Intelligent monitoring and health management system for postoperative drainage liquid of liver, gall and pancreas

Abstract

The invention relates to the technical field of medical monitoring, in particular to an intelligent monitoring and health management system for liver, gall and pancreas postoperative drainage fluid, which comprises a data acquisition module for acquiring core specific indexes of liver, gall and pancreas special drainage fluid, visual images of the drainage fluid, dynamic flow and physiological data of a patient; the intelligent analysis module constructs a hepatobiliary pancreatic postoperative exclusive multi-modal model, deeply couples postoperative disease course time sequence characteristics to identify risks at different stages after the operation, and triggers risk assessment when the operation is abnormal; the early warning module is provided with a light, medium and heavy three-level mechanism, and carries out resource adaptation type grading accurate pushing to a responsible physician and a nurse station terminal in combination with a medical care real-time load and a spatial distance. The health management module fuses individual features of the patient and real-time monitoring data, and constructs a personalized rehabilitation scheme in combination with rehabilitation feedback of the patient; and the data synchronization module transmits model update parameters through an encryption channel, and perfects the complete-cycle health archive of the patient. Therefore, the problems that in the prior art, an early warning mechanism is rigid, and data security and full-period management are insufficient are solved.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, specifically to an intelligent monitoring and health management system for postoperative drainage fluid from hepatobiliary and pancreatic surgeries. Background Technology

[0002] Hepatobiliary and pancreatic surgeries, due to their complex anatomical locations and significant trauma, often require the placement of drainage tubes postoperatively to drain intra-abdominal fluid, blood, and digestive fluids. This is crucial to reduce the risk of serious complications such as infection and anastomotic leakage. Monitoring drainage fluid is a core indicator for assessing postoperative recovery. Current drainage monitoring technology, based on a "drainage device + manual intervention" model, has formed a basic postoperative safety assurance system: the drainage tube, as the core carrier, effectively drains fluid from the surgical area, preventing compression of surrounding organs and tissue necrosis; medical staff regularly observe the color (e.g., bloody, bile-like, purulent), characteristics (turbidity, presence of flocculent matter) of the drainage fluid, and measure the drainage volume using a measuring cup, combining this with clinical experience to determine the presence of abnormalities such as active bleeding or biliary-pancreatic fistula; some clinical institutions have introduced simple electronic devices, such as graduated electronic drainage bags and baseline pressure sensors, to digitally display drainage volume or provide preliminary indications of drainage tube patency, providing basic data support for postoperative infection prevention and treatment plan adjustments. This is an important technical means to ensure patient safety after surgery.

[0003] While existing technologies have established a basic framework for postoperative drainage monitoring, significant shortcomings have become apparent under the demand for refined postoperative care in the hepatobiliary and pancreatic fields: Data collection is limited in scope and lacks specialized focus, only acquiring routine information such as drainage volume and basic characteristics, failing to collect core specific indicators, visual images, and related physiological data specific to the hepatobiliary and pancreatic fields, making it difficult to form a multi-dimensional monitoring system; risk analysis lacks specialized intelligent support, lacking a dedicated multimodal model for postoperative hepatobiliary and pancreatic surgery, and failing to conduct phased risk assessments based on disease progression characteristics, relying on experience-based judgments leading to insufficient accuracy in risk identification; the early warning mechanism is rigid and poorly adaptable to resources, lacking scientific grading standards, failing to consider real-time workload and spatial distance of medical staff, and lacking priority differentiation in early warning pushes, easily causing misallocation of medical resources or delays in critical risks; rehabilitation management plans lack personalization and dynamism, failing to integrate individual patient characteristics with real-time monitoring data, lacking feedback-driven iterative mechanisms, and rehabilitation guidance cannot adapt to the patient's dynamic recovery process; data security and full-cycle management are insufficient, lacking privacy-protected encrypted transmission and storage solutions, data is easily leaked and stored in a scattered manner, making it difficult to synchronously integrate into the patient's full-cycle health record. Summary of the Invention

[0004] This application provides an intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries to address issues such as rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0005] The first aspect of this application provides an intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries, comprising: a data acquisition module, an intelligent analysis module, an early warning module, a health management module, and a data synchronization module; wherein, the data acquisition module is used to collect core specific indicators of hepatobiliary and pancreatic drainage fluid, visualized images of drainage fluid, dynamic flow rate, and patient physiological data; the intelligent analysis module is used to construct a postoperative multimodal model specific to hepatobiliary and pancreatic surgeries, deeply coupled with the postoperative disease progression time sequence characteristics to identify risks at different stages of surgery, perform precise risk identification at different stages, and trigger risk assessment when abnormalities occur; the early warning module is used to set up a three-level mechanism of mild, moderate, and severe risks, and... The system integrates real-time workload and spatial distance of medical staff, introduces dynamic response priorities to accurately match the status of medical and nursing resources, and provides resource-adaptive hierarchical push to responsible physicians and nurse station terminals; the health management module is used to integrate individual patient characteristics with real-time monitoring data, and combine patient rehabilitation feedback to construct a feedback-driven, dynamically iterative, personalized rehabilitation plan, which adapts to the patient's recovery process in real time and provides precise rehabilitation guidance; the data synchronization module is used to transmit model update parameters through an encrypted channel, and adopts federated learning and homomorphic encryption schemes to protect privacy. The original data does not leave the local machine, and only parameters are updated locally, thus improving the patient's full-cycle health record.

[0006] Preferably, the data acquisition module includes a core indicator acquisition unit, an image acquisition unit, a flow monitoring unit, and a physiological data acquisition unit. The core indicator acquisition unit is used to acquire data on amylase, bilirubin, white blood cell count, protein concentration, and pancreatic enzyme activity in the drainage fluid. The image acquisition unit is used to dynamically capture visual characteristics of the drainage fluid's color, turbidity, flocculent distribution, sediment morphology, and presence of blood components. The flow monitoring unit is used to acquire real-time data on the instantaneous flow rate, cumulative drainage volume, and flow rate trend of the drainage fluid. The physiological data acquisition unit is used to acquire data on the patient's body temperature, heart rate, blood pressure, respiratory rate, and blood oxygen saturation.

[0007] Preferably, the intelligent analysis module includes a multimodal data fusion unit, a temporal disease progression modeling unit, a phased risk identification unit, and a risk assessment unit. The multimodal data fusion unit is used to perform cross-modal feature alignment on core specific indicators of drainage fluid, visualized images, dynamic flow, and patient physiological data, extracting high-dimensional correlation features between data. The temporal disease progression modeling unit is used to construct a postoperative multimodal model specific to hepatobiliary and pancreatic surgery based on the pathological mechanism of postoperative rehabilitation, deeply coupling the temporal dependence characteristics and phase-specific patterns of postoperative disease progression. The phased risk identification unit is used to accurately identify drainage-related complications and specific risks of delayed recovery at each stage by combining the differences in pathological characteristics during the acute, recovery, and post-recovery phases, quantifying risk probabilities, and determining the level of data abnormality. The risk assessment unit is used to integrate multi-dimensional feature data for comprehensive judgment based on the data abnormality level as the core judgment benchmark, generating a quantitative risk assessment result.

[0008] Preferably, the early warning module includes a grading standard definition unit, a medical staff status perception unit, a dynamic priority scheduling unit, and a precise push adaptation unit. The grading standard definition unit determines the judgment thresholds and triggering conditions for mild, moderate, and severe early warnings based on the degree of abnormality in postoperative drainage fluid from hepatobiliary and pancreatic surgeries, the range of fluctuations in patients' physiological data, and risk assessment results, thus obtaining the early warning level. The medical staff status perception unit is used to construct a dynamic database of medical resources by collecting real-time data on the current workload, patient volume, and spatial location of medical staff. The dynamic priority scheduling unit is used to establish a dynamic response priority algorithm model by combining the early warning level, real-time workload of medical staff, and spatial distance to accurately match early warning information with the status of medical resources. The precise push adaptation unit pushes the corresponding level of early warning information to the responsible physician's terminal or the nurse station system according to the priority matching results, while also supporting optimization of the information display format based on the terminal type.

[0009] Preferably, the health management module includes an individual characteristic integration unit, a rehabilitation plan dynamic generation unit, and a rehabilitation feedback interaction unit. The individual characteristic integration unit is responsible for organizing patient information such as age, underlying diseases, and surgical type, and synchronously linking drainage fluid monitoring data and real-time dynamic data of physiological indicators to construct a comprehensive patient status dataset. The rehabilitation plan dynamic generation unit, based on the integrated patient status dataset, calls upon a specialized rehabilitation knowledge base to construct an initial personalized rehabilitation plan in conjunction with the patient's recovery stage, and dynamically adjusts the plan content according to changes in real-time monitoring data. The rehabilitation feedback interaction unit collects patient feedback on pain levels, diet, sleep, rehabilitation experiences, and implementation status through an interface, transforming the feedback information into parameters recognizable by a post-hepatobiliary and pancreatic surgery-specific multimodal model, and iteratively optimizing the initial personalized rehabilitation plan.

[0010] Preferably, the data synchronization module includes an encrypted transmission unit, a privacy protection unit, and a parameter archive unit. The encrypted transmission unit establishes a secure and stable encrypted channel for transmitting model update parameters. The privacy protection unit is used to deploy a federated learning and homomorphic encryption scheme, strictly limiting the flow of original data to ensure it does not leave the local machine. The parameter archive unit receives and processes the synchronized model update parameters, iteratively updates local parameters, and integrates relevant data into the patient's health record to continuously improve the patient's full-cycle health data recording.

[0011] The second aspect of this application provides a method for intelligent monitoring and health management of postoperative drainage fluid in hepatobiliary and pancreatic surgeries, comprising: collecting core specific indicators of hepatobiliary and pancreatic drainage fluid, visualized images of drainage fluid, dynamic flow rate, and patient physiological data; constructing a postoperative multimodal model specific to hepatobiliary and pancreatic surgeries based on the core specific indicators, visualized images, dynamic flow rate, and physiological data of the drainage fluid; deeply coupling the postoperative disease progression time sequence characteristics to identify risks at different stages of surgery, performing precise risk identification at different stages, triggering risk assessment when abnormalities occur, and obtaining risk assessment results; based on the risk assessment results, establishing a three-level mechanism of mild, moderate, and severe risks, combining real-time medical and nursing workload and spatial distance, introducing dynamic response priority to accurately match the status of medical and nursing resources, and performing resource-adaptive hierarchical precise push to the responsible physician and nurse station terminals; integrating individual patient characteristics and real-time monitoring data, and combining patient rehabilitation feedback to construct a feedback-driven, dynamically iterative, personalized rehabilitation plan; based on the personalized rehabilitation plan, transmitting model update parameters through an encrypted channel, using federated learning and homomorphic encryption schemes to protect privacy, ensuring that the original data does not leave the local machine, and only updating parameters locally, thereby improving the patient's full-cycle health record.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the method of intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the method of intelligent monitoring and health management system for postoperative drainage fluid of the liver, gallbladder and pancreas as described in the above embodiments.

[0014] The fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing the method of intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects:

[0016] This application's embodiments comprehensively capture core specific indicators, visualized images, dynamic flow rates, and patient physiological data of hepatobiliary and pancreatic drainage fluid through a data acquisition module, providing multi-dimensional and all-scenario data support for postoperative monitoring. Based on a specialized multimodal model constructed using an intelligent analysis module, it deeply couples disease progression time-series characteristics to accurately identify and assess risks at different postoperative stages, overcoming the limitations of traditional monitoring that relies on experience-based judgment. Furthermore, leveraging a three-tiered early warning mechanism (mild, moderate, severe) combined with real-time medical staff workload and spatial distance to introduce dynamic response priorities, it accurately adapts and tiers medical resources, significantly reducing the risk of missed diagnoses of complications. This technology improves medical response efficiency; by integrating individual patient characteristics, real-time monitoring data, and rehabilitation feedback through a health management module, it constructs a feedback-driven, dynamically iterative, personalized rehabilitation plan, providing precise adaptation between rehabilitation guidance and the patient's recovery process; utilizing the encrypted transmission channel and federated learning and homomorphic encryption scheme of the data synchronization module, it updates model parameters and improves the patient's full-cycle health records while ensuring that the original data does not leave the local storage area and that privacy and security are controllable. This effectively improves the rehabilitation effect and safety of patients after hepatobiliary and pancreatic surgery, reduces the workload of medical staff, and provides comprehensive technical support for intelligent, precise, and safe management of specialized postoperative care. Thus, it solves the problems of rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 This is a schematic diagram of the structure of an intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgery, according to an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of a data acquisition module provided according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of an intelligent analysis module provided according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of an early warning module provided according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of a health management module provided according to an embodiment of this application;

[0024] Figure 6This is a schematic diagram of a data synchronization module provided according to an embodiment of this application;

[0025] Figure 7 This is a schematic diagram of the intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery provided according to an embodiment of this application;

[0026] Figure 8 This is a flowchart of a method for intelligent monitoring and health management of postoperative drainage fluid after hepatobiliary and pancreatic surgery according to an embodiment of this application;

[0027] Figure 9 This is a schematic diagram of a method for intelligent monitoring and health management of postoperative drainage fluid after hepatobiliary and pancreatic surgery according to an embodiment of this application;

[0028] Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] The following describes an embodiment of the intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery, with reference to the accompanying drawings. Addressing the problem of rigid early warning mechanisms mentioned in the background section, this application provides an intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery. In this system, a data acquisition module comprehensively captures core specific indicators, visualized images, dynamic flow rates, and patient physiological data of the drainage fluid from hepatobiliary and pancreatic surgeries, providing multi-dimensional and comprehensive data support for postoperative monitoring. Based on a specialized multimodal model constructed using an intelligent analysis module, it deeply couples the temporal characteristics of the disease course to accurately identify and assess risks at different postoperative stages, overcoming the limitations of traditional monitoring that relies on experience-based judgment. Furthermore, leveraging a three-tiered early warning mechanism (mild, moderate, severe) and incorporating real-time workload and spatial distance of medical staff, it introduces dynamic response priorities for medical staff. Precise resource allocation and tiered delivery significantly reduce the risk of missed diagnoses of complications and improve medical response efficiency. The health management module integrates individual patient characteristics, real-time monitoring data, and rehabilitation feedback to construct a feedback-driven, dynamically iterative, personalized rehabilitation plan, precisely matching rehabilitation guidance with the patient's recovery process. Utilizing the encrypted transmission channel and federated learning and homomorphic encryption scheme of the data synchronization module, while ensuring that original data remains locally and privacy is controllable, model parameter updates and the improvement of patients' full-cycle health records are completed. This effectively improves the rehabilitation outcomes and safety of patients after hepatobiliary and pancreatic surgery, reduces the workload of medical staff, and provides comprehensive technical support for intelligent, precise, and safe management of postoperative care in specialized departments. Thus, it solves the problems of rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0031] Figure 1 This is a schematic diagram of the intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery provided in an embodiment of this application.

[0032] This application provides an intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries. The system 10 includes:

[0033] Data acquisition module 100, intelligent analysis module 200, early warning module 300, health management module 400, and data synchronization module 500.

[0034] The system includes: a data acquisition module 100 for collecting core specific indicators of hepatobiliary and pancreatic drainage fluid, visualized images of the drainage fluid, dynamic flow rate, and patient physiological data; an intelligent analysis module 200 for constructing a postoperative multimodal model specific to hepatobiliary and pancreatic surgery, deeply coupling postoperative disease progression characteristics to identify risks at different stages, accurately identifying risks at each stage, and triggering risk assessment when abnormalities occur; an early warning module 300 for setting up a three-tiered mechanism (mild, moderate, and severe), combining real-time medical staff workload and spatial distance, introducing dynamic response priorities to accurately match medical staff resource status, and providing resource-adaptive tiered push notifications to responsible physicians and nurse station terminals; a health management module 400 for integrating individual patient characteristics with real-time monitoring data, combining patient rehabilitation feedback to construct a feedback-driven, dynamically iterative, personalized rehabilitation plan, adapting to the patient's recovery process in real time, and providing precise rehabilitation guidance; and a data synchronization module 500 for transmitting model update parameters through an encrypted channel, using federated learning and homomorphic encryption schemes to protect privacy, ensuring that the original data does not leave the local machine, and only updating parameters locally, thus improving the patient's full-cycle health record.

[0035] It is understood that in this embodiment, the data acquisition module comprehensively captures core specific indicators, visualized images, dynamic flow, and patient physiological data of hepatobiliary and pancreatic drainage fluid, providing multi-dimensional and full-scenario data support for postoperative monitoring. Based on a specialized multimodal model constructed using the intelligent analysis module, it deeply couples disease progression time-series characteristics to accurately identify and assess risks at different postoperative stages, overcoming the limitations of traditional monitoring that relies on experience-based judgment. Furthermore, the early warning module employs a three-tiered mechanism (mild, moderate, and severe), combined with real-time medical staff workload and spatial distance to introduce dynamic response priorities, enabling precise allocation and tiered delivery of medical resources, significantly reducing complications. This approach mitigates the risk of missed diagnoses and improves medical response efficiency. By integrating individual patient characteristics, real-time monitoring data, and rehabilitation feedback through a health management module, it constructs a feedback-driven, dynamically iterative, personalized rehabilitation plan, precisely adapting rehabilitation guidance to the patient's recovery process. Utilizing the encrypted transmission channel and federated learning and homomorphic encryption scheme of the data synchronization module, it updates model parameters and improves the patient's full-cycle health records while ensuring that original data remains locally and privacy is controllable. This effectively enhances the rehabilitation outcomes and safety of patients after hepatobiliary and pancreatic surgery, reduces the workload of medical staff, and provides comprehensive technical support for intelligent, precise, and safe postoperative management in specialized departments. This addresses the problems of rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0036] In this embodiment of the application, the data acquisition module 100 further includes: Figure 2 As shown, the core indicator acquisition unit, image acquisition unit, flow monitoring unit, and physiological data acquisition unit are included.

[0037] The core indicator acquisition unit is used to collect data on amylase, bilirubin, white blood cell count, protein concentration, and pancreatic enzyme activity in the drainage fluid; the imaging acquisition unit is used to dynamically capture the color, turbidity, flocculent distribution, sediment morphology, and visual characteristics of blood components in the drainage fluid; the flow monitoring unit is used to collect the instantaneous flow rate, cumulative drainage volume, and flow rate trend of the drainage fluid in real time; and the physiological data acquisition unit is used to collect data on the patient's body temperature, heart rate, blood pressure, respiratory rate, and blood oxygen saturation.

[0038] It is understood that the embodiments of this application focus on key biochemical parameters such as amylase, bilirubin, white blood cell count, protein concentration, and pancreatic enzyme activity through the core indicator acquisition unit, directly capturing specific signals of the recovery process of liver, gallbladder, and pancreatic function, as well as pancreatic fistula, bile fistula, and infectious complications, providing core biochemical diagnostic basis for early complication identification; the image acquisition unit dynamically captures the visual characteristics of drainage fluid color, turbidity, flocculent distribution, sediment morphology, and presence of blood components, transforming traditional subjective observation into objective quantitative data, accurately capturing subtle fluctuations in the condition, compensating for the lag and individual differences in manual assessment, and providing intuitive morphological support for drainage fluid property analysis; the flow monitoring unit collects instantaneous flow rate, cumulative drainage volume, and flow rate change trends in real time, dynamically reflecting the exudation intensity and healing progress of the surgical area, and providing timely warnings of abnormal drainage conditions, providing accurate flow dynamics data for adjusting drainage plans and judging the recovery status of the surgical area; the physiological data acquisition unit simultaneously acquires the patient's basic vital signs such as body temperature, heart rate, blood pressure, respiratory rate, and blood oxygen saturation, establishing a linkage between local indicators of drainage fluid and the overall physiological state, and conducting collaborative monitoring to avoid the limitations of isolated assessment.

[0039] For example, after a combined hepatobiliary and pancreatic surgery, the patient wears this intelligent monitoring system. The core indicator acquisition unit monitors key data such as amylase (value increased by 30% compared to preoperative level), bilirubin, and white blood cell count in the drainage fluid in real time and uploads them simultaneously. The image acquisition unit uses a built-in high-definition camera to dynamically capture the drainage fluid gradually changing from pale yellow to turbid, with scattered flocculent matter and a small amount of blood. The flow monitoring unit accurately records the instantaneous flow rate of approximately 8 ml / h, with a cumulative drainage volume of 120 ml, while generating a curve showing a slow increase in flow rate. The physiological data acquisition unit continuously collects vital signs data such as the patient's body temperature of 37.8℃, heart rate of 89 beats / min, and blood pressure of 125 / 80 mmHg. After all dimensions of information are integrated and analyzed by the system, a warning of "abnormal drainage fluid characteristics + elevated amylase" is pushed to the medical staff's APP, providing accurate data support for the early screening of pancreatic fistula risk.

[0040] In this embodiment of the application, the intelligent analysis module 200 includes: Figure 3 As shown, the system includes a multimodal data fusion unit, a time-series disease progression modeling unit, a phased risk identification unit, and a risk assessment unit.

[0041] The system comprises the following components: a multimodal data fusion unit for cross-modal feature alignment of core specific indicators of drainage fluid, visual images, dynamic flow, and patient physiological data, and extraction of high-dimensional correlation features between data; a temporal disease progression modeling unit for constructing a dedicated multimodal model for postoperative hepatobiliary and pancreatic surgery based on the pathological mechanism of postoperative rehabilitation, deeply coupling the temporal dependence and stage-specific patterns of postoperative disease progression; a stage-specific risk identification unit for accurately identifying drainage-related complications and specific risks of delayed recovery at each stage by combining the differences in pathological characteristics during the acute, recovery, and post-recovery phases, quantifying risk probabilities, and determining the level of data anomalies; and a risk assessment unit for integrating multi-dimensional feature data for comprehensive assessment based on the level of data anomalies as the core judgment criterion, generating quantitative risk assessment results.

[0042] It is understood that the multimodal data fusion unit in this application's embodiments breaks through the limitations of a single data dimension. Through cross-modal feature alignment and high-dimensional correlation feature extraction, it deeply integrates drainage fluid-specific indicators, visualized images, dynamic flow, and patient physiological data, providing a comprehensive and accurate data foundation for risk assessment. The time-series disease progression modeling unit constructs a dedicated multimodal model based on the postoperative rehabilitation pathological mechanism of hepatobiliary and pancreatic surgery, deeply coupling the time-dependent characteristics and stage-specific patterns of disease progression. This solves the problem of insufficient adaptability of general models to complex postoperative pathological processes, improving pathological fit and prediction accuracy. The phased risk identification unit targets the postoperative acute phase, recovery phase, and rehabilitation phase. Precise risk identification based on pathological characteristics in the recovery period can effectively distinguish stage-specific risks such as drainage-related complications and delayed recovery. Furthermore, by quantifying risk probabilities and determining the level of data anomalies, it provides clear risk indications for clinical intervention. The risk assessment unit uses the level of data anomalies as the core judgment benchmark, integrates multi-dimensional characteristic data for comprehensive judgment, and generates quantitative risk assessment results. This ensures the objectivity and scientific rigor of risk assessment while providing clinicians with intuitive and actionable decision-making support. It enables intelligent analysis throughout the entire process, from data integration to risk warning, improving the accuracy and timeliness of postoperative drainage management in hepatobiliary and pancreatic surgeries and reducing the incidence of postoperative adverse events.

[0043] It should be noted that cross-modal feature alignment was performed on the core specific indicators of drainage fluid, visual images, dynamic flow, and patient physiological data to extract high-dimensional correlation features between data. Using the postoperative time axis as the anchor point, a dynamic time warping algorithm was used to eliminate time offsets caused by differences in data acquisition frequencies, and time synchronization alignment of the core indicators of drainage fluid, image features, flow data, and physiological parameters was achieved. At the same time, semantic correlation mapping between modalities was constructed by combining postoperative pathological priors of hepatobiliary and pancreatic surgery. Subsequently, a multimodal attention fusion network was used to mine key features of single modality through self-attention and capture intermodal correlations through mutual attention. After processing by the hybrid network, a high-dimensional feature set containing multimodal information and correlation patterns was output to support subsequent risk identification.

[0044] The pathological mechanism of postoperative rehabilitation after hepatobiliary and pancreatic surgery refers to the intrinsic law by which the body gradually restores the local structural integrity of the liver, gallbladder, and pancreas and the body's overall physiological homeostasis through a series of pathophysiological processes, including tissue repair, inflammation regulation, metabolic adaptation, and organ function reconstruction, after the trauma of hepatobiliary and pancreatic surgery.

[0045] Multidimensional feature data guides the core indicators of drainage fluid, the visualization features of drainage fluid, the flow rate of drainage fluid, the physiological data features of patients, the postoperative disease stage features, and the individual baseline features of patients.

[0046] By integrating multi-dimensional feature data for comprehensive analysis, a quantitative risk assessment result is generated. First, the core indicators of drainage fluid, imaging / flow characteristics, and patient physiological data are correlated and matched with individual baselines and postoperative disease stages. After standardization, the features are marked to indicate whether they exceed the stage risk threshold. Then, an attention mechanism fusion algorithm is used to weight key features of single-modality and cross-modality, highlighting high-risk contribution features. Subsequently, the fused features are input into a dedicated multimodal model for hepatobiliary and pancreatic surgery, combined with a clinical knowledge base, and an initial quantitative score of 0-100 is generated through time series analysis and probability statistics. Finally, the score is calibrated based on the individual patient situation and mapped to three risk levels: low (0-30 points), medium (31-60 points), and high (61-100 points). The risk quantification value, level, and core inducing factors are output simultaneously to form a complete assessment result.

[0047] A specialized multimodal model was constructed after hepatobiliary and pancreatic surgery. Based on core specific biochemical indicators, visual images, dynamic flow rates, and time-series physiological data of drainage fluid acquired by the acquisition module, a specialized data foundation was built by supplementing multicenter postoperative clinical historical data and individual patient baseline information. After data cleaning and standardization, modal temporal alignment, and missing value processing, a dedicated feature extractor was used to extract nonlinear features of biochemical indicators, specialized semantic features of drainage fluid images, disease stage perception features of time-series data, and dense vector features of individual baselines. A disease stage attention fusion strategy was then used to dynamically adjust the weights of each modality, integrating the multidimensional data... After weighted fusion and optimization of the degree features, the model is input into a multi-task learning framework for simultaneous training of risk level prediction and complication type identification tasks. The model training is completed by combining weighted cross-entropy, FocalLoss loss function and AdamW optimizer. The generalization ability is improved through multi-center federated learning. After lightweight quantization compression, the model is adapted for terminal deployment. At the same time, relying on the encrypted channel of the data synchronization module, clinical feedback and real outcome data are collected. The model is dynamically iterated through incremental learning. The output of the phased risk identification results provides core decision-making basis for the graded push of the early warning module and the formulation of personalized rehabilitation plans in the health management module.

[0048] For example, 48 hours after a pancreaticoduodenectomy (acute phase), the multimodal data fusion unit in the system's intelligent analysis module first analyzed the drainage fluid amylase at 520 U / L (the upper limit of normal serum is 135 U / L, exceeding it by 3.8 times), ultrasound showed a 2.3 cm fluid-filled area around the pancreas, the drainage flow rate was 110 ml / h (not decreasing at a rate of 10%-15% compared to 500 ml 24 hours post-surgery), and physiological data showed a body temperature of 38.6℃ and a white blood cell count of 13.2 × 10⁻⁶. 9 / L performs cross-modal feature alignment and extracts high-dimensional correlation features; the time-series disease progression modeling unit deeply couples the time-series dependence features and stage patterns of the disease progression through a dedicated multimodal model after hepatobiliary and pancreatic surgery, and finds that the decreasing flow trend is inconsistent with the postoperative rehabilitation pathological mechanism; the stage-based risk identification unit combines the differences in pathological features in the acute phase to accurately identify the specific risk of pancreatic fistula, quantifies the risk probability to be 58%, and determines the abnormality level to be intermediate; the risk assessment unit uses this abnormality level as the core judgment benchmark, integrates multi-dimensional feature data for comprehensive judgment, and generates a quantitative risk assessment result: the risk of pancreatic fistula is 58%, and it is necessary to strengthen drainage monitoring and start anti-infection intervention.

[0049] In this embodiment of the application, the early warning module 300 includes: Figure 4 As shown, the system includes a hierarchical standard definition unit, a medical staff status perception unit, a dynamic priority scheduling unit, and a precise push adaptation unit.

[0050] The grading standard definition unit clarifies the judgment thresholds and triggering conditions for mild, moderate, and severe warnings based on the degree of abnormality in postoperative drainage fluid after hepatobiliary and pancreatic surgery, the range of fluctuations in patients' physiological data, and risk assessment results, thus obtaining the warning level. The medical staff status perception unit is used to construct a dynamic database of medical staff resources by collecting data on the current workload, number of patients received, and spatial location of medical staff in real time. The dynamic priority scheduling unit is used to establish a dynamic response priority algorithm model by combining the warning level, real-time workload of medical staff, and spatial distance to accurately match the warning information with the status of medical staff resources. The precise push adaptation unit pushes the warning information of the corresponding level to the responsible physician terminal or nurse station system according to the priority matching results, and also supports optimizing the information display format according to the terminal type.

[0051] It is understandable that the grading standard definition unit in this application provides a standardized and quantitative basis for judging abnormal drainage fluid and patient risk warnings by clearly defining the judgment thresholds and triggering conditions for mild, moderate and severe warnings. This effectively avoids false alarms and missed alarms caused by ambiguous judgment standards, and ensures the accuracy and consistency of warning level classification. The medical staff status perception unit collects multi-dimensional data such as medical staff workload, patient volume and spatial location in real time to build a dynamically updated medical staff resource database. This enables the system to comprehensively and in real time grasp the supply and demand status of medical staff resources, providing key data support for subsequent warning scheduling. Dynamic priority The scheduling unit establishes an algorithm model based on the warning level, real-time workload of medical staff, and spatial distance to accurately match warning information with medical resources. This ensures that high-risk warnings are prioritized for matching with suitable medical resources, improving the efficiency of emergency response in high-risk situations and avoiding delays caused by resource mismatch. The precise push and matching unit pushes warning information to the responsible physician's terminal or nurse station system and optimizes the information display format according to the terminal type. This ensures that the warning information accurately reaches the corresponding responsible parties and reduces the information acquisition cost for medical staff by adapting the display logic to different terminals, thereby improving the convenience and timeliness of warning information processing.

[0052] It should be noted that the dynamic response priority algorithm model formula is as follows:

[0053]

[0054]

[0055]

[0056] in, This is the dynamic response priority coefficient, with a value range of... A larger value indicates a higher response priority. Assigning weights to warning levels; Weighting of healthcare worker workload; Spatial distance weights; This refers to the warning level coefficient. For the workload adaptation coefficient of medical staff; This refers to the spatial distance adaptation coefficient; the current load is a value quantified by the number of patient reception tasks and pending matters collected by the medical staff status sensing unit, with a value range of... The maximum load threshold is a preset quantified value of the maximum workload that medical staff can handle, which can be adjusted according to hospital management needs; the actual distance is the straight-line distance between the location of the medical staff and the patient's ward, collected by the medical staff status sensing unit; the maximum response distance threshold is a preset maximum distance at which medical staff can respond quickly. When the actual distance exceeds this threshold, .

[0057] To construct a dynamic database of medical and nursing resources, the system first connects to the hospital's existing medical and nursing work management system and the mobile work terminals of medical staff to capture real-time workload data such as the current number of patients received, the number of patients under management, and the progress of received and treated alerts. Then, it uses in-hospital positioning technology to collect spatial location information such as the department and ward area where medical staff are located, while simultaneously recording basic attribute data such as the professional qualifications and areas of expertise of medical staff. Subsequently, the collected multi-dimensional data is cleaned, deduplicated, and standardized in format, and a mapping relationship between basic attributes, real-time workload, and spatial location is established. Dynamic data tags are constructed by combining timestamps, ultimately forming a database that supports real-time querying and automatic updates, providing data support for the accurate matching of alert information and medical and nursing resources in subsequent dynamic priority scheduling.

[0058] A dynamic response priority algorithm model was established, constructing an indicator system encompassing three core dimensions: warning level, real-time workload of medical staff, and spatial distance. The warning levels (mild, medium, and severe) were assigned basic priority weights, which were dynamically adjusted based on the quantitative value of abnormal drainage fluid and the fluctuation range of patients' physiological indicators. The real-time workload of medical staff was quantified as a 0-1 load index by multiplying the number of currently pending tasks by the task urgency coefficient and adding the number of pending warnings. Spatial distance was converted into estimated response time based on in-hospital location data. Subsequently, the weights of each dimension were determined using the analytic hierarchy process (AHP), and subjective biases were corrected using the entropy weight method. A priority calculation function was constructed: (Priority = Warning Correction Weight × 0.6 + (1 - Load Index) × 0.2 + (1 - Standardized Response Time Value) × 0.2). Simultaneously, the algorithm accessed a dynamic database of medical staff resources to obtain updated data in real time. A sliding window mechanism was used to dynamically iteratively calculate the matching priority between warning information and each medical staff member, ultimately outputting a ranking result to support accurate push notifications.

[0059] For example, on the 3rd day after pancreaticoduodenectomy, the system detected a 24-hour abdominal drainage volume of 380ml and an amylase level of 2100U / L in the drainage fluid (serum amylase 350U / L, exceeding the normal value by 5 times) through the grading standard definition unit. Simultaneously, the system monitored the patient's heart rate at 115 bpm and blood pressure at 100 / 60 mmHg (a 22% fluctuation from the postoperative baseline of 125 / 75 mmHg). Combined with the risk assessment results, this was classified as a severe warning. The medical staff status sensing unit collected real-time data on the attending physician's current care of two postoperative patients (moderate workload) and their location 30 meters from the patient's ward. At the nursing station, with 3 nurses on duty and 1 patient reception task, a dynamic database of medical resources has been established. The dynamic priority scheduling unit uses an algorithm model to comprehensively match the severe warning level, the physician's medium workload, and the 30-meter proximity to determine the responsible physician as the optimal response target. The precise push and adaptation unit then pushes the warning information to the responsible physician's mobile terminal, optimizing the display of core data in the form of pop-up window + voice: "Severe warning: Patient's drainage fluid 380ml / day, amylase 2100U / L, heart rate 115 beats / min, blood pressure fluctuation 22%, please go to the ward 30 meters away immediately for treatment."

[0060] In this embodiment, the health management module 400 includes, as follows: Figure 5 As shown, there are three units: individual characteristic integration unit, rehabilitation plan dynamic generation unit, and rehabilitation feedback interaction unit.

[0061] The individual feature integration unit is responsible for sorting out information such as patient age, underlying diseases, and surgical type, and synchronously linking drainage fluid monitoring data and real-time dynamic data of physiological indicators to construct a comprehensive patient status dataset. The rehabilitation plan dynamic generation unit is used to construct an initial personalized rehabilitation plan based on the integrated patient status dataset, call the specialized rehabilitation knowledge base, and combine the patient's recovery stage, and dynamically adjust the content of the plan according to changes in real-time monitoring data. The rehabilitation feedback interaction unit collects feedback from patients on their rehabilitation experience and implementation status regarding pain level, diet, and sleep through the interface, and transforms the feedback information into parameters that can be recognized by the postoperative multimodal model specific to hepatobiliary and pancreatic surgery, and iteratively optimizes the initial personalized rehabilitation plan.

[0062] It is understandable that the individual feature integration unit in this application embodiment breaks down the silo barriers between patient basic information, surgical-related data, and dynamic data such as drainage fluid monitoring and physiological indicators. Furthermore, through the systematic sorting and integration of multi-dimensional data, it provides a comprehensive and accurate patient status baseline for rehabilitation plan formulation, avoiding the insufficient targeting issues caused by traditional rehabilitation plans relying on single data. The dynamic rehabilitation plan generation unit, relying on the deep coupling of a specialized rehabilitation knowledge base and a real-time patient status dataset, not only performs personalized adaptation of the initial rehabilitation plan but also completes iterative adjustments through dynamic feedback from real-time monitoring data. This ensures that rehabilitation suggestions always accurately match the patient's recovery process, effectively reducing rehabilitation risks caused by plan rigidity. The rehabilitation feedback interaction unit builds a communication bridge between the patient's subjective feelings and the system's objective management. By converting rehabilitation experiences such as pain levels, diet, and sleep into model-recognizable parameters, it allows the rehabilitation plan to respond to the patient's subjective needs while taking into account objective data, improving the applicability and patient compliance of the rehabilitation plan, accelerating the rehabilitation process of patients after hepatobiliary and pancreatic surgery, and improving rehabilitation prognosis.

[0063] It should be noted that the construction of a comprehensive patient status dataset involves extracting basic patient information by connecting to the hospital's electronic medical record system, including baseline data of individual characteristics such as age, gender, weight, underlying diseases, specific types of hepatobiliary and pancreatic surgeries, and preoperative examination indicators. Simultaneously, real-time monitoring data from the data acquisition module is integrated, covering core indicators of drainage fluid, visual image features, dynamic flow data, and physiological indicators such as patient temperature, heart rate, blood pressure, respiratory rate, and blood oxygen saturation, forming a dynamic monitoring data layer. Subsequently, data fusion technology is used to associate the baseline data and real-time data by timestamp, constructing a three-dimensional data association structure of basic attributes, surgical information, and real-time monitoring. Simultaneously, data cleaning and standardization are performed to form a structured, time-seriesd patient status dataset covering both static characteristics and dynamic states of the patient.

[0064] For example, a 62-year-old patient undergoes laparoscopic choledochotomy and T-tube drainage for choledocholithiasis. The individual characteristic integration unit of the system's health management module first sorts out the patient's age, history of type 2 diabetes, and surgical type. Simultaneously, it correlates dynamic data such as 320ml of T-tube drainage fluid (dark red, no flocculent matter) on postoperative day 1, fasting blood glucose 7.8mmol / L, and heart rate 86 bpm; 410ml of drainage fluid (light red) on postoperative day 3, blood glucose 7.1mmol / L, and heart rate 81 bpm; and 180ml of drainage fluid (light yellow) on postoperative day 7, blood glucose 6.3mmol / L, and heart rate 75 bpm, constructing a complete patient status dataset. The rehabilitation plan dynamic generation unit uses this dataset to call upon the specialist knowledge base and, combined with the characteristics of the postoperative acute phase (0-7 days), generates an initial plan, namely a liquid diet (rice soup, lotus root) for postoperative days 1-3. The patient was initially given a semi-liquid diet (porridge, steamed egg custard) and walked for 5 minutes three times a day in the ward. On postoperative day 3, due to normal drainage fluid characteristics and a steady decrease in blood sugar, the diet was dynamically adjusted to semi-liquid (porridge, steamed egg custard) and the walking time was extended to 10 minutes. On postoperative day 7, the drainage fluid volume decreased significantly and physiological indicators met the standards, so the diet was further adjusted to soft food (soft noodles, fish) and walked for 15 minutes twice a day in the corridor. The rehabilitation feedback interaction unit collected data from the patient through the interface. On postoperative day 1, the patient's NRS pain score was 5, the diet was tolerable, and the nighttime sleep was 3 hours. On postoperative day 3, the pain score was 3, the food intake increased, and the sleep was 5 hours. On postoperative day 7, the pain score was 2, and the diet and sleep returned to normal. After these feedbacks were converted into parameters that the model could recognize, the plan for postoperative days 8-14 was iteratively optimized, namely a low-fat normal diet, 20 minutes of brisk walking every day, and the dosage of hypoglycemic drugs was adjusted to maintain stable blood sugar.

[0065] In this embodiment of the application, the data synchronization module 500 includes, as follows: Figure 6 As shown, there are an encrypted transmission unit, a privacy protection unit, and a parameter file unit.

[0066] The encrypted transmission unit is used to build a secure and stable encrypted channel for transmitting model update parameters. The privacy protection unit is used to deploy federated learning and homomorphic encryption schemes at the core, ensuring that the original data does not leave the local area by strictly limiting the scope of the original data flow. The parameter file unit is used to receive and process the synchronized model update parameters, complete the iterative update of local parameters, and integrate relevant data into the patient's health record to continuously improve the patient's full-cycle health data record.

[0067] It is understood that the embodiments of this application establish a secure and stable encrypted channel through an encrypted transmission unit and dynamically update the encrypted transmission model parameters. This ensures the confidentiality, integrity, and availability of drainage fluid monitoring data and model iteration data during cross-terminal and cross-platform transmission, while also adapting to the transmission needs of complex network environments and avoiding the risk of data transmission interruption or leakage. The privacy protection unit is equipped with a federated learning and homomorphic encryption scheme. By strictly limiting the scope of original medical data flow, the original data does not leave the local area. This not only complies with relevant laws, regulations, and ethical requirements for medical and health data privacy protection, but also aggregates multi-center data resources to complete model collaborative optimization without disclosing sensitive patient information, thus balancing privacy security and model performance improvement. The parameter file unit receives and processes the synchronized model update parameters to perform efficient iterative upgrades of the local model, ensuring continuous optimization of the accuracy of the system's intelligent monitoring. On the other hand, it integrates relevant monitoring data and model iteration records into the patient's health record, constructing a continuous health data chain covering the entire rehabilitation cycle after hepatobiliary and pancreatic surgery, providing complete data support for clinical treatment adjustments, rehabilitation program optimization, and long-term health management.

[0068] For example, a tertiary hospital uses this system to monitor T-tube drainage fluid in patients after liver transplantation. The encrypted transmission unit establishes an end-to-end secure channel using the AES-256 algorithm, combined with TCP / IP protocol optimization to transmit real-time monitoring data such as drainage volume of 20ml-35ml per hour, bilirubin 34.2μmol / L, pH value 7.1-7.3, and amylase 85U / L with a latency of <50ms and a packet loss rate of less than 0.01%. The privacy protection unit deploys a federated learning horizontal federated architecture and Paillier homomorphic encryption scheme, superimposed with dynamic data masking technology, and shares encryption only with three collaborating hospitals. Sensitive data such as the model gradient and feature vectors after decompression, the original drainage fluid characteristics (pale yellow and clear), and infection markers (C-reactive protein 12 mg / L) are all stored in the local hospital data center, reducing privacy budget consumption by 76%. The parameter file unit receives synchronized parameters twice a day and completes local anomaly identification model iteration through gradient descent algorithm, improving accuracy from 90% to 95%. At the same time, data such as the cumulative drainage volume of 860 ml over 7 days postoperatively, turbidity change curve, and parameter update log are integrated into the patient's electronic health record, with hierarchical access permissions set (only authorized medical staff can query), continuously improving the full-cycle diagnosis and treatment data chain from the perioperative period to the recovery period.

[0069] The intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgery proposed in this application comprehensively captures core specific indicators, visualized images, dynamic flow, and patient physiological data of hepatobiliary and pancreatic drainage fluid through a data acquisition module, providing multi-dimensional and full-scenario data support for postoperative monitoring. Based on a specialized multimodal model constructed by an intelligent analysis module, it deeply couples the temporal characteristics of the disease course to accurately identify and assess risks at different postoperative stages, overcoming the limitations of traditional monitoring that relies on experience-based judgment. Furthermore, through a three-tiered early warning mechanism (mild, moderate, and severe), combined with real-time medical staff workload and spatial distance, it introduces dynamic response priorities for precise allocation and tiered delivery of medical resources. This technology significantly reduces the risk of missed diagnoses of complications and improves medical response efficiency. By integrating individual patient characteristics, real-time monitoring data, and rehabilitation feedback through a health management module, it constructs a feedback-driven, dynamically iterative, personalized rehabilitation plan, precisely adapting rehabilitation guidance to the patient's recovery process. Utilizing the encrypted transmission channel and federated learning and homomorphic encryption scheme of the data synchronization module, it updates model parameters and improves the patient's full-cycle health records while ensuring that original data remains locally and privacy is controllable. This effectively enhances the rehabilitation outcomes and safety of patients after hepatobiliary and pancreatic surgery, reduces the workload of medical staff, and provides comprehensive technical support for intelligent, precise, and safe management of postoperative care in specialized departments. Thus, it addresses the problems of rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0070] The following will illustrate the intelligent monitoring and health management system for postoperative drainage fluid after hepatobiliary and pancreatic surgery through a specific embodiment, such as... Figure 7 As shown, it includes:

[0071] A 52-year-old patient underwent a laparoscopic pancreaticoduodenectomy for pancreatic head cancer at a top-tier hospital's hepatobiliary and pancreatic surgery department. This is one of the most challenging surgeries in hepatobiliary and pancreatic surgery, with a 20% risk of postoperative complications such as pancreatic fistula and bile fistula. Accurate monitoring of drainage fluid is directly related to the patient's life safety. After the surgery, the responsible nurse brought a smart device to the ward and connected three drainage tubes—one to the abdominal cavity, one to the pancreatic duct, and one to the bile duct—to the data acquisition terminal of the "Intelligent Monitoring and Health Management System for Postoperative Drainage Fluid in Hepatobiliary and Pancreatic Surgery." This newly implemented system will soon provide comprehensive support for the patient's postoperative recovery.

[0072] The data acquisition module was deployed in just 5 minutes. The responsible nurse connected the disposable smart drainage bag to the patient's drainage tube. The flexible pressure sensor on the outside of the bag automatically aligned, and the miniature spectrometer at the drainage tube interface started operating. Real-time data was immediately displayed on the monitoring screen at the bedside: the abdominal drainage fluid was slightly bloody, with a flow rate of 120 ml / h and an amylase content of 320 U / L; the pancreatic duct drainage fluid was colorless and transparent, with a flow rate of 80 ml / h and an amylase content of 1200 U / L; and the bile duct drainage fluid was golden yellow, with a flow rate of 250 ml / h and an bilirubin content of 18 μmol / L. At the same time, the smart bracelet linked to the monitoring system was already worn by the patient, transmitting real-time physiological data such as heart rate of 72 beats / min, body temperature of 36.5℃, and blood pressure of 125 / 80 mmHg. The high-definition camera above the drainage bag took an image of the drainage fluid every 30 seconds and automatically uploaded it to the system database. “Previously, we had to manually record the amount and color of the drainage fluid every hour, and send samples for testing regularly. Now the system does all of that for us, and the data is accurate.” The nurse in charge explained to the family while entering the patient’s preoperative diabetes history and allergy history into the system. “This data is synchronized in real time, and the doctor can see it in the office.”

[0073] Twelve hours post-surgery, the patient's monitoring data remained stable, and the multimodal model of the intelligent analysis module had completed basic data modeling. This model, trained on 50,000 post-hepatobiliary and pancreatic surgery cases, is specifically designed for the characteristics of the post-pancreatoduodenectomy process, deeply coupling drainage fluid indicators, physiological data, and post-operative temporal characteristics. The system automatically divided the patient's post-operative stage into the "acute exudative phase" and set core monitoring thresholds for this phase: abdominal drainage volume ≤300ml / h, amylase ≤500U / L, pancreatic duct drainage amylase ≤1500U / L, and body temperature ≤38℃. Whenever new data was received, the model performed real-time comparative analysis, generating a "green and safe" assessment result, and dynamically updating the disease progression curve on the doctor's workstation terminal. Before ward rounds, the attending physician reviewed the patient's monitoring report via a mobile app. The system's automatic annotation, "Drainage fluid indicators are consistent with early post-operative patterns, with no abnormal fluctuations," gave him a clear assessment of the patient's recovery status.

[0074] In the early morning of the second day after surgery, a soft beeping sounded in the patient's ward, while the intelligent early warning terminal at the nurses' station emitted a red audible and visual alarm. The nurse on duty immediately checked the screen. The system showed that the patient's pancreatic duct drainage fluid amylase level had surged from 1300 U / L to 2800 U / L within 15 minutes, body temperature had risen to 38.7℃, and heart rate had increased to 95 beats / min. The intelligent analysis module had triggered a risk assessment, identifying it as a "high-risk signal for pancreatic fistula," with a severe warning level. At this time, the system's early warning module automatically initiated a resource adaptation process, retrieving real-time medical and nursing resource data through the hospital's intranet: the attending physician was in the operating room due to an emergency, but the on-duty physician closest to the ward was conducting a consultation on the floor, with their current workload being "only responsible for 1 mild case patient"; the responsible nurse currently had no other critical care duties and was only 20 meters away from the ward. The system immediately sends an emergency alert to the on-duty doctor's mobile phone, including "patient's bed number, pancreatic duct drainage fluid index change curve, and abnormal physiological data values," along with a smart prompt "It is recommended to immediately check the drainage fluid characteristics and activate the pancreatic fistula prevention plan." At the same time, it pushes an execution instruction to the responsible nurse's PDA: "Immediately go to the ward to monitor vital signs and prepare somatostatin injection." The electronic screen at the nurses' station simultaneously displays patient information and treatment priority.

[0075] Upon receiving the alert, the on-duty doctor arrived at the ward within two minutes. The responsible nurse had already rechecked vital signs and prepared the necessary medications. By reviewing the drainage fluid images retrieved by the system, the doctor observed that the pancreatic duct drainage fluid had changed from colorless and transparent to a pale milky white. Combined with real-time data displayed by the system, such as amylase 3100 U / L and C-reactive protein 25 mg / L, a diagnosis of "biochemical pancreatic fistula" was confirmed. He immediately adjusted the treatment plan through the system's health management module. The system automatically integrated individual characteristics such as the patient's diabetes history and postoperative nutritional tolerance to generate a new rehabilitation plan: suspending enteral nutrition infusion and switching to intravenous nutritional support, administering somatostatin continuously at 250 μg / h, monitoring blood glucose eight times daily, and adjusting insulin dosage based on the results. These plans were synchronized in real-time to the nurses' station execution terminal and the patient's family's app. The family could clearly view the reasons for the treatment adjustments and the expected goals through the app, alleviating their anxiety. At the same time, the intelligent analysis module has updated the patient's disease stage to "high-risk period for complications", increased the monitoring frequency of pancreatic duct drainage fluid amylase to once every 5 minutes, and shortened the physiological data sampling interval to 1 minute, forming dynamic monitoring.

[0076] On the 5th day post-surgery, under the precise monitoring and personalized guidance of the system, the patient's pancreatic duct drainage fluid amylase dropped to 800 U / L, and the body temperature returned to normal. The system downgraded the warning level to mild and pushed new rehabilitation guidance through the health management module. Considering the patient's 10-year history of diabetes before surgery, the system specially formulated a "low-sugar, high-protein" stepwise nutritional plan: on the first day, 200ml of rice water was given orally in 4 divided doses. After monitoring and finding no significant fluctuations in postprandial blood glucose, the amount was increased to 400ml on the second day, with the addition of 10g of enteral nutrition powder (diabetic-specific). At the same time, the proportion of intravenous nutrition was gradually reduced according to the drainage fluid volume. The system also pushed rehabilitation training videos to the bedside screen at regular intervals and adjusted the training intensity according to the patient's physical recovery. On the 1st day post-surgery, ankle pump exercises were guided in bed; on the 3rd day, assistance was provided to stand at the bedside; and on the 5th day, the patient was able to walk 50 meters with support. After each training session, the system prompted the patient to input a fatigue score and dynamically adjusted the training duration based on heart rate changes. When a patient reports "wound pain score of 4 when standing", the system immediately pushes the information to the nurses' station, and the nurses promptly adjust the analgesia pump parameters to ensure the smooth progress of rehabilitation training.

[0077] Throughout the treatment process, the data synchronization module ran silently in the background, ensuring patient privacy while updating health records. All patient monitoring data, imaging data, and treatment plans were stored on the department's local server. The system employed a federated learning model, uploading only homomorphically encrypted model update parameters to the hospital's cloud platform via a dedicated encrypted channel. For example, the patient's pancreatic fistula early warning data, after encryption, was used in conjunction with similar case data from other hospitals to optimize the multimodal model. The updated model parameters were then distributed to the department's system via the encrypted channel, increasing the local model's recognition accuracy from the initial 88% to 92%, while the patient's original data never left the hospital's intranet. On the 10th day post-surgery, the patient met the discharge criteria, and the system automatically generated a discharge health record containing "post-operative monitoring data summary, rehabilitation training continuation plan, and follow-up time nodes," which was synchronized to the community health service center's health management platform, providing complete data support for subsequent follow-ups.

[0078] In the second week after discharge, the patient used a simple smart drainage monitoring device provided by the community for follow-up monitoring at home. This device collects drainage fluid flow and color data, transmitting it encrypted via a 5G network to the original hospital's system. During one monitoring session, the system detected a sudden drop in bile duct drainage from 150ml / day to 50ml. It immediately sent a notification to the patient via an app, warning of "Be alert for bile duct obstruction; follow-up examination recommended within 24 hours," and synchronized the data to the attending physician's terminal. After remotely reviewing the data, the attending physician prioritized the patient's follow-up examination, confirming a T-tube stenosis, which was promptly treated, preventing the condition from worsening. The diagnostic and treatment information from this follow-up examination was also added to the patient's full-cycle health record via the data synchronization module.

[0079] In summary, the embodiments of this application clearly demonstrate the practical value of the intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries. It not only replaces medical staff in performing tedious manual monitoring of drainage fluid and physiological data, freeing them from repetitive tasks to focus on precise diagnosis and treatment, but also, through a sensitive risk identification and tiered early warning mechanism, quickly locates problems and matches appropriate medical resources when patients show signs of complications, assisting medical staff in timely intervention. Simultaneously, the system can generate dynamically adjusted personalized rehabilitation plans based on individual patient characteristics, covering the entire process of in-hospital treatment and out-of-hospital monitoring, providing patients with continuous rehabilitation guidance and risk alerts. This not only provides patients and their families with a reassuring medical experience but also effectively improves the overall quality of postoperative diagnosis, treatment, and rehabilitation management.

[0080] Next, referring to the accompanying drawings, a method for intelligent monitoring and health management of postoperative drainage fluid after hepatobiliary and pancreatic surgery, according to an embodiment of this application, is described.

[0081] like Figure 8 As shown, the method of the intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries includes the following steps:

[0082] In step S101, core specific indicators of hepatobiliary and pancreatic drainage fluid, visual images of drainage fluid, dynamic flow rate, and patient physiological data are collected.

[0083] It is understood that the embodiments of this application, by collecting core specific indicators of hepatobiliary and pancreatic drainage fluid, visualized images of drainage fluid, dynamic flow rate, and patient physiological data, can achieve multi-dimensional, full-scenario, and real-time capture of key information related to the postoperative patient's condition. Among them, the core specific indicators directly reflect the pathophysiological nature of the drainage fluid, providing a quantitative basis for judging the severity of the condition and the risk of complications; the visualized images intuitively present the visual characteristics of the drainage fluid's morphology, color, and turbidity, assisting in the rapid identification of visually perceptible abnormal changes; the dynamic flow rate data captures the temporal fluctuation pattern of drainage fluid secretion, and correlates it with key processes such as postoperative wound healing, biliary / pancreatic duct patency, and fluid absorption; the patient's physiological data constructs a baseline of the overall condition, and performs a linkage analysis between local drainage and overall health status. Ultimately, this provides high-quality and highly correlated data source support for the subsequent construction of multimodal models, ensuring the accuracy of risk identification, the personalization of rehabilitation plans, and the continuity of health management throughout the entire cycle, thus guaranteeing the clinical applicability and decision-making reliability of the entire system from the data source.

[0084] In step S102, based on the core specific indicators of hepatobiliary and pancreatic drainage fluid, the visualization of drainage fluid, dynamic flow and physiological data, a multimodal model specifically for hepatobiliary and pancreatic surgery is constructed. The model is deeply coupled with the temporal characteristics of the postoperative disease course to identify risks at different stages after surgery, and to accurately identify risks in stages. When abnormalities occur, risk assessment is triggered to obtain risk assessment results.

[0085] Among them, the postoperative multimodal model for hepatobiliary and pancreatic surgery is a dedicated intelligent model that integrates multi-source data such as imaging, clinical data, and laboratory data of patients after hepatobiliary and pancreatic surgery, and uses deep learning technology to achieve prognosis prediction, complication warning, and personalized diagnosis and treatment support.

[0086] The formula is:

[0087]

[0088]

[0089]

[0090]

[0091] in, This is the original multimodal data; This refers to the time point when the data was collected. The data dimensions include drainage fluid indicators, imaging quantitative values, drainage volume, and patient physiological data. For the first The global mean of the dimensional data; For the first Global standard deviation of dimensional data; For smoothing coefficients; These are the normalized multimodal features; for The fused feature vector at each time step; For modal type index; For postoperative Time of the first Modal stage weighting coefficients; These are the instantaneous feature weight coefficients; for Time of the first Instantaneous feature vectors of a modality; Historical feature weighting coefficients; For historical data collection time points; This is the timing decay coefficient; for Time of the first Historical feature vectors of a mode; The time interval between the current time t and the historical time s is the time difference. The smaller the interval, the closer the historical data is to the current time, and the higher its contribution to the current fusion feature. For a specific moment; for Risk assessment at any given moment; For activation functions; for Risk assessment weight matrix at any given time; for Risk assessment bias at any given time; for Risk level at any time (corresponding to a three-level early warning mechanism of light, medium and heavy); Thresholds are set for mild / moderate risk; Thresholds are used to classify moderate / severe risk.

[0092] It is understood that the embodiments of this application, by deeply integrating core specific indicators of drainage fluid, visualized images, dynamic flow, and patient physiological data, accurately couple the postoperative disease progression time sequence characteristics, enabling phased and precise identification and abnormal trigger-based assessment of risks at different rehabilitation stages. This provides core decision support for dynamic response priority matching of medical resources and precise push notifications to responsible terminals, and lays the data and algorithm foundation for the construction of feedback-driven, dynamically iterative, personalized rehabilitation plans. In addition, with the cooperation of federated learning and homomorphic encryption schemes, the model performance can be continuously optimized through parameter updates while ensuring the privacy and security of the original data without leaving the local area. This fully adapts to the specialized diagnosis and treatment needs after hepatobiliary and pancreatic surgery, improving the accuracy, efficiency, and safety of postoperative risk prevention and health management.

[0093] It should be noted that the deep coupling of postoperative disease course temporal characteristics to identify risks at different postoperative stages is based on the core specific indicators of the acquired drainage fluid, visualized images, dynamic flow, and patient physiological data. It integrates the patient's preoperative basic information, intraoperative intervention parameters, and dynamic monitoring data at different postoperative time points to construct a temporal data sequence. It deeply couples the physiological and pathological characteristics of the postoperative inflammatory period, the high-incidence period of complications, and the stable recovery period. Through multimodal modeling, it integrates the trend of changes in indicators in the temporal dimension with key nodes in the disease course, learns the normal data distribution range and abnormal feature patterns of each stage, accurately identifies specific abnormal signals related to infection, bile leakage, and pancreatic fistula in different disease stages, and performs targeted identification and differential diagnosis of risks at each postoperative stage, avoiding misjudging normal fluctuations in stages as pathological abnormalities.

[0094] Accurate risk identification is performed in stages, and risk assessment is triggered when abnormalities occur. The risk identification is based on the matching degree between the trend of indicator changes and the key nodes of the stage. When the monitoring data deviates from the normal threshold of the corresponding stage, meets the specific characteristics of complications, or shows abnormality in multiple dimensions of indicators, the risk assessment process is automatically triggered. The risk level assessment result is formed by quantifying the degree of abnormality, duration, probability of association with complications, and the patient's individual basic risk factors.

[0095] For example, in a patient who underwent pancreaticoduodenectomy, core data were simultaneously collected using a multimodal model on the 3rd postoperative day: amylase in the drainage fluid was 320 U / L (more than 3 times the upper limit of normal serum; the normal threshold on the 3rd postoperative day should be ≤120 U / L), and total bilirubin was 12 μmol / L (within the normal range). The visual image showed a slightly bloody and turbid appearance (it should be a light red and clear fluid on the 1st-2nd postoperative day). The dynamic flow rate was 45 ml / 24h (significantly lower than 120 ml / 24h on the 1st postoperative day, but still higher than the normal range of 20-30 ml / 24h on the 3rd-5th day). Physiological data included a body temperature of 38.6℃ (normal <37.5℃), a heart rate of 98 beats / min (normal 60-100 beats / min), and a white blood cell count of 13 × 10⁻⁶. 9 / L (normal 4-10×10) 9 The model deeply coupled the time-series characteristics of 1-3 days post-surgery, tracking the continuous increase of amylase from 80 U / L on day 1 and 150 U / L on day 2 post-surgery. The characteristics of the drainage fluid changed from light red and clear to bloody and turbid. In addition, the body temperature and white blood cell count were abnormally synchronized. Combined with the specialty course characteristics of pancreatic fistula in 3-7 days after pancreaticoduodenectomy, it was accurately identified as a moderate risk of pancreatic fistula. The risk assessment results indicated that intervention within 4 hours was required to prevent the condition from progressing to severe pancreatic fistula.

[0096] In step S103, based on the risk assessment results, a three-tier mechanism of mild, moderate and severe is established. Combining the real-time workload of medical staff and spatial distance, a dynamic response priority is introduced to accurately match the status of medical staff resources. Resource-adaptive tiered and precise push is made to the responsible physician and nurse station terminals. The individual characteristics of patients and real-time monitoring data are integrated, and a feedback-driven, dynamically iterative personalized rehabilitation plan is constructed based on the patient's rehabilitation feedback.

[0097] Among them, the status of medical resources refers to the comprehensive representation of the current available quantity, configuration distribution, and operational efficiency of core resources such as medical personnel, medical equipment, and medical supplies in the medical and health scenario.

[0098] It is understood that the embodiments of this application, by real-time sensing of the workload status and spatial distribution distance of medical staff, combined with the three-level risk assessment results of mild, moderate and severe based on drainage fluid monitoring data, and the dynamic response priority mechanism, can achieve precise matching of medical resources with patients' risk needs. This avoids the waste of resources caused by low-risk patients occupying high-quality medical resources, while ensuring that high-risk patients receive timely diagnosis and treatment intervention, effectively reducing the probability of postoperative complications and adverse events. At the same time, reasonable and balanced resource allocation can reduce the workload of medical staff, improve the efficiency of diagnosis and treatment decisions, and, with the privacy protection mechanism of encrypted transmission and federated learning, provide efficient medical resource support for the dynamic iteration of personalized rehabilitation plans while ensuring the security of patient data. Ultimately, this leads to precise, efficient and safe full-cycle health management of postoperative patients.

[0099] It should be noted that a three-tiered mechanism of mild, moderate, and severe warnings is established. This mechanism is based on the degree, duration, and multi-dimensional synergy of abnormalities in core specific indicators of drainage fluid, dynamic flow, visual imaging characteristics, and patient physiological data, combined with the probability of complication association and the patient's individual baseline risk. Mild warnings are for cases where a single non-core indicator slightly deviates from the normal threshold for the corresponding disease stage, the abnormal duration is ≤24 hours, there are no clear complication-specific characteristics, vital signs are stable, and the probability of complication association is <30%, often indicating stage-specific physiological fluctuations or early mild abnormalities. Moderate warnings are for cases where core indicators moderately deviate from the threshold, two or more non-core indicators are abnormal in synergy, the abnormal duration is 24-72 hours, early signs of complications appear, vital signs are basically stable, and the probability of complication association is 30%-70%, indicating a potential risk of complication development. Severe warnings are for cases where core indicators significantly deviate from the threshold, multiple core indicators are severely abnormal, the abnormal duration is >72 hours, the typical characteristics of severe complications are met, vital signs are unstable, and the probability of complication association is >70%, indicating a high-risk condition such as acute infection or severe bile leakage / pancreatic fistula that endangers patient safety.

[0100] By combining real-time workload and spatial distance of medical staff, a dynamic response priority system is introduced to accurately match the status of medical staff resources. The system collects dynamic workload data of responsible medical staff in real time through the hospital information system, medical staff workstations, and mobile terminals. Simultaneously, it obtains the real-time location of medical staff and the physical spatial distance between them and the patient's ward / ICU based on the hospital's positioning system. A basic response weight is set based on a three-tiered warning system (severe > moderate > mild), and a dynamic response priority algorithm is introduced. This algorithm performs multi-dimensional weighted calculations of the warning level weight (severe > moderate > mild), real-time workload coefficient (idle < moderate < busy), and spatial distance coefficient (near < moderate < far) to generate a priority score for each suitable medical staff member. For severe warnings, priority is given to matching responsible physicians and nurses whose workload is idle / moderate and who are the closest in spatial distance. If the core medical staff workload is saturated, the priority is automatically extended to lower-level resources and simultaneously pushed to the nurse station terminal. For moderate warnings, while ensuring response resources for severe warnings, medical staff with moderate workloads and close proximity are matched. For mild warnings, response resources are rationally allocated based on the breaks in medical staff work to avoid consuming core emergency resources.

[0101] Resource-adaptive, tiered, and precise push notifications are sent to responsible physicians and nurse station terminals. Based on risk level assessment results and dynamic response priority ranking, the system first identifies target recipients matching the warning level and patient responsibility. For severe warnings, an emergency warning message containing details of abnormal core indicators, risk level, probability of complication association, patient real-time location, and targeted treatment suggestions is pushed to the mobile terminal of the responsible physician with the highest priority score via pop-up window and voice reminder. Simultaneously, it is displayed at the top of the nurse station terminal screen with a highlighted red icon, triggering the nurse station's audible and visual alarms. For moderate warnings, the message is pushed to the matching responsible physician via text pop-up window and message notification. The terminal for the attending physician and the corresponding ward nurses contains key abnormal data, risk trends, and basic treatment guidelines, which are displayed in the center of the nurse station terminal with a yellow indicator. For mild warnings, a text message is pushed to the terminal of the responsible nurse, along with an explanation of the abnormal indicators and observation suggestions, which is displayed normally in blue on the nurse station terminal. During the push process, the online status of the terminals is checked in real time. If the responsible physician's terminal does not read the message and the timeout occurs, the message is automatically pushed to the terminals of the superior physician and the person in charge of the nurse station. At the same time, the warning treatment status is updated on the nurse station terminal. This achieves precise matching of the push recipients, content, form, warning level, and resource status, ensuring that the tiered warning information is delivered efficiently and the treatment process is traceable.

[0102] By integrating individual patient characteristics with real-time monitoring data and incorporating patient feedback, a feedback-driven, dynamically iterative, personalized rehabilitation plan is constructed. First, individual patient characteristics are integrated with core indicators of drainage fluid, physiological data, and real-time monitoring data of imaging features to build a patient-specific rehabilitation baseline database. Subjective feedback and objective rehabilitation performance during the patient's recovery process are collected, and a feedback analysis model is used to assess the suitability of the current rehabilitation plan to the patient's recovery progress, identifying dimensions requiring optimization. Subsequently, based on the dynamic trends of real-time monitoring data, the core needs of patient rehabilitation feedback, and rehabilitation goals at different postoperative stages, the rehabilitation plan is adjusted in a targeted manner, forming a feedback-driven iterative optimization mechanism. This ensures that each plan update is tailored to the individual patient's recovery characteristics, providing rehabilitation guidance and real-time adaptation to the patient's recovery process, thus promoting the patient's recovery.

[0103] A feedback-driven, dynamically iterative, personalized rehabilitation program is constructed. Based on the patient's individual baseline characteristics, it integrates real-time monitoring data and phased risk and complication identification results. An initial program covering diet, exercise, medication, and follow-up examinations is generated based on hepatobiliary and pancreatic rehabilitation guidelines and a multi-center pathway database. By collecting patient feedback on rehabilitation implementation, medical and nursing recommendations, and follow-up examination results, reinforcement learning algorithms are used to correlate and analyze feedback data with rehabilitation effects, dynamically optimize program parameters, and calibrate rehabilitation priorities in conjunction with the latest risk assessment results.

[0104] For example, on the third day after surgery, a patient's drainage fluid bilirubin concentration reached 120 μmol / L (normal threshold ≤34 μmol / L), and the 2-hour dynamic flow rate suddenly increased to 80 ml / h. The system, combining the turbidity of the drainage fluid with physiological data of a heart rate of 112 beats / min, determined it to be a moderate risk and triggered a response. At this time, the system monitored the real-time status of medical resources: Physician A was currently seeing 4 postoperative patients (including 1 patient requiring special care), with a workload rate of 65%, located in the physician's office on the 3rd floor, 50 meters from the patient's ward; Physician B was seeing 2 ordinary postoperative patients, with a workload rate of 30%, but was on the 5th floor... The operating room is involved in the consultation, 80 meters away; Nurse C is responsible for 3 patients in the same ward, with no emergency treatment tasks, and a load rate of 40%. She is located at the nurse station on the 2nd floor, 20 meters away; The system uses a dynamic response priority algorithm to push treatment instructions to Nurse C's terminal first (priority 1), requiring her to arrive within 5 minutes to collect drainage fluid samples and assess the incision condition. At the same time, a risk warning is pushed to the responsible physician A, who has a moderate load but is closer in space (priority 2), specifying that follow-up treatment should be carried out within 10 minutes. This avoids overloading of high-load medical resources and ensures response timeliness through distance weighting, thus carrying out precise resource-adaptive push.

[0105] In step S104, based on the personalized rehabilitation plan, model update parameters are transmitted through an encrypted channel. Federated learning and homomorphic encryption schemes are used to protect privacy. The original data does not leave the local machine, and only the parameters are updated locally, thus improving the patient's full-cycle health record.

[0106] The encrypted channel transmission model refers to a structured data transmission framework that builds a dedicated transmission link based on encryption algorithms and security protocols, and achieves data confidentiality, integrity, and tamper resistance through data encryption, authentication, and transmission verification.

[0107] The formula is:

[0108]

[0109]

[0110]

[0111]

[0112] in, For public key based Paillier encryption function; The public key for encryption; The core parameters of the public key are composed of large prime numbers. and The product is obtained; This is a core parameter of the public key, and its value is... ; It is a large prime number; For the first Model update parameters for each local node; The modulus of the encryption algorithm is used to determine the range of the ciphertext space. For mathematical modulo operators; For the first Model update parameters for each node Provides cryptographic semantic security while leveraging n-limited random numbers The range of values ​​must be considered to ensure that the encryption result conforms to the specified values. Modulus constraints ensure the legitimacy and privacy of encrypted transmission and decryption; This is the encrypted global parameter set after aggregation. The number of local nodes participating in federated learning; This refers to the local node sequence number; For the first The weight coefficient of each node; Update the parameters of the global model obtained after decryption; For private key based Paillier's decryption function; The hash value is used for integrity verification. This is a SHA-256 hash function, which outputs a 256-bit digest. The parameter transmission timestamp is accurate to milliseconds to prevent replay attacks; A unique identifier for the local node (e.g., patient ID + device number) is used to ensure the binding relationship between parameters and nodes.

[0113] It is understood that the embodiments of this application provide secure transmission guarantees for model iteration under federated learning, shield against the risk of data leakage and tampering through encryption technology, ensure the integrity and confidentiality of parameters, adhere to the principle that the original data does not leave the local area, avoid the leakage of sensitive medical data of patients, comply with medical data security regulations, and support multimodal model optimization, personalized rehabilitation program iteration and full-cycle health record improvement, taking into account both system compliance and reliability.

[0114] It should be noted that federated learning and homomorphic encryption are used to protect privacy. The federated learning distributed architecture is adopted, and the patient's original monitoring data is stored locally and not transmitted across institutions. The central node issues the initial model parameters. After each local node completes local training based on its own data, it encrypts the local gradient and parameter update values ​​through the homomorphic encryption algorithm and uploads them to the central node to complete parameter aggregation in the encrypted domain. The generated global update parameters are then encrypted and sent back to each node.

[0115] For example, the patient bedside monitoring terminal locally stores core indicators such as bilirubin concentration (12.3 μmol / L) and protein quantification (2.1 g / L) of drainage fluid, as well as visualized image features and physiological data such as heart rate (78 beats / min). Based on personalized rehabilitation plans, iteratively generates a 32-dimensional feature weight update vector (mean 0.023, variance 0.0012) and risk identification threshold adjustment parameters (0.68→0.72) for a multimodal model. The data is transmitted through a TLS1.3 encrypted channel conforming to the HIPAA standard + Paillier homomorphic encryption (key length 2048 bits). After Top-K sparsity compression, the parameters are 580KB in size with a transmission latency of 62ms. Combined with differential privacy noise injection of ε=0.45, parameter aggregation is completed among the federated learning nodes of three collaborating hospitals. The original data is stored on the local server throughout the process, and the privacy leakage risk is controlled below 0.028%. The aggregated updated parameters feed back into the local model, improving the accuracy of drainage fluid abnormality identification by 1.7 percentage points.

[0116] The method for intelligent monitoring and health management of postoperative drainage fluid in hepatobiliary and pancreatic surgery proposed in this application comprehensively captures core specific indicators, visualized images, dynamic flow, and patient physiological data of hepatobiliary and pancreatic drainage fluid through a data acquisition module, providing multi-dimensional and full-scenario data support for postoperative monitoring. Based on a specialized multimodal model constructed by an intelligent analysis module, it deeply couples the temporal characteristics of the disease course to accurately identify and assess risks at different postoperative stages, overcoming the limitations of traditional monitoring that relies on experience-based judgment. Utilizing a three-tiered early warning mechanism (mild, moderate, severe) and incorporating real-time medical staff workload and spatial distance to introduce dynamic response priorities, it accurately adapts and grades medical resources. Push notifications significantly reduce the risk of missed diagnoses of complications and improve medical response efficiency. The health management module integrates individual patient characteristics, real-time monitoring data, and rehabilitation feedback to construct feedback-driven, dynamically iterative, personalized rehabilitation plans, providing precise adaptation to rehabilitation guidance and patient recovery progress. Utilizing the encrypted transmission channel and federated learning and homomorphic encryption schemes of the data synchronization module, while ensuring that original data remains locally and privacy is controllable, model parameter updates and the improvement of patients' full-cycle health records are completed. This effectively improves the rehabilitation outcomes and safety of patients after hepatobiliary and pancreatic surgery, reduces the workload of medical staff, and provides comprehensive technical support for intelligent, precise, and safe management of postoperative care in specialized departments. Thus, it solves the problems of rigid early warning mechanisms, insufficient data security, and inadequate full-cycle management in existing technologies.

[0117] The following will illustrate the method of intelligent monitoring and health management system for postoperative drainage fluid in hepatobiliary and pancreatic surgeries through a specific embodiment, such as... Figure 9 As shown, it includes:

[0118] A 52-year-old patient underwent a laparoscopic right hemihepatectomy for hepatocellular carcinoma at a tertiary hospital's hepatobiliary surgery department. Upon returning to the ICU for postoperative monitoring, the head nurse installed an "Intelligent Monitoring and Health Management System for Postoperative Hepatobiliary and Pancreatic Drainage Fluid." This system fundamentally changed the traditional model of drainage tube care, which relied on manual observation, building a precise protective barrier for the patient's postoperative recovery. One hour after the surgery, the nurse connected the intelligent drainage bag, which integrates multiple sensors, to the patient's abdominal drainage tube. The photoelectric sensor on the outside of the drainage bag immediately began working, collecting RGB color data and turbidity parameters of the drainage fluid every 30 seconds. The built-in electromagnetic flowmeter recorded the dynamic flow rate of the drainage fluid in real time with an accuracy of 1 ml, and simultaneously transmitted it to the bedside monitoring terminal. At the same time, a wrist-type physiological monitor wirelessly connected to the terminal continuously collected his heart rate, blood pressure, blood oxygen saturation, and body temperature data, forming a complete data packet every 5 minutes. On the morning of the first postoperative day, the system automatically triggered a reminder to collect drainage fluid samples. Following the instructions, the nurse used a special blood collection tube to obtain a 5ml sample from the sterile sampling port of the drainage bag. After being sent to the laboratory, a dedicated analyzer completed the testing of eight core specific indicators for hepatobiliary and pancreatic diseases, including bilirubin, amylase, and albumin, within 20 minutes. This data, along with images of the sediment after centrifugation of the drainage fluid taken by the laboratory, were imported into the system database through an encrypted interface on the hospital's intranet. At this time, the patient's drainage fluid was slightly bloody, with a 24-hour drainage volume of approximately 350ml and a bilirubin concentration of 12μmol / L. All physiological indicators were stable, and the system terminal displayed an assessment result of "stable in the early postoperative period, low risk level." The attending physician, after viewing the information via a mobile app, marked the order "Continue the current fluid replacement regimen" in the system. This information was simultaneously updated to the patient's electronic medical record.

[0119] At 2:00 AM on the third day after surgery, the patient's bedside monitoring terminal suddenly emitted a slight yellow warning. At this time, he was asleep. The on-duty nurse noticed the abnormal signal on the central monitoring screen at the nurses' station—the system showed that the drainage fluid color had changed from light bloody to golden yellow, and the drainage volume had increased from 15 ml to 48 ml within one hour. At the same time, the bilirubin concentration had risen to 38 μmol / L compared to two hours earlier, and his heart rate had also increased from 72 beats / min to 89 beats / min. This change immediately triggered the system's multimodal model analysis. This model is a proprietary algorithm trained on 120,000 postoperative cases of hepatobiliary and pancreatic surgery in five tertiary hospitals across the country. It can deeply couple the temporal characteristics of the postoperative course of the disease, comparing the currently collected drainage fluid visualization image (golden yellow liquid, no obvious flocculent matter), dynamic flow curve (sudden increase trend), specific indicators (elevated bilirubin), and physiological data (increased heart rate) with a typical risk feature database of 3-5 days after surgery. The model, through time-series analysis, found that the changes in the characteristics and indicators of the drainage fluid matched the features of early bile leakage with a 87% match rate, but serious complications such as fever and hypotension had not yet appeared. Therefore, a "moderate risk (suspected early bile leakage)" assessment result was quickly generated, and a three-level response mechanism was automatically activated. The system backend immediately retrieved the real-time status of medical resources: the attending physician lived in the hospital staff dormitory, only 5 minutes away from the ICU, and had no emergency duties at present; the nearest on-duty nurse was providing basic care to bed 3, which was expected to be completed within 10 minutes; while the ICU chief physician was treating a patient with severe pancreatitis, and the workload was saturated. Based on the dynamic response priority algorithm, the system first pushed a high-priority warning to the attending physician's mobile APP, with the prompt "Sudden change in drainage fluid characteristics, suspected early bile leakage, immediate bedside assessment recommended," and at the same time pushed a secondary reminder to the on-duty nurse, requiring them to complete the patient's vital signs verification within 10 minutes. The attending physician arrived at the ICU within 6 minutes of receiving the notification. Combining the 48-hour drainage fluid index change curve and real-time images provided by the system, the physician performed an abdominal palpation on the patient and issued an emergency ultrasound examination request. The system automatically synchronized the examination request to the ultrasound department's on-call system and marked it with a "priority treatment" label. The ultrasound department completed the bedside examination within 15 minutes and diagnosed "minor bile leak in the liver section". After this diagnosis result was transmitted back to the intelligent monitoring system through an encrypted channel, the model immediately updated the risk assessment to "moderate risk confirmed, intervention required".

[0120] In response to this situation, the system activated the personalized rehabilitation plan generation module based on the patient's individual characteristics. The patient had a 10-year history of hepatitis B, a preoperative liver function classification of A, and an intraoperative blood loss of approximately 200ml. Postoperatively, the patient reported "mild abdominal distension and pain," all of which were incorporated into the system's analysis. The module first retrieved the standard intervention plan for bile leakage after hepatobiliary and pancreatic surgery, and then dynamically adjusted it based on real-time monitoring data: In terms of treatment, it was recommended to adjust the somatostatin dose from 250μg / h to 500μg / h, while increasing intravenous vitamin K1 infusion to protect liver function; in terms of nursing care, based on his weight (68kg) and drainage fluid loss, it was recommended to increase the daily fluid intake by 500ml, including 100ml of albumin to maintain colloid osmotic pressure; in terms of activity guidance, considering the risk of bile leakage, the original "get out of bed 3 times a day" was adjusted to "self-turning over in bed, with limited sitting up," and a timed reminder function was set. After the attending physician confirmed the treatment plan within the system, the nursing station terminal immediately generated a medication execution order and nursing plan. The medication delivery system simultaneously received the medication dispensing notification. The entire process, from risk identification to intervention initiation, took only 28 minutes. On the 5th post-operative day, the patient's bilirubin concentration in the drainage fluid dropped to 18 μmol / L, and the drainage volume stabilized at around 25 ml / h. The system detected significant heart rate fluctuations when the patient attempted bed activity. Combined with the patient's reported dizziness after activity, the system automatically adjusted the activity intensity in the rehabilitation plan to "5 minutes of sitting up in bed each time, twice daily," and added a monitoring item for "blood pressure fluctuations during positional changes." While administering nursing care, the attending nurse recorded "patient's appetite improved, able to eat small amounts of rice porridge" through the system's feedback portal. This information was captured by the model, and the nutritional support module immediately pushed out "semi-liquid diet transition suggestions," including specific content such as controlling daily protein intake to 50g and prioritizing whey protein, forming a feedback-driven dynamic iterative mechanism.

[0121] Throughout the entire diagnosis and treatment process, the system's privacy protection mechanism operates silently in the background. All patient drainage fluid data, physiological indicators, and imaging data are stored on the hospital's local server, and no raw data is transmitted to external platforms. When the health department needs to access this case data for a multi-center clinical study, the system uses a federated learning framework. The lead hospital initiates a model training request, and the system node of the patient's hospital, protected by homomorphic encryption algorithms, only transmits the gradient update parameters of the local model to the joint training platform via a Virtual Private Network (VPN). These parameters are encrypted and cannot be used to deduce the original data content. For example, when training the "sub-model for predicting bile leakage after hepatectomy in hepatitis B patients," the hospital node only contributes the model weight adjustment values ​​and feature correlation parameters, ensuring that the patient's personal information, medical history, and treatment records are completely secure. On the 12th day after surgery, the patient's drainage fluid indicators returned to normal. After evaluation, the drainage tube was removed. The system automatically integrated his postoperative monitoring data, risk intervention records, rehabilitation program iteration process, and follow-up plan to generate a "Full-Cycle Health Management Report after Hepatobiliary and Pancreatic Surgery." This report not only included traditional text records but also included visual content such as dynamic curves of drainage fluid indicators and risk warning timelines, providing a complete basis for subsequent follow-up. Upon discharge, the nurse activated the system's home follow-up module for the patient. He could upload his temperature, diet, and activity status daily via a mobile app. The system then pushed follow-up reminders based on the characteristics of the postoperative recovery sequence. When he reported "decreased appetite" one month after surgery, the system immediately pushed the information to the follow-up physician. Combining this with his baseline monitoring data before discharge, the system determined a "risk of insufficient nutritional intake" and remotely adjusted his diet plan, preventing a recurrence of the condition.

[0122] In summary, this application's embodiments capture abnormalities in drainage fluid and physiological status through multi-dimensional monitoring, rapidly identify risks and initiate tiered responses based on a dedicated model, efficiently match medical resources for early intervention, and avoid serious complications. Furthermore, it can dynamically generate and iterate personalized rehabilitation plans covering treatment, nursing care, and other aspects, based on individual patient characteristics, thus providing feedback. Simultaneously, the system significantly improves medical efficiency, enabling it to focus on individualized treatment. It leverages federated learning and homomorphic encryption to protect patient privacy, providing continuous protection whether the patient is hospitalized or at home, comprehensively optimizing medical quality and supporting rehabilitation.

[0123] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0124] The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0125] When the processor 1002 executes the program, it implements the method for intelligent monitoring and health management system of postoperative drainage fluid after hepatobiliary and pancreatic surgery provided in the above embodiments.

[0126] Furthermore, electronic devices also include:

[0127] Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0128] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0129] The memory 1001 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0130] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0131] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0132] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0133] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for intelligent monitoring and health management of postoperative drainage fluid in hepatobiliary and pancreatic surgery.

[0134] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described method for intelligent monitoring and health management of postoperative drainage fluid in hepatobiliary and pancreatic surgery.

[0135] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0136] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0137] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0138] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0139] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0140] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A system for intelligent monitoring and health management of post- hepatobiliary-pancreatic surgery drainage fluid, characterized in that, The application relates to a liver, gallbladder and pancreas postoperative health management system. The data acquisition module is used for acquiring liver, gallbladder and pancreas postoperative drainage liquid core specificity indexes, drainage liquid visualization images, dynamic flow and patient physiological data. The intelligent analysis module is used for constructing a liver, gallbladder and pancreas postoperative exclusive multi-modal model, deeply coupling postoperative disease course time sequence characteristic identification postoperative different stage risks, and performing stage-by-stage risk accurate identification. The early warning module is used for setting a light, medium and heavy three-stage mechanism, combining medical staff real-time load and space distance, introducing a dynamic response priority to accurately match the medical staff resource state, and performing resource adaptive type stage-by-stage accurate pushing to a responsible physician and a nurse station terminal. The health management module is used for fusing patient individual characteristics and real-time monitoring data, combining patient rehabilitation feedback to construct a feedback-driven dynamic iteration individualized rehabilitation scheme, real-time adapting a patient recovery process, and performing rehabilitation guidance accurate adaptation. The data synchronization module is used for transmitting model update parameters through an encryption channel, adopting a federal learning and homomorphic encryption scheme to protect privacy, and not taking original data out of the local, only updating parameters in the local, and perfecting a patient whole-cycle health record. The data acquisition module comprises a core index acquisition unit, an image acquisition unit, a flow monitoring unit and a physiological data acquisition unit, wherein the core index acquisition unit is used for acquiring amylase, bilirubin, white blood cell count, protein concentration and pancreatic enzyme activity data in the drainage liquid; the image acquisition unit is used for dynamically capturing drainage liquid color, turbidity, flocculent distribution, precipitate morphology and blood component visualization features; the flow monitoring unit is used for real-time acquisition of drainage liquid instantaneous flow, cumulative drainage total amount and flow rate change trend; and the physiological data acquisition unit is used for acquisition of patient body temperature, heart rate, blood pressure, respiration rate and blood oxygen saturation data.

2. The intelligent monitoring and health management system for postoperative biliary drainage fluid according to claim 1, characterized in that, The intelligent analysis module comprises a multi-modal data fusion unit, a time sequence disease course modeling unit, a stage-by-stage risk identification unit and a risk assessment unit, wherein the multi-modal data fusion unit is used for cross-modal feature alignment of drainage liquid core specificity indexes, visualization images, dynamic flow and patient physiological data, and extraction of high-dimensional correlation features among the data; the time sequence disease course modeling unit is used for constructing a liver, gallbladder and pancreas postoperative exclusive multi-modal model according to liver, gallbladder and pancreas postoperative rehabilitation pathological mechanism, and deeply coupling time sequence dependence features and stage specificity rules of postoperative disease course progress; the stage-by-stage risk identification unit is used for accurately identifying specificity risks of drainage related complications and recovery lag in each stage in combination with pathological characteristic differences in postoperative acute stage, recovery stage and post-recovery stage, quantifying risk probability and judging data abnormality level; and the risk assessment unit is used for taking the data abnormality level as a core judgment basis, integrating multi-dimensional feature data for comprehensive judgment, and generating a quantitative risk assessment result.

3. The intelligent monitoring and health management system for postoperative biliary drainage fluid according to claim 1, characterized in that, ​ 4. The intelligent monitoring and health management system for postoperative biliary drainage fluid according to claim 1, characterized in that, The early warning module comprises a hierarchical standard definition unit, a medical care state perception unit, a dynamic priority scheduling unit, and a precise push adaptation unit. The hierarchical standard definition unit determines the judgment threshold and trigger condition of light, medium, and heavy three-level early warning according to the abnormality degree of post-hepatobiliary-pancreatic surgery drainage fluid, patient physiological data fluctuation range, and risk assessment results, and obtains an early warning level. The medical care state perception unit constructs a dynamic database of medical care resources by collecting real-time data of the current workload, reception task volume, and spatial position of medical personnel. The dynamic priority scheduling unit establishes a dynamic response priority algorithm model by combining the early warning level, real-time load, and spatial distance of medical personnel, and performs precise matching of early warning information and medical care resource state. The precise push adaptation unit pushes early warning information of corresponding levels to the terminal of the responsible physician or the nurse station system according to the priority matching results, and supports optimization of information display forms according to terminal types.

5. The intelligent monitoring and health management system for postoperative biliary drainage fluid according to claim 1, characterized in that, The health management module comprises an individual characteristic integration unit, a rehabilitation scheme dynamic generation unit, and a rehabilitation feedback interaction unit. The individual characteristic integration unit is responsible for sorting out patient age, underlying disease, and surgery type information, synchronously associating drainage fluid monitoring data and real-time dynamic data of physiological indicators, and constructing a comprehensive patient state dataset. The rehabilitation scheme dynamic generation unit constructs an initial individualized rehabilitation scheme based on the integrated patient state dataset, calls a specialized rehabilitation knowledge base, and dynamically adjusts the scheme content according to changes in real-time monitoring data. The rehabilitation feedback interaction unit collects feedback on the rehabilitation experience and execution of patients regarding pain degree, diet, and sleep through the interface, converts the feedback information into parameters recognizable by a post-hepatobiliary-pancreatic surgery exclusive multi-modal model, and iteratively optimizes the initial individualized rehabilitation scheme.

6. The intelligent monitoring and health management system for postoperative biliary drainage fluid according to claim 1, characterized in that, The data synchronization module comprises an encryption transmission unit, a privacy protection unit, and a parameter archive unit. The encryption transmission unit builds a secure and stable encryption channel and transmits model update parameters through the encryption channel. The privacy protection unit deploys a federated learning and homomorphic encryption scheme, strictly limits the flow range of raw data, and ensures that raw data does not leave the local area. The parameter archive unit receives and processes the synchronized model update parameters, iteratively updates the local parameters, integrates relevant data into the patient health record, and continuously improves the patient's whole-cycle health data record.

7. A method applied to the liver, gallbladder and pancreas postoperative drainage fluid intelligent monitoring and health management system of any one of claims 1-6, characterized in that, It comprises: Collecting hepatobiliary-pancreatic specialty drainage fluid core specificity indicators, drainage fluid visualization images, dynamic flow, and patient physiological data; Based on the hepatobiliary-pancreatic specialty drainage fluid core specificity indicators, drainage fluid visualization images, dynamic flow, and physiological data, a post-hepatobiliary-pancreatic surgery exclusive multi-modal model is constructed to deeply couple postoperative course time sequence feature recognition and postoperative risk in different stages, perform stage-based risk precise identification, trigger risk assessment when abnormal, and obtain risk assessment results. According to the risk assessment result, a light, medium and heavy three-level mechanism is set, real-time load and spatial distance of medical staff are combined, dynamic response priority is introduced to accurately match the state of medical staff resources, resource adaptive type hierarchical accurate push is carried out to the terminal of the responsible physician and nurse station, individual characteristics and real-time monitoring data of patients are fused, and a feedback-driven dynamic iterative personalized rehabilitation plan is constructed combined with patient rehabilitation feedback; Based on the personalized rehabilitation plan, the model update parameters are transmitted through an encryption channel, the privacy is protected by using federated learning and homomorphic encryption scheme, the original data does not go out of the local, and the local only updates the parameters, and the patient's whole cycle health record is improved.

8. An electronic device, comprising: It comprises: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to realize the method of the liver, gallbladder and pancreas postoperative drainage fluid intelligent monitoring and health management system.

9. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions are executed to realize the method of the liver, gallbladder and pancreas postoperative drainage fluid intelligent monitoring and health management system.

10. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are executed to realize the method of the liver, gallbladder and pancreas postoperative drainage fluid intelligent monitoring and health management system.

Images