A method for comprehensive monitoring of fluid intake and output of a pancreatitis patient

Abstract

The application provides a kind of pancreatitis patient fluid intake and output quantity comprehensive monitoring method, comprising the following steps: obtaining the fluid intake and output quantity data and blood gas analysis data of pancreatitis patient and other data;Set liquid demand quantity prediction model, according to the preprocessed multidimensional monitoring data set and liquid demand quantity prediction model, liquid demand quantity prediction is carried out;Set the time series risk early warning model based on LSTM, according to the preprocessed multidimensional monitoring data set and liquid demand quantity prediction model, risk prediction is carried out;According to the prediction result of risk prediction, generate liquid management early warning information and fluid replacement scheme adjustment suggestion;According to the prediction result of liquid demand quantity prediction, the risk probability of future occurrence of liquid overload, hypovolemia or abdominal cavity high pressure syndrome of patient is predicted.The application can realize the intelligent prediction of individualized liquid demand quantity and the early dynamic early warning of complication risk, generate liquid management suggestion, to assist clinical precise and predictive liquid treatment.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to a method for comprehensive monitoring of fluid intake and output in patients with pancreatitis. Background Technology

[0002] Acute pancreatitis, especially in moderate to severe cases, often involves systemic inflammatory response syndrome and capillary leakage at its core pathophysiology, leading to a sharp decline in effective circulating blood volume and insufficient organ perfusion. Therefore, early and precise fluid resuscitation is a cornerstone treatment aimed at maintaining tissue perfusion and preventing organ dysfunction.

[0003] Currently, mainstream early goal-oriented treatment strategies rely on segmented assessments and empirical adjustments of indicators such as heart rate, mean arterial pressure, urine output, and lactate. This model has several limitations. On the one hand, monitoring data is fragmented and discontinuous, requiring healthcare professionals to rely on timed recordings and manual calculations of fluid intake and output, resulting in delayed information acquisition and difficulty in grasping the patient's dynamic volume status in real time. On the other hand, the decision-making process heavily depends on the clinician's personal experience, lacking objective and quantifiable individualized fluid resuscitation targets, which can easily lead to undertreatment or overtreatment. While existing technologies include independent infusion pumps and monitors, the data are isolated and cannot be integrated for analysis. Some studies have attempted to apply static predictive formulas or simple alarm thresholds, but these cannot handle complex, time-varying clinical scenarios, let alone provide early warnings of evolving risk trends. Summary of the Invention

[0004] This invention provides a comprehensive monitoring method for fluid intake and output in patients with pancreatitis, which can achieve intelligent prediction of individualized fluid requirements and early dynamic warning of complication risks, and generate fluid management suggestions to assist clinicians in carrying out precise and predictive fluid therapy.

[0005] The first aspect of this invention provides a method for comprehensive monitoring of fluid intake and output in patients with pancreatitis, comprising the following steps: Data on fluid intake and output, non-invasive intra-abdominal pressure, vital signs, and blood gas analysis of patients with pancreatitis were acquired as a multidimensional monitoring dataset; the multidimensional monitoring dataset was preprocessed. Set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. Set up an LSTM-based time series risk warning model to predict risks based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. Obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results; Obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, predict the probability of patients developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

[0006] Furthermore, the fluid demand prediction model includes a machine learning-based regression model; the machine learning-based regression model is trained using historical patient data and outputs a predicted individualized fluid demand.

[0007] Furthermore, the process of predicting liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model includes the following steps: Feature variables related to fluid requirements were extracted from the preprocessed data, including heart rate, mean arterial pressure, hematocrit, lactate, urine output, intra-abdominal pressure trend, and blood gas analysis indicators. Input the feature variables into the trained liquid demand prediction model; The model outputs a predicted fluid requirement; the prediction is adjusted based on the patient's real-time fluid balance.

[0008] Furthermore, the LSTM-based time series risk warning model includes one or more LSTM layers, Dropout layers, and fully connected layers, and outputs classification probabilities through sigmoid or softmax activation functions.

[0009] Furthermore, the risk prediction based on the preprocessed multi-dimensional monitoring dataset and the LSTM-based time series risk warning model includes the following steps: The preprocessed time series data is organized into a multidimensional matrix according to time windows and then input into the LSTM model. The model learns sequence dependencies and outputs the probability of the target complication occurring in the future. When the probability exceeds a preset threshold, an early warning signal is triggered.

[0010] Furthermore, the step of generating liquid management early warning information and liquid replenishment plan adjustment suggestions based on the risk prediction results includes the following steps: When a risk of fluid overload is predicted, warning information is generated, including recommendations to reduce the rate of fluid administration and to assess the use of diuretics. When a risk of hypovolemia is predicted, warning information is generated, including suggestions to increase the rate of fluid resuscitation and adjust the composition of the fluid. When a risk of intra-abdominal hypertension syndrome is predicted, early warning information is generated, including recommendations to monitor intra-abdominal pressure and assess the necessity of abdominal drainage.

[0011] Furthermore, the prediction of the patient's future risk of developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome based on the predicted fluid demand includes the following steps: The predicted fluid requirements were integrated with the patient's fluid balance, intra-abdominal pressure, and vital signs data for analysis. Use SHAP values ​​or similar interpretability methods to analyze the contribution of each feature to the prediction results; Based on contribution and clinical threshold, calculate the probability of various complications occurring in a specific future period; Output the risk level and confidence interval.

[0012] The second aspect of this invention provides a comprehensive monitoring system for fluid intake and output in patients with pancreatitis, including a first processing module for acquiring fluid intake and output data, non-invasive intra-abdominal pressure data, and vital sign data of patients with pancreatitis as a multi-dimensional monitoring dataset; and preprocessing the multi-dimensional monitoring dataset. The second processing module is used to set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. The third processing module is used to set up an LSTM-based time series risk warning model and perform risk prediction based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. The fourth processing module is used to obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results of risk forecasting. The fifth processing module is used to obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, to predict the probability of the patient developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

[0013] A third aspect of the present invention provides a computer device, comprising: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory, including executing the above-described method for comprehensive monitoring of fluid intake and output in pancreatitis patients; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

[0014] A fourth aspect of the present invention provides a readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the above-described method for comprehensive monitoring of fluid intake and output in a pancreatitis patient.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention constructs a full-chain system from multi-dimensional data automatic collection and intelligent analysis to clinical decision support, transforming discrete clinical information into continuous, dynamic, and measurable decision-making basis, thereby systematically improving the accuracy and safety of fluid management in severe pancreatitis.

[0016] First, by integrating multi-source data such as bedside monitoring equipment, infusion pumps, smart urine meters, and non-invasive intra-abdominal pressure sensors, this invention achieves automated and continuous comprehensive monitoring of key indicators of the patient's volume status and abdominal environment, breaking through the information barriers and lags caused by traditional manual recording and fragmented assessment.

[0017] Secondly, by introducing and synergistically applying a fluid requirement prediction model and an LSTM-based time-series risk warning model, this invention provides current patients with an objective and quantitative individualized fluid resuscitation benchmark by learning from historical successful experiences. This helps reduce the subjective differences and uncertainties caused by clinicians relying solely on experience. Furthermore, it fully leverages the advantages of deep learning in processing time-series data, enabling the early identification of subtle and complex precursor patterns pointing to fluid overload, hypovolemia, or intraperitoneal hypertension syndrome from the dynamic changes in vital signs and monitoring parameters.

[0018] Finally, this invention automatically transforms the model's prediction results into specific fluid management early warning information and fluid resuscitation plan adjustment suggestions, and quantifies the risk probability of future complications, reducing the risk of clinical errors caused by information overload or judgment delays. It also promotes the homogenization and standardization of treatment through standardized suggestion output.

[0019] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from an examination of the following, or may be learned from the practice of the invention. Attached Figure Description

[0020] Figure 1 The method flowchart provided by the present invention. Detailed Implementation

[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] Example 1 The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The method in this application will be described from the perspective of system implementation below. As shown in the figure, a method for comprehensive monitoring of fluid intake and output in pancreatitis patients includes the following steps: Data on fluid intake and output, non-invasive intra-abdominal pressure, vital signs, and blood gas analysis of patients with pancreatitis were acquired as a multidimensional monitoring dataset; the multidimensional monitoring dataset was preprocessed. Set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. Set up an LSTM-based time series risk warning model to predict risks based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. Obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results; Obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, predict the probability of patients developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

[0023] First, the system can connect to bedside monitoring devices, infusion pumps, smart urine meters, and non-invasive intra-abdominal pressure sensors to acquire real-time data on fluid intake and output, indirectly measured intra-abdominal pressure, and vital signs such as heart rate, blood pressure, and blood oxygen saturation in pancreatitis patients, forming a multi-dimensional monitoring dataset. This multi-dimensional monitoring dataset undergoes preprocessing steps such as data cleaning, time alignment, missing value handling, and standardization to transform it into standardized time-series data suitable for model analysis. Arterial blood gas analysis is an indicator used to assess and measure the body's acid-base balance, and can be used to determine whether a patient has respiratory failure and metabolic disorders.

[0024] The fluid requirement prediction model calculates individualized future fluid requirements for each patient based on preprocessed data using an embedded machine learning algorithm, such as an ensemble learning model trained on historical data. The time-series risk warning model based on long short-term memory networks analyzes the temporal characteristics and dynamic patterns of the input data to predict the probability of target complications such as fluid overload, hypovolemia, or intraperitoneal hypertension in subsequent time periods.

[0025] In the output phase, the calculation results of the above model are transformed into clinically usable information. On the one hand, based on the comparison between the risk probability output by the risk warning model and the preset threshold, corresponding fluid management warning information and specific suggestions for adjusting the fluid replacement plan are automatically generated. On the other hand, by combining the predicted fluid demand results with the patient's real-time status data, the overall risk probability of the patient developing various complications in the future is further analyzed and quantified.

[0026] Example 2 The difference between this embodiment and Embodiment 1 is that the fluid demand prediction model includes a machine learning-based regression model; the machine learning-based regression model is trained using historical patient data and outputs a predicted individualized fluid demand.

[0027] The fluid requirement prediction model, based on a machine learning regression algorithm, learns the complex nonlinear mapping relationship between multidimensional clinical characteristics from a large number of historical pancreatitis patient cases and the fluid requirements ultimately validated clinically. During operation, the model receives preprocessed real-time feature data. Through decision trees, the model performs layer-by-layer filtering and combination calculations on the input features, ultimately outputting a specific, personalized fluid requirement prediction (usually in milliliters). This extracts successful treatment experiences implicit in historical data into a computable model, thus providing an objective quantitative reference benchmark for current patients' fluid resuscitation, reducing the uncertainty and individual differences inherent in judgments based solely on clinical experience.

[0028] Example 3 The difference between this embodiment and Embodiment 2 is that the liquid demand prediction based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model includes the following steps: Feature variables related to fluid requirements were extracted from the preprocessed data, including heart rate, mean arterial pressure, hematocrit, lactate, urine output, intra-abdominal pressure trend, and blood gas analysis indicators. Input the feature variables into the trained liquid demand prediction model; The model outputs a predicted fluid requirement; the prediction is adjusted based on the patient's real-time fluid balance.

[0029] During operation, the system first extracts key clinical features most relevant to fluid resuscitation. For example, vital signs and perfusion indicators, such as heart rate and mean arterial pressure reflecting circulatory status, and lactate levels indicating tissue perfusion and hypoxia, are used. Direct indicators of volume status, such as urine output reflecting renal perfusion and fluid balance, are also included. Hemoconcentration indicators, such as hematocrit, help determine the presence of hemoconcentration and thus assess the degree of fluid deficit. Intra-abdominal pressure indicators, such as intra-abdominal pressure trends, are crucial for assessing the risk of intra-abdominal hypertension and directly affect the safety margins of fluid management. By selecting these features with clear physiological and clinical significance, the model can focus on the core information determining fluid requirements, reducing noise interference and improving the specificity and accuracy of predictions. The blood gas analysis data includes pH, arterial partial pressure of carbon dioxide, arterial partial pressure of oxygen, bicarbonate concentration, and base excess. The system assesses the patient's respiratory / metabolic acid-base imbalance in real time based on indicators such as pH, PaCO2, and HCO3⁻, and uses this status as one of the input features for the fluid requirement prediction model and risk warning model to improve prediction accuracy and clinical relevance.

[0030] The extracted feature variables are combined into a feature vector and input into a trained fluid requirement prediction model, such as a machine learning-based regression model. Internally, the model performs comprehensive calculations on the input features based on the complex nonlinear mapping relationships learned during training. This transforms the current, multidimensional patient state into a specific, quantifiable predicted fluid requirement (typically in milliliters per hour or milliliters per day). The numerical value represents the model's best estimate of the current patient's required fluid supplementation based on historical successful treatment experience.

[0031] The predicted fluid requirement output by the model is a baseline value. The system dynamically adjusts this requirement based on the patient's real-time fluid balance, such as net intake over the past few hours. For example, if the patient is currently in a significantly positive balance, the predicted fluid replacement volume may be appropriately reduced in practice to achieve individualized and dynamic management.

[0032] Example 4 The difference between this embodiment and Embodiment 3 is that the LSTM-based time series risk warning model includes one or more LSTM layers, Dropout layers, and fully connected layers, and outputs classification probabilities through sigmoid or softmax activation functions.

[0033] The LSTM-based time series risk warning model includes one or more LSTM layers, Dropout layers, and fully connected layers, and outputs classification probabilities through sigmoid or softmax activation functions.

[0034] During operation, this model is a deep learning network specifically built for analyzing time-series data, consisting of one or more LSTM (Long Short-Term Memory) layers. Each layer is responsible for processing patient monitoring data input sequentially over time, such as heart rate, urine output, and intra-abdominal pressure over multiple consecutive time periods. The unique gating mechanism within the LSTM unit intelligently determines which long-term information to retain and which temporary information to forget, thereby efficiently capturing complex dynamic changes, trends, and temporal dependencies in vital signs parameters. For example, it can identify patterns with warning significance such as persistently elevated intra-abdominal pressure or progressively decreasing urine output.

[0035] Following the LSTM layer, the network typically incorporates a Dropout layer, which plays a role during model training. This layer randomly and temporarily shuts down a subset of neurons in the network, preventing the model from becoming overly reliant on specific details or random noise in the training data. It is an effective regularization technique. By forcing the network to learn using a different subset of neurons each time it is trained, the model's robustness and generalization ability can be improved, ensuring stable and reliable risk predictions even when encountering new or unseen patient data.

[0036] Finally, after high-level temporal features extracted by LSTM and regularization by the Dropout layer, the data is passed to a fully connected layer for synthesis and transformation. The network's output layer uses specific activation functions to produce the final result. When the model is used to predict a single risk, the sigmoid activation function is used, mapping the output value to between 0 and 1, directly interpreting it as the probability of the complication occurring. When it is necessary to simultaneously assess and distinguish multiple risk categories, the softmax activation function is used, which outputs a probability distribution where the sum of the probabilities of all categories is 1, thus providing a multi-class risk assessment.

[0037] Example 5 The difference between this embodiment and embodiment four is that the risk prediction based on the preprocessed multi-dimensional monitoring dataset and the LSTM-based time series risk warning model includes the following steps: The preprocessed time series data is organized into a multidimensional matrix according to time windows and then input into the LSTM model. The model learns sequence dependencies and outputs the probability of the target complication occurring in the future. When the probability exceeds a preset threshold, an early warning signal is triggered.

[0038] The risk prediction based on the preprocessed multi-dimensional monitoring dataset and the LSTM-based time series risk warning model includes the following steps: organizing the preprocessed time series data into a multi-dimensional matrix according to time windows and inputting it into the LSTM model; the model learns the sequence dependencies and outputs the probability of the target complication occurring in the future period; when the probability exceeds a preset threshold, a warning signal is triggered.

[0039] During operation, the cleaned and aligned continuous monitoring data is first extracted and organized according to a fixed time window to form a multidimensional matrix containing multiple time points and multiple feature dimensions. This transforms the original data stream into a standard input format that the model can process and includes historical context, thus determining the scope that the model can see and learn.

[0040] Subsequently, this time-series data matrix is ​​input into the LSTM-based time-series risk warning model described in Example 4. The LSTM layer within the model processes the matrix step-by-step, automatically learning and memorizing the sequence dependencies and dynamic evolution patterns between data points within the window through its gating mechanism. For example, it might learn that the combination of a sustained decrease in urine output accompanied by a steady increase in intra-abdominal pressure is a strong predictor of intra-abdominal hypertension syndrome. After completing the analysis of the entire time window, the model calculates and outputs a specific probability value—the risk probability of the patient developing the target complication within a specific future time period—through the final fully connected layer and activation function.

[0041] The risk probability output by the model is compared with a preset threshold determined in advance by clinical experts' experience or historical data analysis. This threshold is the decision boundary of the early warning system. When the calculated risk probability exceeds this threshold, the system will automatically trigger an early warning signal, realizing the transformation from continuous risk monitoring to discrete early warning actions. This converts the complex model calculation results into warning information that can be immediately recognized by clinicians, thereby timely reminding medical staff to pay attention to patient risks and initiate assessment or intervention procedures.

[0042] Example 6 The difference between this embodiment and Embodiment 5 is that the step of generating liquid management early warning information and liquid replenishment plan adjustment suggestions based on the risk prediction results includes the following steps: When a risk of fluid overload is predicted, warning information is generated, including recommendations to reduce the rate of fluid administration and to assess the use of diuretics. When a risk of hypovolemia is predicted, warning information is generated, including suggestions to increase the rate of fluid resuscitation and adjust the composition of the fluid. When a risk of intra-abdominal hypertension syndrome is predicted, early warning information is generated, including recommendations to monitor intra-abdominal pressure and assess the necessity of abdominal drainage.

[0043] The system transforms the abstract risk probabilities output by the model into concrete and actionable clinical decision support information. Internally, it has a pre-defined set of clinical response rules tightly bound to different risk types. When the probability of a specific risk output by the time-series risk warning model exceeds a threshold and is determined by the system to be a valid warning, the corresponding rule module is immediately triggered.

[0044] For example, if the model determines that a patient's risk of fluid overload is significantly increased, the system will automatically invoke the rules for this situation. These rules take into account the pathophysiological basis of fluid overload, namely that fluid intake exceeds output, which may lead to increased cardiopulmonary burden. Therefore, the generated warning information will directly include the core measure of recommending a reduction in the rate of fluid infusion, and further suggest evaluating the use of diuretics to actively promote fluid excretion.

[0045] Conversely, if the warning indicates a risk of hypovolemia, the rule base will generate guidance based on the logic of insufficient volume, primarily suggesting increasing the rate of fluid resuscitation. It may also adjust the fluid composition, such as considering the ratio of crystalloid to colloid solutions and electrolyte supplementation, to achieve resuscitation. For the severe risk of intra-abdominal hypertension syndrome, the generated recommendations focus on escalating monitoring and intervention, specifically suggesting increased frequency of intra-abdominal pressure monitoring and assessment of the necessity of abdominal drainage, directly guiding the warning towards key clinical assessment and decision points.

[0046] Example 7 The difference between this embodiment and Embodiment Six is ​​that the prediction of the patient's future risk of developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome based on the prediction results of fluid demand includes the following steps: The predicted fluid requirements were integrated with the patient's fluid balance, intra-abdominal pressure, and vital signs data for analysis. Use SHAP values ​​or similar interpretability methods to analyze the contribution of each feature to the prediction results; Based on contribution and clinical threshold, calculate the probability of various complications occurring in a specific future period; Output the risk level and confidence interval.

[0047] During the work, the recommended fluid requirement from the fluid requirement prediction model is synchronously correlated and comprehensively analyzed with the patient's real-time fluid balance, dynamic intra-abdominal pressure data, and other vital signs to identify potential contradictions or mismatches. For example, when the model recommends a higher fluid replacement volume, but the patient already has a clear positive balance and the intra-abdominal pressure is trending upward, it will initially indicate the risk of fluid overload and intra-abdominal hypertension.

[0048] To clarify the specific impact of various factors on the current risk status, the system introduces interpretable artificial intelligence (XAI) methods, such as calculating the SHAP value. This method can quantitatively reveal the direction and magnitude of each input feature's contribution to the model's final risk assessment, such as high intra-abdominal pressure, low urine output, and high predicted demand.

[0049] The system then integrates the analysis results with predetermined clinical pathophysiological thresholds, and through an integrated risk assessment algorithm, quantifies the independent risk probability of the patient developing each target complication over a given period. To reflect the uncertainty of the prediction, output confidence intervals can be set, and consecutive probability values ​​can be categorized into low, medium, and high risk levels.

[0050] Example 8 A comprehensive fluid intake and output monitoring system for pancreatitis patients includes a first processing module for acquiring fluid intake and output data, non-invasive intra-abdominal pressure data, and vital sign data of pancreatitis patients as a multi-dimensional monitoring dataset; and preprocessing the multi-dimensional monitoring dataset. The second processing module is used to set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. The third processing module is used to set up an LSTM-based time series risk warning model and perform risk prediction based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. The fourth processing module is used to obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results of risk forecasting. The fifth processing module is used to obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, to predict the probability of the patient developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

[0051] Example 9 A computer device, comprising: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory, including executing the above-described method for comprehensive monitoring of fluid intake and output in pancreatitis patients; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

[0052] Example 10 A readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the above-described method for comprehensive monitoring of fluid intake and output in pancreatitis patients.

[0053] In summary, this invention constructs a complete chain system from multi-dimensional data automatic collection and intelligent analysis to clinical decision support, transforming discrete clinical information into continuous, dynamic, and measurable decision-making basis, thereby systematically improving the accuracy and safety of fluid management in severe pancreatitis.

[0054] First, by integrating multi-source data such as bedside monitoring equipment, infusion pumps, smart urine meters, and non-invasive intra-abdominal pressure sensors, this invention achieves automated and continuous comprehensive monitoring of key indicators of the patient's volume status and abdominal environment, breaking through the information barriers and lags caused by traditional manual recording and fragmented assessment.

[0055] Secondly, by introducing and synergistically applying a fluid requirement prediction model and an LSTM-based time-series risk warning model, this invention provides current patients with an objective and quantitative individualized fluid resuscitation benchmark by learning from historical successful experiences. This helps reduce the subjective differences and uncertainties caused by clinicians relying solely on experience. Furthermore, it fully leverages the advantages of deep learning in processing time-series data, enabling the early identification of subtle and complex precursor patterns pointing to fluid overload, hypovolemia, or intraperitoneal hypertension syndrome from the dynamic changes in vital signs and monitoring parameters.

[0056] Finally, this invention automatically transforms the model's prediction results into specific fluid management early warning information and fluid resuscitation plan adjustment suggestions, and quantifies the risk probability of future complications, reducing the risk of clinical errors caused by information overload or judgment delays. It also promotes the homogenization and standardization of treatment through standardized suggestion output.

[0057] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0058] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0059] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0060] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for comprehensive monitoring of fluid intake and output in patients with pancreatitis, characterized in that, Includes the following steps: Data on fluid intake and output, non-invasive intra-abdominal pressure, vital signs, and blood gas analysis of patients with pancreatitis were acquired as a multidimensional monitoring dataset; the multidimensional monitoring dataset was preprocessed. Set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. Set up an LSTM-based time series risk warning model to predict risks based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. Obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results; Obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, predict the probability of patients developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

2. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The fluid demand prediction model includes a machine learning-based regression model; the machine learning-based regression model is trained using historical patient data and outputs a predicted individualized fluid demand.

3. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The process of predicting liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model includes the following steps: Feature variables related to fluid requirements were extracted from the preprocessed data, including heart rate, mean arterial pressure, hematocrit, lactate, urine output, intra-abdominal pressure trend, and blood gas analysis indicators. Input the feature variables into the trained liquid demand prediction model; The model outputs a predicted fluid requirement; the prediction is adjusted based on the patient's real-time fluid balance.

4. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The LSTM-based time series risk warning model includes one or more LSTM layers, Dropout layers, and fully connected layers, and outputs classification probabilities through sigmoid or softmax activation functions.

5. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The risk prediction based on the preprocessed multi-dimensional monitoring dataset and the LSTM-based time series risk warning model includes the following steps: The preprocessed time series data is organized into a multidimensional matrix according to time windows and then input into the LSTM model. The model learns sequence dependencies and outputs the probability of the target complication occurring in the future. When the probability exceeds a preset threshold, an early warning signal is triggered.

6. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The process of generating liquid management early warning information and liquid replenishment plan adjustment suggestions based on the risk prediction results includes the following steps: When a risk of fluid overload is predicted, warning information is generated, including recommendations to reduce the rate of fluid administration and to assess the use of diuretics. When a risk of hypovolemia is predicted, warning information is generated, including suggestions to increase the rate of fluid resuscitation and adjust the composition of the fluid. When a risk of intra-abdominal hypertension syndrome is predicted, early warning information is generated, including recommendations to monitor intra-abdominal pressure and assess the necessity of abdominal drainage.

7. The method for comprehensive monitoring of fluid intake and output in patients with pancreatitis according to claim 1, characterized in that, The method of predicting the risk probability of a patient developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome based on the predicted fluid demand includes the following steps: The predicted fluid requirements were integrated with the patient's fluid balance, intra-abdominal pressure, and vital signs data for analysis. Use SHAP values ​​or similar interpretability methods to analyze the contribution of each feature to the prediction results; Based on contribution and clinical threshold, calculate the probability of various complications occurring in a specific future period; Output the risk level and confidence interval.

8. A comprehensive monitoring system for fluid intake and output in patients with pancreatitis, characterized in that, The system includes a first processing module for acquiring fluid intake and output data, non-invasive intra-abdominal pressure data, and vital sign data of pancreatitis patients as a multi-dimensional monitoring dataset; and preprocessing the multi-dimensional monitoring dataset. The second processing module is used to set up a liquid demand prediction model and predict liquid demand based on the preprocessed multi-dimensional monitoring dataset and the liquid demand prediction model. The third processing module is used to set up an LSTM-based time series risk warning model and perform risk prediction based on the preprocessed multi-dimensional monitoring dataset and liquid demand prediction model. The fourth processing module is used to obtain the prediction results of risk forecasting, and generate liquid management early warning information and liquid replenishment plan adjustment suggestions based on the prediction results of risk forecasting. The fifth processing module is used to obtain the prediction results of fluid demand forecasting, and based on the prediction results of fluid demand forecasting, to predict the probability of the patient developing fluid overload, hypovolemia, or intraperitoneal hypertension syndrome in the future.

9. A computer device, characterized in that, include: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory, including executing a method for comprehensive monitoring of fluid intake and output in pancreatitis patients as described in any one of claims 1 to 7; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

10. A readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by the processor, they implement the steps of the method for comprehensive monitoring of fluid intake and output in pancreatitis patients as described in any one of claims 1 to 7.

Images