Intelligent multi-parameter early warning method and system for postoperative pancreatic fistula risk of pancreas

Abstract

This invention relates to the field of intelligent medical data processing technology, and provides a multi-parameter intelligent early warning method and system for the risk of pancreatic fistula after pancreatic surgery. The method performs dynamic trend analysis on the time series data of amylase in drainage fluid, extracts multi-channel features and identifies abnormal characteristics of drainage fluid appearance images, and quantitatively evaluates drainage volume patterns and systemic inflammatory response indicators. After fusing the multi-dimensional parameters, it outputs the probability of pancreatic fistula occurrence and risk warning level according to the ISGPS grading standard, and finally generates a dynamic assessment report and graded nursing recommendations. Furthermore, it adaptively adjusts the monitoring parameters through a closed-loop feedback mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical data processing technology, and in particular to a multi-parameter intelligent early warning method and system for the risk of pancreatic fistula after pancreatic surgery. Background Technology

[0002] Pancreatic fistula is one of the most common and threatening complications after pancreatic surgery, with an incidence reported in the literature ranging from 3% to 45%, especially after pancreaticoduodenectomy, where the incidence of clinically relevant pancreatic fistula can reach 12% to 30%. The International Study Group for Pancreatic Surgery (ISGPS) defines postoperative pancreatic fistula as a condition where the concentration of amylase in the abdominal drainage fluid is more than three times the upper limit of normal serum amylase on or after the third postoperative day, and classifies it into three grades according to clinical severity: biochemical fistula, grade B pancreatic fistula, and grade C pancreatic fistula. If pancreatic fistula is not properly managed, it can lead to serious complications such as abdominal infection, abdominal hemorrhage, and even sepsis, which is one of the leading causes of death in patients after pancreatic surgery. Therefore, how to accurately predict the occurrence of pancreatic fistula in the early postoperative period and intervene clinically in a timely manner is a key technical problem that urgently needs to be solved in the field of pancreatic surgery.

[0003] Existing pancreatic fistula risk assessment techniques mainly include two categories: intraoperative factor assessment and postoperative single-parameter monitoring. Regarding intraoperative factor assessment, the Fistula Risk Rating (FRS) is currently the most widely used predictive tool, primarily based on four parameters: pancreatic duct diameter, pancreatic texture, surgical pathology type, and intraoperative blood loss. However, the FRS score has significant limitations: firstly, the assessment of pancreatic texture relies heavily on the surgeon's subjective palpation judgment, lacking objective quantitative standards; secondly, the FRS score only reflects the static risk status at the moment of surgery and cannot track dynamic changes postoperatively. Regarding postoperative monitoring, the level of amylase in the drainage fluid on postoperative day 1 (DFA1) is currently commonly used clinically as an early predictive indicator. However, studies have shown that the trend of changes in drainage fluid amylase has higher predictive value than a single time point value, and relying solely on DFA1 may lead to the underestimation of some high-risk patients.

[0004] Chinese patent CN119830610A discloses a virtual surgical simulation method and system based on mixed reality technology. This scheme achieves surgical simulation training by spatially registering the real surgical environment and virtual scene, tracking instrument interaction, and evaluating operations. However, this scheme focuses on intraoperative simulation and does not address the risk monitoring and prediction of postoperative complications. Specifically, this scheme has the following shortcomings: First, its data acquisition focuses on the spatial posture and mechanical parameters during the surgical operation, rather than the postoperative patient's drainage fluid biochemical test data and physiological monitoring data, making it unsuitable for pancreatic fistula risk assessment scenarios; second, its evaluation model is based on the similarity comparison between the surgical operation trajectory and the standard operation sequence, rather than on the time-series dynamic analysis of clinical biochemical indicators, and therefore lacks the ability to predict the probability of postoperative complications; third, its output is an operation evaluation report of the surgical simulation process, rather than pancreatic fistula classification prediction and graded nursing recommendations based on the ISGPS standard, and therefore cannot directly guide postoperative clinical nursing decisions.

[0005] Furthermore, from the perspective of the completeness of monitoring dimensions, existing technologies generally suffer from the problem of relying on a single monitoring parameter. Currently, most pancreatic fistula prediction methods focus only on the single biochemical indicator of drainage fluid amylase, neglecting other parameters with equally important clinical significance, such as changes in the appearance of the drainage fluid, dynamic patterns of drainage volume, and systemic inflammatory response indicators. Clinical practice shows that a gradual change in drainage fluid color from normal bloody to serous is a sign of normal postoperative recovery, while abnormal turbidity, chylous changes, or meat-washing water characteristics often indicate complications such as infection, chylous leakage, or bleeding. A sudden increase or sustained high drainage volume is also an important warning sign of pancreatic fistula development. However, these multidimensional clinical observations have not been effectively integrated and quantified in existing automated prediction systems. At the same time, the prediction output of existing technologies typically only provides a binary classification of pancreatic fistula presence or absence, lacking the ability to perform graded predictions based on international standards and match differentiated nursing recommendations, making it difficult to meet the clinical needs of precise postoperative management.

[0006] In view of this, there is an urgent need for an intelligent early warning technology solution that can integrate multidimensional postoperative monitoring parameters and realize dynamic assessment and graded early warning of pancreatic fistula risk. Summary of the Invention

[0007] This invention provides a multi-parameter intelligent early warning method and system for pancreatic fistula risk after pancreatic surgery, which solves the technical problems in the prior art where pancreatic fistula risk assessment relies on a single parameter and static score, and lacks the ability to perform multi-dimensional dynamic monitoring and graded prediction after surgery.

[0008] In a first aspect, the present invention provides a multi-parameter intelligent early warning method for the risk of pancreatic fistula after pancreatic surgery. The method includes: detecting amylase concentration in drainage fluid samples collected at various monitoring time points after pancreatic surgery to obtain time-series amylase concentration data; performing dynamic trend analysis on the time-series amylase concentration data to generate dynamic trend characteristic data of amylase and amylase abnormal elevation marker information; performing color space conversion and multi-channel feature extraction on the appearance images of drainage fluid collected at various monitoring time points to identify abnormal characteristics appearing during color changes of the drainage fluid, generating drainage fluid characteristic data and abnormal characteristic marker information; and performing pattern analysis on the drainage volume data recorded at various monitoring time points after surgery to identify sudden increase patterns and sustained high-volume patterns of drainage volume, generating drainage volume patterns. Characteristic data, combined with body temperature monitoring data, white blood cell count data, and C-reactive protein detection data, are used to quantitatively assess the systemic inflammatory response and generate inflammatory response assessment data. Based on amylase dynamic trend characteristic data, drainage fluid characteristic data, drainage volume pattern characteristic data, and inflammatory response assessment data, a multi-dimensional parameter fusion prediction model is used to calculate the risk of pancreatic fistula. According to the ISGPS grading standard, the probability of biochemical fistula, the probability of grade B pancreatic fistula, and the probability of grade C pancreatic fistula are output, as well as the bleeding risk warning level and the pancreatic fistula risk warning level. Based on the probability of pancreatic fistula at each level and the bleeding risk warning level and pancreatic fistula risk warning level, a dynamic assessment report of pancreatic fistula risk and graded nursing recommendations are generated, and the sampling frequency and monitoring parameter thresholds at each monitoring time point are adjusted in reverse according to the current risk level.

[0009] Secondly, this invention provides a multi-parameter intelligent early warning system for pancreatic fistula risk after pancreatic surgery. The system includes a drainage fluid amylase dynamic analysis module, a drainage fluid characteristic image analysis module, a drainage volume pattern analysis and inflammation assessment module, a multi-dimensional fusion prediction module for pancreatic fistula risk, and a risk assessment report and nursing suggestion generation module. The drainage fluid amylase dynamic analysis module is used to perform dynamic trend analysis of amylase concentration data at various time points. The drainage fluid characteristic image analysis module is used to identify abnormal characteristics through color space conversion and multi-channel feature extraction. The drainage volume pattern analysis and inflammation assessment module is used to perform drainage volume pattern detection and comprehensive scoring of inflammatory response. The multi-dimensional fusion prediction module for pancreatic fistula risk is used to fuse four-dimensional features through a hierarchical evidence accumulation mechanism and output graded prediction results. The risk assessment report and nursing suggestion generation module is used to output assessment reports and nursing suggestions and perform closed-loop feedback adjustments. Each module corresponds to a step in the above method, forming a deeply coupled collaborative working closed loop through forward data flow and reverse parameter adjustment, realizing multi-dimensional parameter fusion-based dynamic assessment and graded early warning of pancreatic fistula risk.

[0010] The beneficial effects of this invention are as follows: By integrating monitoring parameters from four dimensions—dynamic trend of amylase in drainage fluid, image characteristics of drainage fluid properties, drainage volume change patterns, and systemic inflammatory response indicators—a multi-dimensional parameter fusion model for predicting pancreatic fistula risk is constructed, overcoming the limitations of existing technologies that rely on single parameters or static scoring; dynamic trend analysis replaces single-time-point judgment, improving the sensitivity of early identification of pancreatic fistula risk; through dual-channel color space analysis of drainage fluid images and a temporal consistency verification mechanism, objective quantitative assessment of changes in drainage fluid properties is achieved, compensating for the subjective defects of traditional manual visual observation; individualized drainage volume threshold calculation and multi-mode detection strategies enhance the applicability and accuracy of abnormal drainage volume identification; by outputting differentiated risk probabilities and graded nursing recommendations based on the ISGPS grading standard, precise and standardized management of postoperative complications is achieved; through a closed-loop feedback mechanism that adjusts monitoring frequency and trigger thresholds in reverse according to risk level, the system can automatically increase monitoring density in high-risk states to ensure sensitivity, and reduce sampling frequency in low-risk states to save medical resources, achieving a dynamic balance optimization between early warning effectiveness and resource consumption. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the architecture of the intelligent early warning system for the risk of pancreatic fistula after pancreatic surgery in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the scope of protection of the invention.

[0014] See Figure 1 The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery provided in this embodiment of the invention includes the following steps:

[0015] Step S1: Detect the amylase concentration in drainage fluid samples collected at various monitoring time points after pancreatic surgery to obtain time-series concentration data of amylase in the drainage fluid. Perform dynamic trend analysis on the time-series concentration data of amylase in the drainage fluid to generate dynamic trend characteristic data of amylase and marker information of abnormal elevation of amylase.

[0016] Step S2: Perform color space conversion and multi-channel feature extraction processing on the appearance images of the drainage fluid collected at each monitoring time point, identify abnormal characteristics that appear during the color change of the drainage fluid, and generate drainage fluid characteristic data and abnormal characteristic marker information.

[0017] Step S3: Perform pattern analysis on the drainage volume data recorded at each monitoring time point after surgery, identify the patterns of sudden increase in drainage volume and sustained high volume, generate drainage volume pattern characteristic data, and combine it with body temperature monitoring data, white blood cell count data and C-reactive protein detection data to perform quantitative assessment of systemic inflammatory response and generate inflammatory response assessment data.

[0018] Step S4: Based on the amylase dynamic trend characteristic data, the drainage fluid characteristic data, the drainage volume pattern characteristic data, and the inflammatory response assessment data, a multi-dimensional parameter fusion prediction model is used to calculate the risk of pancreatic fistula. According to the ISGPS grading standard, the probability of biochemical fistula, the probability of grade B pancreatic fistula, and the probability of grade C pancreatic fistula are output, as well as the bleeding risk warning level and the pancreatic fistula risk warning level.

[0019] Step S5: Based on the probability of occurrence of biochemical fistula, the probability of occurrence of grade B pancreatic fistula, the probability of occurrence of grade C pancreatic fistula, the bleeding risk warning level, and the pancreatic fistula risk warning level, generate a dynamic assessment report of pancreatic fistula risk and graded nursing recommendations, and adjust the sampling frequency and monitoring parameter thresholds of each monitoring time point in reverse according to the current risk level.

[0020] In one embodiment of the present invention, the execution process of step S1 involves three steps: acquisition, preprocessing, and dynamic trend analysis of drainage fluid amylase time-series data.

[0021] Regarding the data acquisition phase, the system, after pancreatic surgery with an indwelling abdominal drainage tube, collects drainage fluid samples and detects amylase concentration at preset sampling intervals. Preferably, the sampling time window covers postoperative day 1 (POD1) to postoperative day 7 (POD7), with an initial sampling interval set to 12 hours. Data recorded at each sampling time point includes the amylase concentration value (unit: U / L), sampling timestamp, and drainage tube number. When multiple drainage tubes are in place, the system records the amylase concentration data for each tube and marks the corresponding drainage tube location information for subsequent location correlation analysis. Through the above acquisition process, a time-series dataset of drainage fluid amylase concentration is constructed. ,in For the first Each sampling time point (unit: h, starting from the end of the surgery). This represents the corresponding amylase concentration (unit: U / L). For marking the location of the drainage tube, This represents the total number of samples taken.

[0022] Regarding the preprocessing stage, the system first performs outlier detection and data imputation on the original amylase concentration sequence. Specifically, data points exceeding the normal detection range (set to 0 to 100,000 U / L in this embodiment) are marked as outliers and replaced using linear interpolation of adjacent valid values. For missing data points due to clinical procedures, linear interpolation is also used to complete them. Preferably, when more than three consecutive data points are missing, the analysis results for that time period are marked as low confidence.

[0023] Regarding the dynamic trend analysis step, the system performs the following analysis steps on the preprocessed amylase concentration time series data. First, an exponentially weighted moving average (EWMA) is calculated to obtain a smoothed trend sequence. The EWMA calculation formula is: ,in, For the first EWMA smoothing values ​​at each time point (unit: U / L) For the first Actual amylase concentration values ​​at each time point (unit: U / L). This is the EWMA smoothing value at the previous time point (unit: U / L). This is a smoothing coefficient, ranging from 0.1 to 0.4, with a preferred value of 0.2. Smoothing coefficient The selection criteria are: smaller The value can effectively filter out short-term fluctuation noise while retaining medium- and long-term trend change information. In the clinical verification of this invention, The model achieved a sensitivity of 92.3% in recognizing the sustained increase in amylase levels from day 2 to day 4 post-surgery, while keeping the false alarm rate below 8.5%.

[0024] After obtaining the smoothed trend sequence, the system further calculates the amylase concentration change rate sequence between adjacent time points: ,in, For the first The rate of change in amylase concentration at each time point relative to the previous time point (in %). and The meaning is the same as the definition above. When two or more consecutive time points... All values ​​are positive and At that time, the system determines that amylase is showing an abnormally high trend and generates an abnormally high marker. In this embodiment of the invention, the rate of change threshold... The threshold was set at 15%, and this threshold was selected based on retrospective data validation involving 320 patients who underwent pancreatic surgery. At that time, the average early warning time for clinically relevant pancreatic fistula was 36.7 hours, and the Youden Index reached its maximum value of 0.71.

[0025] Furthermore, the system performs acceleration analysis on the rate of change sequence, calculating the second difference of the rate of change to identify the acceleration inflection point of amylase elevation: ,in, For the first The change in acceleration (in %) at each time point, when From negative to positive and two consecutive time points thereafter At that time, the system determines that an acceleration inflection point of amylase elevation has been detected. This inflection point information, as an important component of the amylase dynamic trend feature data, is passed to step S4 for fusion prediction. Through the above dynamic trend analysis, the output of step S1 includes: amylase EWMA smoothing trend sequence, rate of change sequence, acceleration sequence, abnormal elevation marker (including marker time and severity classification), and acceleration inflection point location information, collectively referred to as amylase dynamic trend feature data.

[0026] In one embodiment of the present invention, the execution process of step S2 includes four steps: acquisition of the appearance image of the drainage fluid, color space conversion, multi-channel feature extraction, and abnormality identification.

[0027] Regarding the image acquisition stage, the system uses a standardized image acquisition unit located next to the drainage fluid collection device to capture images of the appearance of the drainage fluid at various monitoring time points. Preferably, the image acquisition unit includes a standard D55 light source with a color temperature of 5500K and a digital camera with a resolution of no less than 1920×1080 pixels. The drainage fluid sample is placed 5cm in front of a standardized white background for imaging. The original image acquired each time has a resolution of 1920×1080 pixels and a color depth of 24-bit RGB. To eliminate lighting differences between different collection batches, the system simultaneously captures a calibration image containing a standard color chart during each batch of acquisitions and performs white balance correction on the original image based on the color chart reference values.

[0028] Regarding the color space conversion, the system converts the corrected RGB image to both the HSV and Lab color spaces. The RGB to HSV conversion follows a standard color space transformation algorithm, decomposing color information into three independent channels: hue (H), saturation (S), and lightness (V). The RGB to Lab conversion is performed using the XYZ intermediate color space under the D65 standard light source, decomposing color information into three channels: lightness (L), red-green axis (a), and yellow-blue axis (b). The technical advantage of using a dual color space conversion strategy is that the hue channel of the HSV space can intuitively reflect the color evolution of the drainage fluid from bloody (red) to serous (yellow), while the a and b channels of the Lab space can more accurately quantify the color shift caused by turbidity and chylous changes in the drainage fluid.

[0029] Regarding the multi-channel feature extraction stage, the system extracts statistical and texture features for each channel in both the HSV and Lab color spaces. For statistical features, four statistical measures are extracted for each channel: mean, standard deviation, skewness, and kurtosis. In this embodiment, the region of interest (ROI) of the drainage fluid image is divided into a central region and an edge region, with the central region accounting for 60% of the total ROI area. The purpose of this partitioned extraction is that sediment and suspended particles in the drainage fluid tend to concentrate at the bottom and edge regions, and partitioned feature extraction can more accurately capture these local anomalies. For texture features, the system calculates four texture parameters—contrast, correlation, energy, and homogeneity—based on the gray-level co-occurrence matrix (GLCM) to quantify the graininess and non-uniformity in the drainage fluid image. Preferably, the GLCM calculation directions include four directions: 0 degrees, 45 degrees, 90 degrees, and 135 degrees, with a step size of 1 pixel and 64 gray-level quantization levels.

[0030] Regarding the abnormality identification process, the system constructs an abnormality classifier based on the extracted multi-channel features. In this embodiment of the invention, the abnormal characteristics of the drainage fluid are divided into three categories: abnormal turbidity, chylous changes, and meat-washing water-like characteristics. The criterion for determining abnormal turbidity is that the standard deviation of the saturation channels in the HSV space is greater than a preset turbidity threshold. (In this embodiment) Furthermore, the GLCM contrast value was higher than the upper limit of the normal range. The selection of the turbidity threshold was based on the following: analysis of 156 drainage fluid samples showed that the mean standard deviation of the saturation channel for normal serous drainage fluid was 0.065, while the mean standard deviation for drainage fluid containing infectious turbidity rose to 0.185. The optimal boundary point corresponding to the two distributions. The criterion for determining chylous alteration characteristics is that the mean value of the b channel in the Lab space is greater than the preset chylous threshold. (In this embodiment) Meanwhile, the mean value of the L channel is higher than that of normal serous drainage fluid. This threshold setting reflects the yellowish tint and high brightness characteristics caused by the scattering of fat particles in the chyle. The criteria for determining the characteristics of the meat washing water sample are that the mean value of the hue channel in the HSV space is in the red phase range (H value in the range of 0 to 30 degrees or 330 to 360 degrees), and the mean value of the saturation channel is higher than the preset meat washing water threshold. (In this embodiment) The clinical significance of this feature is that the watery drainage from the meat washings suggests the possible presence of hemorrhagic complications or digestive erosion of peripheral blood vessels by pancreatic juice.

[0031] Furthermore, to improve the accuracy of abnormal morphology identification, the system introduces a temporal consistency verification mechanism. Specifically, when a drainage fluid image at a certain time point is determined to have an abnormal morphology, the system retrospectively analyzes the image results from the previous two time points for joint judgment: if similar abnormal features have appeared at previous time points or the abnormal feature scores show an increasing trend, the abnormality marker is confirmed and the confidence level is marked as high; if no abnormal features have appeared at previous time points, the confidence level of the current abnormality marker is marked as pending observation and verified at the next time point. The introduction of this mechanism can effectively reduce the misjudgment rate caused by occasional factors in a single sampling (such as interference from residues after flushing the drainage tube). In the clinical verification of this invention embodiment, after introducing temporal consistency verification, the specificity of abnormal morphology identification increased from 87.6% to 94.2%, while the sensitivity remained above 91.5%.

[0032] In summary, the output of step S2 includes: multi-channel color feature vectors, texture feature vectors, abnormal morphology type labels (normal, abnormal turbidity, chylous change or meat washing water sample) at each time point, and corresponding confidence level labels, collectively referred to as drainage fluid morphology feature data and morphology abnormality label information.

[0033] In one embodiment of the present invention, the execution process of step S3 includes two parallel processing sub-steps: drainage pattern analysis and quantitative assessment of systemic inflammatory response.

[0034] Regarding the drainage volume pattern analysis sub-step, the system performs pattern recognition processing on the drainage volume data recorded at each postoperative monitoring time point. The drainage volume data is recorded in cumulative drainage volume (unit: mL) within each monitoring time interval. The system first processes the original drainage volume sequence... Baseline estimation was performed, and baseline drainage volume was calculated using the median drainage volume at the first three postoperative time points. Based on this, the system performs the following two detection modes.

[0035] For detecting sudden increases in traffic, the system calculates the magnitude of change in traffic at adjacent time points: ,in, For the first Changes in drainage volume at each time point (unit: mL). and The first The and the first The drainage volume (unit: mL) at each time point. When the change at two consecutive time points satisfies At that time, the system determined it to be a sudden surge in traffic. The threshold is the average drainage volume (in mL) at each time point within the previous complete monitoring cycle (24 h). This threshold is set based on the fact that clinical literature shows that when the drainage volume increases by more than 50% of the previous average in a short period of time, it suggests that there may be increased pancreatic juice leakage or ascites formation, which is statistically significantly correlated with the occurrence of pancreatic fistula.

[0036] For detecting persistently high traffic patterns, the system compares the traffic generated at each time point with a preset traffic threshold. Comparison. When the traffic at three or more consecutive time points meets the requirements... At this time, the system determines it to be a sustained high-level mode. In this embodiment of the invention, The determination method is based on individualized calculations according to the type of surgery and the patient's body surface area: ,in, The basic drainage volume threshold (unit: mL) is set at 200 mL / 12 h for pancreaticoduodenectomy and 150 mL / 12 h for distal pancreatectomy. The value is a correction factor for the type of surgery (dimensionless), with a value of 1.0 for pancreaticoduodenectomy and 0.85 for distal pancreatectomy. Patient's body surface area (unit: ), 1.73 This is a reference value for the standard adult body surface area. The criteria for this designation are based on the range of normal postoperative drainage volume reported in clinical literature. The differences reflect the baseline differences in postoperative drainage volume for different surgical types. Through this individualized calculation method, the preset drainage volume threshold can be adapted to patients with different surgical types and body types, avoiding the bias caused by a fixed threshold. In a validation set containing 480 patients, this individualized threshold reduced the false negative rate by 12.3% compared to the fixed threshold.

[0037] Regarding the sub-step of quantitative assessment of systemic inflammatory response, the system integrates body temperature monitoring data, white blood cell count data, and C-reactive protein (CRP) detection data to calculate a comprehensive inflammatory response score. In this embodiment of the invention, the comprehensive inflammatory response score is calculated using a normalized weighted summation method: ,in, The comprehensive score for inflammatory response (dimensionless, ranging from 0 to 100). , and The weighting coefficients for body temperature, white blood cell count, and CRP are respectively, preferably... , , The weighting is based on the fact that CRP has the highest sensitivity to postoperative infection and inflammatory response, followed by white blood cell count, while body temperature has relatively low specificity. , and These are normalized scoring functions for body temperature, white blood cell count, and CRP, respectively, with each function employing a piecewise linear mapping.

[0038] Body temperature scoring function The segmentation is defined as: when body temperature hour ;when hour ;when hour ;when hour The upper limit is truncated to 100. The threshold values ​​for each segment are set with reference to the clinical diagnostic criteria for postoperative infectious complications, with 37.5°C as the starting point for postoperative low-grade fever, 38.0°C as the threshold for moderate fever, and 39.0°C as the alarm threshold for high fever.

[0039] White blood cell count scoring function The segmentation is defined as: when hour ;when hour ;when hour ;when hour The upper limit is cut off at 100. A white blood cell count of 10 × 10⁹ / L is the upper limit of normal reference value, and 15 × 10⁹ / L is the critical value for significantly elevated white blood cell count, indicating infection.

[0040] CRP scoring function The segmentation is defined as: when hour ;when hour ;when hour ;when hour The upper limit is truncated to 100. The segmented thresholds of CRP were set with reference to the results of a study on the diagnostic cutoff values ​​of CRP for clinically relevant pancreatic fistulas from the 3rd to the 5th postoperative day. Among them, 50 mg / L is the upper limit of normal postoperative stress response, 100 mg / L is the warning value for infectious complications, and 200 mg / L is the high-risk threshold for severe infection.

[0041] In addition, the system analyzes the temporal trends of inflammatory markers. Specifically, it calculates the trend of CRP over three consecutive time points: if the CRP value continues to rise and the increase exceeds 20% of the value at the previous time point, it is marked as inflammation progression; if the CRP value continues to decline or the fluctuation range is within 10%, it is marked as inflammation stabilization or resolution. This trend label is passed to step S4 as part of the inflammatory response assessment data.

[0042] In summary, the output of step S3 includes drainage pattern characteristic data (including sudden increase pattern markers, sustained high level pattern markers, and drainage statistics at each time point) and inflammatory response assessment data (including comprehensive inflammatory response score and inflammatory trend markers).

[0043] In one embodiment of the present invention, the execution process of step S4 is based on a hierarchical evidence accumulation mechanism, which is divided into three processing layers from bottom to top: a single-parameter risk scoring layer, a cross-parameter correlation analysis layer, and an ISGPS hierarchical output layer.

[0044] In the single-parameter risk scoring layer, the system independently scores the four-dimensional feature data output from steps S1 to S3. (Amylase dimension risk scoring) The calculation comprehensively considers both the absolute concentration level and dynamic trend characteristics of amylase: ,in, Risk score for amylase dimension (dimensionless, value range 0 to 100). The value is the most recent amylase concentration (unit: U / L). This represents the maximum rate of change within the recent monitoring window (unit: %). To accelerate the cumulative number of times the inflection point occurs (dimensionless). , , These are the weighting coefficients for each sub-feature. The weighting allocation reflects the dominant position of absolute amylase concentration as a core indicator for ISGPS diagnosis, while also taking into account the incremental contribution of dynamic trends and acceleration features to early warning. , and These are the corresponding normalized mapping functions. Using the ISGPS-defined amylase threshold (3 times the upper limit of normal serum amylase, typically corresponding to approximately 300 U / L) as a key reference point: when When the output value is less than 30; The output value increases linearly between 30 and 75; when The output value was between 75 and 100. The upper threshold of 5000 U / L is derived from clinical evidence in research reports showing a significantly increased risk of clinically relevant pancreatic fistula when DFA1 exceeds 5000 U / L. The rate of change threshold set in step S1 Map the reference point. A tiered mapping is applied based on the cumulative number of acceleration inflection points: 0 points for 0 times, 40 points for 1 time, and 80 points or more for 2 times or more.

[0045] Risk score for drainage fluid properties The calculation is based on the type of anomalous trait and the confidence level:

[0046] ,

[0047] in, Risk score for drainage fluid properties (dimensionless, value range 0 to 100). Index for anomalous trait types ( It is abnormally turbid. The changes were chylous. (The sample is from the water used to wash the meat). For the first Risk weights of atypical traits ( , , The meat washing water-like characteristics had the highest weight because they were most strongly correlated with hemorrhagic pancreatic fistula. For the first The probability of detecting an abnormal trait within the current time window (range 0 to 1). The corresponding confidence coefficient is 1.0 for high confidence and 0.5 for unobserved.

[0048] Traffic generation risk score Calculations are based on drainage pattern characteristics: a sudden increase pattern (marked as positive) contributes 40 base points, a sustained high level pattern (marked as positive) contributes 50 base points, and when both are positive, 90 points are awarded, plus additional points for the portion of the absolute drainage volume exceeding a threshold. Inflammation dimension risk score. The comprehensive inflammatory response score output from step S3 is used directly. As a base value.

[0049] In the cross-parameter correlation analysis layer, the system performs adaptive weighted fusion of the four-dimensional single-parameter risk scores: ,in, For the integration of risk scores (dimensionless, ranging from 0 to 100). For the first Single-parameter risk scoring in each dimension For the first Each dimension in postoperative time (Unit: h) Adaptive weights. The dynamic adjustment of adaptive weights follows these principles: in the early postoperative period (POD1 to POD2). (Amylase) (Characteristics) (Traffic generation) (Inflammation); in the mid-postoperative period (POD3 to POD5). , , , In the late postoperative period (POD6 and later). , , , The basis for the dynamic adjustment of weights is that the predictive value of amylase concentration is highest in the early postoperative period, while the indicative value of drainage fluid characteristics and systemic inflammatory response for clinically relevant pancreatic fistulas gradually increases over time after surgery.

[0050] Furthermore, the cross-parameter correlation analysis layer also introduces a synergistic indicator effect detection mechanism. When the risk scores of two or more dimensions simultaneously exceed the synergistic trigger threshold (set to 60 points in this embodiment), the system adds a synergistic bonus to the fused score. ,in, For the set of dimensions that exceed the collaborative triggering threshold, This is a synergistic additive factor, ranging from 10 to 30, with a preferred value of 20. This represents a product operation. Its technical effect is that when multiple dimensions simultaneously present a high-risk state, the joint indication effect is significantly stronger than the sum of the independent indication effects of a single dimension. This synergistic additive term quantifies the nonlinear gain of the multi-parameter joint early warning. In the clinical verification of this invention's embodiments, the prediction specificity for grade B and C pancreatic fistulas improved by 8.3 percentage points after introducing the synergistic effect.

[0051] In the ISGPS hierarchical output layer, the system will integrate the total risk score. The mapping is to the probability of pancreatic fistula occurrence at each level, and the mapping function uses a piecewise Sigmoid transform: ,in, The probability of pancreatic fistula occurring at a specific grade (range 0 to 1). The sigmoid steepness parameter (dimensionless) for this classification controls the steepness of the probability transition from low to high. This represents the center threshold (dimensionless) for this classification, corresponding to a risk score level with a probability of 0.5. The biochemical fistula parameters are... , Grade B pancreatic fistula parameters are: , The parameters for a grade C pancreatic fistula are: , The above parameters were obtained through maximum likelihood estimation of historical data from 520 patients who underwent pancreatic surgery, and the relationships between each level were satisfied. The monotonically decreasing relationship is consistent with the clinical pattern in ISGPS where the severity of each grade increases while the incidence decreases.

[0052] Regarding the determination of bleeding risk warning levels, the system comprehensively considers the detection of characteristics in meat washing water samples and the pattern of sudden increase in drainage volume: High risk is defined as a high-confidence detection of characteristics in meat washing water samples and a positive pattern of sudden increase in drainage volume; medium risk is defined as only one of these positive indicators; and low risk is defined as both negative indicators. Pancreatic fistula risk warning levels are based on a comprehensive assessment of absolute amylase concentration and a persistently high pattern: High risk is defined as a recent amylase concentration exceeding 5000 U / L and a positive pattern of persistently high drainage volume; medium risk is defined as an amylase concentration between 300 and 5000 U / L or only a positive pattern of persistently high drainage volume; and low risk is defined as neither of these conditions is met.

[0053] In one embodiment of the present invention, the execution process of step S5 includes three stages: risk assessment report generation, graded nursing recommendation generation, and closed-loop feedback parameter adjustment.

[0054] Regarding the risk assessment report generation stage, the system integrates the probability of pancreatic fistula occurrence at each level, the bleeding risk warning level, and the pancreatic fistula risk warning level output in step S4 into a structured dynamic pancreatic fistula risk assessment report. Preferably, the report includes the following sections: a patient basic information summary section, recording the patient's surgical type, surgical date, postoperative days, and drainage tube configuration information; a risk probability trend section, displaying the historical change curves and current latest values ​​of the probability of pancreatic fistula occurrence at each level with the time axis as the horizontal axis; a key abnormal event section, listing all events that triggered abnormal markers during the monitoring period, along with their occurrence time and severity; and a comprehensive risk level section, classifying the current comprehensive risk level as low risk (…). ), medium risk ( High risk ) and extremely high risk ( The system is divided into four levels and presented intuitively using color coding.

[0055] Regarding the generation of graded nursing recommendations, the system matches corresponding nursing plan templates based on the comprehensive risk level. Low-risk levels correspond to routine monitoring and nursing plans, including maintaining the patency of the current drainage tube, standardized enteral nutrition support, and regular laboratory testing. Medium-risk levels correspond to enhanced monitoring and nursing plans, including shortening the drainage fluid sampling interval to 6 hours, increasing the frequency of observing drainage fluid characteristics, initiating prophylactic anti-infection measures, and enhancing enteral nutrition support. High-risk and very high-risk levels correspond to emergency intervention nursing plans, including immediately notifying the attending physician and surgical team, preparing for an abdominal CT scan to assess the abdominal drainage status, initiating an escalation of empirical anti-infection treatment, and assessing whether percutaneous drainage or surgical re-exploration is necessary.

[0056] Regarding the closed-loop feedback parameter adjustment step, the system adjusts the monitoring parameters in steps S1 to S3 in reverse order based on the current comprehensive risk level. This closed-loop adjustment mechanism is the core component of this invention for achieving adaptive monitoring, and its adjustment rules are as follows: When the risk level rises from low risk to medium risk, the sampling interval for amylase in the drainage fluid in step S1 is shortened from 12 hours to 6 hours, the acquisition frequency of drainage fluid images in step S2 is doubled accordingly, and the monitoring frequency of body temperature and inflammation indicators in step S3 is shortened from every 12 hours to every 6 hours; when the risk level rises to high risk or very high risk, the amylase sampling interval in step S1 is further shortened to 3 hours, and the amylase change rate threshold in step S1 is adjusted accordingly. The threshold for drainage volume in step S3 was reduced from 15% to 10% to improve sensitivity. The risk of underreporting is reduced by 15%. When the overall risk level remains at a low level for three consecutive monitoring cycles (12 hours each), all monitoring parameters are restored to their initial default values. Through this closed-loop mechanism, the system provides denser monitoring coverage under high-risk conditions to ensure early warning sensitivity, and avoids unnecessary high-frequency sampling under low-risk conditions, achieving a dynamic balance between early warning effectiveness and medical resource consumption.

[0057] In the clinical validation of this invention, retrospective data from 520 patients after pancreatic surgery were used for efficacy evaluation. The incidence of pancreatic fistula in the dataset was 26.5% (14.2% biochemical fistula, 9.0% grade B pancreatic fistula, and 3.3% grade C pancreatic fistula). Compared to the traditional method that only uses the single parameter of amylase in the drainage fluid on the first postoperative day, the multi-parameter intelligent early warning method of this invention improved the AUC value of clinically relevant pancreatic fistula (grade B + grade C) prediction from 0.72 to 0.89, the sensitivity from 78.1% to 93.4%, the specificity from 66.3% to 82.7%, and the average early warning time from 12.4 hours to 38.2 hours. Furthermore, compared to the existing model that only uses the two parameters of amylase in the drainage fluid on the third postoperative day combined with CRP, the AUC value of this invention improved from 0.81 to 0.89, the sensitivity from 85.6% to 93.4%, and the specificity from 72.1% to 82.7%. The above results indicate that the introduction of the drainage fluid characteristics image analysis dimension and the drainage volume pattern analysis dimension provides additional independent information contributions to the prediction model. The synergistic effect of multidimensional parameter fusion and dynamic trend analysis significantly improves the accuracy and timeliness of pancreatic fistula risk prediction.

[0058] To further illustrate the application of the method of this invention in a real clinical scenario, a typical case of a patient after pancreaticoduodenectomy is used as an example. A patient had two abdominal drainage tubes (labeled L1 and L2) placed postoperatively. Monitoring was initiated at an initial sampling interval of 12 hours after system startup. On postoperative day 1 (POD1), the amylase concentration in the drainage fluid from tube L1 was 2800 U / L, and from tube L2 it was 1500 U / L. The drainage fluid appeared bloody, with drainage volumes of 180 mL / 12 h and 120 mL / 12 h, respectively. The patient's temperature was 37.2°C, WBC was 12.3 × 10^9 / L, and CRP was 68 mg / L. Step S1 involved EWMA smoothing of the amylase time-series data to calculate the initial rate of change as baseline data. Step S2, after analyzing the drainage fluid image, determined it to be of normal bloody appearance. Step S3, the drainage volume pattern analysis did not detect any abnormal patterns, and the overall inflammatory response score was 28.4 points. The fusion prediction results in step S4 showed a biochemical fistula probability of 0.62, a grade B pancreatic fistula probability of 0.18, and a grade C pancreatic fistula probability of 0.05, with an overall risk level of medium risk. Step S5 generated enhanced monitoring and nursing recommendations based on the medium risk level and adjusted the sampling interval to 6 hours through closed-loop feedback.

[0059] On postoperative day 3 (POD3), the amylase concentration in tube L1 increased to 4200 U / L, and the rate of change sequence showed a change rate exceeding 15% for two consecutive time points. Acceleration analysis detected one acceleration inflection point. Simultaneously, step S2 detected abnormal turbidity in the drainage fluid image from tube L1 (HSV saturation channel standard deviation was 0.156, exceeding the threshold of 0.12), and the temporal consistency verification confirmed a high confidence level. Step S3 showed that the drainage volume from tube L1 increased to 260 mL / 6 h, triggering a surge pattern marker, with body temperature rising to 38.3°C, CRP rising to 145 mg / L, and the overall inflammatory response score rising to 62.7 points. Step S4 yielded four-dimensional scores: amylase (72 points), trait (45 points), drainage volume (55 points), and inflammation (62.7 points). Because both amylase and inflammation scores exceeded 60 points, a synergistic effect was triggered, resulting in a final fusion risk score of 78.5 points. The probability of grade B pancreatic fistula increased to 0.71, and the probability of grade C pancreatic fistula increased to 0.28, raising the overall risk level to high risk. Step S5 immediately generated emergency intervention nursing recommendations and notified the clinical team. Simultaneously, the sampling interval was further shortened to 3 hours, and the amylase change rate threshold was lowered to 10%. Based on this, the clinical team promptly arranged for an abdominal CT scan and upgraded anti-infection treatment. Ultimately, the patient was diagnosed with grade B pancreatic fistula and recovered after conservative treatment. This case demonstrates that the method of this invention can achieve early warning through multi-parameter synergistic analysis before the clinical manifestations of pancreatic fistula are fully apparent, providing a valuable time window for clinical decision-making.

[0060] It is worth noting that the coupling relationship between the steps in the method of this invention is not only reflected in the forward data flow, but also in the reverse influence of the output result of step S5 on the monitoring parameters of steps S1 to S3. Specifically, when the fusion prediction result of step S4 shows an increased risk level, the closed-loop feedback mechanism of step S5 will synchronously adjust the sampling interval and rate of change threshold of step S1, the image acquisition frequency of step S2, and the drainage threshold and inflammation monitoring frequency of step S3. This reverse adjustment enables steps S1 to S3 to collect and analyze data with higher temporal resolution and lower trigger thresholds under high-risk conditions, thereby improving the data density and sensitivity of the subsequent fusion prediction in step S4. In other words, a positive feedback coupling loop is formed between the steps: the higher the risk, the more intensive the monitoring; the more intensive the monitoring, the more accurate the prediction; the more accurate the prediction, the more timely the nursing intervention; and the more timely the nursing intervention, the better the patient's prognosis. This deeply coupled closed-loop collaborative architecture is one of the core innovations of this invention, which distinguishes it from the existing technology where each monitoring link operates independently and lacks a linkage mechanism.

[0061] See Figure 2 The present invention also provides a multi-parameter intelligent early warning system for pancreatic fistula risk after pancreatic surgery. The system includes a drainage fluid amylase dynamic analysis module 1, a drainage fluid characteristic image analysis module 2, a drainage volume pattern analysis and inflammation assessment module 3, a pancreatic fistula risk multi-dimensional fusion prediction module 4, and a risk assessment report and nursing suggestion generation module 5.

[0062] The drainage fluid amylase dynamic analysis module 1 is used to execute all the processing procedures described in step S1 of the above method embodiments. In its specific implementation, this module includes an amylase data acquisition subunit, a data preprocessing subunit, and a dynamic trend analysis subunit. The amylase data acquisition subunit is communicatively connected to the drainage fluid biochemical detection equipment and is responsible for receiving the amylase concentration detection results at each time point according to the current sampling interval and recording the corresponding timestamp and drainage tube number. The data preprocessing subunit performs outlier detection and missing value imputation on the raw data. The dynamic trend analysis subunit performs EWMA smoothing calculation, rate of change sequence calculation, and accelerated inflection point detection, ultimately outputting amylase dynamic trend characteristic data and abnormal increase marker information.

[0063] The drainage fluid characteristic image analysis module 2 is used to execute all the processing procedures described in step S2 of the above method embodiments. In its specific implementation, this module includes an image acquisition subunit, a color space conversion subunit, a multi-channel feature extraction subunit, and an abnormality identification subunit. The image acquisition subunit includes a standardized D55 light source and digital camera hardware, responsible for acquiring images of the drainage fluid appearance at each monitoring time point. The color space conversion subunit performs dual-channel conversion from RGB to HSV and from RGB to Lab. The multi-channel feature extraction subunit extracts zonal statistical features and GLCM texture features for each channel. The abnormality identification subunit classifies and determines abnormal turbidity, chylous changes, and meat-washing water characteristics based on multi-channel features, and confirms the confidence level of the abnormality markers through a temporal consistency verification mechanism.

[0064] The drainage flow pattern analysis and inflammation assessment module 3 is used to execute all the processing procedures described in step S3 of the above method embodiments. This module includes a drainage flow data receiving subunit, a drainage flow pattern detection subunit, and an inflammation response scoring subunit. The drainage flow data receiving subunit is communicatively connected to the drainage flow metering device to acquire cumulative drainage flow data at each time point. The drainage flow pattern detection subunit performs detection algorithms for sudden increase patterns and sustained high patterns on the drainage flow time-series data. The inflammation response scoring subunit integrates body temperature, white blood cell count, and CRP data, calculates a comprehensive inflammation response score using a piecewise linear normalization function and a weighted summation method, and analyzes the time-series change trend of CRP.

[0065] The multidimensional fusion prediction module 4 for pancreatic fistula risk is used to execute all the processing procedures described in step S4 of the above method embodiments. This module includes a single-parameter risk scoring subunit, a cross-parameter correlation analysis subunit, and an ISGPS graded output subunit. The single-parameter risk scoring subunit calculates independent risk scores for each of the four-dimensional feature data. The cross-parameter correlation analysis subunit performs adaptive weighted fusion and synergistic indicator effect detection. The ISGPS graded output subunit maps the fused risk score to the probability of pancreatic fistula occurrence at each level through piecewise sigmoid transformation, and comprehensively determines the bleeding risk warning level and the pancreatic fistula risk warning level.

[0066] The risk assessment report and nursing suggestion generation module 5 is used to execute all the processing procedures described in step S5 of the above method embodiments. This module includes a report generation subunit, a nursing suggestion matching subunit, and a closed-loop feedback adjustment subunit. The report generation subunit integrates various prediction results into a structured dynamic assessment report containing a risk probability trend curve, a list of key abnormal events, and a comprehensive risk level. The nursing suggestion matching subunit matches the corresponding level of graded nursing suggestions from a preset nursing plan library based on the comprehensive risk level. The closed-loop feedback adjustment subunit sends parameter adjustment instructions to the drainage fluid amylase dynamic analysis module 1, the drainage fluid characteristic image analysis module 2, and the drainage volume pattern analysis and inflammation assessment module 3 according to the current risk level, realizing the dynamic adjustment of sampling frequency and monitoring threshold.

[0067] The data flow relationships between the five modules described above demonstrate a deeply coupled collaborative working mechanism. The drainage fluid amylase dynamic analysis module 1, drainage fluid characteristic image analysis module 2, and drainage volume pattern analysis and inflammation assessment module 3 perform multidimensional data acquisition and analysis in parallel. The output feature data from each module converges to the pancreatic fistula risk multidimensional fusion prediction module 4 for fusion prediction. The fusion prediction results are then transmitted to the risk assessment report and nursing suggestion generation module 5 to generate the final assessment report and nursing suggestions. Simultaneously, the risk assessment report and nursing suggestion generation module 5 adjusts the working parameters of the first three modules through a closed-loop feedback mechanism, forming a complete adaptive closed loop of "acquisition and analysis—fusion prediction—report and suggestion—feedback adjustment." This closed-loop collaborative mechanism enables the system to dynamically optimize monitoring strategies based on the patient's real-time risk status, rationally allocating medical monitoring resources while ensuring early warning sensitivity.

[0068] Preferably, the above modules communicate with each other through standardized data interfaces. The data transmitted through each interface adopts a unified timestamp alignment mechanism to ensure the consistency of data from different monitoring devices and analysis modules in the time dimension. Specifically, the data transmitted from the drainage fluid amylase dynamic analysis module 1 to the pancreatic fistula risk multidimensional fusion prediction module 4 includes amylase EWMA smoothing value, rate of change, acceleration value, and anomaly marker quadruple; the data transmitted from the drainage fluid morphology image analysis module 2 includes multi-channel feature vector, anomaly type label, and confidence triplet; the data transmitted from the drainage volume pattern analysis and inflammation assessment module 3 includes drainage volume pattern marker and inflammation comprehensive score binary. The data transmitted from the pancreatic fistula risk multidimensional fusion prediction module 4 to the risk assessment report and nursing suggestion generation module 5 includes pancreatic fistula occurrence probability vectors of each level, bleeding and pancreatic fistula risk grades, and fusion risk total score. The parameter adjustment instructions transmitted from the risk assessment report and nursing suggestion generation module 5 to the aforementioned three modules through the feedback channel include target sampling interval value, monitoring threshold adjustment ratio, and sensitivity pattern marker. In addition, the system is also equipped with a persistent data storage module to record all intermediate data and final results generated by each module throughout the monitoring period, so as to support postoperative retrospective analysis and continuous optimization and iteration of the predictive model.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-parameter intelligent warning method for post-pancreatic surgery pancreatic fistula risk, characterized in that, The method includes: Amylase concentration was detected in drainage fluid samples collected from patients after pancreatic surgery at various monitoring time points to obtain time-series concentration data of amylase in the drainage fluid. Dynamic trend analysis was performed on the time-series concentration data of amylase in the drainage fluid to generate dynamic trend characteristic data of amylase and information on abnormal elevation of amylase. Color space conversion and multi-channel feature extraction were performed on the appearance images of drainage fluid collected at each monitoring time point to identify abnormal characteristics that appeared during the color change of drainage fluid, and to generate drainage fluid characteristic data and abnormal characteristic marker information. Pattern analysis was performed on the drainage volume data recorded at each monitoring time point after surgery to identify the patterns of sudden increase in drainage volume and sustained high drainage volume, generating drainage volume pattern characteristic data. Combined with body temperature monitoring data, white blood cell count data and C-reactive protein detection data, the systemic inflammatory response was quantitatively assessed, generating inflammatory response assessment data. Based on the amylase dynamic trend characteristic data, the drainage fluid characteristic data, the drainage volume pattern characteristic data, and the inflammatory response assessment data, a multi-dimensional parameter fusion prediction model is used to calculate the risk of pancreatic fistula. According to the ISGPS grading standard, the probability of biochemical fistula, the probability of grade B pancreatic fistula, and the probability of grade C pancreatic fistula are output, as well as the bleeding risk warning level and the pancreatic fistula risk warning level. Based on the probability of occurrence of biochemical fistula, the probability of occurrence of grade B pancreatic fistula, the probability of occurrence of grade C pancreatic fistula, the bleeding risk warning level, and the pancreatic fistula risk warning level, a dynamic assessment report of pancreatic fistula risk and graded nursing recommendations are generated, and the sampling frequency and monitoring parameter thresholds at each monitoring time point are adjusted in reverse according to the current risk level.

2. The multi-parameter intelligent warning method for post-pancreatic surgery pancreatic fistula risk of claim 1, characterized in that, The time-series concentration data of amylase in the drainage fluid includes a sequence of amylase concentration values ​​collected at preset sampling intervals from the 1st to the 7th day after surgery. The preset sampling interval is 12 hours in low-risk conditions, 6 hours in medium-risk conditions, and 3 hours in high-risk conditions.

3. The multi-parameter intelligent warning method for post-pancreatic surgery pancreatic fistula risk of claim 1, characterized in that, The abnormal characteristics include abnormal turbidity, chylous changes, and meat washing water characteristics. The color space conversion converts the RGB color space to the HSV and Lab color spaces. The multi-channel feature extraction includes hue channel feature extraction, saturation channel feature extraction, and lightness channel feature extraction.

4. The multi-parameter intelligent warning method for post-pancreatic surgery pancreatic fistula risk according to claim 1, characterized in that, The surge in drainage volume is defined as a drainage volume increase greater than 50% of the average drainage volume in the previous monitoring period for two consecutive monitoring time points. The sustained high level is defined as a drainage volume greater than a preset drainage volume threshold for three consecutive monitoring time points. The preset drainage volume threshold is determined based on the type of surgery and the patient's body surface area.

5. The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery according to claim 1, characterized in that, The dynamic trend analysis process includes: performing an exponentially weighted moving average calculation on the time-series amylase concentration data of the drainage fluid to obtain a smooth trend sequence of amylase concentration; calculating the amylase concentration change rate sequence between adjacent time points; and determining the trend direction and acceleration based on the amylase concentration change rate sequence to identify the inflection point of abnormal increase in amylase.

6. The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery according to claim 1, characterized in that, The multidimensional parameter fusion prediction model employs a hierarchical evidence accumulation mechanism, comprising: a first layer of single-parameter risk scoring, which independently scores the dynamic trend characteristics of amylase, the characteristics of drainage fluid, the characteristics of drainage volume patterns, and the inflammatory response assessment data; a second layer of cross-parameter correlation analysis, which weights and fuses the single-parameter risk scores of the first layer and analyzes the synergistic indicative effect between parameters; and a third layer of ISGPS grading output, which outputs the probability of pancreatic fistula occurrence at each level based on the cross-parameter correlation analysis results.

7. The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery according to claim 6, characterized in that, The weights of each parameter in the weighted fusion are determined by an adaptive weight adjustment mechanism. The adaptive weight adjustment mechanism dynamically adjusts the weights of each parameter according to the postoperative time. In the early postoperative stage, the weight of the dynamic trend characteristic of amylase is higher than the weight of the characteristic of drainage fluid. In the mid-postoperative stage, the weights of the characteristic of drainage fluid and the weights of the inflammatory response assessment data gradually increase.

8. The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery according to claim 1, characterized in that, The tiered nursing recommendations include routine monitoring and nursing plans for low-risk levels, enhanced monitoring and nursing plans for medium-risk levels, and emergency intervention and nursing plans for high-risk levels. Each level of nursing plan includes recommendations for drainage tube management, nutritional support, and anti-infection treatment.

9. The intelligent early warning method for multi-parameter pancreatic fistula risk after pancreatic surgery according to claim 1, characterized in that, The reverse adjustment includes: when the probability of occurrence of grade B pancreatic fistula or grade C pancreatic fistula exceeds a preset risk threshold, increasing the sampling frequency to a high-risk sampling frequency and reducing the trigger threshold for the abnormal elevation of amylase markers; when the probability of occurrence of pancreatic fistula at each level is lower than a preset safety threshold for multiple consecutive monitoring cycles, restoring the sampling frequency to a low-risk sampling frequency.

10. A multi-parameter intelligent early warning system for the risk of pancreatic fistula after pancreatic surgery, used to implement the multi-parameter intelligent early warning method for the risk of pancreatic fistula after pancreatic surgery as described in any one of claims 1 to 9, characterized in that, The system includes: The drainage fluid amylase dynamic analysis module is used to detect the amylase concentration in drainage fluid samples collected at various monitoring time points after pancreatic surgery, obtain time-series concentration data of drainage fluid amylase, and perform dynamic trend analysis on the time-series concentration data of drainage fluid amylase to generate dynamic trend characteristic data of amylase and amylase abnormal elevation marker information. The drainage fluid appearance image analysis module is used to perform color space conversion and multi-channel feature extraction processing on the appearance images of drainage fluid collected at various monitoring time points, identify abnormal characteristics that appear during the color change of drainage fluid, and generate drainage fluid characteristic data and abnormal characteristic marker information. The drainage pattern analysis and inflammation assessment module is used to perform pattern analysis on the drainage data recorded at various monitoring time points after surgery, generate drainage pattern feature data, and combine it with body temperature monitoring data, white blood cell count data and C-reactive protein detection data to perform quantitative assessment of systemic inflammatory response and generate inflammatory response assessment data. The multidimensional fusion prediction module for pancreatic fistula risk is used to calculate the risk of pancreatic fistula based on the dynamic trend characteristic data of amylase, the characteristic data of drainage fluid, the characteristic data of drainage volume pattern, and the inflammatory response assessment data, through a multidimensional parameter fusion prediction model. According to the ISGPS classification standard, it outputs the probability of occurrence of pancreatic fistula at each level, as well as the bleeding risk warning level and the pancreatic fistula risk warning level. The risk assessment report and nursing recommendation generation module is used to generate a dynamic risk assessment report and graded nursing recommendations for pancreatic fistula based on the probability of occurrence of pancreatic fistula at each level, as well as the bleeding risk warning level and the pancreatic fistula risk warning level. It also adjusts the sampling frequency and monitoring parameter thresholds at each monitoring time point in reverse according to the current risk level.

Images