Methods for identifying the risk of infection in postoperative drainage fluids in orthopedic surgery

Abstract

This invention relates to a method for identifying the risk of infection in postoperative drainage fluid in orthopedic surgery. The method treats drainage fluid as a system output driven by both postoperative local immune regulation and trauma recovery; captures nonlinear time-series curves; models the first-order trend of these curves; analyzes abrupt changes after the rate of change stabilizes; establishes a framework for a normal postoperative recovery path, using dynamic parameters of drainage fluid from non-infected cases to form a standard recovery trend database; compares the Euclidean offset and directional changes between the current patient's drainage parameter behavior vector and the standard trajectory; transforms these changes into a visualized recovery offset map, measuring infection prediction data from a multi-day trend perspective; constructs a framework for identifying the aggregation of local abnormal signals leading to the formation of infection event transitions, deriving clinical infection nodes where quantitative changes lead to qualitative changes; and enables early and accurate identification, significantly advancing the timing of intervention, avoiding misjudgments based on single thresholds, improving identification accuracy and specificity, enhancing model adaptability, and strengthening the interpretability of the identification.

Description

TECHNICAL FIELD

[0001] The present application relates to a method for identifying the risk of postoperative drainage fluid infection in orthopedics. BACKGROUND

[0002] Currently, in clinical practice, the methods for identifying the risk of postoperative drainage fluid infection mainly include routine observation method, sign monitoring method, laboratory index analysis method, bacterial culture method and part of the early warning model relying on static biochemical indicators. These methods have played an auxiliary role in the diagnosis of postoperative infection to some extent, but there are still a series of deficiencies and technical drawbacks to be solved, especially in the face of complex, early and atypical postoperative infections, which show obvious lag, one-sidedness and high misdiagnosis rate. First, the current mainstream method relies on manual observation of the color, properties and volume changes of the drainage fluid, such as color deepening, drainage turbidity, sudden decrease or increase in drainage volume, etc. This method completely depends on the experience of medical staff, is highly subjective, and lacks continuity and quantitative standards, which can easily lead to missed diagnosis and delayed identification at night or when there is insufficient staff. Second, many hospitals still rely on daily scheduled sampling of drainage fluid for laboratory testing of routine bacterial culture and blood index analysis, such as changes in total white blood cell count, C-reactive protein (CRP), procalcitonin (PCT) and other indicators. Although these indicators do have certain trends during infection, their physiological response is highly lagging, and they usually do not show significant abnormalities until the infection has developed to a certain extent, making them unsuitable as early warning signals for infection, especially in some latent infections and periprosthetic microbial permeability inflammation, where related indicators may not show significant changes even 24 hours after infection, resulting in the loss of the best intervention window.

[0003] In addition, although bacterial culture is considered the gold standard, it has a complex operation process, a long time period (usually 24-72 hours), and is prone to false negatives due to delayed sampling, sample contamination, low-density colonies and other factors. Moreover, it cannot reflect the local dynamic changes in real time, and is more of a diagnostic tool than a predictive means. In terms of model-based methods, some studies have attempted to use statistical regression, support vector machines, random forests and other machine learning methods to classify and predict postoperative infections. However, these models generally rely on single-point postoperative data input, such as drainage fluid pH value, white blood cell count and other indicators on the first day after surgery for binary classification training, and lack the ability to perceive the physiological state evolution trend, i.e., they cannot capture the entire process of a parameter moving from normal to abnormal, resulting in a lack of understanding of the time dimension. More seriously, most of these models are black box models, lacking clinical interpretability, and cannot clearly indicate which parameter is shifting behind the current infection risk score, how long it has been shifting, and in which direction it is changing, which directly prevents doctors from making credible decisions based on model results, and even completely discarding model reference when the results are contradictory, severely limiting the practical application of intelligent methods.

[0004] All these methods have one common problem, which is too dependent on single threshold judgment logic. In real clinical practice, a warning value is often set as a trigger condition, such as pH < 6.8, flow rate drop > 50%, CRP > 100 mg / L, etc., but this hard threshold is easily affected by individual differences. Different ages, surgical procedures, surgical areas, underlying conditions, and even the use of drainage tubes and drainage methods after surgery can all lead to significant differences in the scope of threshold application. The result is either a large number of false positives (misjudging as infection, but actually normal fluctuation), or a large number of missed reports (true infection but not reaching the abnormal value standard). In addition, these methods generally cannot identify the coordinated changes between multiple indicators, cannot judge whether the simultaneous directional reversal of two indicators represents a stronger risk, and cannot judge whether a sustained slow drift of a parameter is more alarming than a short-term mutation. Therefore, the ability to judge whether quantitative changes have accumulated into qualitative changes is severely insufficient. In terms of data collection, most hospitals still use manual recording, intermittent sampling, and laboratory analysis mode, which cannot form high-frequency, continuous time series data flow. This directly limits the training and deployment of advanced models based on trend modeling and dynamic identification, and also leads to the inability of the identification system to capture the most critical signal that the physiological trend is changing. Even if some devices support continuous recording such as drainage volume and conductivity, there is a lack of integrated data interface and standardized index mapping mechanism, leading to serious data fragmentation and difficulty in unified modeling analysis. SUMMARY

[0005] The purpose of the present application is to provide a method for identifying the risk of postoperative drainage fluid infection in orthopedics, thereby solving some of the problems and deficiencies pointed out in the background art.

[0006] The present application solves the above-mentioned technical problems by adopting the following technical solution: regarding the drainage fluid as a system output driven by postoperative local immune regulation and wound recovery; capturing the nonlinear time series curve including viscosity, color turbidity, flow rate, and pH indicators; first-order derivative trend modeling; analyzing the change rate including pH drop acceleration or flow rate mutation after stable;

[0007] Establishing a postoperative normal recovery path framework, using the dynamic parameters of non-infected cases to form a standard recovery trend library; comparing the Euclidean offset and directional change between the current patient's drainage parameter behavior vector and the standard trajectory;

[0008] Converting the changes into a visual recovery offset map and measuring the infection prediction data from the trend angle; then building a framework for identifying local abnormal signal aggregation to form an infection event transition, and deriving the clinical infection node of quantitative change leading to qualitative change.

[0009] Further, the nonlinear time series curve construction method comprises:

[0010] In the postoperative drainage channel, a multi-parameter in-situ sensing module is integrated for continuous collection of physical and chemical parameters of the drainage fluid; the in-situ sensing module collects the above-mentioned parameters at a time granularity of minutes, and inputs each data to an edge computing unit after time stamping; the edge computing unit establishes a nonlinear time sequence behavior curve of each parameter, and extracts the change rate, acceleration and abnormal inflection point thereof;

[0011] The behavior curve of each parameter is compared with a postoperative standard recovery model to identify the deviation degree and trend direction thereof; when it is identified that multiple parameters appear a synchronous deviation trend within the same time window, it is judged that it is an early signal of infection risk transition, and an infection risk score result is output.

[0012] Further, the viscosity is measured by a micro-oscillation cantilever beam sensing structure, and the vibration attenuation characteristics of the cantilever beam are used to characterize the viscous behavior of the drainage fluid; the color turbidity is obtained by an optical detection module arranged in the drainage channel, and the optical detection module includes a channel for measuring direct absorption light and a channel for detecting 90-degree scattered light to simultaneously evaluate the color and turbidity.

[0013] Further, the flow rate and pH are obtained by a synchronous collection method, and the time phase difference thereof is analyzed by a behavior curve to determine whether the local environmental acidification is accompanied by a sudden change in flow rate; the standard recovery model is constructed based on the time sequence behavior data of each drainage fluid parameter under the condition that the patient is not infected after surgery, and is used as a control trajectory for subsequent individual identification;

[0014] The above scheme takes the postoperative drainage fluid as a continuous system output under the dynamic control of local immune regulation and tissue repair, constructs a behavior phase shift function model by collecting multiple key indicators (flow rate and pH), judges whether the current patient state deviates from the non-infection standard recovery path, and thus realizes dynamic identification and early warning of the infection risk;

[0015] Unlike the traditional method relying on static threshold judgment, a self-created integral function is introduced to simultaneously investigate the relationship between the flow rate change rate and the directional deviation of pH; the function is designed as follows:

[0016]

[0017] Wherein:

[0018] Ψ(t) represents the overall intensity score of the deviation of the drainage fluid behavior of the current patient from the standard recovery trajectory within a certain time window; the upper and lower limits of the integral t0 and t1 represent the start and end times of the current calculation window, which are usually set to 4 hours or 8 hours continuously after surgery; V(t) represents the flow rate of the drainage fluid at time t, and the derivative The speed of the flow rate change is a key parameter for measuring the stability of the liquid dynamics;

[0019] φ(t) represents the pH trend direction of the drainage fluid, and the dominant phase thereof is obtained by fitting the continuous pH data (for example, continuous acidification can be converted to 180 degrees); θ0(t) represents the trend phase of the pH at the moment in the standard non-infective recovery model, and the data thereof is derived from the recovery curve template established by a large number of non-infective patients;

[0020] The difference φ(t)-θ0(t) between the two represents the deviation angle between the current pH behavior trend and the standard path, which is converted into the consistency degree of the direction by a sine function; if the trend is consistent, the value tends to 0, and if the trend is completely opposite, the value tends to ±1;

[0021] Overall, the integral calculation of the function reflects two characteristics: first, whether a flow rate mutation occurs (that is, is negative or changes sharply); second, whether the change trend of the pH deviates significantly from the standard trajectory; when both occur simultaneously, the absolute value of the product term increases, and the integral result Ψ(t) significantly rises, thereby prompting the risk of local environmental infectivity transition.

[0022] Further, the edge computing unit comprises an adaptive behavior feature extraction module for dynamically updating the baseline behavior curve of the individual patient; the infection risk score result is output to the clinical terminal according to the risk level, and a clinical intervention prompt is triggered when the score value reaches a preset warning threshold.

[0023] Further, the comparison method of the Euclidean deviation degree and the directionality change comprises:

[0024] S1, constructing a postoperative recovery normal path model, comprising collecting drainage fluid multi-parameter time series data of non-infective patients, and classifying and clustering according to the type of surgery, site, and individual characteristics to form a standard recovery trajectory library;

[0025] S2, obtaining a plurality of continuous parameters of the drainage fluid of the target patient after surgery within a set time window to form a multi-dimensional behavior vector;

[0026] S3, aligning the behavior vector of the current patient with the corresponding template in the standard recovery trajectory, and performing point-by-point comparison based on time consistency;

[0027] S4, calculating the Euclidean deviation degree between the current behavior vector and the standard trajectory to quantify the overall deviation degree; simultaneously calculating the trend direction change of each parameter to identify whether there is a change trend opposite to the direction of the standard trajectory, and marking as directionally outlying;

[0028] S5、when the Euclidean deviation exceeds the preset deviation threshold, and the directionality outliers of multiple parameters appear simultaneously, it is determined that the patient has a postoperative infection risk transition trend, and an early warning signal is output.

[0029] Further, the standard recovery trajectory library generates a plurality of representative recovery curve templates through cluster analysis to adapt to differences in different surgical procedures, surgical areas, and patient physiological backgrounds; the drainage fluid parameter collection time window is 4 to 24 hours, and is resampled at a uniform time interval for time alignment with the standard trajectory.

[0030] Further, the behavior vector is composed of a plurality of indicators to form a unified vector structure according to timestamps; the directionality outlier determination includes comparing whether the upward or downward trend of the current indicator is opposite to the trend direction of the corresponding time period of the standard trajectory, and recording an outlier event when the trend is opposite;

[0031] In the above scheme, the flow rate, pH, and turbidity of the key drainage fluid indicators are constructed into a unified time vector structure, and a self-defined directionality function is used to determine whether the change trend of each period of the key drainage fluid indicators is synchronized or reversed with the standard recovery trajectory, so as to identify the behavior deviation event.

[0032] To achieve this goal, a newly constructed trend direction outlier function is introduced, and its expression is as follows:

[0033]

[0034] The formula judges the trend directionality of all n key behavior indicators at the current time point t, and outputs an overall outlier score Δ(t). Its meaning and each variable are as follows:

[0035] Δ(t): represents the trend directionality outlier intensity score between the current patient and the standard recovery model at time point t. The greater the value, the more serious the trend reversal, and the higher the infection risk. n: the number of key drainage fluid parameters monitored, such as flow rate, pH, turbidity, and temperature. i σ(i, t): the trend direction of the current i-th indicator at time point t, defined as a sign function, taking +1 if continuously rising, -1 if falling, and 0 if unchanged. i η(i, t): the change rate absolute value of the current i-th indicator in the period, that is, the trend significance strength, which is used to weight the behavior influence of different indicators. i σ(i, t): the theoretical trend direction of the i-th indicator in the standard recovery trajectory at time t, taking the same value as σ(i, t), that is, +1, -1, or 0. i

[0036] The introduction of the bracket term [1-ξ i (t)] in the formula is to emphasize that when the standard trend is upward (that is, ξ​i (t)=+1), while the current value is decreasing (σ). i (t) = -1), and their product is negative; when the directions are the same, this term tends to zero, and when the directions are opposite, it amplifies the sum of outliers. Therefore, this function Δ(t) essentially focuses on the reverse offset of the trend direction and combines the indicator weights to determine whether the offset is severe enough.

[0037] Furthermore, the Euclidean offset and directional outlier results are used together for risk assessment, and the risk is determined to be an early infection transition tendency only when both conditions are met simultaneously.

[0038] The beneficial effects of this invention are:

[0039] By continuously and dynamically monitoring key physiological parameters in drainage fluid (such as pH, flow rate, turbidity, and viscosity) and modeling nonlinear trends, the system not only identifies whether the indicators are abnormal but also determines whether their changing trends indicate potential infection transitions. Compared to traditional methods relying on body temperature, redness, or laboratory culture results, the warning time is on average 24-48 hours earlier, providing valuable intervention windows for clinicians. By introducing a standard recovery trajectory library and performing hierarchical clustering matching based on factors such as surgical type, anatomical location, age, and physiological background, each patient will receive a personalized postoperative health trajectory comparison template. This is further supplemented by an adaptive behavioral benchmark adjustment mechanism in the edge computing unit to dynamically calibrate the individual recovery path.

[0040] All monitoring indicators are uniformly constructed as time behavior vectors, and calculated through trend-guided functions, integral models, and outlier scoring mechanisms. Each warning is accompanied by a visual trend chart, indicator direction comparison, and risk score curve. Doctors can clearly see which parameters start to deviate and when, and whether the direction of deviation is dangerous, which facilitates medical decision-making and solves the problem of the inability to interpret black box models. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method for identifying the risk of infection in postoperative drainage fluid in orthopedic surgery according to the present invention.

[0042] Figure 2 This is a flowchart of the nonlinear time-series curve construction method of the present invention.

[0043] Figure 3 A flowchart is provided for the comparison method of Euclidean offset and directional change in this invention.

[0044] Figure 4 This is an example diagram of the peroneus longus tendon of the present invention through wound drainage. Detailed Implementation

[0045] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0046] Combined with appendix Figure 1This invention provides a method for identifying the risk of infection in postoperative drainage fluid in orthopedic surgery. The drainage fluid is viewed as a continuously generated system output variable controlled by multiple physiological mechanisms. The state of the drainage fluid is a comprehensive expression of multiple factors, including postoperative local immune response, tissue repair process, cytokine release, and microbial intervention. The key to this method is treating the drainage fluid as an externally observable expression of systemic behavior, rather than simply a biochemical waste fluid. This allows for the revelation of potential physiological abnormalities, especially early signs of infection, through changes in its behavior. The system requires continuous collection of several core parameters of the drainage fluid, particularly viscosity, color / turbidity, flow rate, and pH. These parameters are not randomly selected but are confirmed by medical research to often show abnormal changes before infection occurs. For example, pH typically decreases continuously in the early stages of bacterial infection (local acidification), while flow rate may increase or decrease rapidly due to increased inflammatory exudation. Color and turbidity reflect changes in components such as proteins, cells, and pathogens in the fluid, while viscosity is related to contents such as cellulose and inflammatory exudate. In terms of data acquisition, the system uses an integrated intelligent drainage module to collect data. This module embeds a micro-MEMS sensor array, which can measure the physicochemical parameters of the flowing drainage fluid in real time without affecting the smoothness of drainage. Sensor data is collected every minute, forming a high-density time series. All data is initially cleaned and cached by an edge processing chip. Raw data preprocessing includes sensor self-calibration (drift removal), missing data imputation (based on nearest neighbor time periods), outlier filtering (based on standard deviation multiples), and time normalization (aligning all data to a unified time axis). After data preparation, the system enters the modeling stage. This method does not use traditional static threshold judgment, but instead performs modeling and analysis of nonlinear behavior curves. Specifically, each drainage fluid parameter is treated as a continuous function of time, and its first derivative, i.e., the rate of change at each moment, is calculated to identify trend events. For example, if the first derivative of the pH curve suddenly changes from a slow decrease to a rapid decrease, the system will identify it as accelerated acidification; similarly, if the flow rate curve is stable for a long time but the derivative becomes large or negative and steep at a certain time, it is judged as a sudden flow rate fluctuation. This approach enables the system to not only identify abnormal values ​​but also detect signals that behavior is deviating from its normal trajectory. Based on this, the entire model employs a time-window-based sliding sequence analysis algorithm, with each behavioral indicator forming a separate trend scoring sub-model. These sub-models are then combined into a unified infection risk scoring network. Technically, the model structure is analogous to a shallow LSTM or GRU unit, simple in structure but preserving short-term sequence variation characteristics. During training, the system uses a historical orthopedic postoperative drainage fluid dataset, divided into non-infected and infected groups, and labels the infection occurrence window in each time series.The model aims to identify these windows and uses supervised learning (cross-entropy loss function) to continuously adjust trend-sensitive parameters, ensuring that the final infection prediction output closely matches the true label. The training process employs a rolling time window mechanism to enhance the model's adaptability to features across different time periods; simultaneously, label smoothing and data augmentation strategies are introduced to improve the model's robustness to mild trend shifts. Ultimately, the system no longer relies on whether a single parameter exceeds its limits, but rather identifies trends through the synergy of abrupt changes in multiple indicators, specifically determining whether multiple indicators exhibit combined behaviors such as directional reversal, abnormal acceleration, or synchronous fluctuations within a similar timeframe.

[0047] By establishing a framework for normal postoperative recovery, a standardized recovery trend database composed of dynamic parameters from drainage fluids of numerous non-infected cases is constructed to measure behavioral deviations in current postoperative patients during the actual recovery process, thereby achieving dynamic monitoring and early identification of potential infection risks. The entire method is designed based on the premise that infection often leads to deviations in local physiological states from the normal recovery trajectory. This deviation is not always manifested as a single indicator exceeding the standard, but rather as multiple physiological parameters gradually slipping away from the conventional evolutionary path over time. Therefore, the first step of this method is to construct a standard trajectory library. Specifically, this involves extracting continuous parameters of drainage fluid from a large number of previously postoperative, infection-free patients, including multiple indicators such as pH, flow rate, color turbidity, and viscosity. All data are collected hourly from the first to the fifth postoperative day to form a high-frequency time series. To ensure consistency, these data are first normalized (e.g., by Z-score or min-max scaling) to remove individual baseline differences and differences in indicator dimensions. Next, each indicator is resampled (to ensure consistent time series length for each case), and outliers are handled (e.g., removing data drift caused by collector blockage or patient-specific interventions). Then, hierarchical cluster analysis is performed based on the patient's surgical procedure type (e.g., joint replacement, spinal fixation, internal fracture fixation), surgical site (e.g., hip, knee, lumbar), and underlying pathological factors (e.g., diabetes, presence or absence of implants). Similar recovery trends are grouped together, thus generating multiple standard recovery trajectory templates. Each template is a multi-dimensional, multi-time-point behavioral matrix, serving as a reference for the recovery path. During the current postoperative drainage process, the system continuously collects various parameters of the drainage fluid. Within a set time window (e.g., 6 hours, 12 hours, and 24 hours postoperatively), it generates a behavior vector slice in real time. This vector structure is composed of continuous values ​​of all indicators stitched together according to timestamps, forming a multi-dimensional time series vector. Subsequently, the system aligns this vector point-by-point with the standard template trajectory matched to the corresponding surgical procedure and patient background. It calculates the Euclidean distance between the behavior vector and the standard trajectory in a point-to-point manner, that is, it calculates the overall offset of all sampling points in the multi-parameter space. This process can be understood as the quantification of the distance between the current trajectory and the healthy trajectory in the temporal multi-dimensional space. However, this geometrical deviation is insufficient to reveal the true risk. Therefore, the system further introduces trend directionality analysis: for each parameter curve, its local trend direction is extracted within the current time period (e.g., calculating the difference of the three-point moving average to determine whether it is rising, falling, or flat), and then compared with the trend direction of the corresponding time period in the standard trajectory. When the trend directions are consistent, it is recorded as 0; when the directions are completely opposite, it is recorded as 1. Finally, the directional differences of all parameters within this time window are weighted and summed to obtain a directional outlier score. This score is used to assess whether the current physiological behavior is going in the wrong direction, rather than just being far away.The core of the entire identification model is to fuse the aforementioned Euclidean offset and trend-directed outlier scores. In the algorithm implementation, this part uses a shallow graphical neural network structure to model the model, incorporating the interdependencies between different parameters into the analysis. For example, in the early stages of infection, sudden changes in flow rate and a decrease in pH occur simultaneously; this coordinated change is an important pattern signal. The model is trained using labeled historical datasets, i.e., real postoperative infected and non-infected cases. Supervised learning is used to train the model to identify which offset combinations represent infection risk. A focus loss function is introduced during training to enhance the model's sensitivity to early offset features, and cross-validation is used to optimize the model parameters. The final output is an infection risk score, which is based not only on whether a certain indicator is abnormal, but also on a dynamic judgment made based on the spatial distance and directional differences between the patient's current overall physiological trajectory and the standard healthy trajectory.

[0048] To determine whether the immediate state of the drainage fluid is abnormal, the system tracks data over several consecutive days, transforming the temporal trends of all key parameters into a visualized recovery offset map. This allows for the identification of aggregations of local abnormal signals, and based on this, a transitional framework for infection events is constructed to capture the point where quantitative changes lead to qualitative changes—the turning point from early infection to a clinically recognized infection state. In terms of data acquisition, the system continuously collects multiple physiological parameters of the drainage fluid, including but not limited to pH, flow rate, turbidity, color, and viscosity, via a miniature sensor array embedded in the postoperative drainage pathway. The collection cycle is set to once every minute to every ten minutes, forming a continuous time-series data stream. Each time-series data point is accompanied by a high-precision timestamp. The system performs preliminary data processing in real time at the edge computing end, primarily including time-series alignment, outlier removal (such as extreme jumps and physical interference), data normalization (using zero-mean unit variance or interval mapping), and sliding smoothing (such as local weighted regression or Savitzky-Golay filtering), to ensure that the analyzed signals are stable and comparable trend signals. In terms of data structure, the system extracts key trend features of each indicator in each monitoring cycle (usually every 12 or 24 hours), such as local growth rate, decline rate, peak occurrence time, and change magnitude, and embeds them into a two-dimensional space to construct a recovery offset map. The horizontal axis of this map represents the time axis (number of hours or days post-surgery), and the vertical axis represents the offset intensity or outlier degree of each parameter's trend direction. The map uses heat, color gradient, or marker symbols to represent the degree of deviation of parameter behavior from the standard trajectory. This visualization not only provides doctors with an intuitive reference but also serves as important structural feature data for subsequent model input, used to train the deep trend recognition model. The core innovation of this method lies in constructing an infection transition recognition framework. This framework differs from traditional rule-based models; instead, it is based on a dynamic evolution mechanism to judge the process of potential abnormal behavior aggregation → evolution into infection events. In terms of algorithm design, the system continuously observes hotspot areas in the recovery offset map, especially sudden changes, reversals, and accelerations that occur simultaneously in multiple parameters within similar time points. When multiple indicators show abnormal offset magnitudes or directions within a time window and cluster in the same local time interval, the system identifies this area as an abnormal aggregation zone. These anomalous clusters are further fed into a two-stage model. The first stage employs a graph convolutional network (GCN) structure to construct a graph structure based on the temporal adjacency relationships between anomalous events, extracting the degree of clustering and structural features between signals. The second stage uses a recursive structure (such as Bi-GRU or Transformer encoder) to capture the evolution trend of anomalous events and assess whether they are transitioning to a system state, i.e., from a subclinical anomalous state to a real infection risk. In terms of training, the model uses historical real clinical case data as training samples, where each case is labeled with whether infection occurred, the time of occurrence, and the path of parameter anomalous evolution.Before training, all case data are first transformed into a unified format recovery offset map through the above process. Then, abnormal aggregation segments are extracted and labeled (e.g., the shift in time between the aggregation area and the clinical diagnosis of infection). Training employs a multi-task loss function, which optimizes the accuracy of transition event identification (e.g., using binary cross-entropy) and constrains the offset error of prediction time nodes (using smoothed L1 loss or temporal difference regression loss). During training, the model uses a dynamic learning rate and early stopping strategy to prevent overfitting, and introduces a difficult case reweighting mechanism to improve the ability to identify atypical infection pathways. Finally, the system outputs a dynamic risk score curve and a time estimate of the predicted transition point. This information can not only be fed back to medical staff in real time for early warning, but also form a disease progression map for subsequent decision-making. The greatest advantage of this method is that it no longer relies on a static judgment of whether a certain item is abnormal on a certain day, but instead constructs a mechanism to identify how multiple subclinical abnormalities gradually aggregate over time and trigger risk mutations, thereby achieving high-probability prediction and early warning before the actual clinical diagnosis of infection.

[0049] Example 1:

[0050] In this embodiment, Mr. Zhang, a 65-year-old patient, underwent left knee total joint replacement surgery and started using an intelligent drainage monitoring system on day 0. This system incorporates a multi-parameter in-situ sensing module, including a solid-state pH electrode, a microthermal film flow sensor, a near-infrared optical turbidity probe, and a viscosity microbeam vibration detection unit. The system collects drainage fluid parameters every 5 minutes, continuously recording flow rate (mL / h), pH value (dimensionless), turbidity (relative units), and viscosity (cP). All data are timestamped and sent in real-time to the bedside edge computing unit for processing. Using the first 8 hours of postoperative day 1 as the observation period, a total of 96 time points were collected. The edge computing module cleans the raw data (removing instantaneous spikes), normalizes it (Z-score normalization), and then constructs a nonlinear time-series curve for each parameter. For example, Mr. Zhang's pH value started to decrease continuously from 7.21 in the first hour after surgery, and dropped to 6.96 in the fifth hour. During this process, the first derivative (i.e., the rate of change) of the pH behavior curve gradually decreased from -0.02 to -0.07, and the second derivative of the curve also showed a clear negative inflection point, indicating that acidification was accelerating. The flow rate remained stable for the first 4 hours after surgery (about 28 to 30 mL per hour), but suddenly dropped to 21 mL / h from the fifth hour, accompanied by fluctuations. The first derivative of the flow rate curve suddenly changed from close to 0 to -9, and fluctuated repeatedly in the next 2 hours, with its acceleration showing an abnormal reversal pattern. The turbidity and viscosity curves also showed brief abnormalities in the fifth to sixth hours: the turbidity changed from a linear decrease to a steady increase, and the viscosity increased from 1.2 cP to 1.8 cP. The system synchronously compared the behavioral curves of the above four parameters with the standard recovery model—this model was established based on 102 non-infected patients of the same age and surgical procedure as Mr. Zhang, with no history of infection. The standard model shows that pH should slowly decrease and then stabilize on day 1, flow rate should slowly decrease after 5 hours without a sudden drop, and turbidity and viscosity should gradually decrease without an upward inflection point. The system used the Euclidean distance method to calculate the average behavioral deviation between Mr. Zhang's current data and the standard trajectory, and the result was 2.63 (the average deviation of the standard group was between 0.5 and 1.2), which was significantly higher. At the same time, the trend directionality analysis showed that three of the four indicators were opposite to the standard trajectory in the 5th to 6th hour period (for example, the standard pH stabilized while the actual pH decreased rapidly), and the system marked this window as a directional outlier. Because more than three indicators showed synchronous shifts and reversed trends during this period, the system aggregated and marked this behavior as a potential infection transition node based on the built-in infection transition prediction model. Combining the historical data of previous physiological parameters with the hot spot areas formed in the recovery shift map (i.e., multiple shift points concentrated in one outbreak), the system finally output an infection risk score of 78 / 100, which is far higher than the set threshold of 65. The system immediately sent an alert to the attending physician and nurse's mobile terminal and suggested performing bacterial culture of drainage fluid, rechecking white blood cell count and C-reactive protein, and intervening in advance.The results showed that Mr. Zhang was diagnosed with Staphylococcus aureus infection on the 3rd day. However, since the system detected the abnormal trend on the 1st day, the medical team began to adjust the antibiotic and drainage plan on the 2nd day, avoiding the risk of infection spread and reoperation.

[0051] Viscosity measurement employs a micro-oscillating cantilever beam sensing structure, a miniature MEMS device. Essentially, when a cantilever structure oscillates at a controllable frequency and is immersed in a drainage fluid, the viscous characteristics of the fluid affect the vibration amplitude decay curve of the cantilever. As the viscosity of the drainage fluid increases, the damping force experienced by the cantilever during its movement in the fluid increases, resulting in a faster decay rate of the oscillation signal. By detecting the energy dissipation rate of the oscillation period and its nonlinear decay behavior over time, the system can construct a highly sensitive damping-time response curve and deduce the actual viscosity value of the current fluid. The entire vibration response signal is controlled collaboratively by a chip-integrated accelerometer and drive electrodes, achieving an accuracy of 0.01 cP. The sampling frequency supports 5 Hz. In practical deployment, the system continuously measures with a 1-minute sampling period. The data is denoised and filtered by the preprocessing unit within the sensing chip before being written to the edge computing system for viscosity behavior curve generation and derivative analysis. Meanwhile, color turbidity is acquired by an optical detection module located within the drainage channel. This module consists of two optical paths: a direct-light channel, used to assess the absorption of different wavelengths of light (e.g., 470nm, 525nm, 630nm) by the drainage fluid in the forward direction; this path sensitively reflects changes in the RGB dimensions of the drainage fluid color (e.g., red, pale yellow, pink); and a 90-degree side-scattering light channel, specifically used to measure the scattering intensity of incident light by particles in the liquid (e.g., cell debris, lipid droplets, bacteria, exudates); this path is extremely sensitive to turbidity. The system employs a synchronous emission-synchronous acquisition mechanism to ensure simultaneous measurement of both channels, thereby generating a complete pair of color-turbidity optical vectors in each sampling period. After acquisition, the data first enters the edge module for spectroscopic calculation and light intensity normalization (excluding non-physiological factors such as light source attenuation and optical path contamination), and is assembled into a time series at second-level intervals, then input into the nonlinear curve modeling module. In Mr. Zhang's case, the continuously recorded data from the optical module showed that the drainage fluid was initially stable in color within the absorption peak region of λ1 = 525nm (light yellow) for the first 4 hours post-surgery, with scattered light intensity maintained in the range of 40-43 (low turbidity). However, within the 5th to 6th hour, the direct absorption suddenly shifted towards red, the λ1 peak intensity decreased to 32, and the turbidity scattering channel light intensity increased to 62, exhibiting obvious characteristics of protein and red blood cell leakage and inflammatory cell accumulation. This data was immediately fed into the model. The system first mapped the original absorption-scattering pairs into a color concentration matrix (an analog vector obtained by training a Kohonen self-organizing map network), and then combined it with the accelerated pH decrease and sudden drop in flow rate during the same time period, forming an abnormal superposition in the multi-indicator trend fusion module, which was then pushed to the next stage infection transition identification network.In terms of model structure, the system uses a segmented deep fusion neural network. The bottom layer input consists of the nonlinear derivative sequence of each indicator and trend classification labels. A multi-head attention mechanism is used in the middle to extract the linkage features between different parameters within a specific time window, such as the temporal coordinated burst of pH decrease + turbidity increase. During training, the system uses a large dataset of postoperative drainage fluid from hospitals (containing a complete parameter set of nearly 1400 cases labeled with infection / non-infection tags) for supervised learning. Each sample data point is a continuous time series segment, labeled as whether infected and the hourly offset before infection. The loss function is a dual objective function (infection classification error + time offset prediction error). Training uses the Adam optimizer with dropout and regularization to alleviate overfitting. The final model can accurately identify abnormal aggregation points 12–48 hours before infection, with an average prediction time error of less than 3 hours. Therefore, the abnormal turbidity and sudden increase in viscosity that Mr. Zhang experienced 5 hours post-surgery, as an early signal of local infection physiological response, was accurately captured and aggregated into a trend superposition event by the system. Combined with the transition recognition capability of the trained model, a high-risk score was finally obtained, and an early warning push was proactively triggered.

[0052] Mr. Zhang was continuously monitored starting at 9:00 AM on the first day after surgery. The system began collecting his drainage fluid flow rate V(t) and pH value every 5 minutes. In a 4-hour time window [t0,t1]=[9:00,13:00], a total of 48 time points of data sequence were obtained.

[0053] In the data preprocessing stage, all data points are first subjected to linear interpolation to fill in any minor discontinuities. Noise filtering is performed using wavelet denoising (db4 wavelet, 3-level decomposition). After obtaining the smooth curve, the velocity derivative at each time point is calculated. The dominant phase φ(t) of the pH fitting curve. For example, around 13:00 on the 5th hour after surgery, Mr. Zhang's flow rate decreased from a stable value of V(t)≈30mL / h in the first 30 minutes to 21mL / h, and the derivative... (Unit: mL / h) 2 The pH value dropped significantly from 7.18 to 6.95. In the sliding fit curve, the dominant phase φ(t) reached about 180° around 13:00 (indicating continuous acidification). According to the standard recovery template, the pH should have stabilized 5 hours after the operation. The reference phase of the standard model θ0(t) is about 30° (i.e., the pH maintains a slight increase or remains flat).

[0054] Substitute the above measured data into the formula:

[0055]

[0056] A discrete integration method (trapezoidal rule) is used for numerical approximation, with a sampling point of 5 minutes. The lower limit of integration is t0 = 9:00, and the upper limit is t1 = 13:00. In the data segment near 13:00, the following measurements are defined:

[0057] φ(t)=180°, θ0(t)=30°, then sin(150°)≈0.5

[0058] The product of the individual terms is: |-18·0.5|=9

[0059] Assuming the anomaly lasts for 30 minutes (6 points), the integral contribution of this interval is 9 × 0.5 = 4.5. With each point having a width of 5 minutes (0.083 hours), the total contribution is approximately 6 × 4.5 × 0.083 = 2.24.

[0060] The system simultaneously calculates the normal behavior segments at other time points. Within these normal segments, If φ(t)≈θ0(t), then the product term tends to 0 and contributes very little to the integral value.

[0061] Therefore, the integral result for the entire time window is:

[0062]

[0063] According to the model training data, the average value of Ψ(t) for standard recovered patients ranges from 0.3 to 1.1; while in confirmed infection cases, Ψ(t) is mostly concentrated between 2.0 and 4.5. Therefore, Mr. Zhang's calculated Ψ(t) = 2.6 in this period is already in the high-risk range for infection.

[0064] The range of parameters involved in this model includes: Normal fluctuations are typically between -5 and +5, and fluctuations exceeding ±10 are considered abrupt changes. The sin value of the angle difference φ(t)-θ0(t) converted to radians is generally consistent within ±0.3, and is judged to be in the opposite trend if it exceeds ±0.6. Therefore, the maximum value of its product can theoretically reach 20 (such as extreme abrupt changes + opposite trends). In this system, a value exceeding 8 is considered a strong outlier.

[0065] In terms of model structure, a lightweight convolutional-attention fusion network is used, and the input features are the features within a sliding window. The model uses φ(t), θ0(t), original flow rate and pH value, and their second-order derivative features. The first layer uses two 1D-CNNs to extract the rate of change features, the second layer uses a multi-head self-attention mechanism to extract trend aggregation patterns of temporal phase difference sequences, and the final layer is a fully connected classifier that outputs risk scores. The training process uses actual case data (n=1187, 284 positive samples), achieving a training and validation accuracy of 92.4% and an average early prediction window advance of 38 hours. During training, the function integral value Ψ(t) is used as the continuous risk label for the samples, guiding the model to learn to extract risk transition signals from nonlinear shifts.

[0066] Ultimately, five hours after Mr. Zhang's surgery, the system successfully identified the synergistic reaction of decreased local immune system control, accumulation of tissue metabolites, and abnormal exudate behavior. Using a behavioral phase integral model, it provided an early warning of Staphylococcus aureus infection 36 hours in advance. After culture verification and antibiotic intervention, the patient successfully avoided a second surgery in the surgical area.

[0067] In the first 24 hours after Mr. Zhang's surgery, the system collected nearly 300 complete time-series data points, covering five dimensions: pH, flow rate, turbidity, viscosity, and color. Each data point underwent rapid verification after collection, including edge filtering to remove supraphysiological abnormalities (such as interference from flow rates of 0 or higher than 100 mL / h), noise signal processing (based on time-domain smoothing or frequency-domain noise suppression), and then dynamic Z-score normalization was performed by the standardization module. This process not only considered population statistical data but also integrated Mr. Zhang's individual preoperative physiological data and the mean value of the stable period in the first 6 hours after surgery, ensuring that the standardized curves were both globally comparable and retained individual trend characteristics. Building upon this foundation, the adaptive behavioral feature extraction module built into the edge computing unit is activated. The system uses Mr. Zhang's early stable behavioral segment on the first day after surgery as his initial individual behavioral baseline template. Subsequently, it re-analyzes the changes in the nonlinear behavioral curve over the past six consecutive hours (i.e., a sliding window) every hour, extracting features such as first derivative, second derivative, trend extreme points, inflection points, and acceleration curvature. It then compares these features with the individual's initial template for trend similarity and performs multi-channel alignment with the group standard template matched to the surgical procedure. The system introduces a fusion loss function to balance the offset scores between the individual's true behavioral curve and the group recovery model. Specifically, when Mr. Zhang's indicator continuously deviates from the group average trend but remains consistent with his previous trajectory, the system reduces the sensitivity of his risk score; conversely, when a sudden behavioral change occurs with a reversal trend, the system immediately increases the weight of that behavioral segment in the risk score. The key to this mechanism is that the model's behavioral feature extraction is no longer static template matching, but rather updates the individual recovery trajectory every hour, achieving a truly adaptive prediction system by fine-tuning the baseline model. The entire prediction and scoring module is based on a lightweight stacked residual network (ResidualTemporalCNN+LSTM). The network structure is optimized and deployed on edge computing units to ensure low power consumption and high-speed inference capabilities. The model input is the behavior vector of each indicator over the past 6 hours (including its original value, derivative, directional difference, pH phase, flow velocity gradient, and other 28-dimensional features). The output is a dynamic infection risk score R(t), which ranges from 0 to 100. The model constructs a scoring mechanism through a sigmoid layer and standardized regression, and compares it with a set threshold T. warn=65 The system's score at hour 30 was R(t) = 62, indicating a moderate deviation. At hour 34, R(t) rose to 71, exceeding the warning threshold, immediately triggering the built-in clinical intervention prompt module. The system simultaneously pushed alarm information through nurse terminals and doctors' mobile devices, along with a trend chart of indicators over the past 6 hours, a score of the intensity of the current behavioral deviation, and auxiliary information such as abnormal combinations of indicators (e.g., continuous pH decrease + drastic fluctuations in flow rate + reversal of turbidity trend). It also recommended immediate action, including bacterial culture of drainage fluid, local ultrasound assessment for the presence of an abscess, and consideration of early adjustment of empirical antibiotics. This suggestion was adopted by the clinical team and addressed proactively. Mr. Zhang was subsequently diagnosed with Staphylococcus aureus infection on day 3, but due to the sufficient warning, the infection was controlled in situ and did not spread to the prosthesis, avoiding reoperation and functional impairment. The training process of the risk scoring system is based on a multi-stage transfer learning framework. The base network is first pre-trained on a national public dataset of drainage fluid parameters (n≈4500), and then fine-tuned on a set of annotated postoperative infection cases (n=983, of which 231 were positive) from our hospital to improve the model's sensitivity to local behavioral variations. The training loss function is risk scoring error + alarm time offset penalty + false positive penalty coefficient, the optimizer is AdamW, and the learning rate scheduler dynamically adjusts the interval [1e...]. -4 ,5e -6 During training, the residual gating weights of the model for individual behavioral trajectories are dynamically updated in each epoch, enabling the prediction results to be adjusted in real time according to the patient's performance.

[0068] In summary, the edge computing architecture and adaptive behavior modeling module of this system not only complete continuous, multi-parameter, and high-precision data processing, but also construct a highly robust intelligent infection risk identification method that is truly applicable to orthopedic postoperative scenarios through deep individualized curve management, dynamic scoring output, and closed-loop intervention prompting mechanism. It is not only technically feasible, but also highly practical in clinical practice.

[0069] Example 2:

[0070] Based on Example 1, the system continuously collected four key indicators of Mr. Zhang's drainage fluid during the 8 hours from 2:00 AM to 10:00 AM on the first day after surgery: pH, flow rate, turbidity and viscosity. The sampling frequency was once every 10 minutes, for a total of 48 data points.

[0071] In step S1, the system first selects 187 samples from 1026 non-infected patients in the hospital's existing infection surveillance database, based on their historical data. These samples match Mr. Zhang's surgical type (left knee replacement), location (knee), age group (60-70 years old), and whether an implant was used (yes). Drainage fluid parameters for each patient in the first 48 hours post-surgery are extracted to form a four-dimensional time-series dataset. This data is aligned using standard time axes (e.g., 1 hour post-surgery, 12 hours post-surgery), and K-means clustering is used to group similar recovery trajectories, ultimately forming six standard recovery template curves, each representing a typical recovery path. Mr. Zhang was matched to template category 3, corresponding to characteristics such as pH stabilizing after a slow decrease, gradually decreasing flow rate, slow decrease in turbidity, and viscosity approaching normal levels.

[0072] In the S2 phase, the system extracts continuous behavioral data of Mr. Zhang's four parameters within the current monitoring window (2:00 AM to 10:00 AM), and concatenates them in chronological order into a multi-dimensional behavioral vector X = {x1, x2, ..., X}. 48 Each vector element represents a set of parameters at a certain moment (every 10 minutes), such as [pH(t),V(t),C(t),η(t)], where C represents turbidity and η is viscosity.

[0073] In S3, the system compares Mr. Zhang's behavior vector with the standard vector of his matching template on the same time axis. Perform point-by-point alignment and ensure time consistency to ensure that Mr. Zhang's data at 6 hours post-surgery is compared only with the data at 6 hours post-surgery in the template.

[0074] S4 is the core of this method's computation. The system first calculates the Euclidean offset between the current behavior vector and the standard template, which is in the form of:

[0075]

[0076] Substituting Mr. Zhang's data, we set the mean deviation of each point at any given time to 0.35, the squared deviation to 0.1225, and the sum of the deviations for 48 points to approximately 5.88. Taking the square root, we obtain the Euclidean distance D≈2.42. However, in the non-infected recovery trajectory, this value averages between 1.0 and 1.6, therefore 2.42 is significantly deviated from the population standard.

[0077] Meanwhile, the system determines the trend direction for each set of parameters by calculating the sliding first derivative sign of each indicator within the current window (+1 for rising, -1 for falling, and 0 for stable). This is then compared with the derivative signs of the same indicator in the template. When the trends are opposite (e.g., the current pH is decreasing while the standard should be increasing), it is recorded as a directional outlier. Mr. Zhang's pH, flow rate, and turbidity all showed trends opposite to the standard during this period, with only viscosity showing a consistent trend. The system recorded 3 / 4 of the indicators as directional outliers, and this trend persisted for more than 3 sampling points (i.e., a continuous reverse trend for ≥30 minutes).

[0078] In S5, the system sets the Euclidean offset D = 2.42 and the internally set threshold T. D =2.0 comparison, the result is excessive; the number of directional outlier indicators is ≥3, and the continuous duration is ≥3 points (30 minutes), which meets the abnormal aggregation conditions. The system determines that there is a risk of infection jump in the current period and immediately generates an early warning signal R(t) = 83 / 100, which far exceeds the trigger threshold T. R =65.

[0079] The clinical response mechanism was triggered simultaneously, and the system pushed the scoring curve, trend graph, and parameter offset radar chart to the nurse station terminal. When the on-duty doctor received the push, Mr. Zhang had no visible drainage abnormalities or fever. Microbial testing of the drainage fluid was performed 24 hours in advance, and antibacterial pretreatment was initiated. Ultimately, Mr. Zhang's infection was confirmed positive on the 3rd postoperative day, but the inflammation was controlled in time, preventing the spread of the prosthesis infection. Mr. Zhang did not require a second surgery and was successfully transferred to the rehabilitation department on the 7th postoperative day.

[0080] This comprehensive example demonstrates that all five stages—from standard recovery path modeling, patient-specific behavior vector generation and comparison, Euclidean distance calculation and trend direction analysis, to early warning signal output—possess high engineering feasibility and clinical viability. The adjustable range of the core model's parameters is: Euclidean threshold setting range T. D =1.8-2.2, the trend outlier ratio is determined when ≥50% of the indicators are in opposite directions for more than 3 consecutive time points. The scoring result R(t) is output according to the combination logic and weighting function, and the scoring range is 0 to 100, where 65 is the high sensitivity trigger threshold.

[0081] In Mr. Zhang's postoperative infection risk monitoring, to accurately identify the behavior trajectory of his drainage fluid and judge personalized trend deviations, the system adopted a standard recovery trajectory library construction and time alignment analysis approach. Through cluster analysis, diverse recovery trajectory templates were constructed, and real-time collected patient drainage fluid parameters were mapped to the standard templates, enabling dynamic comparison with the expected recovery path and thus identifying early signs of infection. Firstly, in the data acquisition phase, the hospital infection risk assessment center compiled inpatient surgical data from the past three years, selecting a total of 1412 orthopedic patients who had clearly not developed postoperative infections. These included nine common surgical procedures such as hip replacement, knee replacement, spinal fusion, and humeral internal fixation. The data was tagged according to surgical area (e.g., hip, knee, spine), surgical method (minimally invasive or open), underlying diseases (e.g., diabetes, hypertension), preoperative CRP values, and age group. Each patient had at least 48 hours of continuous postoperative drainage fluid data. Collected parameters included pH, flow rate, turbidity, viscosity, and color spectral absorbance values, with a sampling period of 10 minutes and timestamp recording. In the data preprocessing stage, the system first performs a temporal integrity screening, removing cases with more than 15% missing segments or consecutive abnormal jumps, leaving 1186 valid samples. Subsequently, all data sequences are repaired for minor missing segments using linear interpolation and uniformly resampled. Uniform time intervals (10 minutes, 15 minutes, and 30 minutes) are used to generate sequences of uniform length to ensure all patients have the same temporal dimension after alignment. For example, if a 24-hour window is selected and sampling is performed every 15 minutes, each curve contains 96 sampling points. All parameters are uniformly converted into a matrix structure, forming a 5×96 parameter behavior matrix for each patient. The clustering analysis uses the K-means DTW algorithm based on the dynamic time warping (DTW) distance metric. Since different behavioral patterns have similar forms but different displacements or rates at different times, traditional Euclidean distance is insufficient to reflect the inherent trend of the sequence. DTW allows for non-linear time alignment. The system sets the initial number of clusters K=10 and evaluates the clustering effect using the silhouette coefficient and Davies-Bouldin index, ultimately achieving optimal discriminative power at K=6. Each cluster center represents a standard recovery trajectory template and is labeled with the characteristics of the population it is suitable for. For example, postoperative pH first decreases and then increases - suitable for young patients with fast recovery after knee replacement surgery, or flow rate decreases slowly but turbidity increases slightly in the early stage - suitable for elderly patients with hip implantation, etc.

[0082] When Mr. Zhang entered the postoperative monitoring system, the system first matched Template 3 from the standard trajectory library based on his surgical procedure (left knee replacement, open), age (65 years old), and underlying diseases (hypertension, impaired glucose tolerance). This template represents a recovery pattern where pH steadily decreases within 24 hours postoperatively, flow rate begins to slowly decline after the 8th hour, turbidity rapidly decreases after the 12th hour, and viscosity remains stable. The system collected Mr. Zhang's drainage fluid data in real time, setting the sampling period to every 10 minutes, and selecting a continuous 12-hour window from the 4th to the 16th hour postoperatively as the analysis window, collecting a total of 72 time points for 5 parameters, forming a 5×72-dimensional observation matrix. Since the standard template is a 5×96 structure, the system linearly interpolated and expanded Mr. Zhang's data to make its time dimension completely consistent with the template, forming a 5×96 current behavior matrix. Then, through a time alignment algorithm, the trend direction, numerical offset, and gradient changes of the corresponding time points of the two were synchronously compared, providing input features for subsequent Euclidean offset calculation, trend direction determination, and integral scoring modeling.

[0083] Furthermore, the key trend direction outlier function model in this invention is introduced to accurately identify whether postoperative physiological indicators show trend reversal behavior in local areas, and to infer whether there are behavioral trajectory deviation events as early signals of infection transition. From the first day to the morning of the second day after surgery, the drainage system continuously collected three key parameters: pH, flow rate, and turbidity. The time window length was 24 hours, and the sampling frequency was every 10 minutes, totaling 144 time points. Each time point formed a set of three-dimensional behavioral vectors. To achieve unified modeling, the system sequentially concatenated the pH, flow rate, and turbidity at each time point into a vector x(t) = [x1(t), x2(t), x3(t)], ultimately constructing a 3×144 behavioral matrix. This matrix was then structurally aligned with its matching standard template trajectory x0(t) to form a parallel sequence for trend comparison.

[0084] During the processing, the system first calculates the trend direction σ of each indicator within a sliding window (3 points, i.e., 30 minutes). i (t), if the current point is greater than the center point of the previous window, it is judged as an increase (+1); if it is less than, it is a decrease (-1); if it remains basically unchanged (the absolute value of the change is less than the physiological noise threshold ε, set to 0.02), it is 0. Taking Mr. Zhang's condition from the 20th to the 21st hour after surgery as an example (i.e., from 5:00 to 6:00 the next morning), the pH value decreased from 7.18 to 6.94, σ1(t)=-1; the flow rate decreased from 26mL / h to 18mL / h, σ2(t)=-1; while the turbidity increased from 49 to 61, σ3(t)=+1. The rate of change intensity η iThe values ​​(t) are as follows: pH decreases by 0.24 (η1 = 0.24), flow rate decreases by 8 mL / h (η2 = 0.32 after normalization), and turbidity increases by 12 units (η3 = 0.27).

[0085] The system was compared with the recovery template 3 matched by Mr. Zhang. The standard trend at the 20th hour was as follows: pH entered a stable phase (ξ1(t)=0), flow rate decreased slowly but with minimal change (ξ2(t)=0), and turbidity continued to decrease (ξ3(t)=-1). Substituting this into the outlier function of the trend direction in this invention:

[0086]

[0087] Perform the calculation. At this point, n = 3. Substitute the values ​​into the equations:

[0088] pH term: σ1=-1, η1=0.24, ξ1=0→-1×0.24×(1-0)=-0.24

[0089] Velocity term: σ² = -1, η² = 0.3², ξ² = 0 → -1 × 0.3² × (1 - 0) = -0.3²

[0090] Turbidity terms: σ³=+1, η³=0.27, ξ³=-1→+1×0.27×(1-(-1))=+0.54

[0091] The calculation shows that:

[0092] Δ(t)=-0.24-0.32+0.54=-0.02

[0093] Although Mr. Zhang's flow rate and pH both decreased significantly during this period, and the turbidity changed in the opposite direction (the system template expected a decrease but actually increased), the positive and negative offsets canceled each other out, resulting in an overall outlier score of only -0.02, which is within the neutral offset range set by the system (the low-risk threshold is |Δ(t)|<0.15). Therefore, the system did not trigger a risk warning at this time. However, in the following hour (hours 21-22), the pH continued to decrease to 6.85, σ1=-1, η1=0.09; the flow rate further decreased to 12 mL / h, σ2=-1, η2=0.4; and the turbidity increased rapidly to 75, σ3=+1, η3=0.36. Meanwhile, the standard template's three trends during this period remained ξ1=0, ξ2=0, ξ3=-1. Substituting these values ​​into the formula yields:

[0094] pH value: -1 × 0.09 × (1 - 0) = -0.09

[0095] Flow velocity term: -1 × 0.4 × (1 - 0) = -0.4

[0096] Turbidity term: 1 × 0.36 × (1 - (-1)) = +0.72

[0097] sum:

[0098] Δ(t)=-0.09-0.4+0.72=+0.23

[0099] Because the absolute value exceeds the system's warning threshold (the system defines |Δ(t)|>0.20 as a high-risk trend reversal), the system records this time period as a directional outlier peak, forming a continuous reverse chain with the previous trend, and marks it as a stable behavioral drift zone. Once the outlier score Δ(t) exceeds the high-risk threshold for three consecutive time periods (>30 minutes), the system immediately sends out a warning signal and triggers behavioral drift visualization and doctor-side notification.

[0100] The core advantage of this method lies in its function structure, which does not require checking whether specific values ​​are out of bounds, but rather detects whether physiological behavior is accelerating in the wrong direction. The specific range of its coefficients is set as follows: σ i ∈{-1,0,+1},η i After normalization, the distribution range is [0.0–1.0], where values ​​below 0.05 are not included in outlier scores and are considered as physiological fluctuations; ξ i The range of values ​​is the same as σ. i The system alarm threshold is set to |Δ(t)|>0.20 by default, and can be dynamically adjusted between 0.15 and 0.30 according to the patient's condition to balance sensitivity and specificity.

[0101] During the training phase of this function model, postoperative labeled data (positive infection cases n=203, non-infected cases n=742) were used for cross-validation. Through accuracy optimization and threshold parameter tuning, the time deviation between the outlier function and the system's final infection diagnosis was controlled within an average of 10 hours. Mr. Zhang was finally confirmed to have Staphylococcus aureus infection by culture on the morning of the 3rd day after surgery, but the system successfully issued an early warning in the early morning of the 2nd day. The continuous trend reversal segment identified by this function was pushed to the attending physician's terminal, preventing the spread of local infection and controlling symptoms.

[0102] During postoperative monitoring, the system constructs Mr. Zhang's behavioral vector in real time, splicing various indicators of his drainage fluid (such as pH, flow rate, turbidity, viscosity, etc.) into a multi-dimensional time-series vector according to timestamps in each sampling period. The vector is then aligned point by point using a standard recovery template that matches his surgical procedure, anatomical location, and physiological background. Based on this, the Euclidean offset is calculated to quantify the absolute deviation of Mr. Zhang's overall indicator trajectory within the current time window.

[0103] Between 26 and 28 hours post-surgery, the system measured a cumulative Euclidean offset of 2.36 between Mr. Zhang's behavioral vector and the standard template. This value typically does not exceed 1.4 for patients undergoing this procedure under standard recovery conditions. The system recorded this value as exceeding the warning threshold (set to a range of 2.0 ± 0.1). However, this invention does not immediately issue an alarm. Instead, it simultaneously initiates a directional outlier detection logic. The trend direction outlier function module performs trend sign analysis and standard direction comparison on the three key indicators: pH, flow rate, and turbidity. For three consecutive time periods, it identifies trend direction reversals. For example, pH should have stabilized on the standard trajectory but is currently accelerating downwards; flow rate should have been stable but is currently rapidly declining; and turbidity rebounds from a downward trend. The directional outlier score for each parameter continuously exceeds 0.25 (the system's set threshold for directional anomaly scores is 0.20) and lasts for more than 30 minutes. At this point, the system activates the joint discrimination rule: if the current Euclidean offset is greater than 2.0 and at least two key parameters have directional outlier scores exceeding the high-risk threshold within a continuous time window, the system considers the behavior segment as a critical state of infection risk transition.

[0104] In Mr. Zhang's case, the system first identified a significant difference between his overall physiological state and the standard recovery trajectory through behavioral offset vectors. Then, directional outlier analysis confirmed that these differences were not deviations that were still progressing in a positive direction, but rather deviations in the wrong direction. This double-confirmation logic greatly suppressed the risk of false positives and avoided misreporting due to postoperative fluctuations or parameter noise. The system output a risk score of 87 / 100 based on the joint scoring model. When the score exceeded the joint threshold (set to 80), an early warning mechanism was triggered, identifying the patient as a potential infection transition individual. A push notification was generated on the doctor's terminal, including a behavioral trajectory map of the past 6 hours, highlighted trend reversals, a dynamic Euclidean offset map, and an indicator risk radar chart, along with personalized intervention suggestions, such as immediate bacterial culture of drainage fluid, local ultrasound examination of periprosthetic fluid accumulation, and assessment of whether to adjust antibiotics.

[0105] The underlying model of the entire joint judgment mechanism is a hierarchical decision network structure. The first layer is a trend recognition network (directional outlier function), the second layer is a behavior offset distance scoring network (Euclidean distance discrimination), and the third layer is a joint logistic regression unit or a weighted average fusion layer. During model training, supervised learning is performed using a labeled postoperative monitoring dataset. Each sample data structure includes four labels: Euclidean offset, outlier score, infection occurrence time, intervention timeliness, and whether the final outcome is infection. The model aims to maximize prediction accuracy and early identification window. Training employs 10-fold cross-validation and an early stopping mechanism to control overfitting, ultimately enabling the model to maintain over 95% specificity in uninfected samples and achieve an average early identification time of 36 hours in infected transition samples.

[0106] This mechanism played a crucial role in Mr. Zhang's treatment. Because the system detected the dual signals of trajectory deviation and trend reversal 32 hours before the infection occurred, it promptly sent out warnings. After confirming the local inflammatory state, the doctor immediately initiated empirical antibiotic treatment, successfully controlling the spread of infection and avoiding prosthesis replacement and long-term stagnation in rehabilitation.

[0107] Example 3:

[0108] Combined with appendix Figure 4 The patient's information is as follows: Male, 34 years old, with a history of left tibia and fibula fracture treated with internal fixation with a plate 18 months ago. Due to recurrent leakage from a soft tissue fistula, the implant is scheduled for removal. Intraoperative findings ( Figure 4 a) After plate removal, the peroneus longus tendon is exposed (thick black arrow), with surrounding scar tissue and old sinus tract tissue hyperplasia. Treatment measures ( Figure 4 b): Complete synovectomy + debridement and irrigation of necrotic soft tissue, resulting in a spindle-shaped defect of approximately 10cm × 2cm.

[0109] Using the 1-0 nylon interrupted-release technique, a new suture segment with a spacing of 1-2 fingers is added every 48 hours at the ward dressing table. Figure 4 c, 4d). A Jackson-Pratt (JP) negative pressure drainage tube is placed subcutaneously immediately after each suture. Figure 4 e) The miniature multi-parameter in-situ sensing kit of the present invention is then inserted into the proximal end of the JP tube. This kit can record continuously on a minute-by-minute basis.

[0110] 1. Flow rate (microthermal film sensor);

[0111] 2. pH (solid glass electrode);

[0112] 3. Turbidity / Color (Dual-channel optical probe);

[0113] 4. Viscosity (micro-oscillating cantilever beam).

[0114] Monitoring was initiated immediately after the procedure, and all data were normalized, denoised, and written into the patient's personal behavior vector in real time through the bedside edge computing unit. The system automatically matched a standard recovery template 4 (cluster center of curves of 147 non-infected cases of the same age group and surgical procedure) for the lower limb implant removal + staged closure scenario.

[0115] The Euclidean offset D(t) and the outlier score S(t) of the trend direction are calculated cyclically within a 6-hour sliding window for early warning.

[0116] Key monitoring nodes and model outputs:

[0117]

[0118]

[0119] The triggering decision logic and threshold are both derived from the Euclidean offset + directional outlier joint discrimination framework proposed in the patent, which can output early warning on average 24-48 hours before the clinical diagnosis of infection.

[0120] Potential infection was detected within 24 hours of the second suture, approximately 20 hours earlier than bacterial culture results. An alarm was simultaneously triggered, sending trend heatmaps and risk radar charts to the patient's mobile device, allowing doctors to adjust antibiotics immediately and avoiding suture removal and re-debridement. Even with repeated shortening and disassembly of the drainage path, the sensor module maintained a high-frequency, continuous data stream, demonstrating the portability and robustness of the technology. The wound healed completely in one session, with no reinfection from the implant; the patient was able to bear weight and walk after 6 weeks.

Claims

1. A method of identifying the risk of postoperative drainage fluid infection in orthopedic surgery, characterized in that The method comprises the following steps: The drainage fluid is regarded as a system output driven by postoperative local immune regulation and wound recovery; nonlinear time series curves including viscosity, color turbidity, flow rate, and pH indicators are captured; first-order derivative trend modeling is performed; and the change rate is analyzed, including pH drop acceleration or mutation after flow rate stabilization; A postoperative recovery path framework is established, and a standard recovery trend library is formed using the dynamic parameters of the drainage fluid of non-infected cases; the Euclidean deviation and directional change between the current patient's drainage parameter behavior vector and the standard trajectory are compared; The change is converted into a visual recovery deviation map, and infection prediction data is measured in terms of multi-day trend angles; then a local anomaly signal aggregation is constructed to form an infection event transition framework, and a clinical infection node is obtained from the quantitative change triggering qualitative change; The postoperative drainage fluid is regarded as a continuous system output dynamically controlled by local immune regulation and tissue repair, and by collecting key indicators including flow rate and pH, a behavior phase deviation function model is constructed to determine whether the current patient state deviates from the non-infected standard recovery path, thereby realizing dynamic identification and early warning of infection risk; An integral function is introduced to simultaneously investigate the relationship between the flow rate change rate and the pH directional deviation: Wherein: represents the overall intensity score of the current patient's drainage fluid behavior deviating from the standard recovery trajectory within a certain time window; upper and lower limits of the score and represents the start and end time of the current calculation window; represents the flow rate of the drainage fluid at time , represents the pH trend direction of the drainage fluid; represents the phase of the pH trend that should be in the standard non-infectious recovery model at this time; Flow rate, pH, and turbidity, key drainage fluid indicators, are constructed into a unified time vector structure, and a directional function is used to determine whether the change trend of each period is synchronized or reversed with the standard recovery trajectory, so as to identify behavior deviation events; A newly constructed trend direction outlier function is introduced as follows: wherein, : indicates the trend directionality outlier intensity score between the current patient and the standard recovery model at time point : number of monitored key parameters of the drainage fluid; : the trend direction of the current th indicator at time point : the absolute value of the rate of change of the current th indicator over the period, i.e. its trend significance intensity; : the theoretical trend direction of the indicator in the standard recovery trajectory at time .​​ 2. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 1, characterized in that The nonlinear time series curve construction method comprises: A multi-parameter in-situ sensing module is integrated in the postoperative drainage channel to continuously collect the physical and chemical parameters of the drainage fluid; the in-situ sensing module collects the above-mentioned parameters at a minute-level time granularity, and inputs each data with a time stamp into an edge computing unit; the edge computing unit establishes nonlinear time series behavior curves of each parameter, and extracts the change rate, acceleration, and abnormal inflection point thereof; The behavior curves of each parameter are compared with the postoperative standard recovery model to identify the deviation degree and trend direction; when synchronous deviation trends of multiple parameters are identified in the same time window, it is judged as an early signal of infection risk transition, and an infection risk score result is output.

3. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 2, characterized in that The viscosity is measured by a micro-oscillation cantilever beam sensing structure, and the vibration attenuation characteristics of the cantilever beam are used to represent the viscosity behavior of the drainage fluid; the color turbidity is obtained by an optical detection module arranged in the drainage channel, and the optical detection module comprises a channel for measuring direct absorption light and a channel for detecting 90-degree scattered light to simultaneously evaluate color and turbidity.

4. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 3, characterized in that The flow rate and pH are obtained by synchronous collection, and the time phase difference is analyzed by behavior curve to determine whether local environment acidification is accompanied by flow rate sudden change; the standard recovery model is constructed based on the time series behavior data of each drainage fluid parameter under the condition of non-infection of normal patients after operation, and is used as a control trajectory for subsequent individual identification.

5. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 4, characterized in that The edge computing unit comprises an adaptive behavior feature extraction module for dynamically updating the baseline behavior curve of individual patients; the infection risk score result is output to the clinical terminal according to the risk level, and the clinical intervention prompt is triggered when the score value reaches the preset warning threshold.

6. The method of identifying the risk of postoperative drainage fluid infection in orthopedics of claim 1, wherein The contrast method of the Euclidean offset degree and the directionality change comprises: S1, constructing a postoperative recovery normal path model, comprising collecting multi-parameter time series data of drainage fluid of non-infected patients, and classifying and clustering according to operation type, site and individual characteristics to form a standard recovery trajectory library; S2, obtaining a plurality of continuous parameters of postoperative drainage fluid of a target patient within a set time window to form a multi-dimensional behavior vector; S3, aligning the current patient behavior vector with the corresponding template in the standard recovery trajectory, and performing point-by-point comparison based on time consistency; S4, calculating the Euclidean offset degree between the current behavior vector and the standard trajectory to quantify the overall deviation degree; simultaneously calculating the trend direction change of each parameter, identifying whether there is a change trend opposite to the direction of the standard trajectory, and marking as direction outlier; S5, when the Euclidean offset degree exceeds the preset deviation threshold, and multiple parameters simultaneously appear direction outlier, it is determined that the patient has a postoperative infection risk transition trend, and an early warning signal is output.

7. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 6, characterized in that The standard recovery trajectory library generates a plurality of representative recovery curve templates through cluster analysis to adapt to differences in different operation types, operation areas and patient physiological backgrounds; the drainage fluid parameter collection time window is continuously 4-24 hours, and is resampled according to a uniform time interval for time alignment with the standard trajectory.

8. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 7, characterized in that The behavior vector is composed of a plurality of indexes to form a unified vector structure according to the time stamp; the direction outlier determination comprises comparing whether the upward or downward trend of the current index is opposite to the trend direction of the corresponding time period of the standard trajectory, and recording the outlier event when the trend is reversed.

9. The method of identifying the risk of postoperative drainage fluid infection in orthopedics according to claim 8, characterized in that The Euclidean offset degree and the direction outlier result are jointly used for risk discrimination, and only when both meet the warning conditions, it is determined that there is an early infection transition tendency.

Images