A multi-modal data fusion-based intensive care unit patient monitoring management system

Abstract

The present application relates to the technical field of medical intensive care management system, and particularly relates to a patient monitoring and management system for intensive care unit based on multi-modal data fusion, comprising: a physiological characteristic extraction module, which collects a first characteristic set and processes it into a second characteristic set containing brain electrical entropy value and cerebral perfusion pressure; a pathological state decoupling module, which inversely calculates an observation vector, combines a pharmacological baseline and a drug dose to calculate a pharmacological stress vector, and obtains a pathological vector and generates a pathological index after stripping; an evolution trend prediction module, which predicts risk factors according to time series data of the pathological vector; a closed-loop intervention decision module, which combines the pathological index and the risk factors, calculates a drug dose adjustment amount according to a second mapping relationship, and updates the infusion rate to realize closed-loop control. The present application can strip pharmacological stress, generate a pathological index to accurately judge the condition, actively predict risks, and automatically adjust the drug dose combined with the pathological index, thereby improving the management efficiency and safety of critical patients.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical intensive care management system, and particularly relates to a patient monitoring and management system for intensive care unit based on multi-modal data fusion. BACKGROUND

[0002] In the intensive care unit, especially in the neurological intensive care unit, the monitoring of patients mainly depends on the isolated interpretation of multiple physiological parameters by clinical personnel and the adjustment of treatment intervention based on experience. The presentation of physiological state is often the result of the joint action of internal disease evolution and external drug intervention, which makes it difficult to accurately distinguish the real signal of disease deterioration and the physiological fluctuation caused by drug effect. Clinical decision-making is therefore challenged and cannot clearly determine whether the change in physiological indicators is dominated by pathological factors or caused by pharmacological factors. This ambiguity of information leads to a lack of precise quantitative basis for treatment interventions such as sedative drug dosage titration, which may delay effective response to real disease deterioration. In addition, existing methods generally lack the ability to predict the future evolution trend of pathological state, and monitoring and intervention are mostly in a passive response mode.

[0003] The root cause of the above problems lies in the fact that current monitoring technology has failed to establish a dynamic model that can fuse multi-modal data and deconstruct its internal causal relationship. Specifically, there is a lack of a unified mathematical framework to describe the complex coupling relationship between physiological subsystems and between them and drug intervention. At the same time, existing technical means are difficult to isolate the pharmacological stress component from the observed mixed physiological signals in real time, so as to isolate pure pathological stress information. The final result is that when the patient's physiological state fluctuates, clinical personnel cannot obtain clear and quantitative insights into the nature of disease deterioration in a timely manner, which directly affects the precision and foresight of closed-loop treatment control strategies and poses a potential risk to prevent secondary brain injury and other serious consequences. SUMMARY

[0004] The purpose of the present application is to provide a patient monitoring and management system for intensive care unit based on multi-modal data fusion and method, which solves the problems existing in the background art.

[0005] To solve the above technical problems, the present application provides a patient monitoring and management system for intensive care unit based on multi-modal data fusion, comprising: a physiological feature extraction module, configured to collect a first feature set representing the multi-dimensional physiological state of a patient, and process the first feature set into a second feature set containing electroencephalogram entropy value and cerebral perfusion pressure based on a preset first mapping relationship;

[0006] a pathological state decoupling module configured to: calculate a predicted pharmacological stress vector based on a second set of features and a pharmacological physiological response baseline and a current drug dosage change; obtain a pathological stress vector by stripping the predicted pharmacological stress vector from an observed physiological state change vector calculated from the second set of features; and generate a pathological index representing a degree of pathological deterioration by combining the pathological stress vector and a dynamically adjusted pathological stress threshold;

[0007] an evolution trend prediction module configured to predict and output a risk prediction factor representing a future pathological state of the patient based on time-series data of the pathological stress vector output by the pathological state decoupling module;

[0008] a closed-loop intervention decision module configured to calculate a drug dosage adjustment amount based on the pathological index and the risk prediction factor and a preset second mapping relationship, and update a current drug infusion rate using the drug dosage adjustment amount to achieve closed-loop control.

[0009] Preferably, the first set of features collected by the physiological feature extraction module further includes mean arterial pressure and intracranial pressure; and the first mapping relationship is specifically that the cerebral perfusion pressure is generated by subtracting the intracranial pressure from the mean arterial pressure.

[0010] Preferably, the process of calculating the predicted pharmacological stress vector by the pathological state decoupling module includes: first, obtaining the pharmacological physiological response baseline representing individual drug sensitivity, and then combining the current drug dosage change to predict a theoretical physiological state change caused by the dosage change, and setting the theoretical physiological state change as the predicted pharmacological stress vector.

[0011] Preferably, the process of generating the pathological index by the pathological state decoupling module includes: first, calculating a module length of the pathological stress vector, then obtaining the dynamically adjusted pathological stress threshold, and finally calculating a ratio of the module length of the pathological stress vector to the pathological stress threshold to obtain the pathological index.

[0012] Preferably, the monitoring state division module is further configured to: based on a historical fluctuation range of the second set of features, label a current stage of the patient as one of a stable stage, an observation stage, or an alarm stage; and the pathological state decoupling module dynamically adjusts the pathological stress threshold based on the monitoring state stage output by the monitoring state division module.

[0013] Preferably, the evolution trend prediction module is built-in a long short-term memory network model; the evolution trend prediction module is further configured to integrate time-series data of the pathogenic stress vector with time-series data of auxiliary clinical parameters such as body temperature and inflammation indicators to construct a pathogenic evolution feature dataset; and input the pathogenic evolution feature dataset into the long short-term memory network model, and the output of the long short-term memory network model is the risk prediction factor.

[0014] Preferably, the process of calculating by the closed-loop intervention decision module according to the second mapping relationship comprises: setting a high-priority warning threshold and a medium-priority warning threshold; when the pathogenic index is higher than the high-priority warning threshold, triggering a protective strategy, and setting the drug dose adjustment amount as a first negative value; and when the pathogenic index is lower than the high-priority warning threshold but higher than the medium-priority warning threshold, triggering a conservative strategy, and setting the drug dose adjustment amount as a second negative value.

[0015] Preferably, when the pathogenic index is not higher than the medium-priority warning threshold, the closed-loop intervention decision module is further configured to: acquire a stage risk threshold corresponding to a current monitoring state stage; calculate a deviation between the risk prediction factor and the stage risk threshold by comparing the risk prediction factor with the stage risk threshold; and generate the drug dose adjustment amount based on the deviation.

[0016] Preferably, the system further comprises a baseline calibration module configured to: record a response change of the second feature set after a patient receives a standardized test drug dose at an early stage of hospitalization; process the response change and set the response change as a pharmacological physiological response baseline for calling by the pathogenic state decoupling module.

[0017] Preferably, after obtaining the drug dose adjustment amount, the closed-loop intervention decision module is further configured to: perform summation operation on the current drug infusion rate and the drug dose adjustment amount to generate a new recommended drug infusion rate; and output the new recommended drug infusion rate to a drug infusion device to perform closed-loop regulation.

[0018] Compared with the prior art, the application has the following beneficial effects,

[0019] (1) Through the cooperative work of the physiological feature extraction module, the pathogenic state decoupling module and the baseline calibration module, the core technical problem of confusion between the true state of illness and the drug effect in traditional monitoring is solved. The pathogenic state decoupling module is used to strip the predicted pharmacological stress vector from the observed physiological state change vector, so as to obtain a pure pathogenic stress vector. This process can accurately distinguish between pathogenic and pharmacological factors, and finally generate a pathogenic index that provides an unprecedented objective quantitative index for the degree of illness deterioration. This quantitative evaluation capability significantly improves the accuracy of illness judgment and provides a solid data foundation for subsequent treatment decisions.

[0020] (2) By integrating the evolutionary trend prediction module and the closed-loop intervention decision module, clinical monitoring is elevated from a passive response mode to a new level of proactive prediction and automated intervention. The evolutionary trend prediction module can generate forward-looking risk predictors, enabling the system to anticipate the risk of disease deterioration. The closed-loop intervention decision module combines these risk predictors with real-time pathological indices and automatically calculates drug dosage adjustments based on a preset second mapping relationship to achieve closed-loop control. This proactive and automated intervention approach ensures the timeliness and accuracy of treatment adjustments, effectively improving the management efficiency and safety of critically ill patients. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a logic block diagram of the system of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Example 1:

[0025] Please see Figure 1 The present invention provides a critical care patient monitoring and management system based on multimodal data fusion, comprising: a physiological feature extraction module, used to collect a first feature set representing the multidimensional physiological state of patients, and based on a preset first mapping relationship, process the first feature set into a second feature set containing EEG entropy value and cerebral perfusion pressure;

[0026] The pathological state decoupling module is used to back-calculate the observed physiological state change vector based on the second feature set, and calculate the predicted pharmacological stress vector based on the system's preset pharmacological and physiological response baseline and the current drug dose change; by extracting the predicted pharmacological stress vector from the observed physiological state change vector, the pathological stress vector is obtained, and then combined with the dynamically adjusted pathological stress threshold, a pathological index characterizing the degree of disease deterioration is generated.

[0027] an evolution trend prediction module configured to predict and output a risk prediction factor representing a future pathological state of the patient based on time-series data of the pathological stress vector output by the pathological state decoupling module;

[0028] a closed-loop intervention decision module configured to combine the pathological index and the risk prediction factor, and calculate a drug dosage adjustment amount according to a preset second mapping relationship, and use the drug dosage adjustment amount to update a current drug infusion rate to achieve closed-loop regulation;

[0029] The physiological feature extraction module collects a first feature set of a traumatic brain injury patient in a neuro-intensive care unit through a multi-modal sensor network, and the first feature set includes multiple physiological parameters such as electroencephalogram signal complexity, mean arterial pressure MAP, intracranial pressure ICP, body temperature T, and inflammation index I.

[0030]

[0031] The brain perfusion pressure is calculated as MAP-ICP.

[0032] The mean arterial pressure is calculated as MAP.

[0033] The intracranial pressure is calculated as ICP.

[0034] The electroencephalogram entropy value is extracted through an electroencephalogram signal complexity analysis algorithm to form a first feature set including the electroencephalogram entropy value and the brain perfusion pressure.

[0035] The pathological state analysis module receives the second feature set output by the physiological feature extraction module, and solves an equation through a physiological state vector

[0036]

[0037] The patient individual physiological resilience matrix is calculated as

[0038] The physiological state change vector is calculated as

[0039] The total stress vector borne by the patient is calculated as

[0040] The observed physiological state change vector is calculated inversely as The patient individual physiological resilience matrix is obtained by: at the early stage of patient admission, recording the second feature set response data of the patient after receiving the standardized test drug dose through the baseline calibration module, fitting the mapping relationship between the physiological parameter change and the total stress vector by linear regression or machine learning algorithm, and generating the individualized physiological resilience matrix; or based on the time sequence fluctuation of the real-time physiological data of the patient, dynamically updating the matrix parameters by recursive algorithm;

[0041] The pathological state analysis module is based on the pharmacological physiological response baseline preset by the system and the current drug dose change , through the pharmacokinetics / pharmacodynamics response function:

[0042]

[0043] The pharmacological physiological response baseline is:

[0044] The current drug dose change is:

[0045] The patient's sensitivity coefficient to the sedative drug is:

[0046] The predicted pharmacological stress vector is calculated ;

[0047] The pathological state analysis module calculates:

[0048]

[0049] The pathological stress vector is:

[0050] The observed physiological state total stress vector is:

[0051] The predicted pharmacological stress vector is:

[0052] The pathological stress vector is obtained, combined with the dynamically adjusted pathological stress threshold , generating a pathological index representing the degree of disease deterioration

[0053]

[0054] The pathological index is the degree of deterioration of the patient's current disease;

[0055] The norm of the pathological stress vector is:

[0056] The dynamically adjusted pathological stress threshold is:

[0057] The evolution trend prediction module is based on the pathogenic stress vector output by the pathological state analysis module The evolution trend prediction module is based on the pathogenic stress vector output by the pathological state analysis module

[0058]

[0059] The risk prediction factor is the risk prediction factor;

[0060] The risk weight is set according to clinical experience;

[0061] The prediction value of the LSTM model for each physiological parameter in the future;

[0062] The change amount of the brain electrical entropy value;

[0063] The change amount of the cerebral perfusion pressure;

[0064] The body temperature;

[0065] The inflammation index; The closed-loop intervention decision module combines the pathological index

[0066] And the risk prediction factor According to the preset second mapping relationship, the grading decision calculation is performed; when the pathological index The closed-loop intervention decision module triggers the protective strategy, and calculates the drug dose adjustment amount as

[0067] ,

[0068] Wherein The drug dose adjustment amount;

[0069] The high-priority negative feedback coefficient;

[0070] The current drug infusion rate;

[0071] The alarm period dynamic threshold;

[0072] ​The closed-loop intervention decision module uses the drug dose adjustment amount to update the current drug infusion rate by:

[0073]

[0074] a drug dose adjustment amount;

[0075] a current drug infusion rate;

[0076] a new recommended drug infusion rate;

[0077] calculating a new recommended drug infusion rate to achieve closed-loop regulation;

[0078] In this embodiment, the physiological feature extraction module can accurately collect and process multi-dimensional physiological state information of the patient through multi-modal data fusion technology, significantly improving the comprehensiveness and accuracy of physiological state evaluation compared with traditional single parameter monitoring methods; the pathological state analysis module can effectively distinguish pharmacological stress from pathological stress through decoupling analysis technology, avoiding the problem of confusion between drug effects and pathological changes in traditional methods, and providing more accurate disease assessment basis for clinicians; the evolution trend prediction module can predict the future pathological state evolution trend of the patient based on historical time series data through the LSTM deep learning model, realizing the transition from passive monitoring to active warning compared with traditional static evaluation methods, and effectively reducing the risk of secondary brain injury; the closed-loop intervention decision module can automatically adjust the drug infusion strategy according to the real-time state of the patient through intelligent decision algorithm, significantly improving the accuracy and timeliness of treatment compared with traditional manual adjustment method, reducing the work burden of medical staff, and improving the overall efficiency and safety of intensive care.

[0079] Embodiment 2:

[0080] The first feature set collected by the physiological feature extraction module further includes mean arterial pressure and intracranial pressure; and the first mapping relationship is specifically that the cerebral perfusion pressure is generated by subtracting the mean arterial pressure from the intracranial pressure;

[0081] The process for calculating the predicted pharmacological stress vector by the pathological state decoupling module includes: first, obtaining a pharmacological physiological response baseline preset to represent individual drug sensitivity, and then combining the current drug dose change to predict the theoretical physiological state change caused by the dose change, and setting the theoretical physiological state change as the predicted pharmacological stress vector;

[0082] The first feature set collected by the physiological feature extraction module further includes mean arterial pressure MAP obtained through an invasive arterial catheter and intracranial pressure ICP obtained through an intracranial pressure monitor, and the sampling frequency is set to 100-1000 Hz to ensure the real-time and accuracy of the data; the first mapping relationship in the physiological feature extraction module is specifically implemented by performing a subtraction operation on the mean arterial pressure MAP and the intracranial pressure ICP to generate cerebral perfusion pressure

[0083]

[0084] Cerebral perfusion pressure;

[0085] Mean arterial pressure;

[0086] Intracranial pressure;

[0087] The calculation process is performed in the real-time data processing unit at a millisecond level frequency to ensure the continuity and accuracy of the cerebral perfusion pressure value;

[0088] The process of calculating the predicted pharmacological stress vector by the pathological state analysis module first obtains a pharmacological physiological response baseline preset to represent the overall drug sensitivity , which is calibrated by the data in the 4-6 hour stable period at the beginning of the patient's admission, reflecting the patient's individual basic response characteristics to the standard dose of sedative drugs; the pathological state analysis module combines the current drug dose change , through a pharmacokinetic model:

[0089]

[0090] Pharmacological stress vector;

[0091] Pharmacological physiological response baseline;

[0092] Current drug dose change;

[0093] Sensitivity coefficient of the patient to the sedative drug;

[0094] The predicted pharmacological stress vector , which predicts the theoretical physiological state change caused by the dose change, wherein represents a PK / PD response function based on the Michaelis-Menten equation or linear regression, represents the sensitivity coefficient of the patient to the hemodynamic and neuroinhibitory effect of the sedative drug; the pathological state analysis module sets the theoretical physiological state change as the predicted pharmacological stress vector This is used for subsequent decoupling analysis from the observed total stress vector;

[0095] In this embodiment, the physiological feature extraction module accurately collects mean arterial pressure and intracranial pressure, and calculates cerebral perfusion pressure in real time based on the first mapping relationship. This provides critical care physicians with key indicators for assessing cerebral blood flow perfusion. Compared with traditional indirect estimation methods, it significantly improves the accuracy and reliability of cerebral perfusion status monitoring, playing a particularly important role in the management of intracranial pressure in patients with traumatic brain injury. The pathological state analysis module obtains individualized pharmacological and physiological response baselines and combines them with current drug dosage changes to predict theoretical physiological state changes. This effectively establishes patient-specific drug effect models. Compared with traditional drug effect assessment methods based on population averages, it significantly improves the individualized accuracy of pharmacological stress prediction, providing important technical support for precision medicine. It effectively avoids drug effect assessment bias caused by individual differences and improves the safety and effectiveness of drug treatment in intensive care.

[0096] Example 3:

[0097] The process of generating pathological indices by the pathological state decoupling module includes: first, calculating the magnitude of the pathological stress vector; second, obtaining the dynamically adjusted pathological stress threshold; and finally, obtaining the pathological index by calculating the ratio of the magnitude of the pathological stress vector to the pathological stress threshold.

[0098] It also includes a monitoring status classification module, which is used to: classify the patient's current stage as a monitoring status stage, such as a stable stage, an observation stage, or an alarm stage, based on the historical fluctuation range of the second feature set; and a pathological status decoupling module to dynamically adjust the pathological stress threshold based on the monitoring status stage output by the monitoring status classification module.

[0099] The process of generating pathological indices by the pathological state analysis module first calculates the pathological stress vector. Length of the module This calculation is performed using vector norm operations:

[0100]

[0101] Pathological stress vector The modulus length;

[0102] Let i be the component of the pathological stress vector in the i-th dimension.

[0103] n is the number of dimensions of the vector;

[0104] The pathological state analysis module then obtains the dynamically adjusted pathological stress threshold. This threshold is adjusted in real time according to the monitoring status stage, and the calculation formula is as follows:

[0105] base threshold

[0106] wherein is a dynamically adjusted pathological stress threshold;

[0107] is a threshold coefficient corresponding to the current monitoring state phase;

[0108] The pathological state analysis module finally obtains a pathological index by calculating the ratio of the length of the pathological stress vector to the pathological stress threshold

[0109]

[0110] is a pathological index, which quantitatively represents the degree of deterioration of the patient's current condition;

[0111] is the length of the pathological stress vector;

[0112] is a dynamically adjusted pathological stress threshold;

[0113] The monitoring state division module assesses the stability of the patient's physiological parameters based on the historical fluctuation range of the second feature set by statistical analysis method; the monitoring state division module adopts a moving window standard deviation algorithm, and when the electroencephalogram entropy value or the cerebral perfusion pressure fluctuates by less than 20% of the baseline value within 72 hours, the patient's current phase is determined to be a stable phase, corresponding to ; when the fluctuation standard deviation is between 20%-50% of the baseline value, the patient's current phase is determined to be an observation phase, corresponding to ; when the fluctuation standard deviation exceeds 50% of the baseline value, the patient's current phase is determined to be an alarm phase, corresponding to ; the pathological state analysis module dynamically adjusts the pathological stress threshold based on the monitoring state phase output by the monitoring state division module, and the base threshold is in the stable phase, the base threshold is in the observation phase, and the base threshold is ;

[0114] In this embodiment, the pathological state analysis module can generate a quantitative pathological index by calculating the modulus of the pathological stress vector and performing a ratio operation with the dynamically adjusted threshold value. Compared with the traditional qualitative evaluation method, the objectivity and accuracy of the evaluation of the degree of disease deterioration are significantly improved, a quantifiable disease monitoring index is provided for clinicians, and precise diagnosis and treatment decisions for the intensive care unit are effectively supported. The monitoring state division module can automatically identify the monitoring state stage of the patient through intelligent analysis based on the historical fluctuation range. Compared with the traditional manual judgment method, the consistency and timeliness of the monitoring state evaluation are significantly improved. At the same time, by dynamically adjusting the pathological stress threshold value, individualized and staged pathological evaluation standards are realized, which effectively adapt to the dynamic change characteristics of the critical patient's condition, improve the precision and effectiveness of monitoring management, and provide differentiated monitoring strategy support for patients in different disease stages.

[0115] Embodiment 4:

[0116] The evolution trend prediction module is built-in with a long short-term memory network model. The evolution trend prediction module is further configured to: integrate time series data of the pathological stress vector and time series data of auxiliary clinical parameters such as body temperature and inflammation indicators to construct a pathological evolution feature data set; and input the pathological evolution feature data set into the long short-term memory network model, and the output of the long short-term memory network model is a risk prediction factor.

[0117] The evolution trend prediction module is built-in with a long short-term memory network model LSTM. The model adopts a multi-layer LSTM architecture, including an input layer, a hidden layer and an output layer. The hidden layer is provided with 128 neuron units, which can effectively capture long-term dependence and short-term fluctuation characteristics in time series data. The evolution trend prediction module integrates time series data of the pathological stress vector and time series data of auxiliary clinical parameters such as body temperature T and inflammation indicators I to construct a pathological evolution trend data set. The data set contains continuous monitoring data of the patient in the past 24-48 hours, and the sampling interval is set to 15 minutes to ensure the time series continuity and representativeness of the data.

[0118] When constructing the pathological evolution trend data set, the evolution trend prediction module takes the time series sequence of the pathological stress vector as the main feature, and integrates the time series change of body temperature T, the timing detection result of inflammation indicators I and the monitoring state stage label as auxiliary features. The evolution trend prediction module performs standardization processing on the input data, and adopts a Z-score standardization method for

[0119]

[0120] the original data value;

[0121] the mean value;

[0122] is a standard deviation;

[0123] is a standardized data value;

[0124] The evolution trend prediction module inputs the standardized pathological evolution trend data set into a long short-term memory network model. The LSTM model identifies the change mode of the pathological state through time sequence feature learning and outputs a risk prediction factor ,

[0125] is a risk prediction factor;

[0126] is a risk weight set according to clinical experience, ;

[0127] is a predicted value of the LSTM model for each physiological parameter in ;

[0128] is a change amount of electroencephalogram entropy value;

[0129] is a change amount of cerebral perfusion pressure;

[0130] is a body temperature;

[0131] is an inflammation index;

[0132] In the embodiment, the evolution trend prediction module can effectively process the complex time sequence changes of physiological parameters of the critical patient through the built-in long short-term memory network model. Compared with the traditional linear prediction method, the accuracy and reliability of the pathological state evolution trend prediction are significantly improved, especially in capturing nonlinear change patterns and long-term dependencies. The evolution trend prediction module constructs a multi-dimensional pathological evolution trend data set by integrating the time sequence data of the pathological stress vector and the auxiliary clinical parameters. Compared with the single parameter prediction method, the comprehensiveness and robustness of the prediction model are significantly enhanced, which can more accurately reflect the complex evolution process of the patient's pathological state. The risk prediction factor output by the evolution trend prediction module provides a prospective condition assessment tool for the critical care medicine doctors. Compared with the traditional passive monitoring method, the transformation from reactive treatment to preventive intervention is realized, the risk of secondary brain injury and other serious complications is effectively reduced, the initiative and predictability of critical care are improved, and an important scientific basis for clinical decision-making is provided.

[0133] Embodiment 5:

[0134] The process of the closed-loop intervention decision module performing calculation according to the second mapping relationship comprises: setting a high-priority early warning threshold and a medium-priority early warning threshold; when the pathology index is higher than the high-priority early warning threshold, triggering a protective strategy, and setting the drug dose adjustment amount to a first negative value; when the pathology index is lower than the high-priority early warning threshold but higher than the medium-priority early warning threshold, triggering a conservative strategy, and setting the drug dose adjustment amount to a second negative value;

[0135] When the pathology index is not higher than the medium-priority early warning threshold, the closed-loop intervention decision module is further configured to: acquire a stage risk threshold corresponding to the current monitoring state stage; calculate a deviation between the risk prediction factor and the stage risk threshold by comparison, and generate the drug dose adjustment amount based on the deviation;

[0136] The process of the closed-loop intervention decision module performing calculation according to the second mapping relationship sets the high-priority early warning threshold to 1.8 and the medium-priority early warning threshold to 1.0, which are determined based on a large amount of clinical data statistical analysis and can effectively distinguish different severity of pathological states; when the pathology index of the closed-loop intervention decision module is higher than the high-priority early warning threshold 1.8, the protective strategy is triggered, and the drug dose adjustment amount is set to the first negative value:

[0137]

[0138] The drug dose adjustment amount;

[0139] The high-priority negative feedback coefficient, ;

[0140] The current sedative drug infusion rate;

[0141] The alarm period dynamic threshold, ;

[0142] The pathology index, which quantitatively represents the current degree of patient illness deterioration;

[0143] The strategy aims to quickly reduce the drug dose to reduce pharmacological inhibition; when the pathology index of the closed-loop intervention decision module is lower than the high-priority early warning threshold 1.8 but higher than the medium-priority early warning threshold 1.0, the conservative strategy is triggered, and the drug dose adjustment amount is set to the second negative value:

[0144]

[0145] The drug dose adjustment amount; ​​

[0146] denotes a moderate negative feedback coefficient,

[0147] is a dynamic threshold for the observation period,

[0148] is a pathology index, which quantifies the current degree of patient's disease aggravation;

[0149] The strategy adopts moderate dose adjustment to balance the therapeutic effect and safety;

[0150] The closed-loop intervention decision module obtains a stage risk threshold corresponding to the current monitoring state stage when the pathology index is not higher than a medium priority early warning threshold 1.0, wherein the stage risk threshold corresponding to the stable stage is 0.5, the stage risk threshold corresponding to the observation stage is 1.0, and the stage risk threshold corresponding to the alarm stage is 1.5; the closed-loop intervention decision module compares the risk prediction factor with the stage risk threshold , calculates a deviation between the two, generates a drug dose adjustment amount based on the deviation

[0151]

[0152] is a drug dose adjustment amount;

[0153] denotes a predictive adjustment coefficient,

[0154] : the deviation of the risk prediction factor and the stage threshold;

[0155] : the threshold coefficient corresponding to the current monitoring state stage;

[0156] The adjustment amount is a negative value to reduce the dose when , and the adjustment amount is a positive value to moderately increase the dose when ;

[0157] ​​​​In this embodiment, the closed-loop intervention decision module sets high-priority and medium-priority warning thresholds to establish an intelligent decision-making mechanism for hierarchical response. Compared with the traditional single threshold judgment method, the degree of refinement and adaptability of the intervention strategy is significantly improved, and a differentiated treatment plan can be provided according to the severity of the disease. The closed-loop intervention decision module realizes a gradient treatment mode from aggressive intervention to mild adjustment through the hierarchical triggering of protective strategies and conservative strategies. Compared with the traditional fixed dose adjustment method, the safety and effectiveness of drug treatment are significantly improved, effectively avoiding the problems of over-treatment or under-treatment. When the pathological index is low, the closed-loop intervention decision module calculates the deviation by comparing the risk prediction factor with the stage risk threshold and generates an adjustment amount, realizing a prospective intervention based on predictive evaluation. Compared with the traditional reactive treatment mode, the initiative and predictability of intensive care are significantly improved, the treatment strategy can be adjusted in time before the disease worsens, the occurrence of serious complications is effectively prevented, and the treatment effect and prognosis quality of the patient are improved.

[0158] Embodiment 6:

[0159] Further comprising a baseline calibration module, configured to: record the response change of the second feature set after the patient receives a standardized test drug dose at the early stage of admission; process and set the response change as a pharmacological physiological response baseline for the pathological state decoupling module to call;

[0160] After obtaining the drug dose adjustment amount, the closed-loop intervention decision module is further configured to: sum the current drug infusion rate and the drug dose adjustment amount to generate a new recommended drug infusion rate; and output the new recommended drug infusion rate to the drug infusion device to perform closed-loop regulation.

[0161] The baseline calibration module records the response change of the second feature set after the patient receives a standardized test drug dose within 4-6 hours of the stable period at the early stage of admission; the baseline calibration module uses a standard dose of propofol 1 mg / kg intravenous injection as a standardized test drug dose, and monitors the change amplitude and change pattern of the physiological parameters such as electroencephalogram entropy value , cerebral perfusion pressure , mean arterial pressure MAP, intracranial pressure ICP within 30 minutes after drug administration; the baseline calibration module standardizes and models the response change through a data processing algorithm, processes and sets a pharmacokinetics / pharmacodynamics model as a pharmacological physiological response baseline:

[0162]

[0163] , wherein : the pharmacological physiological response baseline;

[0164] : patient drug sensitivity coefficient;

[0165] : standard dose induced physiological parameter change vector;

[0166] : response time constant;

[0167] The baseline data is called by the pathological state analysis module for subsequent calculation of the pharmacological stress vector;

[0168] The closed-loop intervention decision module obtains the drug dose adjustment amount After that, the current drug infusion rate is summed with the drug dose adjustment amount to generate a new recommended drug infusion rate

[0169]

[0170] The new recommended drug infusion rate;

[0171] The current drug infusion rate;

[0172] The drug dose adjustment amount;

[0173] The closed-loop intervention decision module performs safety verification on the calculation result to ensure that the new recommended drug infusion rate is within the clinical safety range, with a minimum infusion rate of 0.5 mg / kg / h and a maximum infusion rate of 4.0 mg / kg / h. When the calculation result exceeds the safety range, it is automatically limited to the boundary value. The closed-loop intervention decision module outputs the new recommended drug infusion rate to the drug infusion device through a standardized communication protocol. After receiving the instruction, the infusion device automatically adjusts the infusion parameters to achieve precise control and closed-loop regulation of the drug dose. The entire adjustment process is completed within 5 minutes, ensuring the timeliness and continuity of treatment;

[0174] In this embodiment, the baseline calibration module can establish an individual pharmacological physiological response baseline by evaluating the response of the standardized test drug dose at the early stage of patient admission, which significantly improves the individualization accuracy and reliability of pharmacological stress prediction compared with the traditional group average-based drug effect evaluation method, provides an accurate reference standard for subsequent decoupling analysis, and effectively solves the problem of the influence of individual differences on drug effect evaluation; the closed-loop intervention decision module generates a new recommended drug infusion rate by summing the current drug infusion rate and the calculated adjustment amount, and outputs it to the drug infusion device, realizing a complete closed loop from decision calculation to execution control, which significantly improves the accuracy, timeliness and consistency of drug dose adjustment compared with the traditional manual adjustment method, reduces human operation errors and delays, and improves the automation level and treatment efficiency of intensive care; the whole system constructs a highly integrated intelligent intensive care management platform through the complete process of baseline calibration, state monitoring, trend prediction, intelligent decision and automatic execution, which significantly improves the overall monitoring quality and patient safety level of the intensive care unit compared with the traditional scattered monitoring method, and provides important technical support for the digital transformation of intensive care and the development of precision medicine;

[0175] The above is only a preferred embodiment of the present application, not other forms of limitations on the present application, any skilled person in the art can use the above disclosed technical content to make changes or modifications as equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the technical solution of the present application.

Claims

1. A critical care patient monitoring and management system based on multimodal data fusion, characterized in that, include: The physiological feature extraction module is used to collect a first feature set that represents the patient's multi-dimensional physiological state, and based on a preset first mapping relationship, process the first feature set into a second feature set that includes EEG entropy value and cerebral perfusion pressure. The pathological state decoupling module is used to back-calculate the observed physiological state change vector based on the second feature set, and calculate the predicted pharmacological stress vector based on the system's preset pharmacological and physiological response baseline and the current drug dose change; by extracting the predicted pharmacological stress vector from the observed physiological state change vector, the pathological stress vector is obtained, and then combined with the dynamically adjusted pathological stress threshold, a pathological index characterizing the degree of disease deterioration is generated. The evolution trend prediction module is used to predict and output risk prediction factors characterizing the patient's future pathological state based on the time-series data of the pathological stress vector output by the pathological state decoupling module. The closed-loop intervention decision module is used to combine the pathological index with the risk prediction factor and calculate the drug dosage adjustment amount according to the preset second mapping relationship. The drug dosage adjustment amount is used to update the current drug infusion rate to achieve closed-loop control. The first feature set collected by the physiological feature extraction module further includes mean arterial pressure and intracranial pressure; the first mapping relationship is specifically: the cerebral perfusion pressure is generated by subtracting the mean arterial pressure from the intracranial pressure. The characteristic feature is that the process of the pathological state decoupling module for calculating the predicted pharmacological stress vector includes: firstly, obtaining the preset pharmacological physiological response baseline for characterizing individual drug sensitivity, then combining the current drug dose change to predict the theoretical physiological state change caused by the dose change, and setting the theoretical physiological state change as the predicted pharmacological stress vector; The process by which the pathological state decoupling module generates the pathological index includes: firstly, calculating the magnitude of the pathological stress vector; secondly, obtaining the dynamically adjusted pathological stress threshold; and finally, obtaining the pathological index by calculating the ratio of the magnitude of the pathological stress vector to the pathological stress threshold. The evolution trend prediction module has a built-in long short-term memory network model; the evolution trend prediction module is also used to: integrate the time series data of the pathological stress vector with the time series data of auxiliary clinical parameters such as body temperature and inflammatory indicators to construct a pathological evolution feature dataset; input the pathological evolution feature dataset into the long short-term memory network model, and its output is the risk prediction factor; The evolutionary trend prediction module incorporates a Long Short-Term Memory (LSTM) network model. This model employs a multi-layer LSTM architecture, including an input layer, hidden layers, and an output layer. The hidden layers contain 128 neurons. The evolutionary trend prediction module incorporates pathological stress vectors. The time-series data of patients were integrated with the time-series data of auxiliary clinical parameters such as body temperature (T) and inflammatory marker I to construct a pathological evolution trend dataset. This dataset contains continuous monitoring data of patients over the past 24-48 hours, with a sampling interval of 15 minutes to ensure the temporal continuity and representativeness of the data. When constructing the pathological evolution trend dataset, the evolutionary trend prediction module includes pathological stress vectors. The time series data is used as the primary feature, while time-series changes in body temperature (T), timed detection results of inflammatory marker I, and monitoring status stage labels are integrated as auxiliary features. The evolutionary trend prediction module standardizes the input data using the Z-score standardization method. ； The original data value; The mean; Standard deviation; These are the standardized data values; The evolutionary trend prediction module inputs the standardized pathological evolution trend dataset into the Long Short-Term Memory (LSTM) network model. The LSTM model learns and identifies patterns of change in pathological states through temporal features and outputs risk prediction factors. , As a risk predictor; Risk weights set for clinical experience ; For the LSTM model, various physiological parameters are... The predicted value; This refers to the change in brainwave entropy. This represents the change in cerebral perfusion pressure. Body temperature; Inflammation markers; The closed-loop intervention decision module performs calculations based on the second mapping relationship, including: setting a high-priority warning threshold and a medium-priority warning threshold; when the pathological index is higher than the high-priority warning threshold, a protective strategy is triggered, and the drug dosage adjustment amount is set to a first negative value; when the pathological index is lower than the high-priority warning threshold but higher than the medium-priority warning threshold, a conservative strategy is triggered, and the drug dosage adjustment amount is set to a second negative value. The closed-loop intervention decision-making module calculates based on the second mapping relationship, setting a high-priority warning threshold of 1.8 and a medium-priority warning threshold of 1.

0. These two thresholds are determined based on statistical analysis of a large amount of clinical data and can effectively distinguish pathological states of different severity. The closed-loop intervention decision-making module uses pathological indices... When the value exceeds the high-priority warning threshold of 1.8, a protective strategy is triggered, setting the drug dosage adjustment to the first negative value. ； This is for adjusting the drug dosage; The high-priority negative feedback coefficient is... ; This is the current infusion rate of the sedative medication; The dynamic threshold for the alarm period. ; This is a pathological index, which quantitatively represents the degree of deterioration in the patient's current condition; This strategy aims to rapidly reduce drug dosage to alleviate pharmacological inhibition; the closed-loop intervention decision module considers pathological indices. When the dose is below the high-priority warning threshold of 1.8 but above the medium-priority warning threshold of 1.0, a conservative strategy is triggered, setting the drug dosage adjustment to the second negative value. ； This is for adjusting the drug dosage; This represents the coefficient for moderate-intensity negative feedback. ; The dynamic threshold for the observation period. ; This is a pathological index, which quantitatively represents the degree of deterioration in the patient's current condition; This strategy employs moderate dose adjustments to balance therapeutic efficacy and safety; Closed-loop intervention decision module when pathological index When the risk level is not higher than the medium-priority warning threshold of 1.0, obtain the phased risk threshold corresponding to the current monitoring status phase; the stable phase corresponds to... The observation phase corresponds to Alarm phase corresponding The closed-loop intervention decision-making module compares risk prediction factors. With phased risk threshold Calculate the deviation between the two. Based on this deviation, a drug dosage adjustment amount is generated. ； This is for adjusting the drug dosage; Indicates the predictive adjustment coefficient. ; Deviation between risk predictor factors and stage thresholds; : The threshold coefficient corresponding to the current monitoring status stage; when The adjustment amount is set to a negative value to reduce the dosage. Adjust the dosage to a positive value to appropriately increase the dose.

2. A critical care patient monitoring and management system based on multimodal data fusion according to claim 1, characterized in that, It also includes a monitoring status classification module, which is used to: classify the patient's current stage as a monitoring status stage, such as a stable stage, an observation stage, or an alarm stage, based on the historical fluctuation range of the second feature set; and the pathological state decoupling module dynamically adjusts the pathological stress threshold based on the monitoring status stage output by the monitoring status classification module.

3. A critical care patient monitoring and management system based on multimodal data fusion according to claim 1, characterized in that, When the pathological index is not higher than the medium-priority warning threshold, the closed-loop intervention decision module is also used to: obtain the stage risk threshold corresponding to the current monitoring status stage; By comparing the risk prediction factor with the staged risk threshold, the deviation between the two is calculated, and the drug dosage adjustment amount is generated based on the deviation.

4. A critical care patient monitoring and management system based on multimodal data fusion according to claim 1, characterized in that, It also includes a baseline calibration module, which is used to: record the response changes of the second feature set after the patient receives a standardized test drug dose in the early stage of hospital admission; process the response changes and set them as the pharmacological and physiological response baseline for use by the pathological state decoupling module.

5. A critical care patient monitoring and management system based on multimodal data fusion according to claim 1, characterized in that, After obtaining the drug dosage adjustment amount, the closed-loop intervention decision module is further configured to: sum the current drug infusion rate with the drug dosage adjustment amount to generate a new recommended drug infusion rate; and output the new recommended drug infusion rate to the drug infusion device to perform closed-loop control.

Images