Disclosure of Invention
The present invention aims to solve at least some of the technical problems in the above-described technology.
To this end, the invention discloses a vasoactive drug injection system based on AI and invasive blood pressure monitoring, comprising:
The data acquisition module is used for acquiring the systolic pressure Ps, the diastolic pressure Pd and the mean arterial pressure of the arterial blood pressure at the frequency not lower than a preset frequency through the invasive blood pressure sensorSynchronously acquiring an electrocardiosignal E (t), recording a drug accumulation dose D (t) of the vasoactive drug and generating the mean arterial pressure Pm (t) in a time sequence form with a continuous blood pressure waveform function B (t) and a multi-lead electrocardiograph;
The core operation module is used for constructing a hemodynamic prediction model M1 and a vasoactive drug dosage optimization model M2, wherein,
Inputting the mean arterial pressure Pm (t) in time series to the hemodynamic prediction model M1 to obtain a prediction curve for a future preset time
Inputting the mean arterial pressure Pm (t), a predicted deviation amount in time series form, into the vasoactive drug dose optimization model M2And patient-based physiological parameter vectorTo obtain the adjustment quantity of the infusion rate of the vasoactive medicine at the next moment
The injection regulation and control module is used for receiving the next time vascular active drug infusion rate adjustment quantity delta R (t) and generating a flow rate instruction R (t+1) =R (t) +delta R (t);
A flow sensor for monitoring the flow rate Ra (t) of the vasoactive drug injected by the injection regulation module and generating a control signal of an injection driving unit in the injection regulation moduleWherein,
Error signal e (t) =r (t) -Ra (t);
Kp,Ki,Kd is tuning parameter;
the human-computer interaction module is used for displaying arterial blood pressure waveforms and a blood vessel active drug injection rate curve, and popup window alarming and voice prompt of medical staff when predicted blood pressure deviates from a preset safety range value.
According to the vascular active drug injection system based on AI and invasive blood pressure monitoring, disclosed by the invention, more accurate and personalized drug infusion rate adjustment can be provided by collecting arterial blood pressure data and electrocardiosignals in real time and combining the accumulated dose of the vascular active drug, so that the hemodynamic stability of a patient is ensured, and the delay and error caused by manual adjustment are reduced.
In addition, the vasoactive drug injection system based on AI and invasive blood pressure monitoring according to the present disclosure may also have the following additional technical features:
in one embodiment of the present invention, in the data acquisition module, the preset frequency is not lower than 200Hz.
In an embodiment of the present invention, in the core operation module, the future preset time is in a range of 5 to 8 minutes.
In one embodiment of the present invention, in the core computing module, the patient-based physiological parameter vectorComprising the following steps:
Patient age, sex, weight, height, basal heart rate, basal blood pressure, and whether or not suffering from cardiovascular disease.
In one embodiment of the present invention, in the man-machine interaction module, the preset safety range is 5-10 mmhg.
In one embodiment of the invention, in the man-machine interaction module, the data acquisition module is calibrated according to a standard pressure gas cylinder through an operation interface, and an initial flow rate R (0) of the injection regulation module is set.
In one embodiment of the present invention, further comprising:
And the emergency stop module is used for pressing down when the predicted blood pressure deviates from a preset safety range value so as to stop the injection regulation and control module from injecting the vasoactive medicine.
In one embodiment of the present invention, the hemodynamic prediction model M1 updates the rule once within 5-10 s, and the vasoactive drug dose optimization model M2 immediately generates the next time vasoactive drug infusion rate adjustment amount after each update of the hemodynamic prediction model M1
Additional content and advantages of the invention will be set forth in the description which follows, or may be learned by practice of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments.
The disclosed AI and invasive blood pressure monitoring based vasoactive drug injection system will be described with reference to the drawings.
As shown in fig. 1 and 2, a vasoactive drug injection system 100 based on AI and invasive blood pressure monitoring, comprising:
A data acquisition module 101 for acquiring systolic pressure Ps, diastolic pressure Pd, and mean arterial pressure of arterial blood pressure at a frequency not lower than a preset frequency by an invasive blood pressure sensorSynchronously acquiring an electrocardiosignal E (t), recording a medicine accumulation dose D (t) of the vasoactive medicine and generating an average arterial pressure Pm (t) in a time sequence form with a continuous blood pressure waveform function B (t) and a multi-lead electrocardiograph;
it should be noted that the preset frequency is not lower than 200Hz;
The core operation module 102 is used for constructing a hemodynamic prediction model M1 and a vasoactive drug dose optimization model M2, wherein,
Inputting the mean arterial pressure Pm (t) in time series to the hemodynamic predictive model M1 to obtain a predicted curve at a predetermined time in the future
The mean arterial pressure Pm (t) in time series form and the predicted deviation amount are input into a vasoactive drug dose optimization model M2And patient-based physiological parameter vectorTo obtain the adjustment quantity of the infusion rate of the vasoactive medicine at the next moment
It should be noted that the range of the future preset time is 5-8 min;
it should also be noted that the patient-based physiological parameter vectorComprising the following steps:
Patient age, sex, weight, height, basal heart rate, basal blood pressure, and whether or not suffering from cardiovascular disease;
Furthermore, the hemodynamic prediction model M1 -10 s updates the rule once, and after each update of the hemodynamic prediction model M1, the vasoactive drug dose optimization model M2 immediately generates the vasoactive drug infusion rate adjustment amount at the next moment
An injection regulation module 103, configured to receive an infusion rate adjustment Δr (t) of the vasoactive drug at a next time and generate a flow rate command R (t+1) =r (t) +Δr (t);
It should be noted that the preset safety range is 5-10 mmHg;
A flow sensor 104 for monitoring the flow rate Ra (t) of the vasoactive drug injected by the injection regulation module and generating a control signal for the injection drive unit in the injection regulation moduleWherein,
Error signal e (t) =r (t) -Ra (t);
Kp,Ki,Kd is tuning parameter;
the man-machine interaction module 105 is used for displaying arterial blood pressure waveforms and a blood vessel active drug injection rate curve, and popup window alarming and voice prompt of medical staff when predicted blood pressure deviates from a preset safety range value;
It should be noted that, the data acquisition module is calibrated according to the standard pressure gas cylinder through the operation interface, and the initial flow rate R (0) of the injection regulation module is set;
In addition, the AI and invasive blood pressure monitoring based vasoactive drug injection system further comprises:
And the emergency stop module is used for pressing down when the predicted blood pressure deviates from a preset safety range value so as to stop the injection of the vasoactive medicine by the injection regulation and control module.
According to the invention, medical staff calibrate the data acquisition module according to the standard pressure gas cylinder through the operation interface of the man-machine interaction module, so that the acquired blood pressure data is ensured to be accurate and reliable, and meanwhile, the initial flow rate R (0) =5 ml/h of the injection regulation and control module is set, and the injection of the vasoactive medicine for a patient is started;
In the data acquisition module 101, the invasive blood pressure sensor acquires the systolic pressure Ps, the diastolic pressure Pd, and the mean arterial pressure of the arterial blood pressure at a frequency of 200HzSynchronously acquiring an electrocardiosignal E (t), recording a medicine accumulation dose D (t) of the vasoactive medicine and generating an average arterial pressure Pm (t) in a time sequence form with a continuous blood pressure waveform function B (t) and a multi-lead electrocardiograph;
The hemodynamic predictive model M1 in the core operational module 102 receives the mean arterial pressure Pm (t) in time series, and the cardiac electrical signals E (t-30) for the past 30s, E (t) and the drug cumulative dose D (t-30), D (t);
After the hemodynamic prediction model M1 is updated at time t1, the vasoactive drug dose optimization model M2 immediately acquires the current mean arterial pressure Pm(t1) and the predicted deviationPatient-based physiological parameter vectorFor example, patients aged 65 years, men, weight 70kg, height 175cm, basal heart rate 70times/min, basal blood pressure 120/80mmHg, and suffering from cardiovascular disease;
from these data, the vasoactive drug dose optimization model M2 calculates the next time (t1 +1) vasoactive drug infusion rate adjustment amount Δr (t1) =0.5 ml/h;
After receiving Δr (t1), the injection regulation module 103 generates a new flow rate command R (t1+1)＝R(t1)+ΔR(t1) =5+0.5=5.5 ml/h, and controls the injection driving unit to perform vasoactive drug injection according to the new flow rate;
The flow sensor 104 monitors the actual flow rate Ra(t1 of the injection regulating and controlling module for injecting the vasoactive drug in real time, if the actual flow rate is 5.4ml/h, the error signal e (t1)＝R(t1)-Ra(t1) =5.5-5.4=0.1 ml/h, the flow sensor generates a control signal u (t1) of the injection driving unit according to the error signal e (t1) and the tuning parameter Kp,Ki,Kd (assuming Kp＝1,Ki＝0.1,Kd =0.05), and fine tuning is carried out on the injection flow rate to reduce the error, so that the actual flow rate is closer to the command flow rate;
The man-machine interaction module displays arterial blood pressure waveforms and a blood vessel active drug injection rate curve in real time, and medical staff can intuitively observe the blood pressure change and the drug injection condition of a patient.
In the subsequent monitoring process, the predicted blood pressure deviates from a preset safety range value by 5-10 mmHg, for example, the predicted mean arterial pressure is lower than 55mmHg, the man-machine interaction module immediately pops a window for alarming and prompts medical staff by voice, and meanwhile, the scram module can be pressed by the medical staff in emergency so as to stop the injection of the vasoactive medicine by the injection regulation module, thereby ensuring the safety of patients.
And in particular, with respect to two models used in embodiments of the present invention, wherein,
A hemodynamic prediction model M1, which is constructed based on a long-short-term memory network LSTM, and is intended to predict future mean arterial pressure MAP trend of the patient according to historical data;
For the collected arterial blood pressure data (systolic Ps, diastolic Pd, mean arterial pressure)Synchronously acquiring an electrocardiosignal E (t) with a multi-lead electrocardiograph, recording a medicine accumulation dosage D (t) of a vasoactive medicine and generating an average arterial pressure Pm (t) in a time sequence form, firstly cleaning data, removing abnormal values and noise interference, for example, if the blood pressure value at a certain moment is detected to deviate from a normal physiological range obviously and lack continuity with front and rear data, treating the blood pressure value as the abnormal value for correction or elimination, and then carrying out normalization processing on the cleaned data to enable the data with different characteristics to have similar dimensions, thereby facilitating model training;
The LSTM network is formed by stacking a plurality of LSTM layers, each LSTM layer comprises a plurality of memory units, a model can learn the long-term dependence in data by setting the number of proper hidden layers and the number of the memory units, for example, 3 hidden layers are selected and set through experimental comparison, each hidden layer comprises 64 memory units, and the time sequence characteristics of the data such as blood pressure, electrocardiosignals and the like can be effectively captured while the calculation efficiency is ensured;
After the LSTM layer, a fully connected layer is connected, mapping the output of the LSTM layer to the predicted MAP value. The number of neurons of the fully connected layer is determined according to the dimension of the prediction target MAP, here 1;
Dividing the pre-processed time series data (including mean arterial pressure Pm(t-30),…,Pm (t), cardiac signal E (t-30) over 30 seconds,) E (t) and drug accumulation dose D (t-30), D (t) into training, validation and test sets, which can be divided in a ratio of 70%,15%, 15%;
training the model by using a training set, adjusting the weight and bias of the model by a back propagation algorithm to minimize a mean square error MSE between a predicted value and an actual value, wherein the MSE is calculated by a formula,Where n is the number of samples, yi is the actual value,Is a predicted value;
In the training process, an early stopping method is adopted to prevent overfitting, namely training is stopped when a loss function on a verification set is not reduced any more, and a learning rate attenuation strategy is used, so that the learning rate is gradually reduced along with the increase of the training round number, the convergence effect of a model is improved, for example, the initial learning rate is set to be 0.001, and the learning rate is attenuated to be 0.9 times of the original learning rate after 10 epochs pass;
Evaluating the trained model by using a test set, calculating evaluation indexes such as Root Mean Square Error (RMSE), mean Absolute Error (MAE) and the like to measure the prediction performance of the model, wherein the calculation formula of the RMSE is as follows,The calculation formula of the MAE is as follows,
The vasoactive drug dose optimization model M2 is used for constructing a Depth Q Network (DQN) based on reinforcement learning, and aims to determine the optimal vasoactive drug infusion rate adjustment quantity according to the current hemodynamic state and the basic physiological parameters of a patient;
the state space comprises the average arterial pressure Pm (t) at the current moment and the predicted deviation quantity output by the hemodynamic prediction model M1Patient-based physiological parameter vector(Including information of age, sex, weight, height, basic heart rate, basic blood pressure, whether suffering from cardiovascular diseases and the like of the patient), the state variables can comprehensively reflect the current physiological state and hemodynamic variation trend of the patient, and provide basis for adjusting the dosage of the medicine;
continuous state variables (such as blood pressure, heart rate, etc.) are discretized and divided into intervals for DQN model processing, e.g. average arterial pressure is divided into intervals of every 5mmHg and heart rate is divided into intervals of 10 times per minute. For classification variables (such as gender, whether cardiovascular diseases exist or not, etc.), the classification variables are expressed by adopting a single-heat coding mode;
the action space is the adjustment amount DeltaR (t) of the infusion rate of the vasoactive drug, and the adjustment amount is set to be a limited discrete value set, such as { -0.5ml/h, -0.25ml/h,0ml/h,0.25ml/h,0.5ml/h } in consideration of safety and effectiveness in practical clinical application, which means that a certain amount of infusion rate can be selectively reduced, maintained or increased;
the reward function is designed to guide the model to select the action which can enable the hemodynamic state of the patient to be closer to the target range (namely to keep stable), if the predicted mean arterial pressure at the next moment is closer to the preset target range (for example, the normal mean arterial pressure range is 70-105 mmHg), the positive reward is given, otherwise, if the predicted blood pressure deviates from the target range further, the negative reward is given;
The specific form of the reward function can be adjusted and optimized according to practical situations, for example, the reward value is calculated by adopting the following formulaWherein,For the mean arterial pressure at the next moment predicted from the current motion, Ptarget is the target mean arterial pressure (may take the middle value of the target range, such as 87.5 mmHg);
The DQN model learns the optimal strategy by interacting with the environment (here the patient's hemodynamic system), and at each time step t, the model selects an action at, i.e. the drug infusion rate adjustment, according to the current state st, after which the environment will feed back the next state st+1 and the corresponding reward rt;
Storing the experience tuples (st,at,rt,st+1) in an experience playback buffer, during training, randomly sampling a batch of experience data from the buffer for updating a Q-function of the model, the Q-function representing an expected jackpot for taking some action in a given state;
The Q function is approximated using a deep neural network, the input of the network being a state vector, the output being an estimate of the Q for each action, the network parameters being updated by minimizing the mean square error between the target Q and the predicted Q, the calculation formula for the target Q being,Wherein, gamma is a discount factor for balancing the importance of future rewards and current rewards, and the value is usually 0.9-0.99.
In summary, according to the vascular active drug injection system based on AI and invasive blood pressure monitoring disclosed by the invention, more accurate and personalized drug infusion rate adjustment can be provided by collecting arterial blood pressure data and electrocardiosignals in real time and combining the accumulated dose of the vascular active drug, so that the hemodynamic stability of a patient is ensured, and delay and error caused by manual adjustment are reduced.
In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed, mechanically connected, electrically connected, directly connected, indirectly connected via an intervening medium, or in communication between two elements or in an interaction relationship between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the present invention, unless expressly stated or limited otherwise, a first feature "up" or "down" a second feature may be the first and second features in direct contact, or the first and second features in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.