Early monitoring system for postoperative anastomotic leakage in colorectal surgery based on perioperative multi-modal data analysis

Abstract

The present application relates to the field of medical information technology and postoperative monitoring technology, and particularly relates to a postoperative anastomotic leakage early monitoring system based on perioperative multi-modal data analysis, comprising a data acquisition module, which is used to acquire intraoperative physiological data sets and postoperative time series data sets of a target patient; a data processing module, which is used to standardize and time series align the acquired intraoperative physiological data sets and postoperative time series data sets, and generate standardized time series data sets; an analysis and early warning module, which is used to receive the standardized time series data sets, perform risk analysis and generate early warning signals; and a signal output module, which is used to output the risk score value, its change trend and the early warning instruction. The present application realizes the transition from passive monitoring to active prediction, can deeply integrate multi-dimensional information, realizes early, objective and individualized early warning of anastomotic leakage, and significantly improves the clinical decision-making efficiency and patient prognosis.

Description

Technical Field

[0001] This invention relates to the fields of medical information technology and postoperative monitoring technology, specifically to an early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis. Background Technology

[0002] Colorectal surgery is a crucial treatment for colorectal cancer and other intestinal diseases, but anastomotic leakage is one of its most serious and fatal postoperative complications. Once it occurs, it can lead to abdominal infection, sepsis, organ failure, and other complications, significantly increasing reoperation rates and mortality, as well as drastically prolonging hospital stays and increasing medical costs. Therefore, achieving early and accurate warning of anastomotic leakage, and gaining valuable time for clinical intervention, is a critical clinical problem that urgently needs to be solved in the field of colorectal surgery, and is of great significance for improving patient prognosis and enhancing the quality of medical care.

[0003] Currently, the monitoring and diagnosis of anastomotic leakage in clinical practice mainly relies on the following methods: First, traditional clinical observation, where medical staff regularly monitor the patient's vital signs, such as body temperature, heart rate, and the degree of abdominal pain, and observe the characteristics of drainage fluid; second, laboratory tests, including regular testing of blood levels of inflammatory markers such as white blood cell count, C-reactive protein, and procalcitonin; and third, imaging examinations, such as CT scans and gastrointestinal contrast studies used to confirm the diagnosis after signs of infection appear. In addition, intraoperative indocyanine green fluorescence imaging has been applied, allowing surgeons to visually assess blood perfusion in the anastomotic area during surgery to determine the anastomosis location. However, this assessment is mostly qualitative or semi-quantitative, lacking dynamic correlation with the postoperative process.

[0004] Despite the widespread application of the aforementioned technologies, significant drawbacks remain. First, they suffer from high latency: existing monitoring methods are mostly reactive. Significant changes in clinical symptoms and inflammatory markers typically only appear after infection has occurred, by which time the patient may have already developed sepsis, missing the optimal intervention window. Second, they are isolated and subjective: monitoring indicators are usually interpreted in isolation by physicians, relying on their personal experience for comprehensive judgment. There is a lack of an objective tool for in-depth integration and quantitative analysis of multi-dimensional information, easily missing early, weak but crucial abnormal signals. Finally, they lack individualization and prospective capabilities: the initial information provided by intraoperative fluorescence imaging is not effectively used to construct individualized baseline models for patients, and postoperative data cannot be dynamically combined to predict healing trends. Existing methods cannot achieve the leap from "monitoring what has happened" to "predicting what is about to happen." Therefore, developing a system that can integrate intraoperative and postoperative data to achieve dynamic, objective, and early risk warning has become an urgent need in clinical practice. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, this invention provides an early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis, which can effectively solve the problems mentioned in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides an early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis, comprising:

[0008] The data acquisition module is used to acquire intraoperative physiological datasets and postoperative time-series datasets of the target patient;

[0009] The intraoperative physiological dataset includes perfusion characteristic parameters and tissue characteristic parameters of the anastomotic region obtained based on intraoperative fluorescence image analysis;

[0010] The postoperative time-series dataset includes vital signs data, serum inflammatory marker data, and biochemical characteristic parameters of drainage fluid.

[0011] The data processing module is communicatively connected to the data acquisition module and is used to standardize and time-series align the acquired intraoperative physiological dataset and postoperative time-series dataset to generate a standardized time-series dataset.

[0012] The analysis and early warning module is communicatively connected to the data processing module and is used to receive the standardized time-series dataset, perform risk analysis, and generate early warning signals.

[0013] The analysis and early warning module includes:

[0014] The individualized model building unit is used to build a digital model representing the initial healing state of the anastomosis of the target patient based on the intraoperative physiological dataset.

[0015] The data assimilation and update unit is used to take the postoperative time series dataset as dynamic input and dynamically update the internal state variables of the digital model through a data assimilation algorithm;

[0016] The risk score calculation unit is used to calculate and output a real-time risk score based on the updated internal state variables.

[0017] The early warning determination unit is used to compare the risk score and its changing trend with preset static and dynamic thresholds, and generate an early warning instruction;

[0018] The signal output module is communicatively connected to the analysis and early warning module and is used to output the risk score value, its changing trend, and the early warning command to the outside world.

[0019] Furthermore, the perfusion characteristic parameters of the anastomosis region include peak fluorescence intensity, time to peak, area under the curve, and perfusion uniformity index; the tissue characteristic parameters include tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient; the perfusion characteristic parameters of the anastomosis region are extracted by a fluorescence image quantitative analysis algorithm and used to construct the initial state vector of the digital model.

[0020] The vital signs data include heart rate, blood pressure, body temperature, and respiratory rate; serum inflammatory marker data include C-reactive protein, procalcitonin, and interleukin-6; and the biochemical characteristics of the drainage fluid include amylase, bilirubin, lactate dehydrogenase, and pH value. The vital signs data are collected continuously at a sampling frequency of 1 hour.

[0021] Furthermore, the data processing module sequentially executes data cleaning, data standardization, temporal alignment, and feature dimensionality reduction processes to convert the intraoperative physiological dataset and postoperative temporal dataset into a unified, standardized temporal dataset, specifically including:

[0022] The data cleaning unit is used to identify and remove outliers from the intraoperative physiological dataset and the postoperative time-series dataset, and to impute missing values.

[0023] The data standardization unit is used to convert the intraoperative physiological dataset and postoperative time-series dataset into dimensionless standardized data using a standardization algorithm.

[0024] The timing alignment unit is used to align standardized data with different sampling frequencies to a unified timestamp using an interpolation algorithm, with the end time of the operation as the unified time origin.

[0025] The feature reduction unit is used to reduce the dimensionality of standardized and aligned data and extract feature vectors for digital model calculations.

[0026] Furthermore, the data standardization unit adopts the Z-score standardization algorithm; the temporal alignment unit adopts the cubic spline interpolation algorithm; and the feature dimensionality reduction unit adopts the principal component analysis algorithm.

[0027] Furthermore, the individualized model building unit performs the following steps:

[0028] The peak fluorescence intensity, time to peak, area under the curve, perfusion uniformity index, tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient were combined and recorded as the initial physiological characteristics. ;

[0029] The offline-trained support vector regression model is invoked. This model uses a radial basis function as its kernel, with a kernel width γ of 0.3 and an insensitive loss function. It is 0.05;

[0030] The support vector regression model internally stores a set of support vectors. and their corresponding coefficients and bias ;

[0031] Recording of initial physiological characteristics of new input Support vector regression model calculation Kernel function values ​​of all support vectors Then, a linear combination is performed, with the specific formula as follows:

[0032] ;

[0033] in, For continuous prediction values, For the weights of the support vectors, For radial basis kernel functions, For model bias;

[0034] Continuous predicted values The initial leakage probability is obtained by mapping to the [0,1] interval using the Sigmoid function. The specific formula is as follows:

[0035] ;

[0036] in, The initial leakage probability after Sigmoid mapping;

[0037] Finally, the initial leakage probability was determined. tissue oxygenation index Injection uniformity index Tissue scattering coefficient Combined into an internal state vector And write it into the system.

[0038] Furthermore, the data assimilation and update unit uses the internal state vector The initial conditions are updated globally in each extended Kalman filter loop. The four components make The data changes dynamically with postoperative time-series data.

[0039] Furthermore, the data assimilation and update unit operates according to the following steps:

[0040] Postoperative observation records were compiled from the vital signs data, serum inflammatory marker data, and biochemical characteristics of the drainage fluid collected every hour after the operation.

[0041] Using an extended Kalman filter loop, the postoperative observation records are read in one by one, and the internal state value of the anastomosis healing status is updated.

[0042] Furthermore, the risk scoring calculation unit performs the following steps:

[0043] Take the latest updated internal state values ​​to form a risk feature vector, and input the risk feature vector into the online logistic regression model. The model formula is:

[0044] ;

[0045] in, To score risk, For risk feature vectors, and These are the weights and biases, respectively; Mapped to a score of 0–100, the higher the score, the greater the risk.

[0046] Furthermore, the early warning determination unit generates early warning instructions according to the following rules:

[0047] Static threshold determination: When the risk score is above 70 for 30 consecutive minutes, the system issues a Level 1 warning; if it is above 85 for 15 consecutive minutes, it is upgraded to a Level 2 warning.

[0048] Dynamic trend determination: The system calculates the upward slope of the scoring curve every 3 hours. If the upward slope exceeds 5 points per hour, a level 2 warning is triggered.

[0049] Finally, the warning level is pushed to the clinical team.

[0050] The technical solution provided by this invention has the following advantages compared with the known prior art:

[0051] This invention constructs a dynamic analysis framework that combines an intraoperative baseline model with postoperative data assimilation. By utilizing the extended Kalman filter algorithm, it deeply integrates and dynamically updates the initial state of the anastomosis provided by intraoperative fluorescence imaging with continuously acquired physiological data after surgery. This enables the model to simulate the evolution of the anastomosis healing state in real time, changing the existing technology's passive response mode to abnormal indicators that have already appeared. It achieves advanced and accurate prediction of leakage risk, thus gaining a golden window for clinical intervention.

[0052] This invention overcomes the limitations of relying on doctors' subjective experience and interpreting single indicators in isolation. The system automatically integrates multi-dimensional information such as perfusion parameters, tissue characteristics, vital signs, inflammatory markers, and biochemical indicators of drainage fluid, and performs quantitative calculations through machine learning models to finally output an objective and continuous risk score. This score is based on the patient's own intraoperative data to build an initial model, truly realizing the individualization of the assessment plan and greatly improving the accuracy and reliability of the early warning results.

[0053] This invention not only outputs a single risk value, but also combines a dual early warning mechanism of static threshold and dynamic trend judgment. It can identify both continuous high-risk states and keenly capture the rapid escalation trend of risk, and classify the early warning level accordingly. Through integration with the hospital information system, the most critical information is delivered to the clinical team in real time through various methods such as graphical representation and push notifications, which significantly improves the practicality and action-oriented nature of the early warning information and effectively supports the optimization of the clinical decision-making process. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0055] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0056] Figure 2 This is a system workflow diagram of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0058] The present invention will be further described below with reference to embodiments.

[0059] Example:

[0060] Reference Figure 1 An early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis includes:

[0061] The data acquisition module is used to acquire intraoperative physiological datasets and postoperative time-series datasets of the target patient;

[0062] Intraoperative physiological datasets include perfusion characteristic parameters and tissue property parameters of the anastomotic region obtained based on intraoperative fluorescence image analysis;

[0063] Postoperative time-series datasets include vital signs data, serum inflammatory marker data, and biochemical characteristics of drainage fluid;

[0064] The data processing module, which communicates with the data acquisition module, is used to standardize and time-align the acquired intraoperative physiological dataset and postoperative time-series dataset to generate a standardized time-series dataset.

[0065] The analysis and early warning module communicates with the data processing module to receive standardized time-series datasets, perform risk analysis, and generate early warning signals.

[0066] The analysis and early warning module includes:

[0067] The individualized model building unit is used to construct a digital model representing the initial healing state of the anastomosis in the target patient based on intraoperative physiological datasets.

[0068] The data assimilation and update unit is used to dynamically update the internal state variables of the digital model by taking the postoperative time series dataset as dynamic input and using the data assimilation algorithm.

[0069] The risk score calculation unit is used to calculate and output the real-time risk score based on the updated internal state variables;

[0070] The early warning determination unit is used to compare the risk score and its changing trend with preset static and dynamic thresholds and generate early warning instructions.

[0071] The signal output module communicates with the analysis and early warning module and is used to output risk score values, their changing trends, and early warning instructions to the outside world.

[0072] Furthermore, the perfusion characteristic parameters of the anastomotic region include peak fluorescence intensity, time to peak, area under the curve, and perfusion uniformity index; tissue characteristic parameters include tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient; the perfusion characteristic parameters of the anastomotic region are extracted by the fluorescence image quantitative analysis algorithm and used to construct the initial state vector of the digital model.

[0073] Vital signs data include heart rate, blood pressure, body temperature, and respiratory rate; serum inflammatory marker data include C-reactive protein, procalcitonin, and interleukin-6; drainage fluid biochemical parameters include drainage fluid amylase, bilirubin, lactate dehydrogenase, and pH; vital signs data are collected continuously at a sampling frequency of 1 hour.

[0074] Furthermore, the data processing module sequentially executes data cleaning, data standardization, time-series alignment, and feature dimensionality reduction processes to transform the intraoperative physiological dataset and postoperative time-series dataset into a unified, standardized time-series dataset. Specifically, this includes:

[0075] The data cleaning unit is used to identify and remove outliers in the intraoperative physiological dataset and the postoperative time series dataset, and to impute missing values.

[0076] Data standardization unit, used to convert intraoperative physiological datasets and postoperative time-series datasets into dimensionless standardized data using standardization algorithms;

[0077] The timing alignment unit is used to align standardized data with different sampling frequencies to a unified timestamp using an interpolation algorithm, with the end time of the operation as the unified time origin.

[0078] The feature reduction unit is used to reduce the dimensionality of standardized and aligned data and extract feature vectors for digital model calculations.

[0079] Furthermore, the data standardization unit adopts the Z-score standardization algorithm; the temporal alignment unit adopts the cubic spline interpolation algorithm; and the feature dimensionality reduction unit adopts the principal component analysis algorithm.

[0080] Furthermore, the individualized model building unit performs the following steps:

[0081] The peak fluorescence intensity, time to peak, area under the curve, perfusion uniformity index, tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient were combined and recorded as the initial physiological characteristics. ;

[0082] The system calls a support vector regression model that has been trained offline. The support vector regression model uses radial basis functions as the kernel, with a kernel width γ of 0.3 and insensitive loss. It is 0.05;

[0083] The support vector regression model internally stores a set of support vectors. and their corresponding coefficients and bias ;

[0084] Recording of initial physiological characteristics of new input Support vector regression model calculation Kernel function values ​​of all support vectors Then, a linear combination is performed, with the specific formula as follows:

[0085] ;

[0086] in, For continuous prediction values, For the weights of the support vectors, For radial basis kernel functions, For model bias;

[0087] Continuous predicted values The initial leakage probability is obtained by mapping to the [0,1] interval using the Sigmoid function. The specific formula is as follows:

[0088] ;

[0089] in, The initial leakage probability after Sigmoid mapping;

[0090] Finally, the initial leakage probability was determined. tissue oxygenation index Injection uniformity index Tissue scattering coefficient Combined into an internal state vector And write it into the system.

[0091] Furthermore, the data assimilation update unit uses an internal state vector The initial conditions are updated globally in each extended Kalman filter loop. The four components make The data changes dynamically with postoperative time-series data.

[0092] Furthermore, the data assimilation and update unit operates according to the following steps:

[0093] Postoperative observation records were compiled from the vital signs data, serum inflammatory marker data, and biochemical characteristics of the drainage fluid collected every hour after the operation.

[0094] Using an extended Kalman filter loop, the postoperative observation records are read in one by one, and the internal state value of the anastomosis healing status is updated.

[0095] Furthermore, the risk scoring calculation unit performs the following steps:

[0096] Take the latest updated internal state values ​​to form a risk feature vector, and input the risk feature vector into the online logistic regression model. The model formula is:

[0097] ;

[0098] in, To score risk, For risk feature vectors, and These are the weights and biases, respectively; Mapped to a score of 0–100, the higher the score, the greater the risk.

[0099] Furthermore, the early warning determination unit generates early warning instructions according to the following rules:

[0100] Static threshold determination: When the risk score is above 70 for 30 consecutive minutes, the system issues a Level 1 warning; if it is above 85 for 15 consecutive minutes, it is upgraded to a Level 2 warning.

[0101] Dynamic trend determination: The system calculates the upward slope of the scoring curve every 3 hours. If the upward slope exceeds 5 points per hour, a level 2 warning is triggered.

[0102] Finally, the warning level is pushed to the clinical team.

[0103] In one specific embodiment, the system mainly includes a data acquisition module, a data processing module, an analysis and early warning module, and a signal output module. The modules communicate with each other via the hospital's internal network or a secure Internet of Things (IoT) protocol to ensure real-time and secure data transmission.

[0104] The data acquisition module is responsible for collecting patient data from various medical devices and information systems. Specifically, the intraoperative physiological dataset is generated by acquiring images of the anastomosis area using an intraoperative fluorescence imaging system such as an indocyanine green fluorescence imaging system, and extracting a series of key parameters through a built-in quantitative analysis algorithm. These parameters include perfusion characteristic parameters such as peak fluorescence intensity, time to peak, area under the curve, and perfusion uniformity index, as well as tissue characteristic parameters such as tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient, to comprehensively characterize the initial microcirculation state and tissue viability of the anastomosis area. The postoperative time-series dataset is automatically collected from bedside monitors, laboratory information systems, and electronic medical records. It includes vital signs data such as heart rate, blood pressure, body temperature, and respiratory rate collected continuously every hour; serum inflammatory markers such as C-reactive protein, procalcitonin, and interleukin-6 collected every six to twelve hours; and biochemical characteristic parameters of the drainage fluid such as amylase, bilirubin, lactate dehydrogenase, and pH value collected every twelve hours.

[0105] The data processing module receives raw data from the acquisition module and sequentially executes data cleaning, standardization, time-series alignment, and feature dimensionality reduction processes. The data cleaning unit uses box plot statistics to identify and remove outliers and fills in missing values ​​using the k-nearest neighbor interpolation algorithm to ensure data integrity and reliability. The data standardization unit uses the Z-score standardization algorithm to convert various features into dimensionless values ​​with a mean of zero and a standard deviation of one, eliminating the influence of dimensional differences. The time-series alignment unit uses the end time of surgery as a unified time origin and employs cubic spline interpolation to align data from different sampling frequencies to a unified time series at one point per hour. The feature dimensionality reduction unit uses principal component analysis to reduce the dimensionality of high-dimensional features, retaining principal components with a cumulative contribution rate exceeding 95% as feature vectors for subsequent model input, reducing computational complexity while avoiding overfitting.

[0106] The analysis and early warning module, as the core of the entire system, comprises four collaborative units, enabling end-to-end management from initial modeling to dynamic early warning. The individualized model building unit combines seven features extracted during surgery into an initial feature vector and calls a pre-trained offline support vector regression model. This model uses a radial basis function as its kernel function, and its key hyperparameters are optimized on a dataset containing 1200 historical surgical cases through grid search and five-fold cross-validation. The model first calculates initial predicted values, then maps them to initial leakage probabilities using a sigmoid function. Finally, this probability is combined with the tissue oxygenation index, perfusion homogeneity index, and tissue scattering coefficient to form an initial internal state vector, representing the individualized initial state of the patient's anastomosis.

[0107] The data assimilation and update unit employs the Extended Kalman Filter (EKF) algorithm. Starting from the initial state vector, it uses the aligned observation data from each hour post-surgery as dynamic input to recursively update the state vector. The system defines a nonlinear state-space model, including state equations describing the evolution of the state over time and observation equations predicting observations from the state vector. Utilizing the EKF prediction update loop, it combines new observation data with model predictions to achieve dynamic synchronization between the model and the patient's actual state.

[0108] After each state update, the risk score calculation unit inputs the current state vector into an online logistic regression model. The model's weights and biases are trained based on time-series data from multiple postoperative patients, and the output is a risk score ranging from zero to one hundred, with higher scores indicating a greater risk of anastomotic leakage.

[0109] The early warning determination unit combines static thresholds and dynamic trends. The relevant thresholds are based on a retrospective analysis of 500 missed positive cases, and the optimal warning point is determined using ROC curves. The static threshold determination rule is that a risk score above 70 for 30 consecutive minutes triggers a Level 1 yellow warning, and a score above 85 for 15 consecutive minutes triggers a Level 2 red warning. The dynamic trend determination rule is that the system calculates the moving average slope of the score curve every three hours; if the slope exceeds 5 points per hour, the risk is considered to be rapidly escalating, immediately triggering a Level 2 red warning. The generated warning instructions and levels are pushed to the clinical team via the signal output module.

[0110] The signal output module integrates with the hospital's clinical information system or mobile nursing terminal, outputting results in two forms. One is a graphical interface, displaying the patient's risk score curve, current score, and warning status in real time on the nurse station's large screen or the doctor's computer. The other is a push alarm; when a warning is triggered, the system automatically sends an alarm message to the responsible doctor's mobile application or pager, including the patient's identification, warning level, and current score, ensuring timely intervention by the clinical team.

[0111] The overall system workflow can be referenced. Figure 2 First, fluorescence images are acquired during the operation and parameters are extracted to construct the initial state. After the operation, data are continuously acquired, cleaned, standardized, aligned and dimensionality reduced by the processing module. Then, the processed data is sent to the analysis and early warning module to update the state by assimilating it through the extended Kalman filter algorithm. Next, the real-time risk score is calculated based on the latest state. The early warning judgment unit issues an early warning of the corresponding level based on the score and its changing trend. Finally, the results are visualized and pushed to the clinical terminal through the signal output module.

[0112] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis, characterized in that, include: The data acquisition module is used to acquire intraoperative physiological datasets and postoperative time-series datasets of the target patient; The intraoperative physiological dataset includes perfusion characteristic parameters and tissue characteristic parameters of the anastomotic region obtained based on intraoperative fluorescence image analysis; The postoperative time-series dataset includes vital signs data, serum inflammatory marker data, and biochemical characteristic parameters of drainage fluid. The data processing module is communicatively connected to the data acquisition module and is used to standardize and time-series align the acquired intraoperative physiological dataset and postoperative time-series dataset to generate a standardized time-series dataset. The analysis and early warning module is communicatively connected to the data processing module and is used to receive the standardized time-series dataset, perform risk analysis, and generate early warning signals. The analysis and early warning module includes: The individualized model building unit is used to build a digital model representing the initial healing state of the anastomosis of the target patient based on the intraoperative physiological dataset. The data assimilation and update unit is used to take the postoperative time series dataset as dynamic input and dynamically update the internal state variables of the digital model through a data assimilation algorithm; The risk score calculation unit is used to calculate and output a real-time risk score based on the updated internal state variables. The early warning determination unit is used to compare the risk score and its changing trend with preset static and dynamic thresholds, and generate an early warning instruction; The signal output module is communicatively connected to the analysis and early warning module and is used to output the risk score value, its changing trend, and the early warning command to the outside world.

2. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The perfusion characteristic parameters of the anastomotic region include peak fluorescence intensity, time to peak, area under the curve, and perfusion uniformity index; the tissue characteristic parameters include tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient; the perfusion characteristic parameters of the anastomotic region are extracted by a fluorescence image quantitative analysis algorithm and used to construct the initial state vector of the digital model; The vital signs data include heart rate, blood pressure, body temperature, and respiratory rate; serum inflammatory marker data include C-reactive protein, procalcitonin, and interleukin-6; and the biochemical characteristics of the drainage fluid include amylase, bilirubin, lactate dehydrogenase, and pH value. The vital signs data are collected continuously at a sampling frequency of 1 hour.

3. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The data processing module sequentially executes data cleaning, data standardization, time-series alignment, and feature dimensionality reduction processes to convert the intraoperative physiological dataset and postoperative time-series dataset into a unified, standardized time-series dataset. Specifically, this includes: The data cleaning unit is used to identify and remove outliers from the intraoperative physiological dataset and the postoperative time-series dataset, and to impute missing values. The data standardization unit is used to convert the intraoperative physiological dataset and postoperative time-series dataset into dimensionless standardized data using a standardization algorithm. The timing alignment unit is used to align standardized data with different sampling frequencies to a unified timestamp using an interpolation algorithm, with the end time of the operation as the unified time origin. The feature reduction unit is used to reduce the dimensionality of standardized and aligned data and extract feature vectors for digital model calculations.

4. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 3, characterized in that, The data standardization unit uses the Z-score standardization algorithm; the temporal alignment unit uses the cubic spline interpolation algorithm; and the feature dimensionality reduction unit uses the principal component analysis algorithm.

5. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The individualized model building unit performs the following steps: The peak fluorescence intensity, time to peak, area under the curve, perfusion uniformity index, tissue oxygenation index, tissue hemoglobin concentration, and tissue scattering coefficient were combined and recorded as the initial physiological characteristics. ; The offline-trained support vector regression model is invoked. This model uses a radial basis function as its kernel, with a kernel width γ of 0.3 and an insensitive loss function. It is 0.05; The support vector regression model internally stores a set of support vectors. and their corresponding coefficients and bias ; Recording of initial physiological characteristics of new input Support Vector Regression Model Calculation Kernel function values ​​of all support vectors Then, a linear combination is performed, the specific formula is: ； in, For continuous prediction values, For the weights of the support vectors, For radial basis kernel functions, For model bias; Continuous predicted values The initial leakage probability is obtained by mapping to the [0,1] interval using the Sigmoid function. The specific formula is as follows: ； in, The initial leakage probability after Sigmoid mapping; Finally, the initial leakage probability was determined. tissue oxygenation index Injection uniformity index Tissue scattering coefficient Combined into an internal state vector And write it into the system.

6. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The data assimilation and update unit uses the internal state vector The initial conditions are updated globally in each extended Kalman filter loop. The four components make The data changes dynamically with postoperative time-series data.

7. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The data assimilation and update unit operates according to the following steps: Postoperative observation records were compiled from the vital signs data, serum inflammatory marker data, and biochemical characteristics of the drainage fluid collected every hour after the operation. Using an extended Kalman filter loop, the postoperative observation records are read in one by one, and the internal state value of the anastomosis healing status is updated.

8. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The risk scoring calculation unit performs the following steps: Take the latest updated internal state values ​​to form a risk feature vector, and input the risk feature vector into the online logistic regression model. The model formula is: ； in, To score risk, For risk feature vectors, and These are the weights and biases, respectively; Mapped to a score of 0–100, the higher the score, the greater the risk.

9. The early monitoring system for anastomotic leakage after colorectal surgery based on perioperative multimodal data analysis according to claim 1, characterized in that, The early warning determination unit generates early warning instructions according to the following rules: Static threshold determination: When the risk score is above 70 for 30 consecutive minutes, the system issues a Level 1 warning; if it is above 85 for 15 consecutive minutes, it is upgraded to a Level 2 warning. Dynamic trend determination: The system calculates the upward slope of the scoring curve every 3 hours. If the upward slope exceeds 5 points per hour, a level 2 warning is triggered. Finally, the warning level is pushed to the clinical team.

Images