Peritoneal dialysis infection risk early warning method and system

Abstract

The application discloses a peritoneal dialysis infection risk early warning method and system, wherein the method comprises the following steps: continuously collecting the abdominal cavity pressure signal of a patient, and extracting disturbance features representing random fluctuations of the pressure from the pressure signal; calculating the disturbance entropy of the pressure signal and the time derivative thereof according to the disturbance features; dynamically adjusting the time window length of the infection risk prediction according to the disturbance entropy and the time derivative; calculating the progressive risk value of the peritoneal dialysis infection of the patient based on the adjusted time window, the disturbance entropy and the time derivative, and outputting early warning information when the progressive risk value reaches a preset threshold. The application can capture the acceleration features of the random fluctuations of the abdominal cavity pressure in real time, realize the early identification of the acceleration stage of the infection incubation period, and dynamically adjust the prediction window according to the risk, thereby significantly improving the timeliness and sensitivity of the early warning.

Description

Technical Field

[0001] This invention relates to the field of medical and health information processing technology, and in particular to a method and system for early warning of peritoneal dialysis infection risk. Background Technology

[0002] Peritoneal dialysis is one of the main renal replacement therapies for patients with end-stage renal disease, but peritonitis is its most common and serious complication, which can lead to peritoneal failure, increased hospitalization rates, and even death. Therefore, early and accurate prediction of infection risk is of great significance for improving patient prognosis.

[0003] Currently, clinical diagnosis of peritonitis mainly relies on patients' subjective symptoms (such as abdominal pain and cloudy diffusive fluid) and laboratory tests (white blood cell count and bacterial culture in diffusive fluid). However, by this time, the infection has often progressed to a significant stage, missing the optimal intervention window. To provide early warning, existing technologies have developed infection risk warning methods based on intra-abdominal pressure monitoring. For example, intra-abdominal pressure signals are continuously collected using pressure sensors, and statistical quantities such as the mean and standard deviation of pressure are calculated using a fixed time window (such as 24 hours). When these statistical quantities exceed preset thresholds, an alarm is triggered; or an alarm is triggered by directly monitoring a single abnormal change in pressure.

[0004] However, these methods have the following technical drawbacks:

[0005] 1. Fixed time windows cannot adapt to the dynamic changes in the infection process: In the early stage of the incubation period, the pressure fluctuations may be very weak. The statistics of fixed windows (such as 24-hour average) are not sensitive to short-term small changes, resulting in missed reports. When the pressure fluctuations accumulate to a level sufficient to trigger the threshold, the infection period has often already begun, and the early warning is seriously delayed.

[0006] 2. Inability to capture the unique "accelerated random fluctuation" characteristic of the incubation period: Clinical studies have shown that intra-abdominal pressure undergoes a random fluctuation process from stable to gradually increasing before infection occurs, i.e., the accelerated phase of the incubation period. Current technologies only focus on pressure amplitude or simple statistics, failing to effectively characterize this accelerated trend, thus making it difficult to identify risks in advance during the accelerated phase of the incubation period.

[0007] 3. Fixed warning response time and lack of adaptability: Traditional methods use a fixed-length prediction window, and assess risk at the same speed regardless of whether the patient is in a stable period or a period of rising risk. This results in an inability to respond in time when the risk evolves rapidly, while unnecessary false alarms may occur in the stable period due to the window being too short. Summary of the Invention

[0008] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for early warning of peritoneal dialysis infection risk, comprising the following steps:

[0010] The patient's intra-abdominal pressure signals are continuously acquired, and perturbation features characterizing random pressure fluctuations are extracted from the pressure signals. The perturbation entropy of the pressure signals is calculated based on the perturbation features. and its time derivative According to the perturbation entropy and the time derivative Dynamically adjust the length of the time window for infection risk prediction; based on the adjusted time window and the perturbation entropy and its time derivative The system calculates a progressive risk value for peritoneal dialysis infection in patients and outputs an early warning message when the progressive risk value reaches a preset threshold.

[0011] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, the disturbance feature includes the pressure change rate per unit time. and the standard deviation of the disturbance And constitutes the pressure disturbance characteristic vector. ;

[0012] The perturbation entropy Calculated using the following formula:

[0013] ;

[0014] in, The amplitude of the pressure signal in the th... The probability of occurrence within a quantization interval;

[0015] The time derivative It is used to characterize the degree of accelerated change in random stress fluctuations and as a basis for distinguishing the accelerated stage of the incubation period of infection.

[0016] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, wherein: the dynamic adjustment of the time window length for infection risk prediction is achieved through a dynamic compression model of the incubation period:

[0017] ;

[0018] in, This refers to the adjusted time window length. This is the preset initial window length; The standard deviation of the disturbance; This is the compression adjustment factor;

[0019] The compression adjustment coefficient According to the time derivative Adaptive adjustment: when And as it continues to increase, Increase accordingly, thereby increasing the window length Shorten the response time for risk assessment.

[0020] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, the progressive risk value is calculated using a stage transition probability function; the patient's condition is divided into a stable period. Incubation period and the infection period The transition probability is defined as:

[0021] ;

[0022] ;

[0023] in, for , A monotonically increasing function; for monotonically decreasing function A monotonically increasing function;

[0024] The progressive risk value The calculation formula is: ;in, , , These are dynamically adjustable weighting coefficients; This is the acceleration term used to characterize the degree of acceleration of the latency period.

[0025] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, the method performs the following self-evolutionary updates through an end-side smart chip:

[0026] Based on the real-time collected pressure disturbance characteristics and the corresponding infection event labels at specific time points, the compression adjustment coefficient is incrementally updated. ;

[0027] The weighting coefficients are incrementally updated based on the individual patient's historical risk curve and infection time. , , ;

[0028] The update aims to minimize the weighted sum of the false positive rate and the false negative rate within the historical window.

[0029] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, wherein: when the time derivative When the threshold is exceeded and continues to rise, the patient is determined to have entered the accelerated phase of the incubation period, and the following actions are performed simultaneously:

[0030] Increase the compression adjustment coefficient To further shorten the prediction window length Increase the weighting coefficient and To increase the probability of metastasis during the incubation period. and the acceleration term for the progressive risk value Contributions;

[0031] Output an early warning indicator for accelerated incubation period, which corresponds to a specific risk level.

[0032] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, it further includes a risk curve generation and trend prediction step:

[0033] Continuously calculate risk values ​​at different times Forming a risk curve ;

[0034] Calculate the derivative of the risk curve with respect to time. And based on the sign and rate of change of the derivative value, predict the future trend of risk evolution;

[0035] when When the risk exceeds the second threshold, a warning of rapid risk escalation is triggered.

[0036] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, it further includes a threshold adaptive calibration step:

[0037] Collect historical stress disturbance characteristics of patients and the corresponding times of infection events, and use a sliding time window to calculate the disturbance entropy. and progressive risk value Distribution;

[0038] Based on the specified early warning sensitivity, the first percentile and the second percentile of the distribution are selected as the individualized incubation period threshold and the individualized high-risk threshold, respectively; wherein the incubation period threshold is determined based on the risk value distribution during the period when no infection has occurred, and the high-risk threshold is determined based on the risk value distribution within a preset time window before the infection occurs;

[0039] The latency threshold and the high-risk threshold are dynamically updated based on the real-time pressure disturbance trend and used as the preset threshold for the current moment.

[0040] As a preferred embodiment of the peritoneal dialysis infection risk early warning method of the present invention, it further includes an early warning information interaction step:

[0041] When the progressive risk value When the incubation period threshold or high-risk threshold is reached, the edge smart chip triggers a multi-channel warning; where reaching the incubation period threshold corresponds to the incubation period risk level, and reaching the high-risk threshold corresponds to the high-risk level.

[0042] The multi-channel early warning includes mobile terminal notifications, dialysis machine display screen prompts, and push notifications from the medical monitoring platform interface.

[0043] It also outputs the corresponding risk level label and suggested follow-up visit time window based on the risk level.

[0044] This invention also provides a peritoneal dialysis infection risk early warning system, applied to the above method, including:

[0045] The pressure acquisition module is used to continuously acquire intra-abdominal pressure signals and generate pressure disturbance feature vectors.

[0046] An edge-side intelligent chip computing module, integrated into a dialysis device or portable terminal, includes:

[0047] Pressure feature extraction unit for real-time calculation , , and ;

[0048] Window adaptive unit, used to execute dynamic compression model and update compression adjustment coefficients. ;

[0049] The risk progression assessment unit is used to calculate the transition probability and risk value during the execution phase. generate;

[0050] Incremental learning units are used to incrementally update data based on historical stress perturbation data and infection event records. , , , ;

[0051] The risk output module is used to trigger visual, audible, and remote early warnings based on the progressive risk values ​​and output a risk level identifier.

[0052] The data storage module is used to store historical data on stress disturbances, risk curves, and individualized thresholds, and provides data support for the incremental learning unit.

[0053] The beneficial effects of this invention are:

[0054] 1. This invention constructs an adaptive prediction window adjustment mechanism based on a dynamic compression model of the incubation period by continuously collecting intra-abdominal pressure signals and extracting the perturbation entropy and its time derivative. This mechanism dynamically shortens the window length as the compression coefficient increases during the accelerated phase of the incubation period, thereby reducing the response time of risk assessment from several hours in the traditional fixed window to minutes, significantly improving the timeliness and sensitivity of early warning.

[0055] 2. This invention utilizes the self-evolutionary update mechanism of the edge-side intelligent chip to minimize the weighted sum of false alarm and false negative rates. It employs an online stochastic gradient descent algorithm to incrementally update the compression adjustment coefficients and weight coefficients in the progressive risk model, and combines this with a threshold adaptive calibration step. This enables real-time personalized evolution of model parameters and warning thresholds, allowing the system to continuously optimize as the patient's disease progresses, significantly improving the accuracy and individual adaptability of warnings.

[0056] 3. This invention constructs a stage transition probability function that includes the stable period, incubation period, and infection period, and introduces an acceleration term to calculate a progressive risk value. It also combines trend prediction of the derivative of the risk curve, accelerated linkage operation during the incubation period, and multi-channel early warning interaction to form a four-level progressive early warning system from "normal - incubation period - high risk - extremely high risk". This provides medical staff with comprehensive decision support, such as risk level visualization, advanced trend prediction, and suggested follow-up time windows, and realizes closed-loop dynamic management of peritoneal dialysis infection risk from early identification to clinical intervention. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0058] Figure 1 This is a flowchart illustrating the overall process of the peritoneal dialysis infection risk warning method of the present invention.

[0059] Figure 2 This is a flowchart illustrating the dynamic window adaptive adjustment process of the peritoneal dialysis infection risk early warning method of the present invention.

[0060] Figure 3 This is a flowchart illustrating the self-evolutionary update process of the peritoneal dialysis infection risk early warning method of the present invention.

[0061] Figure 4This is a flowchart of the threshold adaptive calibration and multi-level early warning process for the peritoneal dialysis infection risk early warning method of the present invention. Detailed Implementation

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0063] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0064] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0065] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0066] Example 1

[0067] Reference Figure 1-4 This is the first embodiment of the present invention, which provides a method for early warning of peritoneal dialysis infection risk, including the following steps:

[0068] S1: Continuously acquire the patient's intra-abdominal pressure signal and extract the perturbation features that characterize random pressure fluctuations from the pressure signal.

[0069] Specifically, the disturbance characteristics include the rate of change of pressure per unit time. and the standard deviation of the disturbance And constitutes the pressure disturbance characteristic vector. .

[0070] In one specific implementation, peritoneal pressure signal acquisition utilizes a MIK-P300 medical-grade piezoresistive pressure sensor, connected to the proximal end of the peritoneal dialysis catheter via a three-way valve, to monitor intra-abdominal pressure changes in real time. The sensor outputs an analog voltage signal of 0.5–4.5 V, which is converted into a digital signal by a 24-bit high-precision analog-to-digital converter (ADS1220). The sampling frequency is set to 100 Hz, which can completely capture pressure micro-disturbances caused by respiration, changes in body position, and peritoneal micro-inflammation.

[0071] It should be noted that the collected pressure signal (raw pressure signal) First, a median filter (window length L=11 points) is used to remove spike noise caused by patient coughing and tubing vibration; then, a second-order Butterworth low-pass filter (cutoff frequency 5 Hz) is used to filter out high-frequency electrical noise, resulting in a preprocessed pressure signal. All filtering is performed in real time on the edge intelligent chip (e.g., STM32F407), and the time for a single filtering is <0.1 ms.

[0072] In one specific implementation, the calculation of the disturbance characteristics includes: using a 1-minute (6000-point) sliding window with a window step size of 1 second (i.e., outputting a new set of characteristic values ​​every second), calculating the rate of change of pressure per unit time. and the standard deviation of the disturbance Two core perturbation features:

[0073] Rate of change of pressure per unit time : The unit is mmHg / s, where This is the instantaneous pressure value for the current second (the average value of 100 sampling points per second). The instantaneous rate of change of the pressure signal is characterized by its absolute value; the larger the absolute value, the more intense the fluctuation of intra-abdominal pressure.

[0074] Disturbance standard deviation : ;in =60 (meaning the current 1-minute window contains 60) value), For this window The mean; The dispersion of the rate of pressure change within one minute is a key indicator for measuring the magnitude of random pressure fluctuations.

[0075] It should be noted that in the early stages of peritonitis, bacterial metabolites and inflammatory mediators can lead to increased peritoneal capillary permeability and impaired lymphatic drainage, which in turn causes enhanced micro-amplitude, high-frequency, and random fluctuations in intra-abdominal pressure. and This random disturbance is quantified from two dimensions: the rate of change and the amplitude of fluctuation, to provide a basis for subsequent disturbance entropy. The calculation and latency identification provide reliable input features. S2: Calculate the perturbation entropy of the pressure signal based on the perturbation features. and its time derivative ;

[0076] Specifically, perturbation entropy Calculated using the following formula:

[0077] ;

[0078] in, The amplitude of the pressure signal at the th The probability of occurrence within a quantization interval;

[0079] time derivative It is used to characterize the degree of accelerated change in random stress fluctuations and as a basis for distinguishing the accelerated stage of the incubation period of infection.

[0080] In a preferred embodiment, the quantization interval division adopts a fixed interval equal-width quantization method. The specific steps are as follows: First, based on clinical experience and statistical analysis of intra-abdominal pressure data from a large number of dialysis patients, the dynamic range of the pressure signal amplitude is determined to be -5 mmHg to +15 mmHg (covering resting intra-abdominal pressure and pressure fluctuations during dialysate infusion). Then, this range is equally divided into M=64 quantization intervals, each interval having a width of... Then the first The range of values ​​for each interval is:

[0081] ;

[0082] The interval boundaries are fixed and do not require dynamic adjustment, which facilitates rapid indexing by the edge chip.

[0083] In a preferred embodiment, the pressure signal amplitude is at the... The probability of occurrence within each quantization interval The real-time estimation method employs the sliding window histogram approach. Specifically, the steps are as follows: First, take L=6000 points (corresponding to 60 seconds, sampling rate 100Hz), and the perturbation standard deviation... The window remains consistent; then, the histogram is updated using an overlap-preservation method: that is, the latest 100 pressure samples are moved in every second, and the earliest 100 samples are moved out, with incremental updates only performed on the counts of the changed intervals, avoiding a full histogram recalculation and significantly reducing the computational load; then the probability ,in The pressure amplitude within the current window falls on the first Number of sampling points in each interval =6000.

[0084] In a preferred embodiment, the perturbation entropy The entropy is calculated using the natural logarithm (base e), with the unit being nats. The specific calculation formula is as follows:

[0085] ;

[0086] Perturbation entropy The value is updated every second, reflecting the degree of randomness of the pressure signal within the current minute.

[0087] It should be noted that, in order to prevent This can cause computational overflow when At that time, it was agreed .

[0088] In one specific implementation, the time derivative of entropy Using the first-order backward difference approximation, we obtain: The unit is nats / s. A positive derivative indicates that random pressure fluctuations are accelerating, while zero or negative derivatives indicate that the fluctuations are stabilizing or weakening, thus enabling the early detection of rapidly progressing signals. Clinical data has validated that within 24-48 hours before the onset of infection symptoms, If sustained positive growth occurs, its early warning timeliness is significantly better than simply relying on pressure amplitude thresholds or fixed window statistics, and it can be used as a basis for judging the acceleration phase of the incubation period.

[0089] It should be noted that this step calculates the perturbation entropy in real time through fixed-width quantization and incremental updates of the sliding window histogram. and its time derivative ,in As a criterion for identifying the accelerated phase of the incubation period, clinical data has verified that it can show a sustained positive increase 24–48 hours before the onset of infection symptoms, and its early warning timeliness is significantly better than traditional fixed threshold or fixed window statistical methods; the obtained and This will serve as the core input feature, used in subsequent steps for adaptive adjustment of the dynamic compression window coefficient, calculation of stage transition probability, comprehensive determination of the latent acceleration stage, and statistical calibration of individualized early warning thresholds, thus providing a highly sensitive and low-latency early identification basis for the entire peritoneal dialysis infection risk early warning method.

[0090] Reference Figure 2 S3: Based on the perturbation entropy and time derivative The time window length for predicting infection risk is dynamically adjusted.

[0091] Specifically, the time window length for dynamically adjusting the infection risk prediction is achieved through a dynamic compression model of the incubation period:

[0092] ;

[0093] in, This refers to the adjusted time window length. This is the preset initial window length; The standard deviation of the disturbance reflects the severity of the current pressure fluctuation: The larger the value, the more the window is compressed, indicating a faster response to the current high-risk situation; This is the compression adjustment factor.

[0094] Compression adjustment factor According to the time derivative Adaptive adjustment: when And as it continues to increase, Increase accordingly, thereby increasing the window length Shorten the response time for risk assessment.

[0095] In one specific implementation, based on clinical guidelines for peritoneal dialysis and the distribution of infection latency periods in most patients, Set to 24 hours (i.e., 1440 minutes). This value can be manually adjusted by medical staff through the system interface at the start of dialysis (for example, it can be shortened to 12 hours for patients with frequent past infections, and extended to 48 hours for low-risk patients). The system records this setting as a benchmark for subsequent dynamic adjustments.

[0096] In one specific implementation, the compression adjustment factor Using linear positive feedback rules and The binding, its adaptive formula is:

[0097] ;

[0098] in, Updated once per second, in units of nats / s; (Based on training with clinical data) This is the adjustment factor, with a value of 0.5, in units of s / nats.

[0099] This formula guarantees that only when... (i.e., when random fluctuations are accelerating) Only then does it increase; when hour, The window is restored to the basic compression level.

[0100] In a preferred embodiment, the continuously increasing determination mechanism is as follows: a counter records the number of times the condition is met consecutively. When it occurs three times consecutively (i.e., for three consecutive seconds)... When each value is greater than the previous value, the counter increments to 3. At this point, it is determined to be continuously increasing, and the compression adjustment coefficient is updated according to the formula above. If any condition is not met (i.e. If the value is not greater than the previous value, the counter is reset to zero, and the current value is maintained. constant.

[0101] In a preferred embodiment, a minimum window length is set to avoid unreliable statistics due to an excessively short window. =1 hour (60 minutes); at the same time, the window length should not exceed the initial window, i.e. If the calculated value Then let If the calculated value (Theoretically, it won't happen because) ), then let .

[0102] It should be noted that in step S3, by constructing a dynamic compression model of the latency period, the initial window length is... Set as an adjustable 24-hour benchmark and employ a linear positive feedback rule. Compressor adjustment factor With the time derivative of the perturbation entropy Real-time binding, while introducing a trend confirmation mechanism that confirms a continuous upward trend for 3 seconds and =1 hour boundary protection, achieving prediction window length During the accelerated phase of the incubation period of infection ( The adaptive compression of the dynamic window significantly shortens response time when risks evolve rapidly. This will be used as the stage transition probability in subsequent steps. The core input, together with the disturbance characteristics, constructs a progressive risk model, enabling the system to maintain robustness against regular fluctuations while having an advanced ability to capture the acceleration phase of the incubation period, effectively solving the technical problems of delayed early warning and high false alarm rate in traditional fixed-window early warning systems.

[0103] S4: Based on the adjusted time window and perturbation entropy and its time derivative It calculates the progressive risk value of peritoneal dialysis infection in patients and outputs an early warning message when the progressive risk value reaches a preset threshold.

[0104] Specifically, the progressive risk value is calculated using a phase transition probability function; the patient's condition is divided into stable periods. Incubation period and the infection period The transition probability is defined as:

[0105] ;

[0106] ;

[0107] in, for , A monotonically increasing function; for monotonically decreasing function A monotonically increasing function;

[0108] Progressive risk value The calculation formula is: ;in, , , The weighting coefficients are dynamically adjustable, with an initial value set to [value]. , , (Based on patent experience weighting); The acceleration term, used to characterize the degree of acceleration in the latency period, has the physical meaning that when random fluctuations are both intense ( (Big) is accelerating again ( When the risk is high, the risk increases dramatically. All risk values ​​are ultimately linearly mapped to... Intervals facilitate hierarchical classification.

[0109] In one specific implementation, a logistic regression function is used to quantify the transition probability, specifically as follows:

[0110] The probability of transition from the stable period to the latent period: Among them, the coefficient , Bias These values ​​are set based on pre-trained data from several anonymous patient historical data sets. The function relates to... and Monotonically increasing, which aligns with physiological intuition: the more drastic the fluctuations and the stronger the randomness, the higher the probability of entering the incubation period.

[0111] The probability of transmission from the incubation period to the infectious period: Among them, the coefficient , Bias Also from pre-training. This function is related to... Monotonically decreasing (the shorter the window, the higher the risk of progress), regarding Monotonically increasing (the more drastic the acceleration, the higher the risk of progress).

[0112] In a preferred embodiment, progressive risk value It updates every 5 minutes, consistent with the window adjustment cycle, to ensure system timing coordination.

[0113] It should be noted that the progressive risk value This comprehensively reflects the probability and real-time acceleration of a patient's progression from the stable phase to the latent and infectious phases, providing a basis for the self-evolutionary updates of subsequent edge-side smart chips (including compression factor). and weighting coefficients , , (Incremental optimization) provides a real-time, quantitative performance evaluation benchmark.

[0114] Furthermore, such as Figure 3 The method performs the following self-evolutionary updates via the edge-side smart chip:

[0115] Based on the real-time collected pressure disturbance characteristics and the corresponding infection event labels at specific time points, the compression adjustment coefficient is incrementally updated. ;

[0116] The weighting coefficients are incrementally updated based on the individual patient's historical risk curve and infection time. , , ;

[0117] The update aims to minimize the weighted sum of the false positive and false negative rates within the historical window.

[0118] In one specific implementation, to achieve the goal of minimizing the weighted sum of the false positive rate and the false negative rate within the historical window, the loss function is first defined as follows: ;in, This indicates the proportion of periods in the past 7 days during which no infections were detected that triggered an early warning. This indicates the proportion of cases in the past 7 days where no warning was issued within 24 hours prior to the occurrence of an infection; This indicates that clinicians prefer to reduce false alarms (to avoid healthcare worker fatigue). Then, the parameter to be updated is determined as the compression adjustment factor. and the weighting coefficients in the progressive risk value calculation model , , Incremental updates are performed using an online stochastic gradient descent algorithm: a regular update is triggered at 3 AM daily, calculating the numerical gradient of the loss function with respect to each parameter using data from the most recent 24 hours as a batch (each time, a single parameter is perturbed by ±0.01 and the change in loss is evaluated), according to the learning rate. , Adjust the parameter values, among which It is the compression adjustment factor. The learning rate; Weighting coefficient , , The learning rate. For example, with... For example:

[0119] ;

[0120] Meanwhile, whenever a doctor clicks "Confirm Infection" on the dialysis machine interface or mobile terminal, the system immediately triggers an emergency update using the pressure characteristics of the 24 hours prior to the infection event as positive samples and randomly selected characteristics of the uninfected period as negative samples, enabling the model to quickly adapt to the patient's latest disease progression characteristics. All historical samples and parameter versions are stored in external Flash for incremental learning and reuse, thereby enabling real-time and personalized evolution of the model independently on the device side, continuously reducing false positive and false negative rates.

[0121] Furthermore, when the time derivative When the threshold is exceeded and continues to rise, the patient is determined to have entered the accelerated phase of the infection incubation period. When the above conditions are met, the edge chip simultaneously performs the following three actions:

[0122] Increase the compression adjustment coefficient To further shorten the prediction window length Increase the weighting coefficient and To increase the probability of metastasis during the incubation period. and the acceleration term for progressive risk value Contributions;

[0123] Output an accelerated incubation period warning icon, which corresponds to a specific risk level. For example, display a yellow "accelerated incubation period" icon on the dialysis machine screen and push a risk level 2 warning (risk value range 40-60) to the nurse station mobile terminal.

[0124] In a preferred embodiment, the first threshold is determined by analyzing a number of historical dialysis records (e.g., 200 cases) and taking the records within 24 hours prior to the infection. The 5th percentile is used as the first threshold (i.e., the 5th percentile). In this embodiment, it is set as follows: .

[0125] In a preferred embodiment, the compression adjustment coefficient is increased. The method is: to set the current Increase by 0.1 (but not exceeding the upper limit) Make the window length Further shorten.

[0126] In a preferred embodiment, the weighting coefficient is increased. and The method is: to Increase by 0.1, Increase by 0.05 (but) ≤1.0, (≤0.5) to enhance the contribution of the latency period transfer probability and the acceleration term to the total risk value.

[0127] It should be noted that the above method involves setting a first threshold and simultaneously increasing the compression adjustment coefficient. Increase the weighting coefficient and It also outputs early warning indicators, enabling real-time optimization of the prediction window and risk contribution weight during the accelerated phase of the infection incubation period, thereby significantly enhancing the sensitivity and response speed to rapidly progressing risks.

[0128] Based on this, risk curves are generated continuously. And calculate its derivative. This allows for further dynamic prediction of future risk evolution trends, forming a multi-dimensional and progressive risk warning system in conjunction with the aforementioned collaborative operations, providing more comprehensive and advanced decision support for clinical practice.

[0129] Furthermore, risk curve generation and trend prediction include the following steps:

[0130] Continuously calculate risk values ​​at different times Forming a risk curve ;

[0131] Calculate the derivative of the risk curve with respect to time. And based on the sign and rate of change of the derivative value, predict the future trend of risk evolution;

[0132] when When the risk exceeds the second threshold, a warning of rapid increase in risk is triggered, for example, prompting medical staff to pay attention.

[0133] In one specific implementation, the risk curve The method of generation is as follows: the system continuously stores risk values ​​at 5-minute intervals. Forming a time series A preferred implementation: To ensure smooth derivative calculation, a Savitzky-Golay filter (window length 11 points, order 3) is used. Smoothing is performed to obtain a smoothed sequence. .

[0134] In one specific implementation, the smoothed risk derivative The calculation was performed using the first-order central difference method, specifically as follows: ;in, =5 minutes. The unit is risk value / hour (multiply by 12 when converting). The physical meaning of this formula is: to approximate the rate of change (slope) of risk at the current moment by dividing the difference between the smoothed risk values ​​of two consecutive moments by twice the time interval. This is the classic three-point central difference formula in numerical analysis, compared to the first-order backward difference. The central difference has a smaller error and can more accurately reflect the instantaneous trend of change.

[0135] In one specific implementation, the second threshold (denoted as...) The determination is dynamic, based on the patient's past 7 days. Historical data, using the 95th percentile. This threshold is updated daily.

[0136] It should be noted that the above method uses Savitzky-Golay smoothing and three-point central difference to accurately calculate the risk derivative. Based on the patient's historical data over the past 7 days, a second threshold (95th percentile) is dynamically determined, enabling sensitive capture and early warning of rapidly rising risk trends, and significantly improving the system's ability to predict infection progression.

[0137] Further reference Figure 4 It also includes a threshold adaptive calibration step:

[0138] Collect historical stress disturbance characteristics of patients and the corresponding time of infection events, and use a sliding time window to calculate the disturbance entropy. and progressive risk value Distribution;

[0139] Based on the specified early warning sensitivity, the first percentile and the second percentile of the distribution are selected as the individualized incubation period threshold and the individualized high-risk threshold, respectively; wherein the incubation period threshold is determined based on the risk value distribution during the period when no infection has occurred, and the high-risk threshold is determined based on the risk value distribution within a preset time window before the infection occurs;

[0140] The latency threshold and high-risk threshold are dynamically updated based on real-time pressure disturbance trends and used as preset thresholds for the current moment.

[0141] In a preferred embodiment, the incubation period threshold is determined by taking the period during which no infection occurred within the past 90 days. The 95th percentile of the value. If there is insufficient historical data for the patient, the default population value of 30 is used.

[0142] In a preferred embodiment, the high-risk threshold is determined by taking the data from the 24 hours preceding the infection within the past 90 days. The 5th percentile of the value (to ensure high sensitivity). Default value: 60.

[0143] In a preferred embodiment, an individualized extremely high risk threshold is also included: fixed at 80 (based on clinical consensus).

[0144] It should be noted that the above-mentioned individualized thresholds are recalculated every 24 hours and replace the currently used preset thresholds. For new patients in the initial phase (first 14 days), the default population thresholds are used, and the system automatically switches to individualized thresholds after accumulating sufficient data.

[0145] Furthermore, it also includes the steps for exchanging early warning information:

[0146] When progressive risk value When the incubation period threshold or high-risk threshold is reached, the edge smart chip triggers a multi-channel warning; where reaching the incubation period threshold corresponds to the incubation period risk level, and reaching the high-risk threshold corresponds to the high-risk level.

[0147] Multi-channel early warning includes mobile terminal notifications, dialysis machine display screen prompts, and push notifications from the medical monitoring platform interface;

[0148] It also outputs the corresponding risk level label and suggested follow-up visit time window based on the risk level.

[0149] In one specific implementation, the risk level is based on real-time... Compared with the threshold, it is divided into four levels:

[0150] 0≤ <Incubation period threshold: Normal (green), classified as level 0 (normal), no signs of infection, maintain routine monitoring;

[0151] Incubation period threshold ≤ <High-risk threshold: Incubation period risk (yellow), classified as Level 1 (Incubation period risk), indicating that the infection may have entered the incubation period and close monitoring is required;

[0152] High risk threshold ≤ < 80: High risk (orange), classified as Level 2 (high risk), indicating a high risk of infection, and clinical intervention preparation is recommended;

[0153] ≥ 80: Extremely high risk (red), classified as Level 3 (extremely high risk), indicating a very high probability of infection, requiring immediate action.

[0154] In one specific implementation method, the multi-channel early warning is implemented as follows:

[0155] Mobile App: Push notifications via Bluetooth or 4G (if the chip integrates a 4G module), including the risk level and recommended follow-up appointment window (e.g., "Please come to the peritoneal dialysis center for a follow-up appointment within 24 hours").

[0156] Dialysis machine display screen: Sends commands via UART, and a pop-up window displays the risk level and current status. Values ​​and trend curves; the buzzer sounds intermittently when orange / red.

[0157] Medical monitoring platform: Pushes JSON data packets via hospital internal Wi-Fi (MQTT protocol), including patient ID, risk level, and timestamp; the corresponding bed icon changes color and flashes.

[0158] It should be noted that step S4 above, by constructing a complete progressive risk quantification and multi-level early warning system, first calculates the progressive risk value based on the stage transition probability function and acceleration term. The system ensures timing coordination with an update cycle of 5 minutes; furthermore, relying on the self-evolutionary update mechanism of the edge-side intelligent chip, it achieves real-time and personalized evolution of the model; based on this, by setting a first threshold and triggering coordinated execution, it increases... Increase and The system outputs an accelerated incubation period warning indicator, significantly enhancing the sensitivity to detecting rapidly progressing risks; simultaneously, it accurately calculates the risk derivative through Savitzky-Golay smoothing and three-point central difference. The system dynamically determines a second threshold based on patients' historical data, enabling proactive prediction of rapidly rising risk trends. Furthermore, through an adaptive threshold calibration step, individualized incubation period and high-risk thresholds are determined based on the risk value distribution during periods without infection and 24 hours prior to infection, respectively, and are dynamically updated every 24 hours, forming a multi-level early warning threshold system that precisely matches the patient's disease progression characteristics. Finally, through multi-channel early warning interaction, the system achieves visualized, tiered risk level delivery and clinical decision support, thus constructing a peritoneal dialysis infection risk early warning system that integrates real-time quantification, self-evolutionary learning, accelerated phase linkage response, trend prediction, individualized threshold calibration, and multi-level early warning. This significantly improves the lead time and accuracy of incubation period identification, achieving independent, real-time, and personalized dynamic management of infection risk at the end-user level.

[0159] In summary, this invention constructs an adaptive prediction window adjustment mechanism based on a dynamic compression model of the incubation period by continuously acquiring intra-abdominal pressure signals and extracting the perturbation entropy and its time derivative. This mechanism adjusts the window length during the accelerated phase of the incubation period. Furthermore, as the compression coefficient increases, the response time for risk assessment dynamically shortens, reducing it from several hours in a traditional fixed window to minutes, significantly improving the timeliness and sensitivity of early warning. This invention utilizes the self-evolutionary update mechanism of the edge-side intelligent chip, aiming to minimize the weighted sum of false positive and false negative rates. It employs an online stochastic gradient descent algorithm to incrementally update the compression adjustment coefficient and the weight coefficients in the progressive risk model. Combined with a threshold adaptive calibration step (determining individualized incubation period and high-risk thresholds based on the risk value distribution during the non-infection period and 24 hours before infection), it achieves real-time personalized evolution of model parameters and warning thresholds. This allows the system to continuously optimize as the patient's disease progresses, significantly improving the accuracy and individual adaptability of warnings. This invention constructs a stage transition probability function including the stable period, incubation period, and infection period, and introduces an acceleration term to calculate the progressive risk value. It also combines trend prediction of the risk curve derivative and accelerated linkage operations during the incubation period (increasing the risk coefficient). Increase and The system includes multiple warning indicators (output warning signs) and multi-channel warning interaction (mobile terminals, dialysis machine displays, and medical monitoring platforms), forming a four-level progressive warning system from "normal - incubation period - high risk - extremely high risk". This provides medical staff with comprehensive decision support, such as risk level visualization, advanced trend prediction, and suggested follow-up visit time windows, and realizes closed-loop dynamic management of peritoneal dialysis infection risk from early identification to clinical intervention.

[0160] Example 2: This example provides a peritoneal dialysis infection risk early warning method system. This system is applied to the above-mentioned peritoneal dialysis infection risk early warning method and includes:

[0161] The pressure acquisition module is used to continuously acquire intra-abdominal pressure signals and generate pressure disturbance feature vectors.

[0162] An edge-side intelligent chip computing module, integrated into a dialysis device or portable terminal, includes:

[0163] Pressure feature extraction unit for real-time calculation , , and ;

[0164] Window adaptive unit, used to execute dynamic compression model and update compression adjustment coefficients. ;

[0165] The risk progression assessment unit is used to calculate the transition probability and risk value during the execution phase. generate;

[0166] Incremental learning units are used to incrementally update data based on historical stress perturbation data and infection event records. , , , ;

[0167] The risk output module is used to trigger visual, audible, and remote warnings based on progressive risk values ​​and output risk level indicators.

[0168] The data storage module is used to store historical data on stress disturbances, risk curves, and individualized thresholds, and provides data support for the incremental learning unit.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of infection risk in peritoneal dialysis, characterized in that, Includes the following steps: The patient's intra-abdominal pressure signal was continuously acquired, and perturbation features characterizing random pressure fluctuations were extracted from the pressure signal. The perturbation entropy of the pressure signal is calculated based on the perturbation characteristics. and its time derivative According to the perturbation entropy and the time derivative Dynamically adjust the length of the time window for infection risk prediction; based on the adjusted time window and the perturbation entropy and its time derivative The system calculates a progressive risk value for peritoneal dialysis infection in patients and outputs an early warning message when the progressive risk value reaches a preset threshold.

2. The peritoneal dialysis infection risk early warning method as described in claim 1, characterized in that: The disturbance characteristics include the rate of change of pressure per unit time. and the standard deviation of the disturbance And constitutes the pressure disturbance characteristic vector. ; The perturbation entropy Calculated using the following formula: ； in, The amplitude of the pressure signal in the th... The probability of occurrence within a quantization interval; The time derivative It is used to characterize the degree of accelerated change in random stress fluctuations and as a basis for distinguishing the accelerated stage of the incubation period of infection.

3. The peritoneal dialysis infection risk early warning method as described in claim 2, characterized in that: The time window length for dynamically adjusting the infection risk prediction is achieved through a dynamic compression model of the incubation period. ； in, This refers to the adjusted time window length. This is the preset initial window length; The standard deviation of the disturbance; This is the compression adjustment factor; The compression adjustment coefficient According to the time derivative Adaptive adjustment: when And as it continues to increase, Increase accordingly, thereby increasing the window length Shorten the response time for risk assessment.

4. The peritoneal dialysis infection risk early warning method as described in claim 3, characterized in that: The progressive risk value is calculated using a phase transition probability function; the patient's condition is divided into stable periods. Incubation period and the infection period The transition probability is defined as: ； ； in, for , A monotonically increasing function; for monotonically decreasing function A monotonically increasing function; The progressive risk value The calculation formula is: ;in, , , These are dynamically adjustable weighting coefficients; This is the acceleration term used to characterize the degree of acceleration of the latency period.

5. The peritoneal dialysis infection risk early warning method as described in claim 4, characterized in that: The method performs the following self-evolutionary updates via an edge-side smart chip: Based on the real-time collected pressure disturbance characteristics and the corresponding infection event labels at specific time points, the compression adjustment coefficient is incrementally updated. ; The weighting coefficients are incrementally updated based on the individual patient's historical risk curve and infection time. , , ; The update aims to minimize the weighted sum of the false positive rate and the false negative rate within the historical window.

6. The peritoneal dialysis infection risk early warning method as described in claim 5, characterized in that: When the time derivative When the threshold is exceeded and continues to rise, the patient is determined to have entered the accelerated phase of the incubation period, and the following actions are performed simultaneously: Increase the compression adjustment coefficient To further shorten the prediction window length Increase the weighting coefficient and To increase the probability of metastasis during the incubation period. and the acceleration term for the progressive risk value Contributions; Output an early warning indicator for accelerated incubation period, which corresponds to a specific risk level.

7. The peritoneal dialysis infection risk early warning method as described in claim 4, characterized in that: It also includes steps for risk curve generation and trend prediction: Continuously calculate risk values ​​at different times Forming a risk curve ; Calculate the derivative of the risk curve with respect to time. And based on the sign and rate of change of the derivative value, predict the future trend of risk evolution; when When the risk exceeds the second threshold, a warning of rapid risk escalation is triggered.

8. The peritoneal dialysis infection risk early warning method as described in claim 1, characterized in that: It also includes a threshold adaptive calibration step: Collect historical stress disturbance characteristics of patients and the corresponding times of infection events, and use a sliding time window to calculate the disturbance entropy. and progressive risk value Distribution; Based on the specified early warning sensitivity, the first percentile and the second percentile of the distribution are selected as the individualized incubation period threshold and the individualized high-risk threshold, respectively; wherein the incubation period threshold is determined based on the risk value distribution during the period when no infection has occurred, and the high-risk threshold is determined based on the risk value distribution within a preset time window before the infection occurs; The latency threshold and the high-risk threshold are dynamically updated based on the real-time pressure disturbance trend and used as the preset threshold for the current moment.

9. The peritoneal dialysis infection risk early warning method as described in claim 8, characterized in that: It also includes the steps for exchanging early warning information: When the progressive risk value When the incubation period threshold or high-risk threshold in the preset threshold is reached, the edge smart chip triggers a multi-channel early warning. Among them, reaching the incubation period threshold corresponds to the incubation period risk level, and reaching the high risk threshold corresponds to the high risk level; The multi-channel early warning includes mobile terminal notifications, dialysis machine display screen prompts, and push notifications from the medical monitoring platform interface. It also outputs the corresponding risk level label and suggested follow-up visit time window based on the risk level.

10. A peritoneal dialysis infection risk early warning method system, applied to any one of the peritoneal dialysis infection risk early warning methods described in any one of 1-9 above, characterized in that, include: The pressure acquisition module is used to continuously acquire intra-abdominal pressure signals and generate pressure disturbance feature vectors. An edge-side intelligent chip computing module, integrated into a dialysis device or portable terminal, includes: Pressure feature extraction unit for real-time calculation , , and ; Window adaptive unit, used to execute dynamic compression model and update compression adjustment coefficients. ; The risk progression assessment unit is used to calculate the transition probability and risk value during the execution phase. generate; Incremental learning units are used to incrementally update data based on historical stress perturbation data and infection event records. , , , ; The risk output module is used to trigger visual, audible, and remote early warnings based on the progressive risk values ​​and output a risk level identifier. The data storage module is used to store historical data on stress disturbances, risk curves, and individualized thresholds, and provides data support for the incremental learning unit.

Images