Dexmedetomidine anesthesia monitoring method and system based on multi-dimensional electroencephalogram feature fusion

Abstract

The application discloses a dexmedetomidine anesthesia monitoring method and system based on multi-dimensional electroencephalogram feature fusion, which comprises the following steps: collecting at least two channels of original electroencephalogram signals; after the original electroencephalogram signals are preprocessed, multi-dimensional features under different sedation depths are extracted; the multi-dimensional features are weighted and fused to obtain a weighted feature vector; the weighted feature vector is input into a pre-established time sequence attention soft gradient boosting decision tree to obtain a dexmedetomidine depth index; the pre-established time sequence attention soft gradient boosting decision tree is trained by using a weighted feature vector set for training; and the infusion rate of dexmedetomidine is determined according to the dexmedetomidine depth index. By fusing the multi-dimensional features into the time sequence attention soft gradient boosting decision tree, stable, continuous and specific quantitative evaluation of the dexmedetomidine anesthesia state is realized.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, specifically to a method and system for monitoring dexmedetomidine anesthesia based on multidimensional electroencephalogram (EEG) feature fusion. Background Technology

[0002] Intraoperatively and in the intensive care unit, dexmedetomidine, as a highly selective α2-adrenergic receptor agonist, is increasingly widely used due to its unique clinical advantages of "awakenable sedation" and mild respiratory depression. However, the EEG state it induces mimics non-rapid eye movement (NREM) sleep, which differs significantly from the whole-brain inhibition pattern of traditional intravenous anesthetics (such as propofol), posing a new challenge to existing anesthesia depth monitoring techniques.

[0003] Currently, most clinically used anesthesia depth monitoring devices (such as monitors based on the bispectral index BIS) are based on single-band power analysis, bispectral index, or empirical models. These methods have the following inherent limitations in dexmedetomidine monitoring: (1) It has a single feature dimension and is not sensitive to drug-specific brain network reconstruction. The altered consciousness induced by dexmedetomidine is accompanied by multi-dimensional network-level changes, including enhanced functional connectivity in the frontal alpha band, decreased spectral entropy, and increased slow-wave power proportion. Clinical studies show that at the critical point of a modified awake / sedated score of 3 (responsive to calling), the distribution range of the traditional bispectral index (55-75) largely overlaps with that of deep sedation (<60), with a discrimination accuracy of less than 70%. Single-dimensional linear indicators are insufficient to quantify this complex nonlinear brain network remodeling.

[0004] (2) Lack of precise layering capability for “wakeable” states Current monitoring indices lack sufficient resolution in the light to moderate sedation range, failing to effectively distinguish between patients in a state of "awakenable sedation" and "imminent unconsciousness." This directly results in a lack of reliable quantitative data for adjusting dexmedetomidine infusion rates, increasing the risk of intraoperative awareness or oversedation.

[0005] (3) Insufficient dynamic fluctuation suppression capability affects the stability of long-term monitoring.

[0006] Dexmedetomidine maintains a dynamic equilibrium, which may naturally fluctuate within the patient's consciousness boundary. However, existing systems, based on single-window sliding classification, are oversensitive to transient physiological fluctuations or operating room electrosurgical interventions, leading to frequent jumps in anesthesia levels. During prolonged sedation in the ICU, this instability significantly increases the cognitive burden on healthcare staff and the rate of invalid alarms.

[0007] (4) The algorithm model lacks drug-specific optimization.

[0008] The core algorithms of existing commercial indices are mostly trained on EEG databases of traditional drugs such as propofol, mainly capturing general features such as "frequency slowing" and "burst inhibition". They have not been specifically optimized for the "enhanced frontal alpha coherence" and "sleep-like spindle waves" unique to dexmedetomidine, resulting in systematic bias in the monitoring results in the pharmacodynamic model.

[0009] In summary, to achieve accurate monitoring of the depth of dexmedetomidine anesthesia, there is an urgent need for a monitoring scheme that can capture its specific neural oscillation characteristics at the brain network level, has high sensitivity to the "awakenable" boundary, and has robust anti-interference capabilities.

[0010] Based on the above situation, there is an urgent need for an anesthesia depth recognition system that can extract multi-dimensional, network-level features from electroencephalogram (EEG) signals and optimize them for the neurophysiological characteristics of dexmedetomidine. Summary of the Invention

[0011] To address the problems existing in the prior art, this invention provides a method and system for monitoring dexmedetomidine anesthesia based on multidimensional EEG feature fusion. By fusing multidimensional features into a temporal attention soft gradient boosting decision tree, it achieves stable, continuous, and specific quantitative assessment of the dexmedetomidine anesthesia state, thereby overcoming the aforementioned shortcomings in the prior art.

[0012] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0013] According to a first aspect of this application, a method for monitoring dexmedetomidine anesthesia based on multidimensional electroencephalogram (EEG) feature fusion is provided, comprising: Collect raw EEG signals from at least two channels; After preprocessing the raw EEG signals, multidimensional features under different sedation depths were extracted. The multi-dimensional features are weighted and fused to obtain a weighted feature vector; The weighted feature vector is input into a pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors for training. The infusion rate of dexmedetomidine is determined based on the dexmedetomidine depth index.

[0014] In some embodiments of this application, based on the foregoing scheme, the original EEG signal is Prefrontal-frontal lead signals and Signal in the frontal lobe leads.

[0015] In some embodiments of this application, based on the foregoing scheme, the extraction of multi-dimensional features at different sedation depths includes: Within a sliding time window, multi-dimensional features are extracted at different sedation depths: conscious state, moderate sedation, and deep sedation. These multi-dimensional features include spectral entropy and frontal lobe... Band connectivity and relative band energy ratio; The formula for calculating the spectral entropy is:

[0016] in, For power spectral density, This represents the probability distribution after normalization of the power spectral density. Spectral entropy; The formula for calculating the functional connectivity of the frontal α-band is:

[0017] in, The cross spectrum of two signals, To obtain its imaginary part, The symbols are used for identification: 1 represents positive, 0 represents zero, and -1 represents negative. This indicates the calculation of mathematical expectation. For frontal α-band functional connectivity; The relative frequency band energy ratio includes Relative frequency band energy ratio Relative frequency band energy ratio Relative frequency band energy ratio relative frequency band energy ratio or and The formula for calculating the relative frequency band energy ratio summation is as follows:

[0018] in, For the first The relative frequency band energy ratio for , , , or and The sum of for Relative frequency band energy, Relative frequency band energy, Relative frequency band energy, The total energy relative to the frequency band energy This represents the relative frequency band energy ratio.

[0019] In some embodiments of this application, based on the foregoing scheme, the step of weighted fusion of the multi-dimensional features to obtain a weighted feature vector further includes: Based on preliminary experimental data from several healthy volunteers, Spearman correlation coefficients were obtained for each feature and the MOAA / S score. The Spearman correlation coefficients were then normalized to obtain the weight coefficients for each feature.

[0020] In some embodiments of this application, based on the foregoing scheme, the temporal attention soft gradient boosting decision tree includes a multi-head self-attention unit and a soft decision tree base learner. The step of inputting the weighted feature vector into the pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index includes: The weighted feature vector is input into a multi-head self-attention unit to obtain temporal weighted features. The flattened temporal weighted features are then input into a soft decision tree-based learner to obtain the dexmedetomidine depth index.

[0021] In some embodiments of this application, based on the foregoing scheme, the following further methods are also included: A prior reward term and a cost-sensitive loss function are introduced into the split gain calculation of the soft decision tree-based learner. The formula for calculating the prior reward is:

[0022] in, As the reward coefficient, For the final split gain, For the original split gain, Candidate features attempted when evaluating the split of the current node. As multi-dimensional features, This is an indicator function; if the condition inside the parentheses is true, it equals 1; if the condition inside the parentheses is false, it equals 0. The formula for calculating the cost-sensitive loss function is as follows:

[0023] in, This is the actual rating level. To predict the rating level, when and hour, ,otherwise , For cost-sensitive loss values, This represents the total number of training samples.

[0024] In some embodiments of this application, based on the foregoing scheme, determining the infusion rate of dexmedetomidine according to the dexmedetomidine depth index further includes: The dexmedetomidine depth index is continuously obtained within a preset time window. If the dexmedetomidine depth index within the preset time window falls into a new grading interval and the new grading interval is different from the currently displayed grading interval, a new dexmedetomidine infusion rate is determined based on the new grading interval; otherwise, the currently displayed dexmedetomidine infusion rate is maintained.

[0025] According to a second aspect of this application, a dexmedetomidine anesthesia monitoring system based on multidimensional electroencephalogram (EEG) feature fusion is provided, comprising: The electroencephalogram (EEG) signal acquisition module includes several surface electrodes, which are used to acquire raw EEG signals. The analog front-end signal conditioning module includes a preamplifier circuit, a bandpass filter circuit, and a notch filter circuit connected in sequence. The preamplifier circuit amplifies the original EEG signal. The bandpass filter circuit performs bandpass filtering on the amplified original EEG signal, including filtering out baseline drift, electrode polarization potential, anti-aliasing, and filtering out high-frequency electromyographic noise. The notch filter circuit performs notch filtering on the bandpass-filtered original EEG signal, including suppressing power frequency interference and preserving the multi-dimensional characteristics of the original EEG signal at different sedation depths. The analog-to-digital conversion module includes an analog-to-digital conversion processor, which is used to perform analog-to-digital conversion on the raw EEG signal after notch filtering to obtain a digital EEG signal; The data processing control module includes a data processor, used to extract multi-dimensional features from EEG digital signals under different sedation depths, perform weighted fusion of the multi-dimensional features to obtain a weighted feature vector, input the weighted feature vector into a pre-established temporal attention soft gradient boosting decision tree to obtain a dexmedetomidine depth index, and determine the dexmedetomidine infusion rate based on the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors for training. The display and interaction module includes a display for displaying the infusion rate of the dexmedetomidine.

[0026] According to a third aspect of this application, a computer-readable storage medium is provided that stores a computer program thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0027] According to a fourth aspect of this application, an electronic device is provided, comprising: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, cause the one or more processors to implement the method described above.

[0028] The beneficial effects of this application are as follows: (1) The dexmedetomidine anesthesia monitoring method and system based on multidimensional EEG feature fusion provided in this application organically integrates spectral entropy, which characterizes the complexity of brain information, α-band functional connectivity, which characterizes the synergy of frontal lobe networks, and relative frequency band energy, which characterizes the frequency slowing feature. These three dimensions correspond to "information content", "network integration" and "oscillation composition", respectively, and comprehensively depict the full picture of the effects of dexmedetomidine from the neurophysiological level.

[0029] (2) The dexmedetomidine anesthesia monitoring method and system based on multidimensional EEG feature fusion provided in this application proposes a customized temporal attention soft gradient boosting decision tree (TA-sGBDT) for dexmedetomidine, which breaks through the general framework of standard GBDT and is deeply customized from four dimensions: base learner, temporal modeling, feature interaction, and loss function. The soft decision tree makes the entire temporal attention soft gradient boosting decision tree differentiable and parallel to train, improves training efficiency, and makes the output probabilistic, which is more in line with the continuous state transition of "awakenable". The temporal attention module performs weighted fusion of features from multiple consecutive windows. By combining and capturing the gradual evolution of anesthesia depth, the temporal attention soft gradient enhancement decision tree can respond more accurately to changes in the boundary of consciousness. Feature interaction constraints use neurophysiological priors (coordinated changes in spectral entropy, frontal α-band functional connectivity, and relative frequency band energy ratios) as reward terms for split gain, guiding the temporal attention soft gradient enhancement decision tree to learn decision paths that conform to physiological laws. Cost-sensitive loss directly embeds clinical risks into the optimization objective, giving multiple weights to high-risk errors such as "misjudging deep sedation as consciousness", enabling the temporal attention soft gradient enhancement decision tree to prioritize avoiding dangerous misjudgments.

[0030] (3) The dexmedetomidine anesthesia monitoring method and system based on multidimensional EEG feature fusion provided in this application are implemented in the form of software modules. They can be directly embedded into the software architecture of existing anesthesia monitoring equipment without modifying the hardware acquisition front end. They have the advantages of low deployment cost and convenient upgrade, and are suitable for rapid iteration and promotion on the basis of existing medical equipment.

[0031] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. Attached Figure Description

[0032] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are intended to explain the invention, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for monitoring dexmedetomidine anesthesia based on multidimensional electroencephalogram (EEG) feature fusion according to the present invention; Figure 2 This is a schematic diagram of a dexmedetomidine anesthesia monitoring system based on multidimensional electroencephalogram feature fusion according to the present invention; Figure 3 This is a schematic diagram of an electronic device according to the present invention. Detailed Implementation

[0033] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0034] It should be understood that the terms "comprising" and other similar expressions in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, or apparatus that includes a series of steps or units and is not limited to the listed steps or units. Furthermore, "first" and "second" are used to distinguish different objects and are not intended to describe a specific order.

[0035] According to the first aspect of this application, Figure 1 As shown, this embodiment provides a method for monitoring dexmedetomidine anesthesia based on multidimensional EEG feature fusion, including: Step S1: Acquire raw EEG signals from at least 2 channels.

[0036] In some embodiments of this example, the original electroencephalogram (EEG) signal is Prefrontal-frontal lead signals and The frontofrontal lead signal was sampled at a rate of 500Hz. The raw data was stored in 24-bit signed integer format, and each channel's data was stored in a circular buffer via DMA. The buffer size was ( The sample (corresponding to 10 seconds of data) uses a double buffer mechanism to prevent data overwriting.

[0037] In this embodiment, the original EEG signal is subjected to 0.5-45Hz bandpass filtering, 50Hz power frequency notch filtering, and artifact removal. or Those marked as artifacts are not included in the calculation.

[0038] Step S2: After preprocessing the raw EEG signals, extract multi-dimensional features at different sedation depths.

[0039] The extraction of multi-dimensional features at different sedation depths includes: Multidimensional features of sedation depth under conscious state, moderate sedation and deep sedation were extracted within a sliding time window. The multidimensional features include spectral entropy, frontal α-band functional connectivity and relative frequency band energy ratio, which reflect the coordinated changes of EEG signals in three dimensions of complexity, functional connectivity and spectral composition under the action of dexmedetomidine. The formula for calculating the spectral entropy is:

[0040] in, For power spectral density, This represents the probability distribution after normalization of the power spectral density. Spectral entropy; The formula for calculating the functional connectivity of the frontal α-band is:

[0041] in, The cross spectrum of two signals, To obtain its imaginary part, The symbols are used for identification: 1 represents positive, 0 represents zero, and -1 represents negative. This indicates the calculation of mathematical expectation. For frontal α-band functional connectivity; The relative frequency band energy ratio includes Relative frequency band energy ratio , Relative frequency band energy ratio , Relative frequency band energy ratio , Relative frequency band energy ratio or and Relative frequency band energy ratio summation The calculation formula is:

[0042] in, For the first The relative frequency band energy ratio for , , , or and The sum of for Relative frequency band energy, Relative frequency band energy, Relative frequency band energy, The total energy relative to the frequency band energy This represents the relative frequency band energy ratio.

[0043] This embodiment is based on a three-stage EEG change model under the action of dexmedetomidine: ① Awake state, high complexity of brain information (high The frontal network is in a dynamic equilibrium of integration and separation (moderate). ), the spectrum is high frequency ( / ) as the main (low) ); ② Can awaken a sedative state, reducing information complexity ( (Decrease), frontal lobe Enhanced connectivity ( (Increased), showing a trend of "slowing down the frequency" ( (Increase); ③ Deep unconscious state, information complexity is further reduced ( (Approaching low values), frontal α connectivity exhibits a stable directional pattern ( (High and information flow is fixed), the spectrum is dominated by δ / θ slow waves (high) ).

[0044] By integrating information from these three orthogonal dimensions, the fusion index DDI of this invention can clearly distinguish the boundaries of the three stages, especially distinguishing between stage ② and stage ③, thereby achieving precise quantification of the "awakenable" characteristics of dexmedetomidine, and has significant clinical applicability and promotional value.

[0045] In one specific embodiment, the following three core features are calculated within a sliding time window (window length 10 seconds, step size 2 seconds, 50% overlap): In one specific embodiment, the EEG data of 20 healthy volunteers under dexmedetomidine targeted infusion were analyzed using the Modified Observer's Assessment of Alertness / Sedation Scale (MOAA / S) to summarize the typical values ​​(mean ± standard deviation) of features at different depths of sedation: Awake state (MOAA / S=5): , , ; Moderate sedation (MOAA / S=3): , , ; Deep sedation (MOAA / S=1): , (And the direction of information flow is fixed to be dominated by the frontal lobe). .

[0046] Step S3: Perform weighted fusion on the multi-dimensional features to obtain a weighted feature vector.

[0047] In some embodiments of this example, the multi-dimensional features are described. The weighted fusion process yields a weighted feature vector, which also includes: Based on preliminary experimental data from several healthy volunteers, Spearman correlation coefficients were obtained for each feature and the MOAA / S score. The Spearman correlation coefficients were then normalized to obtain the weight coefficients for each feature.

[0048] In one specific embodiment, to reflect the different contributions of various features to the determination of anesthesia depth, this invention introduces a feature weighting mechanism. The weighting coefficients are based on the correlation analysis between features and the clinical gold standard (MOAA / S score) in the pre-experimental data; the higher the correlation, the greater the weight. In this embodiment, based on preliminary experimental data from 20 healthy volunteers, the Spearman correlation coefficients between each feature and the MOAA / S score were calculated, and the weight coefficients were obtained after normalization. Spectral entropy (SpEn):

[0049] Frontal alpha band functional connectivity (wPLI):

[0050] Relative frequency band energy ratio ( ):

[0051] Relative frequency band energy ratio ( ):

[0052] Relative frequency band energy ratio ( ):

[0053] Relative frequency band energy ratio ( ):

[0054] Multiply the original features element-wise with their corresponding weights to obtain the weighted feature vector:

[0055] in, , , , , , These are spectral entropy and frontal lobe, respectively. Band functional connectivity, Relative frequency band energy ratio Relative frequency band energy ratio Relative frequency band energy ratio Weighting of relative frequency band energy proportions.

[0056] To enhance feature stability, a moving average can be applied to the weighted feature vector, taking the average of the past five consecutive windows as the input feature for the current window.

[0057] Step S4: Input the weighted feature vector into the pre-established temporal attention soft gradient boosting decision tree to obtain the Dexmedetomidine Depth Index (DDI). The pre-established temporal attention soft gradient boosting decision tree is trained using the set of weighted feature vectors used for training.

[0058] In some embodiments of this example, the temporal attention soft gradient boosting decision tree includes a multi-head self-attention unit and a soft decision tree base learner. The step of inputting the weighted feature vector into the pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index includes: The weighted feature vector is input into a multi-head self-attention unit to obtain temporal weighted features. The flattened temporal weighted features are then input into a soft decision tree-based learner to obtain the dexmedetomidine depth index.

[0059] Specifically, the weighted feature vector is used as input to construct a continuous... One window (in this embodiment) Temporal feature sequences (corresponding to a 10-second time span) .Will Inputting a multi-head self-attention module yields temporally weighted features. .Will Flattened into a 30-dimensional vector, it serves as the input for subsequent gradient-enhanced temporal attention soft gradient boosting decision trees. This enables the temporal attention soft gradient boosting decision trees to automatically focus on the most important historical moments for the current prediction, effectively integrating temporal information.

[0060] A soft decision tree is used as the base learner, and its internal nodes use the sigmoid function as the routing gate.

[0061] in, The splitting threshold, For temperature parameters, For the sigmoid function, This is a sample. The sample is presented with probability... Enter the right subtree, with Entering the left subtree, the final prediction is a weighted average of the outputs of all leaf nodes, with the weights determined by the product of path probabilities. Soft decision trees are differentiable, supporting parallel training of all base learners, and their outputs are continuous probability values, better reflecting the continuous nature of anesthesia depth.

[0062] In some embodiments of this example, according to the physiological laws governing EEG changes under the influence of dexmedetomidine, a change in consciousness is necessarily accompanied by a decrease in spectral entropy and an increase in functional connectivity of the frontal lobe alpha band. The synergistic change of the relative frequency band energy ratio among the three factors introduces a priori reward term into the calculation of the split gain of the base learner.

[0063] The formula for calculating the prior reward is:

[0064] in, The reward coefficient is denoted as , where For the set of key features, This represents the reward coefficient. When a split involves all three features simultaneously, an additional bonus reward is given to guide the temporal attention soft gradient boosting decision tree to prioritize learning decision paths that conform to physiological laws. For the final split gain, For the original split gain, Candidate features attempted when evaluating the split of the current node. As multi-dimensional features, This is an indicator function; if the condition inside the parentheses is true, it equals 1; if the condition inside the parentheses is false, it equals 0.

[0065] In some implementations of this embodiment, to reflect differences in clinical risk, the output of the temporal attention soft gradient boosting decision tree, dexmedetomidine depth index (continuous value), is mapped to the MOAA / S rank (L_DDI). A cost-sensitive loss function is used as the regression loss, whereby the cost-sensitive loss function is a weighted mean squared error, and the calculation formula is as follows:

[0066] in, This is the actual rating level. To predict the rating level, For cost-sensitive loss values, This represents the total number of training samples.

[0067] When the actual rating level Predicted rating levels hour, ,otherwise This loss allows the temporal attention soft gradient to improve the decision tree's fit to high-risk samples, reducing clinically unacceptable misjudgments.

[0068] In some embodiments of this example, determining the infusion rate of dexmedetomidine based on the dexmedetomidine depth index further includes: The dexmedetomidine depth index is continuously obtained within a preset time window. If the dexmedetomidine depth index within the preset time window falls into a new grading interval and the new grading interval is different from the currently displayed grading interval, a new dexmedetomidine infusion rate is determined based on the new grading interval; otherwise, the currently displayed dexmedetomidine infusion rate is maintained.

[0069] In this embodiment, a multi-window consistency confirmation mechanism is introduced to suppress misjudgments caused by transient fluctuations. This step is implemented in software as a circular buffer of length Nc (Nc=10) to store the DDI values ​​of the most recent 10 windows (20 seconds). Every 2 seconds, the system pushes a new dexmedetomidine depth index into the buffer and checks whether all 10 values ​​in the buffer fall within the same grading interval (the grading interval definition is given in step S5). If all 10 values ​​belong to the same grading interval, and this grading interval is different from the currently displayed grading interval, a state update is triggered; otherwise, the original state remains unchanged. This mechanism ensures that only stable state changes lasting more than 20 seconds can be finally confirmed, effectively suppressing transient misjudgments caused by single electrosurgical interference, electromyography bursts, or poor electrode contact.

[0070] This embodiment also includes state lag control, further introducing a state lag control strategy to prevent frequent state switching in the critical region. A minimum duration threshold is set. After a state transition, the system records the start time of that state. and current time In the following Even if the dexmedetomidine depth index briefly fluctuates into other graded ranges within a certain time period, the system will not allow switching again until the following conditions are met: Furthermore, the current DDI level is not the same as the previous state level.

[0071] The lag control parameters can be adjusted according to clinical needs: Normal mode: (default); Research Model: (For research on rapid changes); ICU mode: (For long-term stable monitoring).

[0072] This embodiment also includes feedback control, which features closed-loop feedback control. It automatically adjusts the dexmedetomidine infusion rate based on the stable dexmedetomidine depth index processed in step S5 and the target anesthesia depth range set by the physician. The feedback control module provides two operating modes: manual mode and automatic mode.

[0073] Manual mode: Only monitors and displays the post-processed dexmedetomidine depth index, without sending control commands. The physician manually adjusts the infusion pump based on the displayed information.

[0074] Automatic Mode: Closed-loop control is activated. The physician sets the target dexmedetomidine depth index range (e.g., 60-70, corresponding to moderate sedation) and the maximum permissible infusion rate via the touchscreen. The feedback control algorithm employs an incremental proportional-integral-derivative (PID) controller, whose input is the deviation between the current post-processing DDI value and the center value of the target range. The output is the infusion rate adjustment amount. :

[0075] in , , The PID parameters are tuned through simulation and clinical trials (in this embodiment). , , The output rate is limited to a safe range (e.g., The controller calculates the adjustment amount every 2 seconds (synchronized with model inference) and sends the command to the target-controlled infusion pump via the RS232 interface. Simultaneously, the system monitors the pump's status feedback in real time and displays the current set rate and cumulative infusion volume on the screen.

[0076] To ensure security, the system has multiple built-in protection mechanisms: If the DDI value exceeds the target range for three consecutive cycles and the deviation continues to increase, an alarm will be triggered.

[0077] If poor signal quality is detected (artifact rate > 20%), automatically switch to manual mode and trigger an alarm.

[0078] If communication with the pump is interrupted for more than 10 seconds, the control will automatically stop and an alarm will sound.

[0079] Doctors can switch back to manual mode at any time via touchscreen or physical buttons to stop automatic control in an emergency and ensure clinical safety.

[0080] In one specific embodiment, the mapping relationship between DDI and MOAA / S scores is as follows: : Awake / Light Sedation (MOAA / S 4-5); Moderate sedation (MOAA / S 3); Deep sedation (MOAA / S 2); Anesthesia / No response (MOAA / S 0-1).

[0081] In this embodiment, a temporal attention soft gradient boosting decision tree is trained by progressively adding base learners. Each tree is a soft decision tree, and the loss function is weighted MSE. Key hyperparameter settings are as follows: Number of trees: 120; Learning rate: 0.08; Maximum depth of soft decision trees: 4; Attention count: 4; Timing window number T: 5; Reward coefficient α: 0.2; Loss function: Weighted MSE (high-risk misjudgment weight 10); After training, the temporal attention soft gradient boosting decision tree outputs a continuous exponent of 0-100—the dexmedetomidine depth index (DDI)—for each new sample.

[0082] By stopping early on the validation set (patience=10 rounds), the root mean square error (RMSE) of the final temporal attention soft gradient boosting decision tree on the test set was 4.8, the coefficient of determination R²=0.92, the Pearson correlation coefficient r=0.95 with MOAA / S score, the AUC for distinguishing awakenable states (MOAA / S≥3 vs≤2) was 0.95, the high-risk misclassification rate (deep sedation misclassified as awake) was 0.8%, and the output curve smoothness (standard deviation of the difference between adjacent window DDI) was 0.7±0.3.

[0083] For continuous consistency verification, a circular buffer of length 10 is implemented in the software to store the DDI values ​​of the most recent 10 windows (20 seconds). Each value is classified according to a DDI classification threshold (>80, 60-80, 40-60, <40). A state update is triggered only when all 10 elements in the buffer belong to the same classification and that classification is different from the currently displayed classification.

[0084] State lag control is implemented, with a minimum duration threshold Tmin set to 30 seconds. After a state transition, the system records the start time, and even if DDI fluctuates briefly within 30 seconds, a second transition is not allowed.

[0085] Feedback control is employed, with the target DDI range set at 60-70 (moderate sedation). An incremental PID controller is used, with a sampling period Ts = 2 seconds. The PID parameters are fine-tuned using the Ziegler-Nichols method to Kp = 0.6, Ki = 0.12, and Kd = 0.08. The control quantity (infusion rate) is limited to 0.2-1.5 μg / kg / h. The system calculates the control quantity every 2 seconds and transmits it to the infusion pump via RS232. Simultaneously, the pump status is monitored; if five consecutive communication failures occur, an alarm is triggered and the system automatically switches to manual mode.

[0086] The transition of a true state of consciousness in the brain network is a slow neurobiological process (time constant > 20 seconds), while operating room interference (electrosurgical, electromyography) is mostly a transient event (< 2 seconds). This invention utilizes this fundamental difference by setting a consistency confirmation threshold of 10 consecutive windows (20 seconds). This mechanism is equivalent to constructing a low-pass filter in the time domain, ensuring that only physiologically persistent changes can trigger state updates. Clinical simulation tests show that, under the superimposed typical electrosurgical interference, the number of false alarms in this embodiment is reduced by 82% compared to the traditional single-window system. Thus, by constructing a temporal attention soft gradient boosting decision tree that coordinates the complexity, functional connectivity, and spectral energy of EEG signals, combined with a multi-window consistency confirmation mechanism, continuous, stable, and quantitative identification of dexmedetomidine-specific anesthesia states is achieved.

[0087] Unlike propofol, which primarily causes a single "slowing of the frequency," dexmedetomidine-induced sedation involves... Frequency band fronting, Power increase Increased power across multiple frequency bands through collaborative reconstruction. This "multi-band collaboration" characteristic means that the power indices of any single frequency band will overlap significantly at points of consciousness transition (such as around MOAA / S 3 points), resulting in a discrimination accuracy of less than 70%. This embodiment will use spectral entropy, which characterizes the complexity of brain information, and spectral entropy, which characterizes the synergy of the frontal lobe network. The band functional connectivity and relative band energy characterizing frequency slowing are organically integrated. These three dimensions correspond to "information content," "network integration," and "oscillation composition," respectively, comprehensively depicting the full picture of dexmedetomidine's effects from a neurophysiological perspective. Experimental data show that in the critical region of MOAA / S scores of 2-3, the AUC (Area Under the Curve) value of the dexmedetomidine depth index in this embodiment reaches 0.93, significantly higher than the 0.75 of the traditional BIS (Bispectral Index) (p<0.01).

[0088] This embodiment leverages the inherent robustness of statistical characteristics to enhance adaptability to complex environments.

[0089] Traditional power metrics are sensitive to signal amplitude; even a brief burst of electromyography (EMG) can lead to misjudgment. The features used in this embodiment (spectral entropy, wPLI, correlation coefficient) are based on the statistical distribution, phase relationship, or sequence correlation of the signal, rather than instantaneous absolute values. For example, wPLI only considers the sign of the phase difference and is insensitive to amplitude changes; the Pearson correlation coefficient undergoes detrending and standardization before calculation, making it naturally resistant to individual outliers. This statistically robust design allows the system to maintain a classification accuracy of over 90% even under harsh conditions with a signal-to-noise ratio as low as -5dB.

[0090] This embodiment pre-determines a specific parameter system optimized for dexmedetomidine to achieve precise matching. Based on previous analysis of EEG characteristics under dexmedetomidine effects, this embodiment found that the theta-α coupling strength in the prefrontal-frontal region is most correlated with the level of consciousness. Therefore, the system pre-determines theta-α coupling analysis for the Fp1-F3 lead pairs and optimizes the sliding window to match the drug effect equilibrium time (10 seconds). This "drug-specific" parameter pre-determining avoids feature redundancy in general models, enabling the temporal attention soft gradient boosting decision tree to accurately capture the "neural fingerprint" of dexmedetomidine.

[0091] Step S5: Determine the infusion rate of dexmedetomidine based on the dexmedetomidine depth index.

[0092] In some embodiments of this example, the dexmedetomidine depth index is continuously obtained within a preset time window. If the dexmedetomidine depth index within the preset time window falls into a new grading interval and the new grading interval is different from the currently displayed grading interval, a new dexmedetomidine infusion rate is determined based on the new grading interval; otherwise, the currently displayed dexmedetomidine infusion rate is maintained.

[0093] Thus, by utilizing the synergistic changes in EEG signals under the action of dexmedetomidine in three dimensions—complexity, functional connectivity, and spectral composition—a multi-dimensional feature fusion model is constructed. Combined with a multi-window consistency confirmation mechanism and a state lag switching strategy, a stable, accurate, and real-time assessment of the dexmedetomidine-specific anesthesia state can be achieved.

[0094] According to the second aspect of this application, such as Figure 2 As shown in the figure, this embodiment provides a dexmedetomidine anesthesia monitoring system based on multidimensional EEG feature fusion, comprising: The electroencephalogram (EEG) signal acquisition module includes several surface electrodes, which are used to acquire raw EEG signals. The analog front-end signal conditioning module includes a preamplifier circuit, a bandpass filter circuit, and a notch filter circuit connected in sequence. The preamplifier circuit amplifies the original EEG signal. The bandpass filter circuit performs bandpass filtering on the amplified original EEG signal, including filtering out baseline drift, electrode polarization potential, anti-aliasing, and filtering out high-frequency electromyographic noise. The notch filter circuit performs notch filtering on the bandpass-filtered original EEG signal, including suppressing power frequency interference and preserving the multi-dimensional characteristics of the original EEG signal at different sedation depths. The analog-to-digital conversion module includes an analog-to-digital conversion processor, which is used to perform analog-to-digital conversion on the raw EEG signal after notch filtering to obtain a digital EEG signal; The data processing control module includes a data processor, used to extract multi-dimensional features from EEG digital signals under different sedation depths, perform weighted fusion of the multi-dimensional features to obtain a weighted feature vector, input the weighted feature vector into a pre-established temporal attention soft gradient boosting decision tree to obtain a dexmedetomidine depth index, and determine the dexmedetomidine infusion rate based on the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors for training. The display and interaction module includes a display for displaying the infusion rate of the dexmedetomidine.

[0095] Specifically, the EEG signal acquisition module uses Ag / AgCl sintered surface electrodes that comply with the IEC60601-2-26 safety standard for medical electrical equipment, and the electrode impedance is maintained below 5kΩ during online monitoring. The electrodes are positioned according to the international 10-20 system standard (focusing on key acquisition areas). , , , , , The leads were positioned, with reference electrodes placed at A1 and A2. This was for accurate calculation of the frontal lobe. Band Functional Connectivity (wPLI) must ensure that it includes and The signal quality of these two pairs of frontofrontal leads is assessed. The electrode wires are double-shielded, 2.5m long, with the inner shield driving the wire to the reference potential and the outer shield grounded, effectively suppressing motion artifacts and environmental electromagnetic interference. The connectors use medical-grade DIN 42802 standard interfaces and feature anti-mis-mating design.

[0096] In the analog front-end signal conditioning module, the preamplifier circuit uses the AD8422 instrumentation amplifier from Analog Devices (ADI), with the following main parameters: Common-mode rejection ratio (CMRR) ≥ 100dB (at gain G=100), input impedance ≥ 10GΩ, input bias current ≤ 1nA, and gain set to 1000 times (40dB). The circuit employs right leg drive (DRL) technology to further suppress common-mode interference. The DRL circuit extracts the common-mode voltage from the reference electrode, amplifies it in reverse, and then drives it to the human body, which can improve the CMRR to over 120dB. The bandpass filter circuit uses a 4th-order Butterworth active filter, composed of two cascaded 2nd-order Sallen-Key units. The high-pass cutoff frequency is 0.5Hz (used to filter baseline drift and electrode polarization potential), and the low-pass cutoff frequency is 45Hz (used for anti-aliasing and filtering high-frequency electromyographic noise). Filter design parameters: passband ripple ≤ 0.5dB, stopband attenuation ≥ 40dB / decade. The notch filter circuit uses a dual-T active notch filter with a center frequency of 50Hz, a Q value of 5, a notch depth ≥40dB, and a bandwidth of ±1.5Hz. This design effectively suppresses power frequency interference while maximizing the preservation of adjacent frequency bands. , The original signal characteristics of ).

[0097] The analog-to-digital conversion module uses TI's technology. The ADS1299 analog-to-digital converter chip is specifically designed for bioelectrical signal acquisition. Its main features include: 8-channel synchronous sampling (meeting the needs of multi-lead connectivity analysis), inter-channel crosstalk ≤ -120dB, built-in programmable gain amplifier (PGA) with selectable gain factors of 1 / 2 / 4 / 6 / 8 / 12 / 24 (set to 12 in this embodiment), sampling rate range of 250SPS to 16kSPS (500SPS in this embodiment), built-in bias drive circuitry and lead dropout detection function, and input reference noise ≤ 1μVpp (at a bandwidth of 70Hz). The ADS1299 communicates with the main control chip via an SPI interface with a clock frequency of 10MHz. Data from each channel is output in 24-bit signed integer format, digitally filtered, and then stored in memory via DMA to ensure no data loss.

[0098] Data processing and control module: Utilizing a high-performance STM32F407ZGT6 ARM Cortex-M4 processor as the main control chip, with a clock frequency of 168MHz, and built-in 1MB Flash and 192KB RAM. Key resource configurations include: 2 DMA controllers for high-speed data transfer, 3 SPI interfaces (one for ADS1299 communication), 2 I²C interfaces (for touchscreen and sensor communication), 1 FSMC interface (for driving the TFT-LCD), 1 Ethernet MAC layer (supporting the IEEE 1588 precise time protocol), and a hardware FPU unit (accelerating Hilbert transform and feature calculation). Considering that steps S2-S5 require real-time wavelet packet decomposition and Hilbert transform, as well as high-dimensional vector operations of the SVM model in determining the dexmedetomidine infusion rate, these calculations involve a large number of floating-point matrix operations. The hardware FPU and DSP instruction set ensure that feature calculation can be completed within 50ms under a 10-second window and 2-second step size real-time processing constraint, meeting the needs of clinical real-time monitoring.

[0099] To meet real-time requirements, the system uses the FreeRTOS real-time operating system for task scheduling. The main task divisions are as follows: Data acquisition task (highest priority, period 2ms): Read raw data from ADS1299 and store it in a circular buffer.

[0100] Preprocessing task (high priority, 10ms cycle): Performs filtering, power frequency notch filtering, and artifact detection.

[0101] Feature extraction task (medium priority, 2s cycle): Perform frequency band decomposition, envelope extraction, spectral entropy, frontal α-band functional connectivity, and relative frequency band energy ratio calculation.

[0102] Model inference task (medium priority, period 2s): Execute TA-sGBDT model inference and output the dexmedetomidine depth index.

[0103] Feedback control task (medium priority, 2s cycle): Determine the infusion rate of dexmedetomidine based on the dexmedetomidine depth index.

[0104] This task prioritization is designed based on the real-time requirements of the algorithm: feature extraction and model inference (2-second cycle) are computationally intensive but can tolerate a certain delay; while data acquisition (2-ms cycle) must respond in real time, so it is given the highest priority to ensure no data loss.

[0105] The display and interaction module performs display and communication tasks (low priority, 100ms cycle): updating the UI and sending network data. It uses a 7-inch TFT-LCD capacitive touchscreen with a resolution of 1024×600 and a brightness of 800 cd / m², and supports operation with gloves and wiping with medical disinfectants. The display interface includes: Real-time waveform area: Displays the raw EEG waveform (channels can be switched); Trend chart area: Displays the trends of the fused dexmedetomidine depth index and various sub-indicators (spectral entropy, frontal α-band functional connectivity, and relative frequency band energy ratio) (last 30 minutes); Anesthesia grade zone: Displays the current anesthesia grade and its confidence level using a bar chart; Signal quality indication area: Displays electrode impedance, power frequency interference level, and probability of motion artifacts.

[0106] Alarm zone: Flashing red indicates abnormal conditions (such as lead detachment, epileptic discharge). Human-computer interaction supports both touch and physical button operation. The physical buttons include power button, freeze button, print button and knob (for parameter adjustment).

[0107] In some embodiments of this example, a data transmission module is also included. This data transmission module is used for data transmission and supports two transmission modes: Wired transmission: Employs the W5500 hardware TCP / IP protocol stack chip, connecting to the hospital's local area network via an RJ45 interface. Supports TCP Server / Client mode, with data packet format conforming to the HL7 (Health Level 7) medical information exchange standard, facilitating integration with the Hospital Information System (HIS). Wireless transmission: Employs the ESP32-WROOM-32 module, supporting 2.4GHz Wi-Fi (802.11b / g / n) and Bluetooth 5.0. In Wi-Fi mode, data can be pushed to the cloud server via the MQTT protocol; in Bluetooth mode, it can pair with mobile monitoring terminals (tablets, smartphones) for portable monitoring. Wireless transmission uses AES-128 encryption to ensure patient data security.

[0108] According to a third aspect of this application, this embodiment provides a computer-readable storage medium having a computer program stored thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0109] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0110] According to the fourth aspect of this application, such as Figure 3 As shown, an electronic device is provided, comprising: One or more processors; Memory is used to store executable instructions for the processor, which, when executed by one or more processors, cause one or more processors to implement the methods described above.

[0111] Electronic devices are manifested in the form of general-purpose computing devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including memory and processor).

[0112] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of a computer system, connecting all parts of the computer system through various interfaces and lines.

[0113] Memory can be used to store computer programs and / or modules. The processor implements various functions of the computer system by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0114] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and memory) containing computer-usable program code.

[0115] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0116] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0117] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0118] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0119] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dexmedetomidine anesthesia monitoring system based on multidimensional electroencephalogram (EEG) feature fusion, characterized in that, include: The electroencephalogram (EEG) signal acquisition module includes several surface electrodes, which are used to acquire raw EEG signals. The analog front-end signal conditioning module includes a preamplifier circuit, a bandpass filter circuit, and a notch filter circuit connected in sequence. The preamplifier circuit amplifies the original EEG signal. The bandpass filter circuit performs bandpass filtering on the amplified original EEG signal, including filtering out baseline drift, electrode polarization potential, anti-aliasing, and filtering out high-frequency electromyographic noise. The notch filter circuit performs notch filtering on the bandpass-filtered original EEG signal, including suppressing power frequency interference and preserving the multi-dimensional characteristics of the original EEG signal at different sedation depths. The analog-to-digital conversion module includes an analog-to-digital conversion processor, which is used to perform analog-to-digital conversion on the raw EEG signal after notch filtering to obtain a digital EEG signal; The data processing control module includes a data processor for extracting multi-dimensional features at different sedation depths from EEG digital signals, performing weighted fusion of the multi-dimensional features to obtain a weighted feature vector, inputting the weighted feature vector into a pre-established temporal attention soft gradient boosting decision tree to obtain a dexmedetomidine depth index, and determining the dexmedetomidine infusion rate based on the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors. Within a sliding time window, multi-dimensional features at sedation depths of awake state, moderate sedation, and deep sedation are extracted. These multi-dimensional features include spectral entropy, frontal lobe... Band connectivity and relative band energy ratio; The formula for calculating the spectral entropy is: in, For power spectral density, This represents the probability distribution after normalization of the power spectral density. Spectral entropy; The formula for calculating the functional connectivity of the frontal α-band is: in, The cross spectrum of two signals, To obtain its imaginary part, The symbols are used for identification, with 1 representing positive and 0 representing zero. 1 represents negative. This indicates the calculation of mathematical expectation. For frontal α-band functional connectivity; The relative frequency band energy ratio includes Relative frequency band energy ratio Relative frequency band energy ratio Relative frequency band energy ratio relative frequency band energy ratio or and The formula for calculating the relative frequency band energy ratio summation is as follows: in, For the first The relative frequency band energy ratio for , , , or and The sum of for Relative frequency band energy, Relative frequency band energy, Relative frequency band energy, The total energy relative to the frequency band energy This refers to the relative frequency band energy ratio; The display and interaction module includes a display for displaying the infusion rate of the dexmedetomidine.

2. The system according to claim 1, characterized in that: The original EEG signal is Prefrontal-frontal lead signals and Signal in the frontal lobe leads.

3. The system according to claim 1, characterized in that, The step of weighted fusion of the multi-dimensional features to obtain a weighted feature vector further includes: Based on preliminary experimental data from several healthy volunteers, Spearman correlation coefficients were obtained for each feature and the MOAA / S score. The Spearman correlation coefficients were then normalized to obtain the weight coefficients for each feature.

4. The system according to claim 1, characterized in that: The temporal attention soft gradient boosting decision tree includes a multi-head self-attention unit and a soft decision tree base learner. The step of inputting the weighted feature vector into the pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index includes: The weighted feature vector is input into a multi-head self-attention unit to obtain temporal weighted features. The flattened temporal weighted features are then input into a soft decision tree-based learner to obtain the dexmedetomidine depth index.

5. The system according to claim 4, characterized in that, Also includes: A prior reward term and a cost-sensitive loss function are introduced into the split gain calculation of the soft decision tree-based learner. The formula for calculating the prior reward is: in, As the reward coefficient, For the final split gain, For the original split gain, Candidate features attempted when evaluating the split of the current node. As multi-dimensional features, This is an indicator function; if the condition inside the parentheses is true, it equals 1; if the condition inside the parentheses is false, it equals 0. The formula for calculating the cost-sensitive loss function is as follows: in, This is the actual rating level. To predict the rating level, when and hour, ,otherwise , For cost-sensitive loss values, This represents the total number of training samples.

6. The system according to claim 1, characterized in that, The step of determining the infusion rate of dexmedetomidine based on the dexmedetomidine depth index further includes: The dexmedetomidine depth index is continuously obtained within a preset time window. If the dexmedetomidine depth index within the preset time window falls into a new grading interval and the new grading interval is different from the currently displayed grading interval, a new dexmedetomidine infusion rate is determined based on the new grading interval; otherwise, the currently displayed dexmedetomidine infusion rate is maintained.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program includes executable instructions that, when executed by a processor, enable the acquisition of raw EEG signals from at least two channels. After preprocessing the raw EEG signals, multidimensional features at different sedation depths were extracted, including: Within a sliding time window, multi-dimensional features are extracted at different sedation depths: conscious state, moderate sedation, and deep sedation. These multi-dimensional features include spectral entropy and frontal lobe... Band connectivity and relative band energy ratio; The formula for calculating the spectral entropy is: in, For power spectral density, This represents the probability distribution after normalization of the power spectral density. Spectral entropy; The formula for calculating the functional connectivity of the frontal α-band is: in, The cross spectrum of two signals, To obtain its imaginary part, The symbols are used for identification, with 1 representing positive and 0 representing zero. 1 represents negative. This indicates the calculation of mathematical expectation. For frontal α-band functional connectivity; The relative frequency band energy ratio includes Relative frequency band energy ratio Relative frequency band energy ratio Relative frequency band energy ratio relative frequency band energy ratio or and The formula for calculating the relative frequency band energy ratio summation is as follows: in, For the first The relative frequency band energy ratio for , , , or and The sum of for Relative frequency band energy, Relative frequency band energy, Relative frequency band energy, The total energy relative to the frequency band energy This refers to the relative frequency band energy ratio; The multi-dimensional features are weighted and fused to obtain a weighted feature vector; The weighted feature vector is input into a pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors for training. The infusion rate of dexmedetomidine is determined based on the dexmedetomidine depth index.

8. An electronic device, characterized in that, include: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, enable the one or more processors to acquire raw EEG signals from at least two channels; After preprocessing the raw EEG signals, multidimensional features at different sedation depths were extracted, including: Within a sliding time window, multi-dimensional features are extracted at different sedation depths: conscious state, moderate sedation, and deep sedation. These multi-dimensional features include spectral entropy and frontal lobe... Band connectivity and relative band energy ratio; The formula for calculating the spectral entropy is: in, For power spectral density, This represents the probability distribution after normalization of the power spectral density. Spectral entropy; The formula for calculating the functional connectivity of the frontal α-band is: in, The cross spectrum of two signals, To obtain its imaginary part, The symbols are used for identification, with 1 representing positive and 0 representing zero. 1 represents negative. This indicates the calculation of mathematical expectation. For frontal α-band functional connectivity; The relative frequency band energy ratio includes Relative frequency band energy ratio Relative frequency band energy ratio Relative frequency band energy ratio relative frequency band energy ratio or and The formula for calculating the relative frequency band energy ratio summation is as follows: in, For the first The relative frequency band energy ratio for , , , or and The sum of for Relative frequency band energy, Relative frequency band energy, Relative frequency band energy, The total energy relative to the frequency band energy This refers to the relative frequency band energy ratio; The multi-dimensional features are weighted and fused to obtain a weighted feature vector; The weighted feature vector is input into a pre-established temporal attention soft gradient boosting decision tree to obtain the dexmedetomidine depth index. The pre-established temporal attention soft gradient boosting decision tree is trained using a set of weighted feature vectors for training. The infusion rate of dexmedetomidine is determined based on the dexmedetomidine depth index.

Images