In some implementations, the raw sampling signal is acquired using a pressure sensor and / or a flow sensor, and the raw sampling signal before the flushing pulse injection is buffered in real time through a sliding time window. The statistical mean within the sliding time window is calculated as the steady-state baseline value. Then, the steady-state baseline value is subtracted point by point from the raw sampling signal within the preset acquisition time window to eliminate the DC bias component and obtain the debiased dynamic waveform.

[0058] It should be noted that slow-change drift exists during the operation of blood purification equipment (such as sensor zero-point drift, changes in loop temperature, or slow changes in fluid viscosity). If baseline correction is not performed first, it will lead to attenuation fitting deviation and reduce the stability of smoothness assessment. Therefore, using the sliding window mean as the baseline can achieve online adaptive correction without introducing offline calibration.

[0059] Then, the debiased dynamic waveform is decomposed into multiple scales using discrete wavelet transform, the high-frequency coefficients after decomposition are filtered using a soft threshold function, and the flushing response signal sequence that retains transient edge features is reconstructed through inverse wavelet transform.

[0060] Regarding the specific implementation details of discrete wavelet transform, wavelet basis functions (such as Daubechies wavelet or Symlet wavelet) that match the transient edge features of the flushing response signal sequence and a preset decomposition level L can be selected. L-level wavelet decomposition is then performed on the bias-free dynamic waveform to obtain the set of high-frequency detail coefficients {D} from level 1 to level L. j} and the low-frequency approximation coefficient A of the Lth layer L The number of decomposition layers L can be determined based on the sampling frequency and the duration of the flushing pulse, ensuring that at least one layer of detail coefficients corresponds to a frequency band that covers the main transient frequency range induced by the flushing pulse. Subsequently, the high-frequency detail coefficients D of each layer are... j A soft thresholding function is used for coefficient shrinkage. This threshold can be adaptively determined based on the noise level, for example, by estimating the noise standard deviation based on the absolute median of the detail coefficients in the highest frequency layer, and setting the threshold for each layer accordingly (either a uniform threshold or a layer-dependent threshold that varies with the number of layers). This suppresses random noise while avoiding excessive attenuation of transient edges. Finally, the thresholded detail coefficients of each layer are compared with the low-frequency approximation coefficient A. L The common input wavelet inverse transform is used to reconstruct the signal, and the reconstructed time-domain waveform is output as the flushing response signal sequence.

[0061] Therefore, while suppressing random noise and high-frequency interference, the transient edges and attenuation patterns excited by the flushing pulse are preserved as much as possible, thereby avoiding peak passivation or phase lag caused by traditional low-pass filtering, thus obtaining a flushing response signal sequence with higher signal-to-noise ratio and less morphological distortion.

[0062] Regarding the implementation details of extracting the target response fragment in step S130, in some examples of embodiments of this application, discrete gradient analysis is performed on the flushing response signal sequence to calculate the first-order difference between adjacent sampling points to generate a gradient sequence.

[0063] Here, the flushing pulse injection introduces a more stable event feature—a "slope abrupt change"—into the response signal. Compared to directly finding inflection points in the amplitude domain, the gradient domain is less sensitive to slow drift and baseline shift, and is more suitable for locating event trigger boundaries in the presence of noise. Specifically, the response signal sequence can be differentially analyzed point-by-point according to the sampling time sequence to obtain a gradient sequence reflecting the local rate of change. The differential values ​​can then be scaled and normalized according to the sampling period to ensure that the gradient threshold has portability across sampling rates. Thus, the occurrence of flushing events is transformed from an amplitude change problem into a rate of change abrupt change detection problem, providing a criterion basis with higher signal-to-noise ratio and less susceptibility to baseline drift interference.

[0064] Then, gradient mutation detection is performed, and the moment when the magnitude of the gradient sequence first exceeds the preset trigger noise threshold is marked as the flushing start moment.

[0065] It should be noted that the transient disturbances induced by the flushing pulse typically manifest as significant slope jumps within a short period, while the gradient fluctuations caused by operating noise and sampling jitter have limited amplitudes. Therefore, a clear and consistent trigger point can be established on the time axis through the "first threshold exceedance". Here, the trigger noise threshold can be set based on the gradient statistics of the stable interval before flushing injection, for example, by adding a certain number of standard deviations to the absolute mean of the gradient sequence in the stable interval as the threshold, in order to reduce false triggering. In this way, online positioning of the flushing event start point can be achieved without relying on external synchronization signals or additional hardware trigger lines, enhancing the consistency and repeatability of start time identification under different operating conditions.

[0066] Then, based on the start time of flushing, the maximum point of the flushing response signal sequence is searched backward to determine the peak value of the response signal and its corresponding peak time.

[0067] In some implementations, a local maximum point can be located as the peak value within a preset search window (the window length of which can be related to the pulse duration and the expected decay time) after the start of the flushing. This provides a stable anchor point for segment truncation and ensures that the decay analysis targets the same type of dynamic process rather than mixing in rising segments or other irrelevant fluctuations.

[0068] Then, using the peak moment as a reference, the system moves backward until the signal amplitude decays to below a preset proportional coefficient of the peak value of the response signal, and marks this moment as the end of decay.

[0069] In some implementations, the preset scaling factor can be set according to the expected signal-to-noise ratio and fitting stability, and a monotonicity constraint can be superimposed during the search process to avoid misjudgments caused by noise bounce. For example, it can be required that after reaching the scaling threshold, the scaling factor remains below the threshold for a certain number of consecutive points. This ensures that the attenuation segments obtained under different detection cycles and filter states have a consistent effective information density, reduces the tail noise introduced by excessively long segments that causes deviations in the fitting, and avoids unstable parameter estimation caused by excessively short segments.

[0070] Subsequently, signal data within the time interval from the peak time to the end of decay is extracted, and the peak value of the response signal is used to normalize the amplitude of the extracted data. The processed data sequence is then used as the target response segment for smoothness assessment.

[0071] It should be noted that the smooth assessment focuses more on the decay pattern and recovery speed, rather than the absolute amplitude; normalization can suppress the interference of pulse amplitude fluctuations, sensor gain differences and individual operating condition differences on the morphological analysis, so that subsequent model fitting and feature parameter estimation can focus more on dynamic differences.

[0072] Specifically, each sampling point of the truncated segment can be divided by the peak amplitude (with baseline subtraction added if necessary to eliminate residual DC terms), thereby mapping the target response segment to a uniform scale; and the peak value and normalized scale factor can be saved as auxiliary information for subsequent anomaly diagnosis or multidimensional feature fusion. Thus, a target response segment with uniform scale, consistent boundaries, and complete main information segments is obtained.

[0073] Regarding the implementation details of estimating the dynamic response parameters in step S140, in some examples of embodiments of this application, the dynamic response parameters include an equivalent time constant and an initial response amplitude.

[0074] More specifically, based on the lumped parameter theory of fluid networks, the blood filter is modeled as a first-order inertial system with fluid resistance characteristics and membrane compliance characteristics.

[0075] It should be noted that after the flushing pulse introduces transient disturbances on the input side of the blood filter, the effective flow channel, equivalent fluid resistance, and membrane / circuit compliance inside the filter jointly determine the "recovery rate" of the pressure (or flow) response from the peak to the steady state. This recovery process can be abstracted as a dynamic behavior with a "single dominant time scale".

[0076] Here, the blood filter can be viewed as a simplified fluid network composed of equivalent resistance and equivalent compliance, causing its free response after pulse excitation to exhibit an approximately exponential decay characteristic. Blockage or membrane fouling leads to an increase in equivalent resistance and an elastic change in effective volume, thereby altering the decay rate and decay pattern. Thus, the complex microscopic blockage state within the filter is transformed into a macroscopic kinetic index that can be stably characterized by a few parameters (such as the equivalent time constant).

[0077] The target response segment was fitted to a single exponential pressure decay function using a system identification algorithm to characterize the free response behavior of the blood filter after flushing pulse excitation.

[0078] Equation (1)

[0079] In the formula, Indicates the target response segment at time [time]. Pressure amplitude, This represents the initial response amplitude triggered by the flushing action. This indicates the peak time corresponding to the peak value of the response signal. This represents the equivalent time constant.

[0080] Here, the target response segment is fitted with a single exponential pressure decay function to characterize the free response after flushing pulse excitation. In equation (1), the peak time is used as the reference value. As the starting point of decay, in Within the attenuation range, calculate the pressure amplitude of the target response segment. It can be the amplitude after baseline correction.

[0081] Alternatively, the original amplitude can be retained for estimation first, and then a normalized segment can be used to enhance the stability of the fit. Where, This represents the initial response magnitude at the point where decay begins (i.e., (Amplitude at time), and utilize the equivalent time constant This reflects how quickly the response returns to a steady state.

[0082] Therefore, the "waveform attenuation pattern" is compressed into and Two engineering-significant quantitative parameters allow the dynamic differences caused by blockage to be stably output in parametric form, no longer relying on manual observation; among them It is often more sensitive to early local blockages. This can reflect the efficiency of excitation transmission and the change in transient amplitude.

[0083] A logarithmic linearization transformation is performed on the single exponential pressure decay function to construct the corresponding linear regression model:

[0084] Equation (2)

[0085] Among them, time deviation Set as the independent variable, and use the logarithmic pressure amplitude. The data points are set as the dependent variable, and then the least squares method is used to perform linear fitting on the transformed data points to obtain the slope of the regression line. and intercept .

[0086] Based on the coefficient correspondence between the linear regression model and the fitting results, the equivalent time constant is calculated. and initial response amplitude .

[0087] Here, to reduce the computational complexity of nonlinear exponential fitting and improve the stability of online parameter calculation, a logarithmic linearization transformation is performed on equation (1) to obtain the linear regression form shown in equation (2). Specifically, exponential decay in the logarithmic domain can be transformed into a function of time deviation. The linear relationship, where As an independent variable As the dependent variable, the intercept of the line corresponds to The slope corresponds to .

[0088] Specifically, the condition that satisfies the condition can be selected within the attenuation segment. The effective sampling points are subjected to logarithmic transformation, and a small number of sampling points near the peak are optionally removed (to reduce the disruption of linearity caused by transient ringing or sampling saturation near the peak), thus making the logarithmic domain data more consistent with the linear assumption. Therefore, the problem of "nonlinear exponential parameter estimation" is transformed into the problem of "linear fitting to find the slope and intercept," which reduces the computational burden and the instability caused by local optima or initial value sensitivity, enabling reliable output even under the computational constraints of real-time operation of the blood purification equipment. and .

[0089] In some implementations, after constructing a linear regression model, the least squares method can be used to linearly fit the transformed data points to obtain the slope of the regression line. With intercept And the parameter calculation is completed based on the above coefficient correspondence.

[0090] Specifically, reasonable validity constraints can be set on the fitting results to avoid non-physical interpretations (e.g., requiring...). To ensure Furthermore, the fitting residuals or coefficients of determination can be used as consistency checks to trigger re-extraction or refitting mechanisms in case of noise anomalies or segment truncation bias. Simultaneously, when amplitude normalization is used in the upstream step to enhance fitting stability, the recorded peak scaling factor can be used to recover... The physical dimensions of the time constant allow for the simultaneous validity of both "stable time constant estimation" and "interpretable initial amplitude". Therefore, a repeatable algorithm outputs the equivalent time constant and initial response amplitude, enabling online calculation of dynamic response parameters and cross-condition comparability. Furthermore, it provides a quantitative characterization of phenomena such as "slower decay and delayed recovery" during filter clogging.

[0091] Regarding the implementation details of estimating the dynamic response parameters in step S140, in some examples of embodiments of this application, the dynamic response parameters also include a response symmetry index.

[0092] More specifically, determine the total duration of the target response segment on the time axis. The target response segment is divided into associated first half intervals of equal duration in the time domain. and the second half of the interval .

[0093] For example, the total duration of the target response segment on the time axis is first determined. This duration can be determined by the time difference between the start and end sampling times of the target response segment (e.g., the time span from the peak time to the end of decay), with the segment start point as... The reference zero point is used to uniformly map the target response fragments to... Within the time domain; then the time domain is divided into the first half interval. and the second half of the interval It should be noted that when micropore occlusion or local contamination occurs in the blood filter, a "fluid stagnation tail" often appears at the tail of the response decay, causing the energy proportion in the latter half to increase.

[0094] Energy integration is performed on the target response segments, and the ratio of the cumulative energy in the first half of the interval to the cumulative energy in the second half of the interval is calculated to construct a response symmetry index. Computational model:

[0095] Equation (3)

[0096] In the formula, Indicates the target response segment at time [time]. The normalized pressure amplitude, for example, the amplitude normalized to the peak amplitude; Indicates at time The instantaneous energy intensity of a transient fluid disturbance can be understood as a characterization of the "energy intensity" of the transient disturbance at that moment (the squared amplitude is used to highlight the energy accumulation contribution of the tail segment and suppress the influence of positive and negative signs), thereby enabling the analysis of... By integrating the first and second halves separately, we obtain a comparison of the energy proportions in the two segments; therefore The physical meaning can be summarized as "whether the response energy is more concentrated in the first half".

[0097] More specifically, the integral can be approximated by summation in the discrete sampling domain: for each sampling point calculate The first and second halves are summed separately and then multiplied by the sampling interval. (Right now This yields a numerical approximation consistent with continuous integrals; furthermore, the normalized... Perform light smoothing (e.g., suppress high-frequency noise) to avoid the amplification effect of noise spikes on squared energy.

[0098] Thus, without significantly increasing computational complexity, we obtain a quantitative index that is more sensitive to tailing and relatively insensitive to baseline drift. This enables the "abnormal energy accumulation in the second half" to be output stably in the form of a single scalar, providing a directly usable input for the identification of blockage mechanisms and the fusion of multiple indicators.

[0099] Furthermore, by utilizing response symmetry indicators The fluid retention tailing effect induced by micropore blockage of fiber bundles within the blood filter was quantitatively characterized. This fluid retention tailing effect is negatively correlated with the effective flux of the blood filter, meaning that as the dissipation process of the flushing fluid within the filter is prolonged, the response symmetry index... The values ​​show a monotonic decreasing trend.

[0100] Here, the response symmetry index is used. The fluid retention tailing effect induced by micropore occlusion of fiber bundles within the blood filter was quantitatively characterized. When the effective flux of the filter decreases and local channels are restricted, the dissipation path of flushing disturbances within the filter is lengthened, resulting in a more pronounced tailing and delayed energy delivery in the latter half of the response decay. This increases the integral term in the latter half and drives the ratio. Decrease; therefore, It can be used as a specific characterizing factor for "tail dissipation extension" and is related to the equivalent time constant. To form complementarity: More emphasis is placed on overall recovery speed, while It places more emphasis on the energy distribution before and after the process and the degree of tailing.

[0101] In some implementations, it is possible to As part of the dynamic response parameters and , These factors are used together to construct the flow index or determine the grading threshold; for example, when the filter is in a relatively unobstructed state, the response energy is more concentrated in the first half. It is usually large and stable, but as the congestion develops and the tail intensifies, It exhibits a monotonically decreasing trend and can reflect the prolonged dissipation caused by local micropore occlusion in advance. Therefore, it improves the detection capability of tailing phenomena caused by early local blockage / membrane fouling, reduces the risk of missed detection in the tail of multiple time scales when relying solely on a single decay model, and thus enhances the sensitivity, robustness and interpretability of fluent assessment.

[0102] Regarding the implementation details of calculating the patency index in step S150, in some examples of embodiments of this application, the initial response amplitude measured by the blood filter in its initial healthy operating state is obtained as the standard reference amplitude. The measured equivalent time constant is used as the reference time constant. .

[0103] It should be noted that different filter batches, installation conditions, loop configurations, and sensor gains may lead to individual differences in absolute amplitude and absolute decay time. Directly using absolute parameters can easily introduce incomparability. Therefore, by using the baseline parameters of the same filter under healthy conditions as a reference, subsequent parameter changes can be attributed to filter flux decay, compliance changes, or kinetic changes caused by local blockage.

[0104] Specifically, and The results can be obtained by performing one or more standard flushing pulses during the initial stable operation phase after filter installation, or during a preset benchmark calibration procedure; when performing multiple flushing pulses, multiple estimates can be taken. and The statistical mean or median is used as and This is to reduce the impact of occasional noise and short-term operating condition fluctuations on the reference.

[0105] A multi-dimensional feature vector containing response rate, effective capacity, and morphological symmetry dimensions is constructed based on dynamic response parameters, and a smoothness index is calculated using a linear weighted fusion algorithm. :

[0106] Equation (4)

[0107] In the formula, Reference time constant Equivalent time constant of the present The ratio of represents the fluid conduction rate of the blood filter relative to its initial state; The normalized amplitude characterizes the effective transient compliance capacity of the blood filter; In response to the symmetry index; These are the weight coefficients for the corresponding feature dimensions, and they satisfy the normalization constraint. .

[0108] Among them, the smoothness index Used to comprehensively quantify the current patency status of the blood filter, and The higher the value, the closer the hydrodynamic performance of the blood filter is to its initial healthy operating state.

[0109] In equation (4), the attenuation is faster when the filter is more unobstructed. Smaller, i.e. Larger), and the initial response amplitude is closer to the benchmark ( Closer to 1), weaker tailing effect ( (larger), the weighted average of the three makes Maintained at a high level. Although filter clogging typically leads to an increase in static flow resistance, at the transient dynamic level, micropore blockage and the formation of deposits on the membrane surface significantly increase the system's viscous damping and reduce the membrane's effective elasticity, resulting in excessive dissipation of the transient response energy of pulse excitation, manifested as an increase in the initial response amplitude. Relative health baseline A decay occurs. Therefore, when filter patency decreases, this ratio... reduce. In response to the symmetry index, the focus is on reflecting the tailing effect and the degree of energy delay: enhanced tailing often leads to... The rate of decline is thus expanded from a single measurement (such as transmembrane pressure trend) to a comprehensive characterization covering three complementary types of information: "velocity-capacity-morphology," which can better distinguish different degradation modes such as early local blockage, overall resistance increase, or prolonged tail dissipation.

[0110] Regarding the explanation of the aforementioned feature weight coefficients, weight normalization can make... The numerical scale is more stable, threshold settings are easier to set, and cross-device migration is facilitated, while avoiding meaningless amplification or shrinkage of the exponent due to changes in the absolute value of the weights. Specifically, each weight coefficient can be set based on empirical rules or data-driven methods, for example, increasing it in scenarios that emphasize "early sensitivity". and To enhance the response to slower decay and enhanced tailing; to improve performance in scenarios that place greater emphasis on "effective flux changes". This enhances the response to amplitude decay and allows for necessary numerical normalization or amplitude limiting of each characteristic term to prevent any one term from dominating the overall index under abnormal noise.

[0111] In this embodiment, the smoothness index Used to comprehensively quantify the current patency status of the blood filter, and A higher value indicates that the blood filter's hydrodynamic performance is closer to its initial healthy operating state. When the filter is unobstructed, the response decays more quickly. Increased), and the initial response amplitude is closer to the reference ( Closer to 1), weaker tailing effect ( (larger), the weighted average of the three makes Maintaining at a high level; as blockage or membrane fouling progresses, slower attenuation, amplitude decay, or enhanced tailing will drive [the membrane]. This decrease provides a continuous quantitative characterization of smooth degradation.

[0112] Therefore, compared to passive pressure trend indicators, which only show significant anomalies when blockages are severe, the adopted... It can reflect changes through differences in dynamic response in the early stage of local blockage, supporting real-time, standardized, and thresholdable online monitoring; and because it includes reference amplitude and multi-dimensional feature fusion, it can reduce the interference of individual differences and operational fluctuations on single indicators, improve the ability to suppress false alarms and the robustness under different operating conditions.

[0113] Regarding the implementation details of solving the blood filter patency test results in step S150, in some examples of embodiments of this application, during the baseline establishment phase after device startup, the initial... The patency index sequence obtained from the second flushing test was used to calculate its arithmetic mean. and standard deviation and the arithmetic mean Set as the initial reference mean;

[0114] Equation (5)

[0115] Equation (6)

[0116] In the formula, This indicates the preset baseline sampling number. This indicates the first stage of the baseline establishment phase. The flow index was obtained from the second flushing test.

[0117] Here, during the baseline establishment phase after device startup, the initial data is buffered and statistically analyzed. The patency index sequence obtained from the second flush test This establishes a statistical baseline for the filter under "initial health / stable operation" conditions. Specifically, the arithmetic mean in equation (5) describes the central position of the initial flow level, and the standard deviation in equation (6) describes the dispersion of the flow index under noise and minor operating condition fluctuations in the initial stage. In this way, replacing the fixed threshold with a statistically significant "individualized baseline" significantly reduces the impact of individual differences caused by different filter batches, different loop configurations, and different sensor gains.

[0118] During the dynamic monitoring phase of blood purification equipment operation, the reference mean value updated at the previous moment is used. and standard deviation Build the dynamic alarm threshold for the current moment. :

[0119] Equation (7)

[0120] In the formula, The statistical safety factor is used to reflect risk tolerance.

[0121] Here, the dynamic alarm threshold will be... The setting is "several standard deviations below the reference mean," so that an anomaly detection is triggered when the fluent index shows a statistically significant decrease. Used to map risk tolerance A larger value indicates a more conservative threshold, a lower probability of false alarms, but a slower response time. A smaller threshold indicates a more sensitive threshold, allowing for earlier detection of downward trends, but it may also be more susceptible to noise.

[0122] In some implementations, it is possible to It can be set as a configurable parameter to adapt to different treatment modes or different alarm strategies, and can be adjusted accordingly. A lower limit is set to avoid frequent alarms caused by the threshold getting too close to the mean under extremely low noise estimation. Thus, the alarm threshold is tied to the statistical fluctuation level of the device itself, realizing a threshold mechanism that adapts to baseline drift and self-calibrates with noise level, avoiding missed or false alarms caused by fixed thresholds under individual differences and changes in operating conditions.

[0123] The smoothness index calculated at the current moment With dynamic alarm threshold Compare;

[0124] like If so, the blood filter is determined to have abnormal patency;

[0125] like If the blood filter is found to be functioning normally, the reference mean is iteratively updated using an exponentially weighted moving average algorithm to generate a reference mean for the next time step. :

[0126] Equation (8)

[0127] In the formula, This is a forgetting factor used to adjust the sensitivity to new data in order to track the baseline drift trend of blood filter performance.

[0128] As described before, the smoothness index Obtained by the fusion of multidimensional dynamic response parameters, it can reflect whether the filter's dynamic performance is close to its initial healthy state, and when clogging or membrane fouling progresses. It typically shows a downward trend. This is determined by comparing it to the lower bound of a reference mean updated over time. By comparison, statistically significant drops caused by kinetic degradation can be detected in advance, before significant transmembrane pressure anomalies occur in the filter, thus improving early warning capabilities and reducing reliance on human experience.

[0129] In equation (8), the forgetting factor is used. A trade-off is struck between "new observations" and "historical mean" to ensure that the reference mean smoothly tracks the slow drift of filter performance without being swayed by single noises or short-term fluctuations; among which The larger, More sensitive to new data and faster at tracking, but may be more prone to jitter; The smaller, It is smoother and more noise-resistant, but has a slower response. Therefore, long-term adaptive baseline maintenance is achieved without introducing complex model training, enabling dynamic thresholding. It can gradually adjust to slow variables such as system operating environment and liquid temperature / viscosity changes, thereby improving the consistency, sustainability and anti-false alarm capability of smooth detection under long-term operation.

[0130] In some examples of embodiments of this application, after the blood filter patency test result is output in step S150, further decision-making operations can be performed.

[0131] When the blood filter patency is determined to be normal, maintain the current flushing pulse injection cycle. It should be understood that flushing pulses are an active excitation operation, which, although used to obtain kinetic response information, may lead to increased loop disturbances, increased measurement burden, or additional adjustments to treatment flow control if the patency is stable. Therefore, maintaining the established injection cycle can achieve a balance between adequate monitoring and minimizing operational disturbances when patency is stable.

[0132] When a blood filter patency abnormality is determined, the reference mean updated at the previous time step is used. and standard deviation Constructing critical failure thresholds :

[0133] Equation (9)

[0134] In the formula, The limit safety factor is given, and the following conditions are met: .

[0135] Here, after outputting the smoothness detection result in step S150, in order to further classify the "abnormality" and trigger differentiated handling, the reference mean updated at the previous time step is used. with standard deviation Constructing critical failure thresholds .

[0136] In equation (9), the dynamic alarm threshold is... Based on this, a more stringent limit safety factor is introduced. and demand Thus This becomes a lower critical lower bound, representing a higher risk of failure. When When the flow rate falls below this lower bound, it can be considered that the bleaching capacity has progressed from a general anomaly to a more severe state that is closer to failure.

[0137] More specifically, and These can correspond to "early warning sensitivity" and "protection conservatism" respectively, and can be adjusted as configurable parameters according to treatment mode or risk level; simultaneously, Continuing to serve as a benchmark for fluctuation scales ensures that threshold grading maintains consistent statistical significance across different noise levels. This expands the single anomaly determination into a two-tiered threshold system that coordinates early warning and failure assessment. This allows the system to maintain early sensitivity while providing more reliable triggering conditions for severe congestion, avoiding over-handling or under-protection due to relying solely on a single threshold.

[0138] like If the blood filter is in a subcritical warning state, it will automatically generate a monitoring frequency adjustment command to shorten the injection cycle of the flushing pulse and perform encrypted monitoring, while outputting a first-level prompt message suggesting adjustment of the anticoagulant dosage.

[0139] In some implementations, the blockage of the blood filter may be in the early progression or localized stage, and there is still a window of opportunity to prevent deterioration through enhanced monitoring and early intervention. Specifically, the system can automatically generate monitoring frequency adjustment instructions, shortening the flushing pulse injection cycle to a preset encrypted cycle, in order to obtain more patency index observations on a shorter time scale, thereby improving the reliability of trend judgment and reducing misjudgments caused by occasional noise; at the same time, it outputs a first-level prompt message, prompting the operator to pay attention to the coagulation trend of the filter or check the current anticoagulation environment, but without directly triggering mandatory protection actions, thus maintaining the gradual nature of the treatment.

[0140] like If the blood filter is found to be in a severely clogged state, an automatic equipment protection control command will be generated to control the blood purification equipment to suspend ultrafiltration operation or reduce the blood pump flow rate. At the same time, a second-level audible and visual alarm will be triggered to recommend replacing the blood filter.

[0141] It should be noted that a patency index falling below the critical threshold indicates a severe degradation in filter kinetics. Continuing to maintain the original operating parameters may lead to transmembrane pressure exceeding physical safety limits or even equipment failure risks such as hollow fiber rupture. Therefore, the system triggers device-level protective controls (such as pausing ultrafiltration or reducing its rate) by describing the physical integrity of the extracorporeal circulation loop and simultaneously triggering an alarm to prompt for consumable replacement. This process aims to illustrate the fault-prevention operating logic of the equipment and its components, and does not involve diagnostic or treatment decisions for the patient's disease.

[0142] For example, the system automatically generates equipment protection control commands to control the blood purification equipment to suspend ultrafiltration or reduce the blood pump flow rate, thereby reducing loop load, inhibiting further risk escalation, and simultaneously triggering a second-level audible and visual alarm, suggesting replacement of the blood filter or immediate intervention. The specific actions of suspending ultrafiltration or reducing the blood pump flow rate can be configured with different levels of protection strategies to ensure consistency with equipment safety logic and ease of clinical confirmation. Thus, a safety protection closed loop is promptly introduced after severe degradation is identified, avoiding merely providing warnings and missing opportunities for risk control. This allows monitoring results to be directly translated into actionable equipment-side risk mitigation actions, thereby improving patient safety and system reliability.

[0143] Figure 2 A schematic diagram illustrating the operational mechanism of an example of a blood filter patency detection method based on flushing pulse response according to an embodiment of this application is shown.

[0144] like Figure 2 As shown, during the operation of the blood purification equipment, a preset volume of flushing pulse is injected into the input side of the blood filter through the pulse flushing injection unit, which induces transient fluid disturbance inside the blood filter corresponding to the flushing pulse; the pressure sensor array synchronously collects the pressure response signal or flow response signal under the action of the transient fluid disturbance, forms a flushing response signal sequence corresponding to the flushing pulse, and inputs the flushing response signal sequence to the pulse response analysis module.

[0145] The pulse response analysis module processes the flushing response signal sequence to extract the target response segment characterizing the attenuation process and estimate the dynamic response parameters. Then, the patency index calculation module generates a patency index based on the dynamic response parameters. Subsequently, the adaptive control / alarm module compares the patency index with a preset threshold or a dynamic alarm threshold, outputs the patency status of the blood filter, and triggers corresponding alarm / action strategies when an abnormality in patency is detected. For example, it generates encrypted monitoring instructions to adjust the flushing pulse injection cycle, or generates equipment protection control instructions to perform risk suppression operations such as pausing ultrafiltration or reducing the blood pump flow rate, thereby realizing online assessment and graded treatment of blood filter patency.

[0146] To verify the distinguishability of the proposed method under different degrees of clogging, a simplified dynamic response simulation model of the blood filter was established, and a comparative experimental analysis was conducted on the attenuation response under flushing pulse excitation. Specifically, the blood filter was equivalent to a first-order inertial system with fluid resistance and compliance characteristics, and different equivalent time constants were used. Characterizing different flow states: for example, setting a normal filter as... Partial congestion set as Severe congestion set as This is to illustrate the trend of slower system response and enhanced decay tail caused by increased congestion. In the simulation, the flushing pulse is approximated as a unit impulse input (to characterize the triggering effect of "short-term active excitation" on the system's free response). Therefore, assuming normalized system gain, the impulse response of the first-order system can be expressed as: Based on this, different values ​​can be obtained through numerical calculation. Pressure response curve under the specified conditions.

[0147] After obtaining the response curves for each group, the characteristic quantities related to the decay rate (such as those derived from the unobstructedness index construction logic of this application) can be extracted from the response curves according to the logic. The derived response rate term is used to calculate the flow index, thereby comparing the variation patterns of the flow index under different levels of congestion.

[0148] The above experimental setup allows us to verify the consistency chain of "increased time constant - slower decay - decreased smoothness index" while controlling the model complexity, thus providing an interpretable theoretical benchmark for the application of this method.

[0149] Figure 3 The simulation curve of the blood filter flushing pressure response versus time, obtained based on a first-order system impulse response model, is shown. The horizontal axis represents time (s), and the vertical axis represents the pressure response (normalized amplitude). The figure shows the response results under three equivalent unobstructed states: the equivalent time constant for a normal filter. (Blue line) indicates partial blockage. (Orange line) Severe blockage corresponds to (Green line). In this simulation setup, the flushing pulse is approximated as a unit impulse input. The impulse response of the first-order system is calculated, such that the response amplitude of each curve at the initial moment varies with... The amplitude increases and then decreases (for example, the blue line initially has the highest amplitude, followed by the orange line, and the green line has the lowest amplitude), and then decays exponentially thereafter.

[0150] Furthermore, from Figure 3 The indicated decay patterns show that: the normal filter (blue line) decays the fastest, with the response dropping rapidly within a short time; as the degree of clogging increases and the equivalent time constant increases, the decay process of the partially and severely clogged curves (orange and green lines) slows down significantly, and the tailing of the response becomes more pronounced, indicating that the dissipation process of the flushing disturbance within the filter is prolonged, and the system dynamic response becomes "hysteretic." Therefore, Figure 3 The simulation results confirm that by quantifying the flushing response decay characteristics (e.g., by using parameters such as the equivalent time constant), different flow states can be distinguished and a basis for calculating the flow index can be provided.

[0151] Figure 4Simulated graphs of the blood filter patency index under different blockage ratios are shown, where the horizontal axis represents the blockage ratio (e.g., increasing from 0% to 80%) and the vertical axis represents the patency index. In the simulation settings of this application embodiment, the equivalent time constant is first estimated based on the flushing response curves corresponding to each clogging ratio. Press again The smoothness index is calculated, thus mapping the rate of flushing response decay into a comparable quantitative indicator.

[0152] like Figure 4 It can be seen that as the congestion ratio gradually increases, the equivalent time constant... The corresponding increase leads to a smoothness index It exhibits a monotonically decreasing trend, showing an overall approximately linear (or piecewise approximately linear) decay relationship. This result indicates that increased clogging leads to an increase in the filter's equivalent fluid resistance and a slower dynamic response, thereby causing... It can quantitatively characterize the decline in unobstructedness, providing a simulation basis for subsequent threshold-based unobstructedness determination and trend monitoring.

[0153] Figure 5 This diagram illustrates a comparison of ROC (Receiver Operating Characteristic) curves when different monitoring indicators are used to determine the clogging status of blood filters. More specifically, under simulation conditions, the ROC curves of the patency index proposed in this application and the transmembrane pressure (TMP) monitoring indicator of the baseline method are compared when determining the clogging status of blood filters. To obtain this comparison result, 100 filter samples were constructed in this embodiment, with 50 unclogging samples as category 0 and 50 clogging samples as category 1. For each sample, a prediction score based on the patency index and a prediction score based on the transmembrane pressure were generated. Then, the corresponding true positive rate and false positive rate were obtained by scanning with different discrimination thresholds, and the ROC curves were plotted.

[0154] like Figure 5 As shown, the ROC curve corresponding to the patency index (the dark blue curve in the figure) is generally closer to the upper left corner, and its AUC (Area Under the Curve) is close to 1.0, indicating that this index has a stronger discriminative ability on simulated samples. In contrast, the AUC of the ROC curve obtained by simply using the transmembrane pressure prediction score is about 0.82, indicating relatively weaker discriminative performance. This comparison result shows that the patency index constructed based on the dynamic response characteristics extracted by flushing pulse excitation can provide a more sensitive and accurate clogging identification capability in the early stages of filter performance degradation, thus outperforming the traditional transmembrane pressure monitoring method that relies on static pressure trends.

[0155] Experimental results show that controlled flushing pulses can rapidly induce transient dynamic responses in the blood filter during device operation, with repeatable differences in decay rates and tailing patterns under different clogging states. System identification based on flushing response segments allows for stable estimation of dynamic response parameters such as equivalent time constant, initial response amplitude, and response symmetry, enabling the construction of a patency index that is more sensitive to changes in patency. Compared to traditional passive monitoring indicators such as transmembrane pressure, the patency index demonstrates stronger discriminative power (e.g., higher AUC) in simulation comparisons and is expected to reduce false alarm probability. This indicates that the method has the potential to identify filter patency decline trends earlier, providing a basis for timely filter replacement, increased monitoring, or adjustments to anticoagulation strategies, thereby reducing the risk of unplanned downtime and improving resource utilization efficiency during treatment.

[0156] From the limitations of current related technologies, patency assessment often relies on a posteriori signs such as increased transmembrane pressure and decreased solute clearance, or focuses monitoring on vascular channels rather than the filter itself, making it difficult to form quantifiable and graded online conclusions in the early stages of local micro-blockage or membrane fouling. In contrast, the main innovations of this application are: first, achieving "actively induced response" through flushing pulses, replacing simple trend observation with controlled excitation, thereby improving observability from a mechanistic perspective; second, introducing dynamic modeling and feature extraction, mapping the filter state to multi-dimensional features such as time constant, amplitude normalization, and symmetry, avoiding insufficient sensitivity caused by relying solely on average pressure changes; third, through a patency index and adaptive threshold / trend analysis framework, the evaluation results simultaneously consider individual baseline differences and online stability, and can output graded warnings and corresponding treatment actions, thus forming a closed-loop "detection-judgment-decision" chain. The above improvements do not require the introduction of additional complex hardware, but only enhance the algorithm level based on existing sensing and control capabilities, thus possessing strong engineering feasibility.

[0157] Future work can be carried out in the following two aspects: First, different blood equivalent viscosity, different anticoagulation conditions and controllable blockage degree can be introduced into the extracorporeal circulation simulation platform to collect flushing response data to test the robustness of parameter estimation and threshold strategy under complex working conditions, and further calibrate the weight settings of each feature dimension; Second, small-scale validation can be carried out under compliant conditions to focus on evaluating the balance between "early trend identification - false alarm control - intervention benefits", and to examine the impact of different filter models and loop configurations on the patency index baseline.

[0158] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0159] Figure 6 A structural block diagram of an example of a blood filter patency detection system based on flushing pulse response according to an embodiment of this application is shown.

[0160] like Figure 6 As shown, the blood filter patency detection system 600 based on flushing pulse response includes a flushing pulse control unit 610, a signal synchronization acquisition unit 620, a target fragment extraction unit 630, a dynamic parameter analysis unit 640, and a patency detection and evaluation unit 650.

[0161] The flushing pulse control unit 610 is used to control the fluid circuit of the blood purification device to inject a preset volume of flushing pulse into the input side of the blood filter during the operation of the blood purification device, so as to induce transient fluid disturbances corresponding to the flushing pulse inside the blood filter.

[0162] The signal synchronization acquisition unit 620 is used to respond to the injection of the flushing pulse by using a pressure sensor and / or a flow sensor to synchronously acquire the response signal of the blood filter under the transient fluid disturbance within a preset acquisition time window, so as to generate a flushing response signal sequence.

[0163] The target segment extraction unit 630 is used to identify the flushing start time from the flushing response signal sequence and extract a signal segment containing the attenuation process after the flushing start time from the flushing response signal sequence to obtain a target response segment for smoothness assessment.

[0164] The dynamic parameter analysis unit 640 is used to analyze the attenuation characteristics of the blood filter to the flushing pulse based on the target response segment, so as to estimate the dynamic response parameters reflecting the current hydrodynamic characteristics of the blood filter; the dynamic response parameters include at least an equivalent time constant.

[0165] The patency detection and evaluation unit 650 is used to calculate a patency index that characterizes the patency of the blood filter based on the dynamic response parameters, and compare the patency index with a preset threshold to output the patency detection result of the blood filter.

[0166] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions, which can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the above-described blood filter patency detection methods based on flushing pulse response.

[0167] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described blood filter patency detection methods based on flushing pulse response.

[0168] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a blood filter patency detection method based on flushing pulse response.

[0169] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0170] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting blood filter patency based on flushing pulse response, characterized in that, The method includes: During the operation of the blood purification equipment, the fluid circuit of the blood purification equipment is controlled to inject a preset volume of flushing pulse into the input side of the blood filter, so as to induce transient fluid disturbances corresponding to the flushing pulse inside the blood filter. In response to the injection of the flushing pulse, the blood filter's response signal under the transient fluid disturbance is synchronously acquired within a preset acquisition time window using a pressure sensor and / or a flow sensor to generate a flushing response signal sequence. The flushing start time is identified from the flushing response signal sequence, and a signal segment containing the attenuation process after the flushing start time is extracted from the flushing response signal sequence to obtain a target response segment for unobstructed assessment. Based on the target response fragment, the attenuation characteristics of the blood filter to the flushing pulse are analyzed to estimate the dynamic response parameters reflecting the current hydrodynamic characteristics of the blood filter; the dynamic response parameters include at least an equivalent time constant. The patency index, which characterizes the patency of the blood filter, is calculated based on the dynamic response parameters, and the patency index is compared with a preset threshold to output the patency detection result of the blood filter.

2. The method according to claim 1, characterized in that, The fluid circuit controlling the blood purification device injects a preset volume of flushing pulses into the input side of the blood filter, including: The current base flow rate command of the flow-driven pump in the fluid circuit is obtained in real time. The instantaneous flow rate increment is calculated based on the preset volume and preset pulse duration, and the instantaneous flow rate increment is superimposed on the current basic flow rate command to construct a composite flow control command; The composite flow control command is used to drive the flow drive pump to perform instantaneous acceleration, thereby maintaining the basic operating flow rate while superimposing a square wave or step flow pulse on the input side of the blood filter. In response to the injection of the flushing pulse, the blood filter's response signal under the transient fluid disturbance is synchronously acquired within a preset acquisition time window using a pressure sensor and / or a flow sensor to generate a flushing response signal sequence, including: The raw sampling signal is acquired using the pressure sensor and / or the flow sensor, and buffered based on a sliding time window before the flushing pulse is injected. The statistical mean of the original sampled signal within the sliding time window is calculated as the steady-state baseline value; The steady-state baseline value is subtracted from the original sampled signal within the preset acquisition time window to eliminate the DC bias component and obtain the debiased dynamic waveform. The debiased dynamic waveform is decomposed into multiple scales using discrete wavelet transform, and the high-frequency coefficients after decomposition are filtered using a soft threshold function. The flushing response signal sequence, which retains transient edge features, is then reconstructed by inverse wavelet transform.

3. The method according to claim 2, characterized in that, The step of identifying the flushing start time from the flushing response signal sequence and extracting a signal segment containing the attenuation process after the flushing start time from the flushing response signal sequence to obtain a target response segment for unobstructedness assessment includes: Discrete gradient analysis is performed on the flushing response signal sequence to calculate the first-order difference between adjacent sampling points to generate a gradient sequence; Perform gradient mutation detection and mark the moment when the amplitude of the gradient sequence first exceeds a preset trigger noise threshold as the flushing start moment; Based on the start time of the flushing, the maximum point of the flushing response signal sequence is searched backward to determine the peak value of the response signal and its corresponding peak time; Using the peak moment as a reference, the system moves backward until the signal amplitude decays to below a preset proportional coefficient of the peak value of the response signal, and marks this moment as the decay end moment. The signal data within the time interval from the peak time to the end time of attenuation is extracted, and the amplitude of the extracted data is normalized using the peak value of the response signal. The processed data sequence is used as the target response segment for smoothness assessment.

4. The method according to claim 3, characterized in that, The dynamic response parameters include the equivalent time constant and the initial response amplitude; The step of analyzing the attenuation characteristics of the blood filter to the flushing pulse based on the target response fragment to estimate the dynamic response parameters reflecting the current hydrodynamic characteristics of the blood filter includes: Based on the lumped parameter theory of fluid networks, the blood filter is modeled as a first-order inertial system with fluid resistance characteristics and membrane compliance characteristics; The target response segment was fitted to a single exponential pressure decay function using a system identification algorithm to characterize the free response behavior of the blood filter after flushing pulse excitation. ， In the formula, This indicates that the target response segment is at time [time]. Pressure amplitude, This represents the initial response amplitude triggered by the flushing action. This indicates the peak time corresponding to the peak value of the response signal. Indicates the equivalent time constant; A logarithmic linearization transformation is performed on the single exponential pressure decay function to construct the corresponding linear regression model: ， Among them, time deviation Set as the independent variable, and use the logarithmic pressure amplitude. The data points are set as the dependent variable, and then the least squares method is used to perform linear fitting on the transformed data points to obtain the slope of the regression line. and intercept ; Based on the coefficient correspondence between the linear regression model and the fitting results, the equivalent time constant is calculated. and initial response amplitude .

5. The method according to claim 4, characterized in that, The dynamic response parameters also include a response symmetry index; The estimation of the response symmetry index includes: Determine the total duration of the target response segment on the time axis. The target response segment is then divided in the time domain into associated first half intervals of equal duration. and the second half of the interval ; Energy integration is performed on the target response segment, and the ratio of the cumulative energy in the first half of the interval to the cumulative energy in the second half of the interval is calculated to construct a response symmetry index. Computational model: ， In the formula, This indicates that the target response segment is at time [time]. The normalized pressure amplitude, Indicates at time The instantaneous energy intensity of transient fluid disturbances; Using the aforementioned response symmetry index The fluid retention tailing effect induced by micropore blockage of fiber bundles inside the blood filter is quantitatively characterized; wherein, the fluid retention tailing effect is negatively correlated with the effective flux of the blood filter, so that when the dissipation process of the flushing fluid in the filter is prolonged, the response symmetry index... The values ​​show a monotonic decreasing trend.

6. The method according to claim 5, characterized in that, The calculation of the patency index, which characterizes the patency of the blood filter, based on the dynamic response parameters includes: The initial response amplitude measured by the blood filter under initial healthy operating conditions is used as the standard reference amplitude. And the measured equivalent time constant is used as the reference time constant; Based on the dynamic response parameters, a multi-dimensional feature vector is constructed, including response rate, effective capacity, and morphological symmetry dimensions. The smoothness index is then calculated using a linear weighted fusion algorithm. : ， In the formula, Reference time constant Equivalent time constant of the present The ratio of represents the fluid conduction rate of the blood filter relative to its initial state; The normalized amplitude characterizes the effective transient compliance capacity of the blood filter. This refers to the response symmetry index; These are the weight coefficients for the corresponding feature dimensions, and they satisfy the normalization constraint. ; Among them, the smoothness index Used to comprehensively quantify the current patency status of the blood filter, and The higher the value, the closer the hydrodynamic performance of the blood filter is to its initial healthy operating state.

7. The method according to claim 6, characterized in that, The step of comparing the patency index with a preset threshold to output the blood filter patency test result includes: During the baseline establishment phase after device startup, buffering and statistics are initially performed. The patency index sequence obtained from the second flushing test was used to calculate its arithmetic mean. and standard deviation and the arithmetic mean Set as the initial reference mean; ， ， In the formula, This indicates the preset baseline sampling number. This indicates the first stage of the baseline establishment phase. The flow index was obtained from the second flushing test; During the dynamic monitoring phase of blood purification equipment operation, the reference mean value updated at the previous moment is used. and the standard deviation Build the dynamic alarm threshold for the current moment. : ， In the formula, The statistical safety factor is used to reflect risk tolerance. The smoothness index calculated at the current moment With the dynamic alarm threshold Compare; like If so, the blood filter is determined to be malfunctioning; like If the blood filter is found to be functioning normally, the reference mean is iteratively updated using an exponentially weighted moving average algorithm to generate a reference mean for the next time step. : ， In the formula, This is a forgetting factor used to adjust the sensitivity to new data in order to track the baseline drift trend of blood filter performance.

8. The method according to claim 7, characterized in that, After comparing the patency index with a preset threshold to output the blood filter patency test result, the method further includes: When the blood filter is determined to be functioning properly, maintain the current flushing pulse injection cycle; When the blood filter is determined to have abnormal patency, the reference mean updated at the previous time step is used. and the standard deviation Constructing critical failure thresholds : ， In the formula, The limit safety factor is given, and the following conditions are met: ; like If the blood filter is determined to be in a subcritical warning state, a monitoring frequency adjustment command is automatically generated to shorten the injection cycle of the flushing pulse and to perform encrypted monitoring. Simultaneously, a first-level warning message is output to check the filter's coagulation risk. like If the blood filter is found to be in a severely clogged state, an automatic equipment protection control command is generated to control the blood purification equipment to perform an operation to suspend ultrafiltration or reduce the blood pump flow rate. At the same time, a second-level audible and visual alarm is triggered to recommend replacing the blood filter.

9. A blood filter patency detection system based on flushing pulse response, characterized in that, The system includes: The flushing pulse control unit is used to control the fluid circuit of the blood purification equipment to inject a preset volume of flushing pulse into the input side of the blood filter during the operation of the blood purification equipment, so as to induce transient fluid disturbances corresponding to the flushing pulse inside the blood filter; A signal synchronization acquisition unit is used to respond to the injection of the flushing pulse by using a pressure sensor and / or a flow sensor to synchronously acquire the response signal of the blood filter under the transient fluid disturbance within a preset acquisition time window, so as to generate a flushing response signal sequence. The target segment extraction unit is used to identify the flushing start time from the flushing response signal sequence and extract a signal segment containing the attenuation process after the flushing start time from the flushing response signal sequence to obtain a target response segment for smoothness assessment. A dynamic parameter analysis unit is used to analyze the attenuation characteristics of the blood filter to the flushing pulse based on the target response segment, so as to estimate the dynamic response parameters reflecting the current hydrodynamic characteristics of the blood filter; the dynamic response parameters include at least an equivalent time constant; The patency detection and evaluation unit is used to calculate a patency index that characterizes the patency of the blood filter based on the dynamic response parameters, and compare the patency index with a preset threshold to output the patency detection result of the blood filter.

Images