Propofol drug delivery system and method based on fast electroencephalogram characteristics, medium and product

Abstract

The invention discloses a propofol drug delivery system and method based on appreciation electroencephalogram characteristics, a medium and a product, and relates to the field of intelligent medical systems, and the propofol drug delivery system comprises a signal collection module used for collecting original electroencephalogram signals; the signal processing module is used for mapping the real-time brain function state of the patient to a neural state space containing an appreciation related state and determining a neural state; the prediction calculation module is used for predicting a future evolution trajectory and a state transition probability of a neural state in a neural state space; the control decision module is used for generating a drug administration intervention strategy of propofol by taking the brain function state staying in the target appreciation area as a target; and the infusion execution module is used for adjusting the target control infusion pump according to the administration intervention strategy so as to guide and maintain the appreciation state. By implementing the application, the patient can be prevented from being in a poor sedative state as much as possible, and the operation prognosis effect is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical systems, and in particular to a propofol administration system, method, medium and product based on euphoria electroencephalogram features. BACKGROUND

[0002] In modern anesthesiology and sedation medicine, the comfort of patient's medical experience is increasingly valued. Studies have found that propofol, as a widely used intravenous anesthetic drug, not only has good sedation and anesthetic effects, but also can produce euphoria and relaxation in some patients at subanesthetic doses, improve the patient's medical experience, reduce preoperative anxiety and improve satisfaction. However, the euphoric effect of propofol has high individual variability, and its occurrence and maintenance require precise dose control.

[0003] In related technologies, an anesthetic administration system based on electroencephalogram monitoring is mainly used to realize automatic drug infusion control. The electroencephalogram signal of the patient is collected by an electroencephalogram monitoring device such as a BIS monitor or an entropy index monitor, electroencephalogram parameters such as the bispectral index and spectral entropy are extracted, the depth of sedation is analyzed based on the electroencephalogram parameters, and the adjustment amount of the drug infusion rate to maintain a specified depth of sedation is calculated, and the rate of anesthetic drug such as propofol is adjusted by a target-controlled infusion pump. That is, the related technology analyzes the depth of sedation by BIS, entropy index and other parameters, controls the anesthetic administration based on the depth of sedation, and maintains the stability of anesthesia to reduce the workload of medical staff.

[0004] However, in some short-term anesthesia or sedation processes with high requirements for patient state (such as gastrointestinal endoscopy, plastic surgery, etc.), patients exhibit different states at the same depth of sedation index (for example, BIS value of 70), and two states may occur: one is a euphoric state of relaxation, pleasure and cooperation, and the other is a suboptimal sedation state of restlessness and anxiety, which significantly increases the difficulty of surgery and affects the prognosis of surgery. SUMMARY

[0005] The present application provides a propofol administration system, method, medium and product based on euphoria electroencephalogram features, which is used to avoid the patient being in a suboptimal sedation state as much as possible and optimize the prognosis of surgery.

[0006] In a first aspect, the present application provides a propofol administration system, comprising: a signal acquisition module configured to acquire an original electroencephalogram signal of a patient; a signal processing module configured to map a real-time brain function state of the patient to a neural state space containing a state related to euphoria based on a multidimensional feature dynamic pattern of the original electroencephalogram signal, and determine a neural state; a prediction calculation module configured to predict a future evolution trajectory and a state transition probability of the neural state in the neural state space according to the neural state and a dynamic model representing a drug effect of propofol; a control decision module configured to generate a propofol administration intervention strategy based on the future evolution trajectory and the state transition probability, so as to maintain the brain function state in a target euphoria region; and an infusion execution module configured to adjust a target-controlled infusion pump according to the administration intervention strategy, so as to guide and maintain the euphoria state.

[0007] In the above embodiment, the propofol administration system can monitor the electroencephalogram features that can reflect euphoria in real time, and predict the changes in drug effect, so as to achieve precise administration control. The propofol administration system can not only actively induce and maintain the euphoria state of the patient, but also avoid the poor sedation such as restlessness and anxiety caused by relying only on the depth of sedation, thereby improving the patient's experience and satisfaction in medical treatment, and realizing emotion-guided personalized and comfortable medical treatment.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the signal processing module comprises: a pattern extraction unit configured to synchronously extract brain network topology features, cross-band coupling features and time-domain oscillation pattern features from the original electroencephalogram signal, and construct a high-dimensional feature vector; and a state mapping unit configured to match a real-time evolution pattern of the high-dimensional feature vector with preset patterns of states in the neural state space, and determine a neural state with the highest confidence.

[0009] In the above embodiment, the propofol administration system uses the pattern extraction unit to deeply mine deep information related to euphoria in the electroencephalogram signal from multiple dimensions, and constructs a comprehensive high-dimensional feature vector. In addition, the state mapping unit can accurately match these complex feature patterns with preset neural states such as euphoria, sedation and anxiety, thereby improving the recognition ability and specificity of the system for the euphoria state, providing a reliable basis for subsequent precise administration control, and ensuring the accuracy of the control target.

[0010] In combination with some embodiments of the first aspect, in some embodiments, the signal processing module further comprises: a confidence monitoring unit configured to monitor a classification confidence of the neural state; and a decision suspension unit configured to suspend the administration intervention when the classification confidence is lower than a preset safety threshold, until the classification confidence recovers to above the preset safety threshold.

[0011] In the above embodiment, the propofol administration system introduces a confidence monitoring and decision suspension mechanism. When the electroencephalogram signal quality is poor or the patient state pattern is ambiguous, resulting in a decrease in the confidence of the neural state classification, the system can automatically suspend the intervention to avoid making incorrect administration decisions based on uncertain information, thereby enhancing the robustness and safety of the propofol administration system and ensuring that automatic control is only performed when the patient state is sufficiently determined, thereby effectively preventing potential administration risks caused by signal interference and other factors.

[0012] In combination with some embodiments of the first aspect, in some embodiments, the propofol administration system further comprises an external event monitoring module configured to monitor an event signal indicative of a strong surgical stimulus; the prediction calculation module is further configured to switch to a disturbance emergency prediction model to recalculate the state transition probability and predict a state transition trajectory caused by the stimulus when the event signal is received; the disturbance emergency prediction model superimposes the instantaneous impact effect of the strong stimulus on the neural state; and the control decision module is further configured to generate an immediate compensation administration strategy aimed at resisting the influence of the stimulus and pulling the neural state back to the target euphoria region.

[0013] In the above embodiment, the propofol administration system can switch to an emergency prediction model in advance by monitoring the strong surgical stimulus signal, estimate the impact of the stimulus on the patient's neural state, and generate a compensation administration strategy, so that the propofol administration system can intervene before the stimulus actually occurs and affects the patient, effectively resist the influence of the stimulus, suppress state fluctuations to the minimum range, and more stably maintain the patient in the target euphoria region, thereby ensuring the smoothness of the surgical process and the comfort of the patient.

[0014] In combination with some embodiments of the first aspect, in some embodiments, the propofol administration system further comprises a bias calculation module configured to compare the predicted neural state determined by the prediction calculation module and the actual neural state determined by the signal processing module to determine a data cumulative bias; and a model calibration module configured to adjust the patient-specific pharmacodynamic parameters in the dynamic model to align the model prediction results with the actual observation results when the data cumulative bias exceeds a preset bias threshold.

[0015] In the above embodiment, the bias calculation and model calibration module of the propofol administration system can monitor the difference between the prediction model and the actual response of the patient in real time, and dynamically adjust the individualized pharmacodynamic parameters in the model according to the cumulative bias, so that the prediction model can continuously adapt to the specificity of the patient during use, making the prediction model more and more close to the individual true situation, thereby improving the accuracy of prediction and the precision of control.

[0016] In some embodiments of the first aspect, the control decision module comprises: a global phase identification unit configured to identify, according to the historical sequence of the neural state, whether the current process is in the euphoria induction phase or the euphoria maintenance phase; an induction period strategy generation unit configured to generate, when the current process is identified as being in the euphoria induction phase, a path planning type of drug administration strategy with the optimization objective of minimizing state transition time and avoiding entering an undesirable intermediate state; and a maintenance period strategy generation unit configured to generate, when the current process is identified as being in the euphoria maintenance phase, a perturbation suppression type of drug administration strategy with the optimization objective of minimizing drug consumption and maximizing state stability.

[0017] In the above embodiments, the global phase identification unit of the propofol administration system can intelligently distinguish between the induction period and the maintenance period and invoke the corresponding strategy generation unit. The induction period pursues rapid and smooth arrival at the target state, while the maintenance period focuses on stable and economical maintenance of the state, making the entire drug administration process more efficient and optimized, achieving euphoria quickly and maintaining it with the lowest drug consumption and minimal state fluctuation, thereby improving the overall performance of the control.

[0018] In some embodiments of the first aspect, the induction period strategy generation unit comprises: a phase design subunit configured to decompose the total drug administration process of the induction period into multiple-stage micro-dose administration trips; a response evaluation subunit configured to, after each micro-dose administration trip is executed, evaluate the transition rate and transition direction of the neural state through the prediction calculation module to determine the patient's immediate drug sensitivity; and a plan adjustment subunit configured to adjust the dose and time interval of the subsequent micro-dose administration trip according to the patient's latest immediate drug sensitivity.

[0019] In the above embodiments, the propofol administration system administers the drug in micro-dose steps and evaluates the patient's response in real time. The system can accurately grasp the patient's sensitivity to the drug at the current time and continuously optimize the subsequent drug administration plan, thereby improving the individualization and safety of the induction process and avoiding too fast or too slow induction due to individual differences, achieving a more precise and safe euphoria state achievement path.

[0020] In a second aspect, the embodiments of the present application provide a propofol administration method based on hedonic electroencephalogram features, the propofol administration method comprising: collecting original electroencephalogram signals of a patient; mapping a real-time brain function state of the patient to a neural state space containing hedonic related states based on a multi-dimensional feature dynamic pattern of the original electroencephalogram signals, and determining a neural state; predicting a future evolution trajectory and state transition probability of the neural state in the neural state space according to the neural state and a dynamic model representing the efficacy of propofol; generating a propofol administration intervention strategy based on the future evolution trajectory and the state transition probability, with the goal of maintaining the brain function state in a target hedonic region; and adjusting a target-controlled infusion pump according to the administration intervention strategy to guide and maintain the hedonic state.

[0021] In a third aspect, the embodiments of the present application provide a computer readable storage medium comprising instructions for running on a propofol administration system.

[0022] In a fourth aspect, the embodiments of the present application provide a computer program product comprising instructions for running on a propofol administration system.

[0023] It can be understood that the propofol administration method based on hedonic electroencephalogram features provided in the second aspect, the computer storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all executed on the propofol administration system provided in the embodiments of the present application. Therefore, the beneficial effects that can be achieved are referred to the beneficial effects in the corresponding method, which will not be described here.

[0024] The one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Since the multi-dimensional feature dynamic pattern of the original electroencephalogram signals is used to map the brain function state of the patient to the neural state space containing the hedonic state, and a closed-loop control framework is used for prediction and decision-making based thereon, the propofol administration system can directly take the hedonic feeling of the patient as the core control target beyond the traditional sedation depth monitoring. The propofol administration system accurately identifies the complex electroencephalogram pattern related to the hedonic feeling through the signal processing module, which is different from the traditional technology relying only on a single index such as BIS; the prediction calculation module uses a dynamic model to predict state evolution, so that the control decision module can generate an active and accurate administration strategy rather than passive and lagging adjustment. The problem that the existing technology cannot distinguish between the hedonic and suboptimal sedation states, resulting in large differences in patient experience and inability to achieve emotional state closed-loop control, is avoided.

[0025] 2. Due to the adoption of an external event monitoring module, a disturbance emergency prediction model, and an immediate compensatory dosing strategy, this propofol delivery system possesses a proactive disturbance response capability. When the external event monitoring module detects signals indicating strong stimuli, such as surgical incision or colonoscopy passing through a bend, the prediction calculation module immediately switches to a disturbance emergency prediction model specifically superimposed with the stimulus impact effect, accurately predicting the trajectory of impending drastic state fluctuations. The control decision module can generate and execute an immediate compensatory dosing strategy before the stimulus actually affects the patient, enhancing the stability of euphoric state maintenance and the smoothness of the surgical procedure, thereby improving the system's safety and clinical applicability.

[0026] 3. By employing a deviation calculation module that continuously compares the predicted neural state with the actual neural state, and a model calibration module that can adjust patient-specific pharmacodynamic parameters when the deviation exceeds a threshold, this propofol delivery system constructs a dynamic and adaptive learning and calibration mechanism. This mechanism ensures that the dynamic model characterizing propofol efficacy is no longer a fixed population model, but rather can continuously self-optimize and be personalized based on each patient's unique physiological response during use. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of a propofol delivery system based on euphoric EEG characteristics in an embodiment of this application. Figure 2 This is another schematic diagram of the propofol delivery system based on euphoric EEG characteristics in the embodiments of this application; Figure 3 This is a schematic flowchart of a propofol administration method in an embodiment of this application; Figure 4 This is a schematic diagram of the physical device structure of a propofol delivery system according to an embodiment of this application.

[0028] Explanation of reference numerals in the attached figures: 101. Signal Acquisition Module; 102. Signal Processing Module; 103. Prediction Calculation Module; 104. Control Decision Module; 105. Infusion Execution Module; 1021. Pattern Extraction Unit; 1022. State Mapping Unit; 1023. Confidence Monitoring Unit; 1024. Decision Pause Unit; 1041. Global Stage Identification Unit; 1042. Induction Period Strategy Generation Unit; 10421. Stage Design Subunit; 10422. Response Evaluation Subunit; 10423. Plan Adjustment Subunit; 1043. Maintenance Period Strategy Generation Unit; 106. External Event Monitoring Module; 107. Deviation Calculation Module; 108. Model Calibration Module. Detailed Implementation

[0029] The terminology used in the following examples of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the description of the application, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0030] Hereinafter, the terms "first", "second" are used only for the purpose of description, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise stated.

[0031] In the specific implementation scenarios proposed in the present application, for example in the performance of fiber colonoscopy or orthopedic surgery requiring high patient cooperation and comfort, the clinical goal is not only to achieve basic sedation, but also to actively guide and maintain the patient in a relaxed and happy state of mind and body, so as to optimize the surgical conditions and patient prognosis.

[0032] To achieve this goal, the present application constructs a closed-loop control logic based on the quantification of brain function state, and the key technical terms involved are explained as follows. The neural state space is a multi-dimensional coordinate system constructed by mathematical methods, and each coordinate axis is defined by different quantitative EEG features, such as orthogonal or non-orthogonal bases composed of brain network topology features (such as clustering coefficient in graph theory), cross-frequency coupling features (such as theta-gamma phase amplitude coupling strength) and time-domain oscillation mode features (such as instantaneous power of alpha band) and the like. This space is essentially different from the one-dimensional scale in the prior art which only uses a single sedation depth index (such as BIS value) to describe the brain state; in the neural state space, each point uniquely corresponds to a comprehensive brain function state defined by a combination of multi-dimensional EEG features. Therefore, this space can finely distinguish between states with different physiological manifestations at the same sedation depth index (for example, BIS value is 70), such as happy cooperation and dysphoric anxiety.

[0033] The target happy region is a pre-labeled subspace or volume in this neural state space, and its boundary and center are trained and labeled by a large amount of clinical data through machine learning algorithms (such as clustering analysis and support vector machine). If the real-time brain function state of the patient falls into this region, it is recognized by the system as having reached the ideal happy state.

[0034] The future trajectory is the prediction of the continuous path of the patient's neural state point in the space under a certain drug administration strategy within a future time window (e.g. 60 seconds), based on the current neural state (the starting point in the space) and the pharmacokinetics / pharmacodynamics (PK / PD) dynamic model representing the drug effect of propofol. The state transition probability is a quantitative evaluation of this predicted path, specifically the likelihood of transitioning from the current state point to the target euphoria region or any other predefined region within a certain future time.

[0035] In summary, the scheme of the present application builds and uses this set of neural state space and prediction system, so that the administration decision of propofol is no longer based on a vague and single sedation depth index, but is changed into a forward-looking optimal control problem with the purpose of accurately reaching and maintaining in the target euphoria region defined by multiple characteristics, which solves the fundamental problem of closed-loop control of emotional state that cannot be achieved by the prior art.

[0036] The system provided by the present embodiment is described below. Please refer to Figure 1 , which is a system structure diagram of the propofol administration system based on euphoric electroencephalogram features in the embodiment of the present application.

[0037] Figure 1 In the embodiment, the propofol administration system comprises: The signal acquisition module 101 is configured to acquire the original electroencephalogram signal of the patient.

[0038] The original electroencephalogram signal refers to the weak voltage fluctuation signal directly acquired by the electrode worn on the patient's head (such as the forehead) without processing, which reflects the electrical activity of the cerebral cortex neurons.

[0039] Specifically, before clinical application, medical staff will paste a special electroencephalogram acquisition electrode sheet on the designated position of the patient's forehead, and connect it to the signal acquisition module 101 through the lead wire. The signal acquisition module 101 is built-in with high-precision bioelectric signal amplifier and analog-to-digital converter. During the whole sedation or anesthesia process, the signal acquisition module 101 continuously captures the microvolt-level voltage changes on the surface of the patient's scalp at a high sampling rate (e.g. 256Hz or higher), and converts it into a digital signal stream, which is transmitted to the signal processing module 102 in real time, providing a continuous and high-quality data source for subsequent state analysis.

[0040] In some embodiments, the acquisition of raw EEG signals can be achieved in various ways: optionally, a wired EEG acquisition device that meets medical standards is used, the electrode is connected to the host through a physical cable to ensure the stability and anti-interference of signal transmission; the steps include cleaning the patient's forehead skin, pasting disposable electrodes, and connecting the lead wire to the amplifier. Optionally, a wireless Bluetooth or Wi-Fi EEG acquisition headband is used to provide patients with greater freedom and comfort; the steps include charging the wireless module, connecting the module to the propofol administration system host through pairing, and wearing and fixing the headband for the patient. It can be understood that the acquisition of EEG signals can also be achieved by integrating a multi-parameter module in a monitor, which is not limited here.

[0041] In some embodiments, there is electromyographic (EMG) interference, especially in the case of light sedation, the patient's involuntary facial muscle twitching produces high-frequency artifacts, which seriously pollutes the raw EEG signal, leading to misjudgment of the euphoric state. To this end, the signal acquisition module 101 can integrate a real-time impedance monitoring function to ensure good contact between the electrode and the skin to reduce external noise. At the same time, in the signal preprocessing stage, an independent component analysis (ICA) algorithm is used, which can decompose the mixed EEG signal into multiple independent source signal components. By identifying and removing components with typical EMG signal characteristics (such as high frequency and burstiness), the remaining pure EEG components are reconstructed, effectively separating and removing EMG interference.

[0042] The signal processing module 102 is used to map the real-time brain function state of the patient to a neural state space containing euphoric-related states based on the multi-dimensional feature dynamic pattern of the raw EEG signal, and determine the neural state.

[0043] Among them, the multi-dimensional feature dynamic pattern refers to the combination rule of multiple different dimensional features (such as frequency domain, time domain, connectivity, etc.) extracted from the EEG signal changing over time. The neural state space is a pre-constructed mathematical model for describing the state of the brain.

[0044] Specifically, after the signal processing module 102 receives the raw EEG signal from the signal acquisition module 101, it first performs preprocessing, including filtering, denoising and artifact rejection. The signal processing module 102 will extract multiple key features from the processed signal in parallel, for example, using fast Fourier transform to calculate the power of the Gamma frequency band, analyzing the frontal lobe α wave power asymmetry (frontal lobe lateralization), and calculating the phase synchronization between different brain regions to represent the coupling strength of the reward network. These features form a high-dimensional feature vector, and the signal processing module 102 inputs the real-time value of this vector into a pre-trained classifier (such as a support vector machine or a deep neural network), which maps the vector to the neural state space and outputs the current most likely neural state, such as euphoria, anxiety, sedation or wakefulness.

[0045] In some embodiments, the mapping of the brain functional state can be achieved in various ways: optionally, a template matching based method is adopted; the steps are as follows: a standard brain electrical characteristic template library of euphoria state is established, the similarity of the current characteristic vector and each template in the library is calculated in real time, and the state corresponding to the template with the highest similarity is selected as the output. Optionally, a machine learning model based method is adopted; the steps are as follows: a large amount of labeled (such as through patient subjective score) electroencephalogram data is used to train a classification model, the real-time extracted feature vector is input into the model, and the classification result is directly output. It can be understood that other time sequence models such as hidden Markov model can also be used for state mapping, which is not limited here.

[0046] It should be noted that the training phase of the classification model requires a large amount of “electroencephalogram-state” paired data. Specifically, by recruiting subjects, under the condition of giving propofol and strict guardianship, the multi-lead original electroencephalogram signals and the instant feedback of their subjective feelings (for example, through handheld device buttons, 1 represents euphoria, 2 represents ordinary sedation, 3 represents restlessness and anxiety, etc.) are synchronously collected, or the objective state is labeled by trained anesthesiologists according to the micro-expression and body language of the patient. The collected electroencephalogram signals are preprocessed and feature extracted (such as the brain network topological features and cross-band coupling features mentioned above), forming a high-dimensional feature vector sequence. These time-stamped feature vector sequences and their corresponding state labels (euphoria, anxiety, sedation, wakefulness, etc.) jointly constitute the training data set. The goal of training is to minimize the classification cross-entropy loss function (Cross-Entropy Loss) between the model's predicted state and the true label, and continuously adjust the internal weight parameters of the model through the back propagation algorithm and the gradient descent optimizer (such as Adam) until the model reaches the preset classification accuracy (such as more than 90%) on the independent validation data set.

[0047] The main structure of the model can be a hybrid neural network, for example, stacked by a convolutional neural network (CNN) and a long short-term memory network (LSTM). Specifically, the input is a high-dimensional feature vector matrix (dimension: feature number x time point number) within a time window (such as the past 10 seconds). The CNN layer is responsible for capturing the local dynamic pattern and combination relationship of the features in a short time, and extracting higher level abstract features. Subsequently, these abstract feature sequences are sent to the LSTM layer, and the recurrent structure of the LSTM can learn and remember the long-term dependence relationship of the feature sequence over time, which is crucial for distinguishing the dynamic evolution process of the state. The last layer of the model is a Softmax activation function layer, which converts the output of the LSTM into a probability distribution vector, and each element of the vector corresponds to the prediction probability of each neural state such as “euphoria”, “anxiety”, “sedation”, “wakefulness”, etc.

[0048] In practical application, the system inputs the feature vector matrix of the latest time window obtained after processing the real-time collected electroencephalogram signal into the CNN-LSTM model which has been trained. The model performs a forward propagation calculation, and finally generates a probability distribution vector at the output end, for example, [0.85, 0.05, 0.10, 0.00], which represents that the system determines that the confidence of the current state being “euphoria” is 85%, “anxiety” is 5%, and “calm” is 10%. The system will select the state with the highest probability (in this case, “euphoria”) as the currently determined “neural state”, and output the highest probability value (0.85) as the “classification confidence” for subsequent modules.

[0049] In some embodiments, the euphoria state has great individual differences, and a general model may not be able to accurately identify the euphoria electroencephalogram features of all patients. For this purpose, the signal processing module 102 can include a personalized calibration process. During the baseline recording stage before the start of drug administration, the system guides the patient to perform a simple task of relaxation or recalling a pleasant experience, and records the electroencephalogram data at this time as the personalized euphoria tendency baseline of the patient. In the early induction stage, the system allows the patient to provide feedback through a button or the like, and these feedbacks will be used as instant labels to fine-tune the pre-trained model online, so that the state mapping model quickly adapts to the unique electroencephalogram expression pattern of the current patient, and improves the accuracy of state recognition.

[0050] The prediction calculation module 103 is used to predict the future evolution trajectory and state transition probability of the neural state in the neural state space according to the neural state and the dynamic model representing the effect of propofol.

[0051] The dynamic model representing the effect of propofol refers to a mathematical model that describes the quantitative relationship between the change of propofol blood concentration or effect site concentration and the movement of state points in the neural state space. The future evolution trajectory is the predicted movement path of the state point in a future period of time (such as 30 seconds).

[0052] Specifically, the prediction calculation module 103 receives the current neural state determined by the signal processing module 102 (i.e. a starting point in the state space), and at the same time obtains the current and planned propofol infusion rate from the infusion execution module 105. It inputs these infusion information into a pharmacokinetics / pharmacodynamics (PK / PD) model to calculate the effect site concentration change curve of propofol in the future period of time. Then, using the dynamic model, the concentration change curve is converted into the expected motion vector of the neural state in the state space. Through numerical integration, the prediction calculation module 103 can depict the future evolution trajectory of the state point and calculate the probability of the trajectory passing through different state regions (such as from the calm region to the euphoria region), i.e. the state transition probability.

[0053] In some embodiments, the prediction of future evolution trajectories can be achieved in various ways: optionally, a Kalman filter based method can be used; the steps of which are: modeling the evolution process of neural state as a linear dynamic system, using Kalman filter to fuse model prediction and actual observation, updating and predicting the state at the next time in real time. Optionally, a recurrent neural network (RNN) or long short-term memory network (LSTM) based method can be used; the steps of which are: using the historical drug administration sequence and neural state sequence as input, training a deep learning model to directly predict the future state sequence. It can be understood that other model predictive control (MPC) based algorithms can also be used for trajectory prediction, which is not limited here.

[0054] It should be noted that the dynamic model representing the efficacy of propofol is to mathematically associate the change in drug concentration with the movement of the state point in the neural state space, and the training of this model requires a comprehensive data set containing drug infusion history, drug concentration change and neural state evolution trajectory. First, using a standard pharmacokinetic (PK) model (such as the Marsh or Schnider model), the effect-site concentration (Ce) curve changing with time is calculated according to the basic information of the patient such as weight, age and the actual propofol infusion record. At the same time, using the state mapping model described above, the actual movement trajectory of the patient's neural state in the multidimensional space within the same time period is obtained, i.e. a series of state vectors S(t). The goal of training is to establish a function that describes the relationship between the change rate of the state vector (i.e. the derivative dS / dt) and the current state vector S(t) and the effect-site concentration Ce(t). Specifically, it is to find a function f, such that the root mean square error (RMSE) between the predicted trajectory S_pred(t) obtained by integrating the differential equation dS / dt=f(S(t), Ce(t), P) and the actual trajectory S(t) is minimized. Where P represents a set of parameters to be learned, representing the individual sensitivity of drug efficacy. This process is essentially a system identification problem, which can be solved by a nonlinear least squares method or a neural network based method (such as neural differential equation Neural ODE).

[0055] The dynamic model can be mathematically expressed as a system of Ordinary Differential Equations (ODEs): dS / dt = f(S, Ce, P). Here, S is an N-dimensional vector representing the patient's coordinates in the N-dimensional neural state space; dS / dt is the rate of change of the state vector with respect to time, i.e., the velocity vector of the state point in space; Ce is the current effect-site drug concentration; P is the set of individualized pharmacodynamic parameters (e.g., the half maximal effective concentration EC50 of the drug effect and the slope γ of the effect curve, etc.). The function f itself can be a parameterized non-linear function or directly fitted by a small neural network that takes the current state S and concentration Ce as inputs and outputs the velocity vector dS / dt of the state change. This velocity vector indicates the direction and speed at which the state point moves in the neural state space under the current drug concentration.

[0056] In the prediction phase, the prediction calculation module inputs three pieces of information: the current neural state S_current determined by the signal processing module, the current and future planned dosing strategy provided by the infusion execution module, and the updated individualized parameters P by the model calibration module. The system first calculates the effect-site concentration curve Ce_future(t) for a future period of time (e.g., 60 seconds) based on the dosing strategy and the PK model. Then, using the numerical integration method (e.g., the fourth-order Runge-Kutta method), the system solves the above ordinary differential equation system dS / dt = f(S, Ce_future(t), P) with S_current as the initial condition, to obtain the predicted evolution trajectory of the neural state in space within the next 60 seconds. The state transition probability is estimated by running Monte Carlo simulation with Gaussian noise added to the model, running hundreds of predictions, and calculating the proportion of trajectories that finally fall into the target euphoria region.

[0057] In some embodiments, standard population PK / PD models cannot accurately reflect individual differences in drug metabolism and sensitivity, leading to deviations in predictions. To address this, the propofol dosing system introduces a model calibration mechanism that continuously compares the predicted trajectory of the prediction calculation module 103 with the actual observed state trajectory of the signal processing module 102. When the cumulative error between the two exceeds a pre-set threshold, a parameter optimization algorithm (such as Bayesian estimation) is started to automatically adjust the individualized parameters (such as the central compartment volume k10 and the effect-site elimination rate constant ke0) in the PK / PD model, making the model output more closely match the actual response of the patient, achieving individualization and precision of the prediction model.

[0058] The control decision module 104 is used to generate a propofol dosing intervention strategy based on the future evolution trajectory and the state transition probability, with the goal of maintaining the brain functional state in the target euphoria region.

[0059] The drug intervention strategy refers to a series of specific instructions for controlling the infusion rate or target concentration of the target-controlled infusion pump in the next period of time.

[0060] Specifically, the control decision module 104 receives the future evolution trajectory and state transition probability from the prediction calculation module 103. The core goal is to generate a drug intervention strategy, so that the predicted future evolution trajectory under the strategy can enter and stay in the target euphoria region in the neural state space with the maximum probability. For example, if the predicted trajectory shows that the current dose will cause the state to deviate from the euphoria region to the excessive sedation region, the control decision module 104 will calculate a drug dose that needs to be reduced, so that the new predicted trajectory can be pulled back into the euphoria region. The control decision module 104 usually uses an optimization algorithm such as model predictive control (MPC) to find the optimal drug control sequence under the premise of meeting safety constraints (such as the BIS value not lower than 60).

[0061] In some embodiments, the drug intervention strategy can be generated in various ways: optionally, a proportional-integral-derivative (PID) control algorithm is used; the steps are: calculate the error between the current neural state and the center point of the target euphoria region, and calculate the adjustment amount of the infusion rate according to the proportional, integral and differential terms of the error. Optionally, fuzzy logic control is used; the steps are: establish a rule base based on expert experience (for example, if the state is close to the euphoria region and the speed is fast, then slightly reduce the dose), and generate control instructions through fuzzy reasoning. It can be understood that reinforcement learning and other artificial intelligence methods can also be used to allow the controller to learn the optimal drug intervention strategy through interaction with a virtual patient model, which is not limited here.

[0062] In some embodiments, how to balance multiple control objectives, such as quickly reaching the euphoria state, avoiding drug overdose, and saving drug amount. For this, the control decision module 104 uses a control strategy with a multi-objective optimization function. The optimization function will consider multiple performance indicators at the same time and assign different weights to them, for example: J = w1 * (distance between state and target) + w2 * (time to reach target) + w3 * (total drug amount).

[0063] By adjusting the weights w1, w2, and w3, the doctor can customize the preference of the control strategy according to the clinical needs (such as pursuing fast induction or stable maintenance), so that the generated drug intervention strategy can achieve the best balance between different objectives.

[0064] The infusion execution module 105 is used to adjust the target-controlled infusion pump according to the drug intervention strategy to achieve the guidance and maintenance of the euphoria state.

[0065] Among them, target-controlled infusion (TCI) pump is a kind of intelligent infusion pump which can reach and maintain the preset plasma concentration or effect-site concentration by automatically adjusting the infusion rate according to the built-in pharmacokinetic model.

[0066] Specifically, the infusion execution module 105, as the final executor of the control instructions, receives the specific drug intervention strategy generated by the control decision module 104, for example: “set the target effect-site concentration to 1.8 μg / ml”; Or “infuse at a rate of 50 ml / h for the next 1 minute”.

[0067] The infusion execution module 105 converts these high-level instructions into underlying communication protocol commands that the target-controlled infusion pump can recognize and execute, and sends them to the target-controlled infusion pump through a data interface (such as an RS-232 serial port). After receiving the instructions, the target-controlled infusion pump immediately adjusts the speed of its internal motor to accurately push the propofol in the syringe as required by the instructions, changing the amount of drug entering the patient's body.

[0068] In some embodiments, the adjustment of the target-controlled infusion pump can be achieved in various ways: optionally, a concentration target control mode is adopted; the steps are: the control decision module 104 generates a target effect-site concentration value, and the infusion execution module 105 sends this value to the TCI pump, which calculates and executes the infusion rate sequence required to reach the concentration by itself. Optionally, a rate direct control mode is adopted; the steps are: the control decision module 104 directly calculates the infusion rate (ml / h) to be executed at each time, and the infusion execution module 105 sends this series of rate instructions to the infusion pump for execution in sequence. It can be understood that a hybrid control mode can also be used, using rate control during the induction period and switching to concentration control during the maintenance period, which is not limited here.

[0069] In some embodiments, communication delay or interruption may cause the execution instructions to lag or fail, posing a safety hazard. For this purpose, the infusion execution module 105 is built with a communication handshake and safety monitoring protocol. After sending each instruction, the infusion execution module 105 waits for a signal from the target-controlled infusion pump indicating that the instruction has been received and executed. If no confirmation is received within a preset timeout period, the infusion execution module 105 will trigger an alarm and immediately notify the control decision module 104 to suspend decision-making. At the same time, the infusion execution module 105 will continuously monitor the actual infusion volume and pressure reported by the infusion pump, and if it finds that the instruction is not consistent or an abnormality such as blockage occurs, it will also immediately trigger an alarm and take safety measures such as switching to a preset basic infusion rate to ensure patient safety.

[0070] The system provided by the present embodiment is further described below. Please refer to Figure 2Fig. 2 is another schematic diagram of a system structure of a propofol administration system based on the euphoria electroencephalogram feature in the embodiments of the present application.

[0071] The signal acquisition module 101 is configured to acquire the original electroencephalogram signal of the patient.

[0072] The signal acquisition module 101 is configured to acquire the original electroencephalogram signal of the patient.

[0073] The signal processing module 102 is configured to map the real-time brain function state of the patient to the neural state space containing the euphoria-related state based on the multi-dimensional feature dynamic pattern of the original electroencephalogram signal, and determine the neural state.

[0074] The signal processing module 102 is configured to map the real-time brain function state of the patient to the neural state space containing the euphoria-related state based on the multi-dimensional feature dynamic pattern of the original electroencephalogram signal, and determine the neural state.

[0075] In some embodiments, the signal processing module 102 comprises: a pattern extraction unit 1021 configured to synchronously extract the brain network topology feature, the cross-band coupling feature and the time-domain oscillation pattern feature from the original electroencephalogram signal, and construct a high-dimensional feature vector; and a state mapping unit 1022 configured to match the real-time evolution pattern of the high-dimensional feature vector with the preset pattern of each state in the neural state space, and determine the neural state with the highest confidence.

[0076] The brain network topology feature refers to a network structure quantitative index obtained by analyzing the functional connection relationship between multi-lead electroencephalogram signals by graph theory method, and is used to represent the efficiency of brain information integration and separation; the cross-band coupling feature refers to a specific phase or amplitude locking relationship between different frequency electroencephalogram activities, which is an important neural mechanism for realizing complex cognitive function; the time-domain oscillation pattern feature refers to the dynamic change characteristics of the specific frequency electroencephalogram in the time dimension, such as the instantaneous change of power and frequency; the high-dimensional feature vector is an ordered set composed of the above-mentioned multiple features of different dimensions, which is a comprehensive snapshot of the brain function state at a specific moment; and the preset pattern refers to a typical feature vector template representing a specific neural state (e.g., euphoria) which is fixed in the model of the state mapping unit 1022 by learning a large amount of labeled data in the model training stage.

[0077] Specifically, after receiving the raw EEG signal stream from the signal acquisition module 101, multiple parallel algorithms in the pattern extraction unit 1021 are activated simultaneously. First, it calculates the phase lag index or weighted phase lag index between each electrode signal to construct a functional connectivity matrix, and extracts the clustering coefficient, global efficiency and other topological features from it. Second, it uses bicoherence analysis or modulation index calculation to quantify the modulation strength of the Theta band phase on the Gamma band amplitude as a cross-band coupling feature. At the same time, it obtains the instantaneous power and frequency of Alpha and Beta waves through Hilbert-Huang Transform as time-domain oscillation pattern features. These dozens of extracted feature values are combined into a high-dimensional feature vector in real time. The vector is then input into the state mapping unit 1022. The state mapping unit 1022 may be a pre-trained deep neural network or a support vector machine classifier inside, which performs mathematical operations and comparisons between the input real-time feature vector and the internal stored preset patterns of each neural state, and finally outputs a classification result, i.e. the neural state with the highest current confidence (e.g. euphoria), accompanied by a confidence score quantifying the reliability of the classification.

[0078] Optionally, non-supervised feature learning based on autoencoder and K-Nearest Neighbors (KNN) algorithm are used for state mapping. The steps include: using a deep autoencoder network to train a large amount of unlabeled EEG data to learn low-dimensional compressed features that can effectively represent the state; during operation, extract low-dimensional feature vectors from the real-time EEG data through the encoder part; in the pre-constructed labeled feature point library, find the K nearest points to the current feature vector, and determine the current neural state according to the category of the K points. It can be understood that other advanced signal processing and machine learning techniques can also be combined to achieve this, which is not limited here.

[0079] In some embodiments, the extracted feature dimensions can be too high, containing redundant or irrelevant features, which not only increases the computational burden, but also reduces the classification performance and generalization ability of the state mapping unit 1022 due to the curse of dimensionality. To this end, a feature selection and dimension reduction submodule is added between the pattern extraction unit 1021 and the state mapping unit 1022. This submodule uses algorithms such as Principal Component Analysis (PCA) or Minimum Redundancy Maximum Relevance (mRMR). Before the system is put into use, an offline data set is used to analyze all candidate features, automatically selecting a subset of features that are most relevant to the euphoria state classification and have the lowest redundancy among each other. In real-time operation, the pattern extraction unit 1021 still extracts all features, but only the features selected by the submodule are sent to the state mapping unit 1022. In this way, the effectiveness of the input features is guaranteed, the model complexity is reduced, and the computational efficiency of the entire signal processing module 102 and the accuracy and robustness of state recognition are improved.

[0080] In some embodiments, the signal processing module 102 further includes a confidence monitoring unit 1023 for monitoring the classification confidence of the neural state, and a decision suspension unit 1024 for suspending the drug intervention when the classification confidence is below a preset safety threshold, until the classification confidence recovers above the preset safety threshold.

[0081] Among them, the classification confidence of the neural state is a quantitative indicator (usually between 0 and 1) output by the state mapping unit 1022, which represents the certainty or degree of confidence of the model for the current classification result (such as euphoria). The preset safety threshold is a critical value (for example, 0.7) set in advance according to the clinical safety requirements and the performance of the model, which is used to judge whether the classification result is reliable enough for closed-loop control decision; the suspension of drug intervention refers to temporarily suspending the drug adjustment instructions generated by the control decision module 104, so that the infusion pump returns to a preset basic maintenance state or keeps the current infusion rate unchanged.

[0082] Specifically, in each control cycle, the state mapping unit 1022 calculates and outputs a corresponding confidence score while determining the neural state. The score is transmitted to the confidence monitoring unit 1023 in real time. The core logic of the confidence monitoring unit 1023 is a simple comparator that continuously compares the input real-time confidence with the internally stored preset safety threshold. When the real-time confidence is greater than or equal to the threshold, the system is considered reliable and no additional action is taken. However, once the confidence falls below the threshold due to strong electromyographic interference, electrode detachment, or the patient entering a fuzzy transition state that the model has not fully learned, the confidence monitoring unit 1023 will immediately generate a logic high "pause" trigger signal and send it to the decision pause unit 1024. Upon receiving this signal, the decision pause unit 1024 immediately broadcasts an interrupt instruction to the control decision module 104 and the infusion execution module 105, forcing them to ignore the automatic control output of the current cycle, and at the same time triggering an audible and visual alarm on the user interface to prompt medical personnel that the current system is in an uncertain state and requires manual attention. This pause state will remain until the subsequent electroencephalogram signal quality improves, causing the classification confidence to rise again and stabilize above the safety threshold, at which point the confidence monitoring unit 1023 will cancel the pause signal and the decision pause unit 1024 will release the lock on the system, restoring the automatic closed-loop control process to normal.

[0083] In some embodiments, there may be a jitter phenomenon, i.e., the classification confidence frequently fluctuates in a small range around the preset safety threshold, causing the system to quickly and repeatedly switch between automatic control and pause states. This jitter not only affects the stability of control, but also causes confusion to clinicians. To address this, the confidence monitoring unit 1023 uses a double-threshold logic with hysteresis. It sets two thresholds: a higher recovery threshold (e.g., 0.75) and a lower pause threshold (e.g., 0.70). The system only enters the pause state when the confidence is below the pause threshold. Once in the pause state, the system must wait for the confidence to rise above the higher recovery threshold before it can be released from the pause. By establishing a buffer zone between the two thresholds, the interference caused by small fluctuations in confidence around the threshold can be effectively filtered out, ensuring the stability and decisiveness of system state switching, and avoiding unnecessary frequent start and stop.

[0084] The prediction calculation module 103 is used to predict the future evolution trajectory and state transition probability of the neural state in the neural state space according to the neural state and the dynamic model representing the propofol efficacy.

[0085] The same as the prediction calculation module 103 described above, which will not be repeated here.

[0086] The control decision module 104 is configured to generate a propofol administration intervention strategy based on the future evolution trajectory and the state transition probability, so as to maintain the brain function state in the target euphoria region.

[0087] The control decision module 104 is configured to generate a propofol administration intervention strategy based on the future evolution trajectory and the state transition probability, so as to maintain the brain function state in the target euphoria region.

[0088] In some embodiments, the control decision module 104 comprises a global phase identification unit 1041 configured to identify whether the current process is in a euphoria induction phase or a euphoria maintenance phase according to a historical sequence of neural states; an induction period strategy generation unit 1042 configured to generate a path planning type administration strategy aiming to minimize state transition time and avoid entering an undesirable intermediate state when the current process is identified to be in the euphoria induction phase; and a maintenance period strategy generation unit 1043 configured to generate a disturbance suppression type administration strategy aiming to minimize drug consumption and maximize state stability when the current process is identified to be in the euphoria maintenance phase.

[0089] The historical sequence of neural states refers to a set of neural state labels determined by the signal processing module 102 in a time sequence within a past period of time (e.g., 5 minutes); the euphoria induction phase refers to the entire process from the start of administration to the first stable entry of the patient's neural state into the target euphoria region; the euphoria maintenance phase refers to the subsequent process aimed at continuously maintaining the patient's neural state in the target euphoria region after successful induction; the path planning type administration strategy is a control strategy oriented to optimal path, which aims to plan and execute an optimal trajectory from the starting point to the ending point (target region) in the state space; and the disturbance suppression type administration strategy is a control strategy oriented to stability and robustness, which aims to resist internal and external disturbances and maintain the state in a small balance point or region.

[0090] Specifically, after receiving a new neural state, the control decision module 104 first makes a macro-stage decision by the global stage identification unit 1041. The global stage identification unit 1041 maintains a time series buffer storing the last N neural state data points. It makes a decision by analyzing the overall trend of the sequence: if the sequence shows that the state points have a clear direction of movement in the neural state space, continuously approaching the target hedonic region from the awake region, and have not yet stabilized in the target region, it is determined that the current is the hedonic induction stage. On the contrary, if the sequence shows that the state points have been in the target hedonic region for most of the time in the past period, and fluctuate around a certain center point in a small range, it is determined that the current is the hedonic maintenance stage. The decision result is used as a switch signal to selectively activate the subsequent strategy generation unit. If it is determined to be the induction stage, the induction period strategy generation unit 1042 is activated, which calls the model predictive control (MPC) algorithm in it, and sets the optimization objective function to minimize the weighted sum of the transition time and the penalty term for deviating from the optimal path. If it is determined to be the maintenance stage, the maintenance period strategy generation unit 1043 is activated, which also uses MPC, but its optimization objective function is switched to minimize the weighted sum of state fluctuation, drug consumption and control action frequency.

[0091] In some embodiments, stage identification and strategy generation can be implemented in various ways: optionally, Hidden Markov Model (HMM) based stage identification is used. The steps include: 1. regarding the induction and maintenance stages as two hidden states, and regarding the observed neural state sequence as the observation sequence; 2. training an HMM model, and using the Viterbi algorithm to decode the most likely hidden state sequence in real time to identify the current stage. Optionally, a reinforcement learning (RL) based method is used in strategy generation. The steps include: training two independent RL agents for the induction and maintenance stages respectively; the output of the global stage identification unit 1041 is used to select which agent to activate; the activated agent outputs a dosing action that maximizes its long-term cumulative reward (the reward function corresponds to the optimization objective of the stage) according to the current neural state. It can be understood that expert system rules can also be used for stage identification, or fuzzy logic control can be used for strategy generation, which is not limited here.

[0092] In some embodiments, the judgment of phase switching is not clear or lagging, especially in the transition period between the end of induction and the beginning of maintenance, which can cause the system to switch to the conservative maintenance strategy too early, prolonging the induction time, or too late, causing unnecessary drug overshoot and state fluctuations at the end of the induction period. To this end, the global phase recognition unit 1041 adopts a composite judgment criterion based on sliding window statistical analysis and state transition probability. It not only analyzes the historical sequence of neural state labels, but also analyzes the stability of the speed and direction of state point movement in space, as well as the probability of future staying in the target area output by the prediction calculation module 103. A smooth, fuzzy logic decision system will integrate these indicators: only when the weighted scores of "the length of time the state has been in the target area", "the state point movement speed has dropped to a low level" and "the prediction probability of future short-term stability in the target area has risen to a high level" exceed a preset switching threshold, the phase switching from induction to maintenance will be made. This multi-index fusion decision mechanism makes the phase recognition more robust and accurate, ensuring that the strategy switching is done at the most appropriate time.

[0093] In some embodiments, the induction period strategy generation unit 1042 includes: a phase design subunit 10421 for decomposing the total drug administration process in the induction period into multiple stages of micro-dose administration trips; a response evaluation subunit 10422 for evaluating the transfer rate and transfer direction of the neural state after each micro-dose administration trip is executed to determine the patient's immediate pharmacodynamic sensitivity; a plan adjustment subunit 10423 for adjusting the dose and time interval of subsequent micro-dose administration trips according to the patient's latest immediate pharmacodynamic sensitivity.

[0094] Wherein, the multiple stages of micro-dose administration trips refer to the total induction dose theoretically completed at one time is divided into multiple small, independent administration observation units, each giving only a part of the total dose; the transfer rate and transfer direction of the neural state are the position vector changes of the neural state point in the neural state space caused by micro-dose administration, the modulus represents the rate, and the direction represents the main brain function effect dimension of drug action; the immediate pharmacodynamic sensitivity is a dynamic updated, individualized parameter quantifying the current patient's response to unit dose of propofol.

[0095] Specifically, when the induction period strategy generation unit 1042 is activated, first, by the stage design subunit 10421, according to the patient's basic information such as weight, age, etc., combined with an initial population pharmacokinetic model, a preliminary plan consisting of, for example, 5 micro-dose administration trips is initially planned. After the first trip (for example, a bolus of 0.3 mg / kg propofol, waiting for 45 seconds) is executed, the response evaluation subunit 10422 immediately starts working. It records the starting point and ending point coordinates of the patient's neurological state point in the state space within these 45 seconds, and calculates the displacement vector between the two points. The length and direction of the vector are input into an evaluation model to inversely calculate an updated instantaneous pharmacodynamic sensitivity parameter (for example, the slope γ value of the effect compartment drug concentration-effect relationship curve). This latest sensitivity parameter is immediately passed to the plan adjustment subunit 10423. The responsibility of the plan adjustment subunit 10423 is to use this new parameter to re-plan all subsequent micro-dose administration trips that have not yet been executed. For example, if the evaluation finds that the patient's sensitivity is higher than expected, it may proportionally reduce the dose of all subsequent trips, and may also extend the observation interval. This closed loop is iterated with the completion of each micro-dose trip until the patient's state successfully enters the target euphoria region.

[0096] In some embodiments, in the early induction period, due to the possible large deviation of the pharmacokinetic model from the true situation of the individual, the initial micro-dose planning may be inappropriate, resulting in the patient's reaction being too fast or too slow, or even entering an undesirable intermediate state (such as agitation). In this regard, the stage design subunit 10421 introduces a safety path constraint when making the initial plan. Not only the dose is planned, but also a safety corridor from wakefulness to euphoria is planned in advance in the neurological state space, avoiding known undesirable state regions (such as the agitation region). In subsequent plan adjustment, the optimization algorithm of the plan adjustment subunit 10423 not only considers the sensitivity, but also must ensure that the predicted trajectory of the adjusted plan always remains within this safety corridor. If any adjustment scheme will cause the trajectory to deviate from the safety corridor, the scheme will be rejected, and the system will select a suboptimal but safer alternative. This adaptive induction strategy with safety path constraint greatly improves the safety and stability of the induction process, ensuring that the target is efficiently achieved while effectively avoiding potential risks.

[0097] The infusion execution module 105 is used to adjust the target-controlled infusion pump according to the drug intervention strategy to achieve the guidance and maintenance of the euphoric state.

[0098] The same as the infusion execution module 105 described above, which will not be repeated here.

[0099] The external event monitoring module 106 is used to monitor the event signal indicating the strong stimulation of the surgery.

[0100] The event signal refers to a signal that can predict an operation that will cause strong physiological or psychological stimulation to the patient, such as a doctor pressing the foot switch of the electrosurgical unit, the voice instruction "prepare to pass the liver curve" issued by the endoscope operator, or the preset surgical step marker integrated with the hospital information system (HIS).

[0101] Specifically, the external event monitoring module 106 is connected to other devices or information systems in the operating room through various interfaces. It can be a physical signal receiver for receiving start signals from high-energy instruments such as electrosurgical units and laser devices; it can also be a voice recognition unit that captures and recognizes key instructions spoken by surgeons or anesthesiologists through a microphone; it can also be integrated with the hospital's surgical anesthesia information system through a network interface to obtain real-time key node information of the surgical process. Once these preset event signals are monitored, the external event monitoring module 106 will immediately send an alarm signal with the event type and intensity level to the prediction calculation module 103 and the control decision module 104.

[0102] In some embodiments, the event signal can be monitored in various ways: optionally, through a physical interface for monitoring; the steps are as follows: connect the start control line (such as the foot switch) of the surgical device to the digital input port of the external event monitoring module 106 through an isolation relay, and monitor the level change of the port. Optionally, through voice recognition monitoring; the steps are as follows: deploy a directional microphone aimed at the lead surgeon and run a voice recognition engine pre-trained with specific surgical instructions (such as "cut skin" and "dilate"). Once the key words are recognized, it is determined that an event has occurred. It can be understood that the movement of surgical instruments can also be predicted by video analysis, which is not limited here.

[0103] In some embodiments, the event signal is false or missed. For example, ordinary conversation between doctors may be misidentified as an instruction, or a non-standard operation process may cause the preset event signal to not appear. In this regard, the external event monitoring module 106 adopts a multi-modal information fusion strategy. It will simultaneously receive and analyze information from multiple sources (such as device signals, voices, heart rate and blood pressure change rates from vital sign monitors). Through a decision fusion algorithm (such as a Bayesian network or evidence theory), the probability of a strong stimulation event occurring is comprehensively determined. Only when multiple evidences collectively point to a high probability event, the external event monitoring module 106 will issue an alarm, greatly reducing the false positive rate and the false negative rate, and improving the reliability of event monitoring.

[0104] The prediction calculation module 103 is also used to switch to the disturbance emergency prediction model when receiving the event signal, recalculate the state transition probability, and predict the state transition trajectory caused by the stimulation.

[0105] Where, the perturbation emergency prediction model is a mathematical term superimposed on the basis of the pharmacodynamic model, representing the instantaneous impact effect of strong stimulation on the neural state. The state transition trajectory specifically refers to the rapid and large-scale jump change path that the neural state may take under the action of strong stimulation.

[0106] Specifically, when the prediction calculation module 103 receives the alarm signal from the external event monitoring module 106, it will immediately suspend the use of the conventional prediction model. It will load a perturbation emergency prediction model that matches the event type and intensity. The model adds an impact function to the original state evolution equation, such as a pulse term proportional to the stimulation intensity and acting for a short time, to simulate the strong activation of the brain arousal system by pain stimulation. Based on this new model and the current drug delivery rate, the prediction calculation module 103 re-performs numerical integration calculation to generate a new future evolution trajectory that considers the impact of stimulation, i.e., the state transition trajectory. This trajectory usually shows that the neural state rapidly deviates from the euphoria region and transitions to an anxiety or arousal state in a short time.

[0107] It should be noted that the perturbation emergency prediction model is an extension of the conventional pharmacodynamic model, specifically for quantifying the impact of external strong stimulation on the neural state. Training this model requires special data that includes explicit stimulation events. In clinical data collection, the occurrence time (t_stim) and type / intensity (I) of strong stimulation events need to be accurately labeled, such as surgical incision, colonoscopy through the sigmoid colon bend, etc. The training data includes the drug delivery history, effect site concentration curve Ce(t), and actual neural state trajectory S(t) for a period of time before and after the stimulation occurs. The goal of training is to fit an additional impact function g(I, t, t_stim) based on the conventional dynamic model. Specifically, by comparing the deviation of the actual state trajectory at the time of stimulation with the difference between the conventional model prediction trajectory, the parameters of the impact function are optimized. The optimization criterion is to minimize the error between the trajectory predicted by the "conventional model + impact function" and the actual trajectory within the stimulation event window. By training on data from multiple stimulation types and multiple patients, a set of impact function parameters corresponding to different stimulations can be obtained.

[0108] The mathematical expression of the perturbation emergency prediction model is: dS / dt = f(S, Ce, P) + g(I, t, t_stim). Where f(S, Ce, P) is the aforementioned regular pharmacodynamic model. The newly added g(I, t, t_stim) is the impact function term, which is non-zero only in a short time around the stimulation occurrence time t_stim. This function can be modeled as a pulse with a certain direction and amplitude. For example, g can be a directional vector V_stim multiplied by a time-varying scalar pulse function h(t; t_stim, τ). The directional vector V_stim is a fixed vector pointing to the "wakefulness" or "anxiety" region in the neural state space, and its direction is learned through training data. The amplitude of the scalar pulse function h(t) is proportional to the stimulation intensity I, and its shape can be a Gaussian function or a double exponential function, and τ determines the duration of the impact. Therefore, this model mathematically describes that when a stimulus occurs, the neural state point will experience an additional, short but strong, "force" pushing it towards the undesirable region.

[0109] When the external event monitoring module detects a signal indicating a strong stimulus at time t_event, the prediction calculation module estimates the actual occurrence time t_stim of the stimulus (possibly t_stim = t_event + estimated delay). Subsequently, the system immediately switches to this emergency model. It uses the complete differential equation containing the impact term g, combined with the current drug administration plan, to perform forward integration calculations. Due to the presence of the impact term g, the calculated "state transition trajectory" will show that the neural state point suddenly experiences a rapid and large jump near t_stim, with the direction determined by V_stim and the severity of the jump determined by the stimulation intensity I. This predicted jump trajectory provides quantitative basis for the control decision module to formulate accurate compensation drug strategies.

[0110] In some embodiments, the switching and prediction of the model can be achieved in various ways: optionally, a parameter switching method is used; the steps are: different sets of impact model parameters are pre-labeled for different types of stimuli, and after receiving the event signal, the corresponding parameter set is selected and loaded into the model for calculation according to the event type. Optionally, a model superposition method is used; the steps are: the regular prediction model and the stimulus response model are regarded as two independent sub-models, and when the event signal is received, the outputs of the two sub-models are weighted and summed to obtain the final state transition trajectory. It can be understood that machine learning models can also be used to directly predict the state transition trajectory by learning a large amount of electroencephalogram data under stimulation events, which is not limited here.

[0111] In some embodiments, different patients exhibit vastly different responses (pain tolerance) to the same stimulus, making accurate prediction difficult with a fixed perturbation model. To address this, the perturbation emergency prediction model possesses online adaptive capabilities. During the surgery, after each strong stimulus event, the prediction calculation module 103 records the actual observed deviation from the target state and compares it with the transition trajectory predicted by the model. Using this deviation, the prediction calculation module 103 adjusts the patient's stimulus sensitivity coefficient, i.e., the gain parameter in the impulse function, online using a recursive least squares algorithm.

[0112] The control decision module 104 is also used to generate an immediate compensatory drug delivery strategy based on state transition trajectories, which aims to resist the effects of stimuli and pull the neural state back to the target euphoric region.

[0113] Among them, the immediate compensation dosing strategy is a short-term, rapid prophylactic dosing regimen with a dose usually higher than the maintenance dose. Its purpose is to rapidly increase the drug concentration in the effect room when the stimulus occurs in order to counteract the arousal effect brought about by the stimulus.

[0114] Specifically, when the control decision module 104 receives the state transition trajectory generated by the prediction calculation module 103, which indicates that an unfavorable transition is about to occur, it re-runs its optimization algorithm (such as MPC) with the new optimization objective of "minimizing the magnitude and time of the state's deviation from the euphoric region." The algorithm calculates an optimal pre-emptive propofol booster dose (Bolus) and / or a temporarily increased target concentration. The goal of this strategy is that when the stimulus reaches its peak effect, the sedative effect of the drug also reaches a peak sufficient to counteract the stimulus, so that the two opposing trajectories cancel each other out. Ultimately, the patient's actual neural state trajectory only produces a small fluctuation near the euphoric region, rather than being excluded from the region.

[0115] In some embodiments, an immediate compensation strategy can be generated in several ways: Optionally, a feedforward control strategy can be employed, the steps of which are: based on the predicted transition amplitude, the required compensation dosage is directly calculated using a preset gain function, and an execution command is immediately issued. Optionally, a rolling optimization strategy based on optimal control can be employed, the steps of which are: during a short window period before the stimulus occurs, prediction and optimization calculations are repeatedly performed, adjusting the dosing plan at each small step, in order to find the dosing regimen that makes the state most stable throughout the entire stimulus period. It is understood that a rule-based expert system can also be used to select the most suitable compensation scheme from a preset emergency plan library based on the stimulus type and the patient's current state; this is not limited here.

[0116] In some embodiments, it is necessary to ensure that the compensatory administration can effectively suppress the stimulation, and will not cause the patient to fall into a deep sedation state after stimulation, and cause the risk of respiratory depression and the like due to excessive dose. For this purpose, the control decision module 104 adopts a planning method with a safety fallback path when generating the compensation strategy. It not only plans the administration strategy during stimulation, but also plans the dose rapid rollback strategy after the end of stimulation. It will ensure that the total drug exposure during the entire compensation rollback process is within a safe range, and the predicted minimum BIS value is always maintained above the safety threshold.

[0117] The deviation calculation module 107 is configured to compare the predicted neural state determined by the prediction calculation module 103 and the actual neural state determined by the signal processing module 102, and determine a data accumulation deviation.

[0118] The predicted neural state is an estimated value of the patient's brain state at a future time (for example, at time t+1) made by the prediction calculation module 103. The actual neural state is a true state value analyzed by the signal processing module 102 from the real-time acquired electroencephalogram at the same time (at time t+1). The data accumulation deviation is a quantitative indicator of the persistent and systematic difference between the two values over a period of time.

[0119] Specifically, the deviation calculation module 107 is activated in each control cycle of the propofol administration system. It obtains the state prediction value made in the last cycle from the prediction calculation module 103, and waits until the current time to obtain the actual state observation value at this time from the signal processing module 102. Both values are vectors in the neural state space. The deviation calculation module 107 calculates the Euclidean distance or other forms of measurement between the two vectors as the instantaneous deviation. Then, it calculates the average deviation over a period of time (for example, the past 5 minutes) by a sliding window average or an exponential weighted moving average method. This result is the data accumulation deviation. This value can reflect whether there is a persistent and non-random error between the prediction model and the reality.

[0120] In some embodiments, the data accumulation deviation can be determined in various ways: optionally, the root mean square error (RMSE) is calculated; the steps are: calculating the square of the difference between the predicted value and the actual value at each time point, averaging the squared differences in a time window, and then taking the square root. Optionally, the Mahalanobis distance is calculated; the steps are: considering the correlation between different feature components in the state vector, calculating the statistical distance of the predicted point to the actual point distribution, which is not sensitive to feature scale. It can be understood that other statistical test methods can also be used to determine whether the deviation is, which is not limited here.

[0121] In some embodiments, the system needs to distinguish the systematic bias of the model from the transient bias caused by un-modeled random perturbation (e.g. slight body movement of the patient, transient emotional fluctuation), to avoid over-reaction to random noise and frequent adjustment of the model. For this purpose, the bias calculation module 107 employs a double-threshold decision logic. It monitors both the transient bias and the cumulative bias. Only when the cumulative bias continuously exceeds a higher calibration threshold, the model calibration is triggered. While if only the transient bias occasionally exceeds a lower warning threshold, but the cumulative bias is still within normal range, the bias calculation module 107 only marks it as a random perturbation, without triggering calibration, to ensure the stability and necessity of model calibration.

[0122] The model calibration module 108 is used to adjust the patient-specific pharmacokinetic / pharmacodynamic parameters in the dynamic model, to align the model prediction with the actual observation, when the cumulative bias exceeds the preset bias threshold.

[0123] The patient-specific pharmacokinetic / pharmacodynamic parameters refer to those parameters in the PK / PD model that are different due to individual physiological differences (e.g. weight, age, liver and kidney function, genotype, etc.), such as drug elimination rate constant (k10), conversion coefficient between effect-site concentration and electroencephalographic effect (EC50), etc.

[0124] Specifically, the model calibration module 108 is normally in a dormant state, and is monitored by the bias calculation module 107. Once it receives the trigger signal of the cumulative bias exceeding the threshold from the bias calculation module 107, the model calibration module 108 is activated. After activation, it starts a parameter optimization algorithm, such as extended Kalman filter (EKF), unscented Kalman filter (UKF) or Bayesian parameter estimation. The algorithm takes the drug administration history in the recent period, the predicted state sequence and the actual state sequence as input, and aims to "minimize the error between prediction and actual observation", to inversely calculate and update one or more patient-specific pharmacokinetic / pharmacodynamic parameters in the dynamic model. The updated parameters are immediately pushed to the prediction calculation module 103, so that it uses the model that better fits the current patient to make prediction at the beginning of the next control cycle.

[0125] In some embodiments, the pharmacodynamic parameters can be adjusted in various ways: optionally, using gradient descent method; the steps of which are: defining a loss function (i.e. prediction error) with respect to the model parameters, calculating the gradient of the loss function with respect to each parameter, and updating the parameters in the opposite direction of the gradient with small steps. Optionally, using particle filtering or Markov Chain Monte Carlo (MCMC) method; the steps of which are: maintaining a set of parameter particles (i.e. multiple sets of possible parameter values) with weights, updating the weight of each particle according to the new observation data, and resampling, and finally taking the mean of the particles with high weights as the updated parameters. It can be understood that a pre-built parameter adaptation atlas can also be used to directly select a set of initial parameters according to the patient's basic information and preliminary response characteristics, and then fine-tune, which is not limited here.

[0126] In some embodiments, the parameter adjustment process can cause the model to be unstable, or when multiple parameters are adjusted at the same time, the problem of parameter identifiability occurs, that is, multiple different parameter combinations can explain the same observation data. In this regard, the model calibration module 108 adopts a hierarchical and constrained parameter adjustment strategy. First, it will prioritize the adjustment of the key parameters (such as ke0) that have the greatest impact on the model output according to the sensitivity analysis results of the parameters. Second, during the optimization process, reasonable physiological constraints will be imposed on the value range of the parameters to avoid their updating to unrealistic values. Finally, it can use regularization techniques to add a penalty term to the optimization objective to penalize the drastic changes of the parameters, making the parameter adjustment process smoother and more stable, and ensuring the robustness and reliability of the model calibration.

[0127] The method provided by the present embodiment is described in the following flow. Please refer to Figure 3 , a flowchart of the propofol administration method in the embodiment of the present application.

[0128] S301, collect the original electroencephalogram signal of the patient.

[0129] This step aims to continuously obtain a real-time data stream reflecting the brain electrical activity of the patient in a non-invasive manner, including the correct placement of the electrodes, the inspection of the skin impedance, the amplification and digitization of the signal. This step is the starting point of the entire closed-loop system, and the quality of the collected signal directly determines the accuracy of all subsequent analysis and decision-making.

[0130] S302, based on the multi-dimensional feature dynamic pattern of the original electroencephalogram signal, map the real-time brain function state of the patient to a neural state space containing euphoria-related states, and determine the neural state.

[0131] The system performs complex processing and analysis on the original signal, extracts deep features that can distinguish fine emotional states such as euphoria and anxiety from multiple dimensions, and uses an intelligent model to determine these features as neural states that clinicians can understand.

[0132] S303, according to the neural state and the dynamic model representing the efficacy of propofol, predicting the future evolution trajectory and state transition probability of the neural state in the neural state space.

[0133] This step combines the current state and the principle of pharmacokinetics to deduce how the patient's brain state will evolve under the current dosing regimen.

[0134] S304, based on the future evolution trajectory and state transition probability, generating a propofol dosing intervention strategy to maintain the brain function state in the target euphoria region.

[0135] The system generates an optimal dosing regimen based on the prediction of the future. The goal of this regimen is very clear: to guide and maintain the patient in the ideal euphoria state.

[0136] S305, according to the dosing intervention strategy to adjust the target controlled infusion pump to achieve the guidance and maintenance of euphoria state.

[0137] The system is responsible for accurately implementing the decision. It converts the abstract dosing strategy into specific control instructions for the physical device (target controlled infusion pump), and ensures that these instructions are accurately executed through reliable communication to complete the regulation of drug concentration in the patient's body.

[0138] In the embodiments of the present application, a complete set of closed-loop control method based on euphoric electroencephalogram features is adopted, so that the emotional state of the patient can be accurately regulated. This method accurately identifies the euphoria state by extracting multi-dimensional features from the electroencephalogram signal, predicts the state evolution trend using a dynamic model, and generates an active, individualized, and phased dosing strategy accordingly. It also has the ability to respond to external stimuli and model self-calibration. It effectively solves the problem that the existing technology can only control the depth of sedation, cannot identify and guide euphoria, and has poor control effect when facing individual differences and surgical stimulation, thereby realizing a major technical breakthrough from traditional sedative anesthesia to emotion-oriented precise anesthesia with the core of improving patient comfort experience.

[0139] The propofol dosing system in the embodiments of the present application will be described from the perspective of hardware processing. Please refer to Figure 4 , which is a schematic diagram of an entity device structure of the propofol dosing system in the embodiments of the present application.

[0140] It should be noted that, Figure 4The structure of the propofol delivery system shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0141] like Figure 4 As shown, the propofol delivery system includes a CPU 401, which can perform various appropriate actions and processes according to a program stored in ROM 402 or a program loaded from storage portion 408 into RAM 403, such as performing the methods described in the above embodiments. RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via bus 404. I / O interface 405 is also connected to bus 404.

[0142] The following components are connected to I / O interface 405: input section 406 including audio input devices, push-button switches, etc.; output section 407 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 408 including hard disks, etc.; and communication section 409 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 409 performs communication processing via a network such as the Internet. Drive 410 is also connected to I / O interface 405 as needed. Removable media 411, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 410 as needed so that computer programs read from them can be installed into storage section 408 as needed.

[0143] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by CPU 401, it performs the various functions defined in the present invention.

[0144] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0145] Specifically, the propofol administration system in the embodiment includes a processor and a memory, and the memory stores a computer program. When the computer program is executed by the processor, the propofol administration method provided in the above embodiment is implemented.

[0146] As another aspect, the application further provides a computer-readable storage medium. The storage medium can be included in the propofol administration system described in the above embodiments, or can exist independently and not be assembled into the propofol administration system. The storage medium carries one or more computer programs. When the one or more computer programs are executed by a processor of the propofol administration system, the propofol administration system implements the propofol administration method provided in the above embodiments.

[0147] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0148] In the above embodiments, according to the context, the term "when" can be interpreted as meaning "if" or "after" or "in response to determining" or "in response to detecting". Similarly, according to the context, the phrase "upon determining" or "if detecting (the stated condition or event)" can be interpreted as meaning "if determining" or "in response to determining" or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)".

Claims

1. A propofol administration system based on euphoric electroencephalographic features, characterized by, The propofol administration system comprises: a signal acquisition module, configured to acquire original electroencephalogram signals of a patient; a signal processing module, configured to map a real-time brain function state of the patient to a neural state space containing euphoria-related states based on a multi-dimensional feature dynamic pattern of the original electroencephalogram signals, and determine a neural state; a prediction calculation module, configured to predict a future evolution trajectory and a state transition probability of the neural state in the neural state space according to the neural state and a dynamic model representing a propofol pharmacodynamic effect; a control decision module, configured to generate a propofol administration intervention strategy based on the future evolution trajectory and the state transition probability, with a target of maintaining the brain function state in a target euphoria region; an infusion execution module, configured to adjust a target-controlled infusion pump according to the administration intervention strategy, so as to guide and maintain the euphoria state.

2. The propofol administration system of claim 1, wherein, The signal processing module comprises: a pattern extraction unit, configured to synchronously extract brain network topological features, cross-band coupling features and time-domain oscillation pattern features from the original electroencephalogram signals, and construct a high-dimensional feature vector; a state mapping unit, configured to match a real-time evolution pattern of the high-dimensional feature vector with preset patterns of states in the neural state space, and determine a neural state with the highest confidence.

3. The propofol administration system of claim 2, wherein, The signal processing module further comprises: a confidence monitoring unit, configured to monitor a classification confidence of the neural state; a decision suspension unit, configured to suspend the administration intervention when the classification confidence is lower than a preset safety threshold, until the classification confidence recovers to above the preset safety threshold.

4. The propofol administration system of claim 1, wherein, The propofol administration system further comprises: an external event monitoring module, configured to monitor an event signal indicating a strong surgical stimulus; the prediction calculation module is further configured to switch to a disturbance emergency prediction model to recalculate the state transition probability and predict a state transition trajectory triggered by the stimulus when the event signal is received; the disturbance emergency prediction model superimposes an instantaneous impact effect of the strong stimulus on the neural state; the control decision module is further configured to generate an immediate compensation administration strategy aiming to resist the influence of the stimulus and pull the neural state back to the target euphoria region based on the state transition trajectory.

5. The propofol administration system of claim 1, wherein, The propofol administration system further comprises: a bias calculation module, configured to compare a predicted neural state determined by the prediction calculation module with an actual neural state determined by the signal processing module, and determine a data accumulation bias; a model calibration module, configured to adjust a patient-specific pharmacodynamic parameter in the dynamic model when the data accumulation bias exceeds a preset bias threshold, so as to align the model prediction result with the actual observation result.

6. The propofol administration system of claim 1, wherein, The control decision module comprises: a global phase identification unit, configured to identify whether a current process is in a euphoria induction phase or a euphoria maintenance phase according to a historical sequence of the neural state; an induction period strategy generation unit, configured to generate a path planning administration strategy aiming to minimize a state transition time and avoid entering an undesirable intermediate state when it is identified that the current process is in the euphoria induction phase; The maintenance period strategy generation unit is configured to generate a perturbation suppression type drug delivery strategy aiming at minimizing drug consumption and maximizing state stability when it is identified that the current process is in the euphoria maintenance stage.

7. The propofol administration system of claim 6, wherein, The induction period strategy generation unit comprises: a stage design subunit configured to decompose the total drug delivery process of the induction period into multiple stage micro-dose delivery trips; a response evaluation subunit configured to evaluate the transfer rate and transfer direction of the neural state through the prediction calculation module after each micro-dose delivery trip is executed, so as to determine the instant drug efficacy sensitivity of the patient; a plan adjustment subunit configured to adjust the dose and time interval of the subsequent micro-dose delivery trip according to the latest instant drug efficacy sensitivity of the patient.

8. A propofol administration method based on euphoric electroencephalographic features, characterized by, The propofol delivery method applied to the propofol delivery system of claim 1 comprises: collecting the original electroencephalogram signal of the patient; mapping the real-time brain function state of the patient to the neural state space containing the euphoria related state based on the multi-dimensional feature dynamic pattern of the original electroencephalogram signal, and determining the neural state; predicting the future evolution trajectory and state transfer probability of the neural state in the neural state space according to the neural state and the dynamic model representing the drug efficacy of propofol; generating the drug delivery intervention strategy of propofol based on the future evolution trajectory and the state transfer probability, aiming at maintaining the brain function state in the target euphoria region; adjusting the target controlled infusion pump according to the drug delivery intervention strategy to realize the guidance and maintenance of the euphoria state.

9. A computer-readable storage medium comprising instructions, characterized in that, The instructions run on the propofol delivery system of any one of claims 1-7.

10. A computer program product, characterised in that, The computer program product runs on the propofol delivery system of any one of claims 1-7.

Images