Cardiac intensive care intelligent early warning system

Abstract

The invention discloses an intelligent early warning system for cardiac intensive care, and the system comprises a treatment input module which is used for receiving the input parameters of a cardiac treatment drug and basic physical sign data of a patient; the data real-time acquisition module is used for acquiring ECG data, kidney marker data and respiratory function data of a patient; the exception handling module is configured to output a causal effect estimated value and an exception probability value through a pre-trained DML model, and output predicted concentration of a preset marker and predicted data of preset function data through a pre-trained NODE model; and the closed-loop control module is used for updating the pre-trained DML model and the pre-trained NODE model. Therefore, on one hand, heart treatment drug-related and non-drug-related sexual organ dysfunction can be accurately distinguished, clinical misjudgment is avoided, on the other hand, continuous time trajectory prediction of kidney and respiratory function indexes can be realized, and potential abnormality can be warned in advance.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an intelligent early warning system for intensive cardiac care. Background Art

[0002] In the field of cardiac intensive care, improving cardiac function through intravenous infusion of cardiac therapeutic drugs is a common clinical approach, but such drugs may have potential effects on the functions of organs such as the kidneys and lungs.

[0003] Existing monitoring systems usually rely on the regular collection of kidney markers such as blood creatinine and urea nitrogen, and respiratory function indicators such as blood oxygen saturation and tidal volume, combined with manual experience to make abnormal judgments. This discrete detection method makes it difficult to capture the dynamic changing trends of indicators; when organ dysfunction is detected, traditional technologies lack the means to accurately distinguish between drug-related and non-drug-related causes, which can easily lead to excessive drug discontinuation or missed diagnosis of other causes; and traditional machine learning models are mostly based on short-term status assessments based on discrete data at fixed time intervals. They are unable to model the evolution of physiological indicators from a continuous time dimension, and their ability to warn of potential risks in the future is limited. Summary of the Invention

[0004] The present invention aims to solve the technical problems in the above-mentioned technologies at least to some extent.

[0005] To this end, the present invention discloses an intelligent early warning system for critical care cardiac care, comprising:

[0006] A treatment input module is used to receive input parameters of cardiac treatment drugs set by clinical doctors and basic vital signs data of patients;

[0007] Real-time data acquisition module, used to collect patients' ECG data, renal marker data and respiratory function data;

[0008] Exception handling module, configured as:

[0009] When the clearance rate of a preset marker in the renal marker data is less than a first threshold or the preset function data in the respiratory function data is less than a second threshold, inputting the input parameters of the cardiac therapeutic drug, the ECG data, the renal marker data, and the respiratory function data into a pre-trained DML model, respectively, and outputting a causal effect estimate and an abnormal probability value of the cardiac therapeutic drug;

[0010] when the concentration of the preset marker in the kidney marker data is less than or equal to a first threshold value or the preset function data in the respiratory function data is greater than a second threshold value, inputting the concentration of the preset marker in the kidney marker data and the preset function data in the respiratory function data into a pre-trained NODE model, and outputting predicted concentration of the preset marker in the kidney marker data and predicted data of the preset function data in the respiratory function data at the next time according to a preset period;

[0011] a closed-loop control module configured to update the first threshold value and the second threshold value each time the cardiac treatment drug is used, and configured to update the third threshold value and the fourth threshold value each time the dose of the cardiac treatment drug is adjusted.

[0012] According to the cardiac intensive care intelligent early warning system disclosed by the present application, on the one hand, the cardiac treatment drug relatedness and non-drug relatedness organ dysfunction can be accurately distinguished, and clinical misjudgment can be avoided, and on the other hand, the continuous time trajectory prediction of the kidney and respiratory function indicators can be made, and potential abnormalities can be early warned.

[0013] In addition, the cardiac intensive care intelligent early warning system disclosed by the present application can also have the following additional technical features:

[0014] In an embodiment of the present application, the pre-trained DML model is configured to:

[0015] a processing network configured to input an input parameter of the cardiac treatment drug and output a processed variable hidden vector after processing through 3 fully connected layers ;

[0016] a result network configured to input the ECG data, the kidney marker data and the respiratory function data, and generate an abnormality probability after processing through an attention layer after extracting time sequence features through a 1D-CNN . .

[0017] In an embodiment of the present application, the causal effect loss function of the pre-trained DML model is specifically:

[0018] wherein, is an estimated value of the causal effect of the cardiac treatment drug, is a cardiac treatment drug sensitive index mask vector labeled by a medical expert.

[0019] In an embodiment of the present application, the decision logic of the pre-trained DML model is specifically:

[0020] when the estimated value of the causal effect of the cardiac treatment drug Greater than the fifth threshold and the abnormal probability If the pressure is greater than a sixth threshold, it is determined that the cardiac drug affects the kidneys and lungs, and a warning signal is output, and a stop signal is output to the cardiac drug supply system;

[0021] When the causal effect estimate of the cardiac drug Less than or equal to the fifth threshold and the abnormal probability When it is greater than a sixth threshold, it is determined that the cardiac treatment drug has not affected the kidneys and lungs, and a warning signal is output.

[0022] In one embodiment of the present invention, the pre-trained NODE model is configured as follows:

[0023] The dynamic system modeling layer is used to input a state vector consisting of the concentration of the preset marker in the kidney marker data and the preset functional data in the respiratory function data, and after being processed by a preset differential equation, output the predicted concentration of the preset marker in the kidney marker data and the predicted data of the preset functional data in the respiratory function data at the next moment according to the preset period.

[0024] In one embodiment of the present invention, the preset differential equation in the pre-trained NODE model is specifically:

[0025] ,in, is the concentration of the preset marker in the kidney marker data, is the preset function data in the respiratory function data, The predicted concentration of the preset marker in the kidney marker data, It is the predicted data of the preset function data in the respiratory function data.

[0026] In one embodiment of the present invention, the decision logic of the pre-trained NODE model is specifically as follows:

[0027] When the predicted concentration of the preset marker in the kidney marker data Greater than the seventh threshold or the predicted data of the preset function data in the respiratory function data When it is less than the eighth threshold, an early warning signal is output.

[0028] When the predicted concentration of the preset marker in the kidney marker data The predicted data of the preset function data in the respiratory function data is less than or equal to the seventh threshold value When the value is greater than or equal to an eighth threshold, a maintenance operation signal is output to the cardiac drug supply system.

[0029] In one embodiment of the present invention, the integrator of the pre-trained NODE model is RK4, the initial value of the integration step is 15 minutes, and the error threshold is , the initial value of the prediction time window is 12 hours.

[0030] In one embodiment of the present invention, the predetermined differential equation It is parameterized by a 2-layer residual network.

[0031] In one embodiment of the present invention, the state vector composed of the concentration of the preset marker in the kidney marker data and the preset function data in the respiratory function data includes a timestamp feature.

[0032] Additional contents and advantages of the present invention will be given in the following description or can be understood through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The technical solutions and beneficial effects of the present invention will become apparent and easily understood from the following contents in conjunction with the accompanying drawings, in which:

[0034] Figure 1 A flowchart of the intelligent early warning system for intensive cardiac care of the present invention;

[0035] Figure 2 This is another workflow diagram of the intelligent early warning system for intensive cardiac care of the present invention. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0037] The following will describe the intelligent early warning system for critical care of heart disease disclosed by the present invention with reference to the accompanying drawings. Figure 1 and Figure 2 As shown:

[0038] An intelligent early warning system for cardiac intensive care, comprising:

[0039] A treatment input module is used to receive input parameters of cardiac treatment drugs set by clinical doctors and basic vital signs data of patients;

[0040] Real-time data acquisition module, used to collect patients' ECG data, renal marker data and respiratory function data;

[0041] Exception handling module, configured as:

[0042] When the clearance rate of a preset marker in the renal marker data is less than a first threshold or the preset function data in the respiratory function data is less than a second threshold, inputting the input parameters of the cardiac therapeutic drug, the ECG data, the renal marker data, and the respiratory function data into the pre-trained DML model, respectively, and outputting a causal effect estimate and an abnormal probability value of the cardiac therapeutic drug;

[0043] Pre-trained DML model, configured as:

[0044] Processing network, used to input the input parameters of cardiac therapeutic drugs, and output the processed variable latent vector after processing through 3 layers of fully connected layers ;

[0045] The resulting network is used to input ECG data, renal marker data, and respiratory function data, and extracts time series features through 1D-CNN and Splicing, generating anomaly probabilities after processing by the attention layer ;

[0046] The causal effect loss function of the pre-trained DML model is:

[0047] ,in, For cardiac medications, the causal effect estimate is Mask vectors of cardiac drug sensitivity indicators annotated by medical experts;

[0048] The decision logic of the pre-trained DML model is as follows:

[0049] When the causal effect of cardiac medications is estimated Greater than the fifth threshold and abnormal probability If the value is greater than a sixth threshold, it is determined that the cardiac medication affects the kidneys and lungs, and a warning signal is output, and a stop signal is output to the cardiac medication supply system;

[0050] When the causal effect of cardiac medications is estimated Less than or equal to the fifth threshold and abnormal probability If it is greater than the sixth threshold, it is determined that the cardiac treatment drug has not affected the kidneys and lungs, and a warning signal is output;

[0051] When the concentration of the preset marker in the kidney marker data is less than or equal to the first threshold or the preset functional data in the respiratory function data is greater than the second threshold, the concentration of the preset marker in the kidney marker data and the preset functional data in the respiratory function data are input into the pre-trained NODE model, and the predicted concentration of the preset marker in the kidney marker data and the predicted data of the preset functional data in the respiratory function data at the next moment are output according to a preset period;

[0052] The pre-trained NODE model is configured as follows:

[0053] a dynamic system modeling layer, configured to input a state vector consisting of the concentration of a preset marker in the renal marker data and the preset functional data in the respiratory function data, and output, after processing by a preset differential equation, predicted concentrations of the preset marker in the renal marker data and predicted functional data of the preset functional data in the respiratory function data at the next moment according to a preset period;

[0054] The preset differential equations in the pre-trained NODE model are:

[0055] ,in, is the concentration of the preset marker in the kidney marker data, It is the preset function data in the respiratory function data. Predicted concentrations of pre-set markers in the renal marker data, Predicted data of preset functional data in the respiratory function data;

[0056] The decision logic of the pre-trained NODE model is as follows:

[0057] When the predicted concentration of the preset marker in the renal marker data Greater than the seventh threshold or the predicted data of the preset function data in the respiratory function data When it is less than the eighth threshold, an early warning signal is output.

[0058] When the predicted concentration of the preset marker in the renal marker data The predicted data of the preset function data in the respiratory function data is less than or equal to the seventh threshold value When the value is greater than or equal to the eighth threshold, a maintenance signal is output to the cardiac drug supply system;

[0059] The integrator of the pre-trained NODE model is RK4, the initial value of the integration step is 15 minutes, and the error threshold is , the initial value of the prediction time window is 12 hours;

[0060] Preset differential equation Parameterized by a 2-layer residual network;

[0061] The state vector composed of the concentration of the preset marker in the renal marker data and the preset function data in the respiratory function data includes a timestamp feature;

[0062] The closed-loop control module is configured to update the first threshold and the second threshold each time the cardiac therapy drug is used, and to update the third threshold and the fourth threshold each time the dosage of the cardiac therapy drug is adjusted.

[0063] In an embodiment of the present invention, the treatment input module receives cardiac medication input parameters (including dosage and duration) and patient basic vital signs data (such as age, weight, and baseline vital signs) specified by clinical physicians. The real-time data acquisition module continuously collects patient ECG data (sampling frequency ≥ 250 Hz), renal marker data (such as creatinine concentration and creatinine clearance), and respiratory function data (blood oxygen saturation SpO2 and tidal volume) through bedside monitoring equipment, and synchronizes the data to the system database.

[0064] Specifically:

[0065] When creatinine clearance is less than Or blood oxygen saturation is less than When , activate the DML model;

[0066] The processing network uses a 3-layer fully connected layer (64-32-16, ReLu activation), inputs drug parameters and patient correction values ​​(such as weight correction coefficient), and outputs a latent vector ; Results The network extracted the temporal features of ECG, kidney and respiratory indicators through 1D-CNN (3 convolution kernels with size 5-3-3), and compared them with After splicing, the abnormal probability is generated by the attention layer ;

[0067] In the causal effect loss function In , BCE is binary cross entropy;

[0068] like and , it is determined that the cardiac drug correlation is abnormal, triggering an early warning and automatically cutting off the cardiac drug infusion pump;

[0069] like and , it is determined that the non-cardiac treatment drug-related abnormality is triggered, triggering an early warning and prompting the clinician to investigate other causes;

[0070] When creatinine is less than or equal to And the blood oxygen saturation is greater than or equal to When , the NODE model is activated;

[0071] Dynamic system modeling layer establishes equations , where the state vector contains the creatinine concentration and blood oxygen saturation ;

[0072] Adaptive RK4 integrator is used, the integration step is initialized to 0.25h (15 minutes), and the error threshold is set to ;

[0073] The default prediction window is 12 hours, and the prediction status is output every 0.5 hours. ;

[0074] If at any time or , then trigger an early warning and mark the earliest abnormal time point;

[0075] If all predicted points are within the normal range, maintain the current cardiac medication infusion and proceed to the next round of monitoring (10-minute intervals);

[0076] For example, in a patient with heart failure, when dopamine (dose ) After 4 hours, the system collected creatinine clearance , triggering the DML model;

[0077] Processing network output Hidden vector, resulting in the network generating anomaly probability , causal effect estimate The system determines drug-related renal damage, automatically cuts off the infusion pump, and prompts "dopamine may cause abnormal renal function";

[0078] At the same time, the NODE model is in the initial stage of medication (creatinine ) Predict that creatinine may rise to within 12 hours , issuing an early warning 6 hours in advance, providing a time window for clinical adjustment of treatment plans.

[0079] In addition, for the closed-loop control module, before each medication, the DML model parameters are fine-tuned based on the patient's latest baseline data (such as the average creatinine value and basic SpO2 within 24 hours) to improve individual adaptability;

[0080] After each round of monitoring, if an abnormal event occurs, the system updates the first threshold (60ml / min) and the second threshold (94%) based on patient population data (such as the creatinine clearance distribution of patients with the same drug class and disease). When adjusting the drug dosage, the initial state of the NODE model is automatically reset, and real-time monitoring is triggered.

[0081] Abnormal event data (including warning types, handling measures, and indicator change curves) are stored in a dedicated database. Every 100 cases of data trigger iterative training of the DML and NODE models to optimize causal effect estimation and prediction accuracy.

[0082] In summary, according to the intelligent early warning system for heart intensive care disclosed by the application, on the one hand, the heart treatment drug correlation and non-drug correlation organ function abnormalities can be accurately distinguished, and clinical misjudgment can be avoided; on the other hand, the continuous time trajectory prediction of the kidney and respiratory function indexes can be made, and potential abnormalities can be early warned.

[0083] Although the embodiments of the present application have been shown and described above, it should be understood by those skilled in the art that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. An intelligent early warning system for intensive cardiac care, characterized in that: include: A treatment input module is used to receive input parameters of cardiac treatment drugs set by clinical doctors and basic vital signs data of patients; Real-time data acquisition module, used to collect patients' ECG data, renal marker data and respiratory function data; Exception handling module, configured as: When the clearance rate of a preset marker in the renal marker data is less than a first threshold or the preset function data in the respiratory function data is less than a second threshold, inputting the input parameters of the cardiac therapeutic drug, the ECG data, the renal marker data, and the respiratory function data into a pre-trained DML model, respectively, and outputting a causal effect estimate and an abnormal probability value of the cardiac therapeutic drug; When the concentration of the preset marker in the kidney marker data is less than or equal to a first threshold or the preset functional data in the respiratory function data is greater than a second threshold, the concentration of the preset marker in the kidney marker data and the preset functional data in the respiratory function data are input into a pre-trained NODE model, and the predicted concentration of the preset marker in the kidney marker data and the predicted data of the preset functional data in the respiratory function data at the next moment are output according to a preset period; A closed-loop control module is configured to update the first threshold and the second threshold each time the cardiac therapy drug is used, and to update the third threshold and the fourth threshold each time the dosage of the cardiac therapy drug is adjusted.

2. The intelligent early warning system for intensive cardiac care according to claim 1, characterized in that: The pre-trained DML model is configured as follows: Processing network, used to input the input parameters of the cardiac therapeutic drug, and output the processed variable latent vector after processing through 3 layers of fully connected layers ; The result network is used to input the ECG data, the renal marker data and the respiratory function data, and extract the time series features through 1D-CNN and Splicing, generating anomaly probabilities after processing by the attention layer .

3. The intelligent early warning system for intensive cardiac care according to claim 2, characterized in that: The causal effect loss function of the pre-trained DML model is specifically: ,in, is the estimated causal effect of the cardiac drug, Mask vectors of cardiac drug sensitivity indicators annotated by medical experts.

4. The intelligent early warning system for intensive cardiac care according to claim 3, characterized in that: The decision logic of the pre-trained DML model is as follows: When the causal effect estimate of the cardiac drug Greater than the fifth threshold and the abnormal probability If the pressure is greater than a sixth threshold, it is determined that the cardiac drug affects the kidneys and lungs, and a warning signal is output, and a stop signal is output to the cardiac drug supply system; When the causal effect estimate of the cardiac drug Less than or equal to the fifth threshold and the abnormal probability When it is greater than a sixth threshold, it is determined that the cardiac treatment drug has not affected the kidneys and lungs, and a warning signal is output.

5. The intelligent early warning system for intensive cardiac care according to claim 1, characterized in that: The pre-trained NODE model is configured as follows: The dynamic system modeling layer is used to input a state vector consisting of the concentration of the preset marker in the kidney marker data and the preset functional data in the respiratory function data, and after being processed by a preset differential equation, output the predicted concentration of the preset marker in the kidney marker data and the predicted data of the preset functional data in the respiratory function data at the next moment according to the preset period.

6. The intelligent early warning system for intensive cardiac care according to claim 5, characterized in that: The preset differential equation in the pre-trained NODE model is specifically: ,in, is the concentration of the preset marker in the kidney marker data, is the preset function data in the respiratory function data, The predicted concentration of the preset marker in the kidney marker data, It is the predicted data of the preset function data in the respiratory function data.

7. The intelligent early warning system for intensive cardiac care according to claim 6, characterized in that: The decision logic of the pre-trained NODE model is as follows: When the predicted concentration of the preset marker in the kidney marker data Greater than the seventh threshold or the predicted data of the preset function data in the respiratory function data When it is less than the eighth threshold, an early warning signal is output. When the predicted concentration of the preset marker in the kidney marker data The predicted data of the preset function data in the respiratory function data is less than or equal to the seventh threshold value When the value is greater than or equal to an eighth threshold, a maintenance operation signal is output to the cardiac drug supply system.

8. The intelligent early warning system for intensive cardiac care according to claim 5, characterized in that: The integrator of the pre-trained NODE model is RK4, the initial value of the integration step is 15 minutes, and the error threshold is , the initial value of the prediction time window is 12 hours.

9. The intelligent early warning system for intensive cardiac care according to claim 6, characterized in that: The preset differential equation It is parameterized by a 2-layer residual network.

10. The intelligent early warning system for intensive cardiac care according to claim 5, characterized in that: The state vector composed of the concentration of the preset marker in the kidney marker data and the preset function data in the respiratory function data includes a timestamp feature.

Images