Anesthesia depth real-time evaluation method and system based on deep learning

Through deep learning technology and multimodal signal fusion, the problems of insufficient signal fusion, insufficient real-time, poor individual differences adaptability and weak anti-interference ability in the anesthesia depth assessment are solved, and high-precision and real-time anesthesia depth assessment are achieved. Personalized adaptation improves surgical safety.

CN118948218AActive Publication Date: 2024-11-15BEIJING STOMATOLOGY HOSPITAL CAPITAL MEDICAL UNIV

Patent Information

Application Number
CN202411086531.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-15
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

The existing anesthesia depth assessment methods have problems such as insufficient signal fusion, insufficient real-time, poor adaptability to individual differences, insufficient interpretability and weak anti-interference ability.

Method used

Using a deep learning-based real-time evaluation method for anesthesia depth, the signal fusion is fusion by obtaining multimodal physiological signals (EEG, ECG, PPG, BP, RESP), and using a hierarchical attention fusion network to perform signal fusion, and through dynamic time windows and multi-task learning frameworks Characterization adjustment and prediction are performed to generate anesthesia depth staging charts, providing interpretability analysis and drug adjustment recommendations.

Benefits of technology

It improves the accuracy and real-time performance of the in-depth evaluation of anesthesia, enhances the adaptability and anti-interference ability to individual differences, provides comprehensive patient status information, reduces the risk of insufficient or excessive anesthesia, and improves surgical safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118948218A_ABST
    Figure CN118948218A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of medical treatment, in particular to an anesthesia depth real-time evaluation method and system based on deep learning, and the method comprises the following steps: obtaining a multi-modal physiological signal of a patient, the multi-modal physiological signal comprising an electroencephalogram, an electrocardiogram, a photoelectric volume pulse wave, a blood pressure and a respiratory signal; the multi-modal physiological signals are preprocessed; fusing the preprocessed multi-modal physiological signals by using a hierarchical attention fusion network to obtain fusion features; wherein the hierarchical attention fusion network comprises a signal level attention layer, a time scale attention layer and a frequency component attention layer; performing time scale adjustment on the fusion features based on a dynamic time window to obtain adjusted features; through the multi-mode signal fusion and deep learning technology, the anesthesia depth evaluation accuracy is improved to 95% and exceeds 10% in the prior art, the risk of insufficient or excessive anesthesia can be effectively reduced, and the operation safety is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a real-time evaluation method and system for anesthesia depth based on deep learning. Background Art

[0002] Real-time assessment of anesthesia depth is crucial for surgical safety and patient prognosis. With the advancement of medical technology, anesthesia depth monitoring methods have evolved from a single indicator to multimodal analysis.

[0003] Early assessment of anesthesia depth mainly relied on clinical signs, such as blood pressure, heart rate, and pupillary response. However, these indicators often lag behind the actual changes in the state of anesthesia, making it difficult to achieve precise control. In the 1990s, electroencephalogram (EEG)-based monitoring technologies, such as the BIS index, began to be widely used. These methods evaluate the depth of anesthesia by analyzing the spectral characteristics of EEG, significantly improving the accuracy of monitoring.

[0004] In recent years, with the development of machine learning technology, researchers have begun to try to combine multiple physiological signals to obtain a more comprehensive assessment of the anesthetic state. For example, some studies have combined EEG with electrocardiogram (ECG) and used neural network models for analysis, which has achieved certain results. However, these methods still have the following technical problems:

[0005] 1. Insufficient signal fusion: Most existing methods still focus on a single signal (usually EEG) and fail to fully utilize the complementary information of multimodal signals.

[0006] 2. Lack of real-time performance: Complex algorithms often lead to processing delays and are unable to meet the rapidly changing needs during surgery.

[0007] 3. Poor adaptability to individual differences: Most existing models are static models, which are difficult to adapt to the individual differences of different patients during anesthesia.

[0008] 4. Lack of explainability: Many machine learning models are like black boxes, making it difficult for anesthesiologists to provide intuitive and understandable decision-making basis.

[0009] 5. Weak anti-interference ability: The operating room environment is complex, and existing methods are not able to resist factors such as electrosurgical knife interference and body movement.

[0010] In view of the above problems, there is an urgent need for an anesthesia depth assessment method that can comprehensively utilize multimodal physiological signals, has real-time processing capabilities, can adapt to individual differences, provide interpretable results, and has strong anti-interference capabilities. Summary of the invention

[0011] The purpose of the present invention is to provide a real-time assessment method and system for anesthesia depth based on deep learning to solve the above-mentioned problems existing in the prior art.

[0012] The present invention provides a real-time evaluation method for anesthesia depth based on deep learning, comprising the following steps:

[0013] Acquiring a multimodal physiological signal of a patient, wherein the multimodal physiological signal includes an electroencephalogram (EEG), an electrocardiogram (ECG), a photoplethysmogram (PPG), a blood pressure (BP) and a respiratory signal (RESP); preprocessing the multimodal physiological signal; the method further includes:

[0014] The pre-processed multimodal physiological signals are fused using the hierarchical attention fusion network HAFN to obtain fusion features;

[0015] The hierarchical attention fusion network includes a signal-level attention layer, a time-scale attention layer, and a frequency component attention layer;

[0016] Adjusting the time scale of the fusion feature based on the dynamic time window DTW to obtain an adjusted feature;

[0017] The dynamic time window uses a reinforcement learning method to adaptively adjust the size of the observation window according to the current signal characteristics and historical status;

[0018] The adjusted features are input into a multi-task learning framework to simultaneously predict anesthesia depth classification, EEG frequency band energy ratio, autonomic nervous system balance index, and drug concentration;

[0019] Based on the output of the multi-task learning framework, the patient's anesthesia depth is determined and an anesthesia depth staging map is generated.

[0020] Specifically, the step of preprocessing the multimodal physiological signal includes:

[0021] The multimodal physiological signal is subjected to adaptive wavelet denoising; wherein the optimal wavelet basis function and threshold are automatically selected through an unsupervised learning method; compressed sensing sampling is performed on the denoised signal; wherein, the sparsity of the signal is utilized to perform dimensionality reduction sampling through a random projection matrix; the sampled signal is subjected to online anomaly detection and correction; wherein a local anomaly factor (LOF) algorithm is used to detect and correct signal anomalies in real time.

[0022] Specifically, the processing steps of the hierarchical attention fusion network HAFN include: applying signal-level attention to the input multimodal physiological signal to obtain signal-level fusion features; applying time-scale attention to the signal-level fusion features to obtain time-scale fusion features; applying frequency component attention to the time-scale fusion features to obtain the final fusion features.

[0023] Specifically, the implementation steps of the dynamic time window DTW include: defining the state space, including the current signal characteristics and the historical window size; defining the action space, including the amplitude of adjusting the window size; defining the reward function based on signal stability and information volume; using the reinforcement learning model of the Actor-Critic architecture to learn the optimal window adjustment strategy by interacting with the environment.

[0024] Specifically, the processing steps of the multi-task learning framework include: setting the main task as anesthesia depth classification; setting auxiliary tasks including EEG frequency band energy ratio prediction, autonomic nervous system balance index prediction and drug concentration estimation; building a shared feature extraction network and task-specific output layer; defining a multi-task loss function, comprehensively considering the loss and weight of each task; and optimizing the performance of all tasks simultaneously through the back-propagation algorithm.

[0025] Specifically, the following feature extraction steps are also included:

[0026] The phase synchronization index PSI is calculated to quantify the synchronization between different brain regions or different physiological signals;

[0027] Calculate the multi-scale entropy MSE to evaluate the complexity of the signal at different time scales;

[0028] Extract symbolic dynamic features SDF, convert continuous signals into symbol sequences, and extract the dynamic characteristics of the system.

[0029] Specifically, the multi-task learning framework adopts the following optimized model architecture:

[0030] Use dynamic depthwise separable convolution to dynamically adjust the size and number of groups of the convolution kernel according to the complexity of the input features; introduce the attention mechanism in the residual connection to realize attention-guided jump connection; adopt adaptive activation function to automatically adjust the parameters of the activation function according to the needs of different layers.

[0031] Specifically, the following post-processing steps are also included: calculating the probability distribution of anesthesia depth; detecting abnormal conditions and generating alarms; performing interpretable analysis to quantify the contribution of different input features to the prediction results; and generating anesthesia adjustment recommendations based on the prediction results and interpretable analysis.

[0032] Specifically, the steps for generating the anesthesia depth staging diagram include: dividing the anesthesia depth into six stages: awake, mild sedation, moderate sedation, deep sedation, general anesthesia and excessively deep anesthesia; using different colors to represent different anesthesia stages, and the depth of the color indicates the credibility of the stage; using time as the horizontal axis and the anesthesia depth stage as the vertical axis to draw a dynamically changing bar graph.

[0033] A real-time evaluation system for the depth of anesthesia for implementing the method, the system comprising:

[0034] A multi-source signal acquisition module, used to acquire multi-modal physiological signals of patients;

[0035] The signal preprocessing module is used to perform denoising, compressed sensing sampling and anomaly detection on the collected signals;

[0036] Feature extraction module, used to calculate time-frequency features, phase synchronization index, multi-scale entropy and symbolic dynamics features;

[0037] Deep learning inference module, including hierarchical attention fusion network, dynamic time window adjustment unit and multi-task learning framework;

[0038] Results analysis and presentation module, used to generate anesthesia depth staging diagrams and decision-making recommendations;

[0039] Among them, the deep learning reasoning module is implemented using edge computing equipment to ensure real-time performance; the output of the result analysis and display module is used to guide the multi-source signal acquisition module to adjust the acquisition parameters to form a closed-loop control.

[0040] The present invention has the following beneficial effects:

[0041] 1. High-precision assessment: Through multimodal signal fusion and deep learning technology, the accuracy of anesthesia depth assessment has been increased to 95%, 10 percentage points higher than the existing technology. This high-precision assessment can effectively reduce the risk of insufficient or excessive anesthesia and improve surgical safety.

[0042] 2. Real-time performance: With edge computing and FPGA acceleration, the system response time is less than 100ms, which can capture rapid changes in anesthesia status in a timely manner. This feature is particularly important for emergency situations during surgery.

[0043] 3. Personalized adaptation: Dynamic time window technology can adaptively adjust according to the individual characteristics of the patient and the stage of anesthesia, improving adaptability to different patients and different surgical stages.

[0044] 4. Comprehensive evaluation: The multi-task learning framework simultaneously predicts the depth of anesthesia, the energy ratio of the EEG frequency band, the autonomic nervous system balance index and the drug concentration, providing the anesthesiologist with comprehensive patient status information.

[0045] 5. Interpretability: The anesthesia depth staging diagram shows the changes in the anesthesia status in an intuitive way, enhances the interpretability of decisions, and helps doctors make more reasonable clinical decisions.

[0046] 6. Strong anti-interference ability: Adaptive signal preprocessing and multimodal fusion significantly improve the system's resistance to common interferences in the operating room, ensuring the stability and reliability of the evaluation results.

[0047] 7. Auxiliary decision support: The system not only provides anesthesia depth assessment, but also gives anesthetic drug adjustment suggestions based on the results of multi-task learning, providing strong support for precise anesthesia management.

[0048] In summary, the real-time anesthesia depth assessment method provided by the present invention effectively solves the problems faced by the existing technology through innovative algorithm design and system integration, and makes important contributions to improving anesthesia safety, optimizing anesthesia management process, and promoting the development of personalized precision anesthesia. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0050] Figure 2 This is the logic block diagram of the hierarchical attention fusion network HAFN of the present invention.

[0051] Figure 3 It is a logic block diagram of the dynamic time window DTW technology of the present invention.

[0052] Figure 4 This is a logic block diagram of the multi-task learning framework of the present invention.

[0053] Figure 5 A logic diagram is generated for the anesthesia depth staging diagram of the present invention.

[0054] Figure 6 This is a logic block diagram of the feature extraction module of the present invention.

[0055] Figure 7 It is a logic block diagram of the phase synchronization index PSI calculation of the present invention.

[0056] Figure 8 This is a logic block diagram of the multi-scale entropy MSE calculation of the present invention.

[0057] Fig. 9 It is the logic block diagram of the symbolic dynamics feature SDF extraction of the present invention.

[0058] Fig.10 It is a line graph of the anti-interference capabilities of the present invention and comparative examples 1-3.

[0059] Fig.11 It is a 3D surface diagram of anesthesia depth-time-EEG frequency of the present invention and comparative examples 1-3. DETAILED DESCRIPTION

[0060] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following is a detailed description of the real-time assessment method and system of anesthesia depth based on deep learning proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0061] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0063] like Figure 1-11 As shown, the real-time evaluation method of anesthesia depth based on deep learning provided by the present invention comprises the following steps:

[0064] Step S1: Acquire multimodal physiological signals of a patient.

[0065] In an operating room or intensive care unit, a variety of medical sensors are used to synchronously collect multimodal physiological signals of patients, specifically, the multimodal physiological signals include electroencephalogram (EEG), electrocardiogram (ECG), photoplethysmogram (PPG), blood pressure (BP) and respiratory signal (RESP).

[0066] Preferably, in one embodiment of the present invention, the EEG signal uses a 10-20 system to place electrodes, focusing on the frontal region (Fp1, Fp2, F3, F4), with a sampling rate of 250Hz. This is because the EEG signal in the frontal region is most sensitive to changes in the depth of anesthesia. The ECG signal uses a standard 12-lead with a sampling rate of 500Hz to capture subtle changes such as heart rate variability. The PPG signal uses a finger clip sensor with a sampling rate of 100Hz to evaluate changes in peripheral vascular resistance. The BP signal measures non-invasive blood pressure every 3 minutes to monitor the effects of anesthesia on the cardiovascular system. The RESP signal uses a chest strap sensor with a sampling rate of 25Hz to monitor changes in respiratory rate and depth.

[0067] Step S2: preprocessing the multimodal physiological signal.

[0068] In order to improve the accuracy of subsequent analysis, this step preprocesses the collected original signal. This step includes the following sub-steps:

[0069] S2.1 Adaptive wavelet denoising, firstly, the adaptive wavelet denoising technology is applied to each physiological signal. The specific implementation is as follows:

[0070] a) Calculate the wavelet coefficients of the signal:

[0071] W = WaveletTransform(x)

[0072] Wherein, x is the original signal, and WaveletTransform represents the wavelet transform function. The present invention preferably uses discrete wavelet transform (DWT) because it performs well in terms of computational efficiency and signal reconstruction.

[0073] b) Adaptively select the threshold based on the local characteristics of the signal

[0074]

[0075] Wherein, σ is the estimated value of the noise standard deviation, and N is the signal length. The present invention adopts the MAD (Median Absolute Deviation) method to estimate the noise standard deviation because it is more robust to outliers.

[0076] c) Perform soft threshold processing on the wavelet coefficients:

[0077] W denoised =sign(W)·max(0,|W|-λ)

[0078] Soft thresholding can produce smoother results and reduce artifacts compared to hard thresholding.

[0079] d) Perform inverse wavelet transform to obtain the denoised signal:

[0080] x denoised =InverseWaveletTransform(W denoised )

[0081] Preferably, in one embodiment of the present invention, the sym8 wavelet is selected as the base wavelet for the EEG signal, and the decomposition layer number is 5. This selection can effectively remove high-frequency electromyographic interference and 50Hz power frequency interference, while retaining the δ, θ, α, and β frequency band information related to the depth of anesthesia in the EEG signal. For ECG signals, the db4 wavelet is selected, and the decomposition layer number is 4, which can effectively remove baseline drift and high-frequency noise, and retain the morphological characteristics of the QRS complex.

[0082] S2.2 Compressed sensing sampling, in order to reduce the pressure of data storage and transmission, while retaining the key information of the signal, the present invention uses compressed sensing technology to sample the denoised signal. The specific steps are as follows:

[0083] a) Construct a random projection matrix Φ:

[0084]

[0085] Among them, M is the dimension after sampling, N is the dimension of the original signal, and usually M << N. In the present invention, the selection of M is based on experimental results and is usually 20% - 30% of the length of the original signal.

[0086] b) Multiply the signal x by Φ to obtain the sampling result:

[0087] y = Φx

[0088] c) Restore the original signal through L1 optimization:

[0089]

[0090] Here, the orthogonal matching pursuit (OMP) algorithm is used to solve the above optimization problem because it achieves a good balance between computational speed and reconstruction accuracy.

[0091] Preferably, in an embodiment of the present invention, for EEG signals, the sampling rate can be reduced from 250 Hz to 62.5 Hz; for ECG signals, it can be reduced from 500 Hz to 125 Hz. This reduction in the sampling rate can significantly reduce the amount of data while ensuring that the key features of the signal are not lost. Especially for EEG signals, a sampling rate of 62.5 Hz can still retain frequency components up to 30 Hz, which is sufficient to cover the main frequency bands related to the depth of anesthesia.

[0092] S2.3 Online anomaly detection and correction. To identify and process outliers in the signal in real time, such as artifacts caused by body movement, the present invention uses the local outlier factor (LOF) algorithm for online anomaly detection. The specific steps are as follows:

[0093] a) Calculate the local outlier factor of each data point:

[0094]

[0095] Among them, N k (p) represents the k-nearest neighbors of point p.

[0096] b) If the LOF value exceeds the preset threshold, mark this point as an anomaly:

[0097] if LOF k (p) > threshold:

[0098] mark p as anomaly

[0099] c) Use interpolation or prediction methods to correct the outlier points:

[0100] x corrected = Interpolate(x anomaly)

[0101] Preferably, in one embodiment of the present invention, the k value is selected as 5 and the threshold is set to 2.0. The selection of these parameters is based on the statistical analysis of a large amount of clinical data, which can effectively identify abnormal values ​​caused by sudden movement of the patient or electrode detachment, while not misjudging normal physiological fluctuations. For data points marked as abnormal, cubic spline interpolation is used for correction to maintain signal continuity.

[0102] Step S3: Use the Hierarchical Attention Fusion Network (HAFN) to fuse the preprocessed signals.

[0103] In the assessment of anesthesia depth, different physiological signals have different focuses on the reflection of the anesthesia state. For example, EEG mainly reflects the activity of the cerebral cortex, while ECG and BP reflect the changes in the autonomic nervous system. In order to comprehensively and accurately assess the depth of anesthesia, the present invention designs an innovative hierarchical attention fusion network (HAFN). The processing steps of HAFN include:

[0104] First, signal-level attention is applied to the input multimodal physiological signal X to obtain the signal-level fusion feature X s :

[0105] a) Calculate attention weights: Apply signal-level attention

[0106]

[0107] b) Obtain signal-level fusion features:

[0108] X s =A s X

[0109] Among them, W s and V s is the learnable parameter matrix.

[0110] Preferably, in one embodiment of the present invention, W s and V s The initial value of is generated by the Xavier initialization method, and the initial learning rate is set to 0.001. This initialization method can make the network converge more easily in the early stages of training, especially when processing high-dimensional signals such as EEG. The signal-level attention mechanism can automatically learn the importance of different physiological signals at different stages of anesthesia. For example, in the stage of light anesthesia, the alpha wave activity of EEG may be more important, while in deep anesthesia, the delta wave of EEG and the heart rate variability of ECG may be more indicative.

[0111] S3.2 Time scale attention, during anesthesia, signal changes at different time scales have different meanings for the depth of anesthesia. For example, a short period of EEG burst suppression may indicate deep anesthesia, while a long period of reduced heart rate variability may indicate a sustained anesthesia state. Therefore, we focus on the signal-level fusion feature X s , and obtain the time scale fusion feature X t

[0112] a) Calculate attention weights:

[0113]

[0114] b) Obtain time scale fusion features:

[0115] X t =A t X s

[0116] Preferably, in one embodiment of the present invention, the time scale attention mechanism considers three time scales: short-term (5 seconds), medium-term (30 seconds) and long-term (2 minutes). This design can capture the characteristics of various time scales, from rapid EEG changes to slow blood pressure trends, so as to comprehensively assess the depth of anesthesia.

[0117] S3.3 Frequency component attention, different frequencies of EEG activity correspond to different depths of anesthesia. For example, the enhancement of α waves (8-13Hz) usually indicates light anesthesia or imminent awakening, while the dominance of δ waves (0.5-4Hz) indicates deep anesthesia. To this end, we fused the time scale feature X t Apply frequency component attention to obtain the final fusion feature X f :

[0118] a) For X t Perform Fourier Transform

[0119] F(X t )=FFT(X t

[0120] b) Calculate the attention weight:

[0121]

[0122] c) Obtain frequency component fusion features:

[0123] X f =A f F(X t )

[0124] Preferably, in one embodiment of the present invention, the EEG signal focuses on four frequency bands: δ (0.5-4 Hz), θ (4-8 Hz), α (8-13 Hz) and β (13-30 Hz). The frequency component attention mechanism can automatically learn the importance of different frequency bands in anesthesia depth assessment and dynamically adjust the weights according to different anesthesia stages.

[0125] Step S4: Perform additional feature extraction.

[0126] In order to more comprehensively describe the patient's anesthesia state, the present invention also introduces the following feature extraction steps:

[0127] S4.1 Calculate the phase synchronization index (PSI). The phase synchronization index is used to quantify the synchronization between different brain regions or different physiological signals. The calculation formula is as follows:

[0128]

[0129] in, and is the phase of the two signals.

[0130] Preferably, in one embodiment of the present invention, the PS1 between the left and right hemispheres (Fp1 and Fp2) of the prefrontal lobe in the EEG signal is calculated. This index can reflect the functional connectivity strength between the cerebral hemispheres, and exhibits characteristic changes under different anesthesia depths. For example, under deep anesthesia, the PSI value usually decreases, indicating a weakening of brain functional connectivity.

[0131] S4.2 Calculate the multiscale entropy (MSE). Multiscale entropy is used to evaluate the complexity of the signal at different time scales. The calculation formula is as follows:

[0132]

[0133] Here, τ is the scaling factor, m is the embedding dimension, and r is the similarity tolerance.

[0134] Preferably, in one embodiment of the present invention, the MSE of 20 time scales is calculated for the EEG signal, the embedding dimension m=2, and the similarity tolerance r=0.15×std(signal). Multi-scale entropy can capture the changes in the complexity of EEG signals during anesthesia. Generally, as the depth of anesthesia increases, the MSE value decreases, reflecting the simplification of brain activity.

[0135] S4.3 Extract symbolic dynamic features (SDF), which extract the dynamic features of the system by converting continuous signals into symbol sequences. The specific steps are as follows:

[0136] a) Convert the signal into a symbol sequence S = {s1, s2, ..., s n},s i ∈A(symbol set);

[0137] b) Calculate subsequence probability w is a subsequence of length m.

[0138] Preferably, in one embodiment of the present invention, the RR interval sequence of the ECG signal is subjected to symbolic dynamics analysis. Three symbols are used (0 for decrease, 1 for unchanged, and 2 for increase), and the sequence length m=3. This analysis can reveal the fine structure of heart rate variability and reflect the regulatory state of the autonomic nervous system. For example, under deep anesthesia, patterns such as "222" (continuous increase) and "0000" (continuous decrease) may decrease, while alternating patterns such as "101" and "010" may increase, reflecting the decrease in heart rate variability.

[0139] By introducing these additional features, the present invention can describe the patient's physiological state from multiple perspectives and improve the comprehensiveness and accuracy of anesthesia depth assessment. f Together, they are passed as input to the subsequent dynamic time window adjustment and multi-task learning framework, further enhancing the performance of the system.

[0140] Step S5: Adjust the time scale of the fused features based on the dynamic time window (DTW) technology.

[0141] The speed of change of anesthesia depth is different at different stages of surgery. For example, during the anesthesia induction period, the depth changes faster; while during the maintenance period, the change is relatively slow. In order to adapt to this dynamic characteristic, the present invention uses the dynamic time window (DTW) technology to adjust the time scale of the fusion feature. The specific steps are as follows: a) Define the state space s t , including the current signal characteristics and the historical window size. b) Define the action space a t , including the magnitude of adjusting the window size (increase / decrease / maintain). c) Define the reward function r t , based on signal stability and information content:

[0142] r t =α·StabilityMeasure(X f )+β·InformationGain(X f )

[0143] Among them, StabilityMeasure measures the stability of the signal, and InformationGain measures the information gain brought by the new data. α and β are weight coefficients determined by experiments.

[0144] d) Use the Actor-Critic architecture reinforcement learning model to learn the optimal window adjustment strategy by interacting with the environment:

[0145] Update value function:

[0146] V(s t )=V(s t )+η[r t +γV(s t+1 )-V(s t )]

[0147] Update strategy function:

[0148]

[0149] Preferably, in one embodiment of the present invention, the initial time window is set to 30 seconds, and dynamic adjustment is allowed between 10 seconds and 2 minutes. The learning rate η is set to 0.01, and the discount factor γ is set to 0.99. This setting can strike a balance between maintaining evaluation stability and responsiveness to rapid changes.

[0150] Step S6: Input the adjusted features into the multi-task learning framework.

[0151] In order to comprehensively evaluate the anesthesia state, the present invention adopts a multi-task learning framework to simultaneously predict multiple related indicators. Specifically, it includes:

[0152] a) Build a shared feature extraction network f shared and a task-specific output layer f taski ;

[0153] b) Define the multi-task loss function:

[0154] L=λ1L depth +λ2L eeg +λ3L ans +λ4L drug +R(θ)

[0155] Among them, L depth is the anesthesia depth classification loss, L eeg is the EEG band energy ratio predicted loss,

[0156] L ans is the autonomic nervous system balance index predicting loss, L drug is the estimated loss of drug concentration, R(θ)

[0157] is the regularization term.

[0158] c) Optimize the performance of all tasks simultaneously through the back-propagation algorithm:

[0159]

[0160] Preferably, in one embodiment of the present invention, the shared network adopts an improved MobileNetV3 structure to balance computational efficiency and feature extraction capabilities. The weight ratio λ1 to λ4 is set to 2:1:1:1, reflecting the dominance of the anesthesia depth classification task. The learning rate η adopts a cosine annealing strategy with an initial value of 0.001. This multi-task learning framework can utilize the correlation between tasks to improve the generalization ability and robustness of the model.

[0161] Step S7: According to the output of the multi-task learning framework, the patient's anesthesia depth is determined and an anesthesia depth staging map is generated.

[0162] Based on the output of the multi-task learning framework, this step performs the final anesthesia depth assessment and visualization. Specifically, it includes:

[0163] a) Calculate the probability distribution of anesthesia depth:

[0164] P(depth|X)=softmax(f depth (X)

[0165] Among them, f depth is the anesthesia depth classifier and X is the input feature.

[0166] b) The depth of anesthesia is divided into six stages: awake, mild sedation, moderate sedation, deep sedation, general anesthesia and excessively deep anesthesia.

[0167] c) For each time point t, the anesthesia depth stage is determined based on the output of the multi-task learning framework t and the corresponding probability p t .

[0168] d) Generate anesthesia depth stage diagram: with time t as the horizontal axis, anesthesia depth stage t Different colors are used on the vertical axis to represent different anesthesia stages. The depth of the color indicates the reliability of the stage, which is calculated as follows:

[0169] color_intensity=base_color×p t

[0170] Wherein, base_color is a predefined basic color. Preferably, in one embodiment of the present invention, the color coding is as follows: awake: blue (#0000FF); mild sedation: light green (#90EE90); moderate sedation: green (#008000); deep sedation: dark green (#006400); general anesthesia: yellow (#FFFF00); too deep anesthesia: red (#FF0000).

[0171] This color coding scheme intuitively reflects the changes in the depth of anesthesia, helping anesthesiologists quickly identify the patient's status. For example, when the color gradually changes from green to yellow, it indicates that the doctor may need to reduce the dose of anesthetic drugs to prevent excessive anesthesia.

[0172] In addition, the present invention also combines the outputs of other tasks to provide a more comprehensive anesthetic state assessment:

[0173] e) EEG frequency band energy ratio: The energy ratio of the four frequency bands δ, θ, α, and β is displayed in the form of a pie chart, reflecting the brain activity pattern.

[0174] f) Autonomic Nervous System Balance Index: This is displayed in a dashboard style, indicating the relative activity of the sympathetic and parasympathetic nervous systems.

[0175] g) Estimated drug concentrations: Estimated plasma and effect-site concentrations of major anesthetic drugs (e.g., propofol, remifentanil) are presented in line graph form.

[0176] By integrating this information, anesthesiologists can have a more comprehensive understanding of the patient's anesthetic status and make more accurate medication decisions.

[0177] In order to implement the above method, the present invention also provides a real-time evaluation system for the depth of anesthesia. Figure 2 As shown, the system includes the following modules:

[0178] 1. Multi-source signal acquisition module 1, which is responsible for collecting the patient's multimodal physiological signals, including EEG, ECG, PPG, BP and RESP.

[0179] Preferably, in one embodiment of the present invention, the module includes: an 8-lead EEG acquisition device with a sampling rate of 250Hz; a 12-lead ECG monitor with a sampling rate of 500Hz; a finger-clip PPG sensor with a sampling rate of 100Hz; a non-invasive BP monitor that measures every 3 minutes; a chest strap respiratory monitor with a sampling rate of 25Hz; all devices are connected to the central processing unit through a medical-grade data acquisition interface to ensure the synchronization and integrity of the data.

[0180] Signal preprocessing module 2, which performs denoising, compressed sensing sampling and anomaly detection on the collected signals.

[0181] Preferably, in one embodiment of the present invention, the module is implemented using FPGA (field programmable gate array) to ensure real-time performance. The FPGA model is XilinxArtix-7, with a clock frequency of 200MHz, and can process multiple signals in parallel.

[0182] Feature extraction module 3, which is responsible for calculating time-frequency features, phase synchronization index, multi-scale entropy and symbol dynamics features.

[0183] Preferably, in one embodiment of the present invention, the module adopts GPU acceleration, uses the NVIDIA Jetson Xavier NX module, has 384 CUDA cores, and can efficiently process large-scale parallel computing tasks.

[0184] Deep learning reasoning module 4, which includes a hierarchical attention fusion network, a dynamic time window adjustment unit and a multi-task learning framework.

[0185] Preferably, in one embodiment of the present invention, the module also uses the NVIDIA Jetson Xavier NX module and uses its deep learning accelerator (DLA) for efficient reasoning. The model is optimized using TensorRT, and the quantization accuracy is FP16 to balance accuracy and speed.

[0186] Result analysis and presentation module 5, which is responsible for generating anesthesia depth staging diagrams and decision recommendations.

[0187] Preferably, in one embodiment of the present invention, the module uses a 12-inch high-resolution touch screen (1920x1080 resolution), uses the Qt framework to develop the user interface, supports real-time data visualization and interactive operations. The interface refresh rate is 60Hz to ensure a smooth user experience.

[0188] Through the collaborative work of the above modules, the system can achieve real-time assessment of anesthesia depth at the millisecond level, provide timely and accurate decision support for anesthesiologists, and significantly improve the safety and effectiveness of anesthesia management.

[0189] The advantages and applications of the present invention are as follows:

[0190] 1. High-precision assessment: Through multimodal signal fusion and deep learning technology, the accuracy of anesthesia depth assessment has been increased to more than 95%, significantly surpassing existing technologies. This high-precision assessment can effectively reduce the risk of insufficient or excessive anesthesia and improve surgical safety.

[0191] 2. Real-time performance: With edge computing and optimized algorithm design, the system response time is reduced to less than 100ms, which can capture rapid changes in anesthesia status in a timely manner and provide timely support for emergency treatment.

[0192] 3. Personalized adaptation: Dynamic time window technology can be adaptively adjusted according to the individual characteristics of the patient and the stage of anesthesia, significantly improving the adaptability to different patients and different surgical stages, and the adaptability to individual differences is improved by more than 40%. This is especially important for anesthesia for special populations (such as elderly patients, children or obese patients).

[0193] 4. Comprehensive evaluation: The multi-task learning framework simultaneously predicts the depth of anesthesia, the energy ratio of the EEG frequency band, the autonomic nervous system balance index, and the drug concentration, providing anesthesiologists with comprehensive patient status information and supporting precise anesthesia management. For example, in patients with impaired liver function, the prediction of drug concentration can help doctors adjust the dosage more accurately.

[0194] 5. Strong anti-interference ability: Innovative signal preprocessing and multimodal fusion technology significantly improves the system's resistance to common interference in the operating room. In the presence of electrosurgical interference, the accuracy rate drops by no more than 5%. This is especially important in electrosurgery, such as laparoscopic surgery or neurosurgery.

[0195] 6. Interpretable design: The anesthesia depth staging diagram shows the changes in the anesthesia state in an intuitive way, enhances the interpretability of decision-making, and helps doctors make more informed clinical decisions. Especially in teaching hospitals, this visualization tool can help train young anesthesiologists.

[0196] 7. Drug use optimization: Based on accurate anesthesia depth assessment, this method can help optimize the use of anesthetic drugs, reduce the average amount of drugs by 20%-30%, and reduce the risk of drug-related adverse reactions. This can not only reduce the patient's recovery time, but also reduce medical costs.

[0197] 8. Prevention of intraoperative awareness: Through precise control of anesthesia depth, this method can effectively prevent intraoperative awareness events, and no cases of intraoperative awareness occurred in clinical trials. This is especially important for high-risk patients (such as cardiac surgery patients).

[0198] 9. Economic benefits: This approach can significantly reduce medical costs by optimizing drug use and reducing complications. Preliminary estimates show that each operation can save about 5%-10% of the total cost.

[0199] 10. Remote monitoring support: The system supports remote transmission and real-time analysis of data, allowing experts to remotely assist in anesthesia management of complex cases. This is particularly valuable in areas where medical resources are unevenly distributed.

[0200] The real-time anesthesia depth assessment method and system of the present invention can be widely used in various surgical and medical scenarios, including but not limited to:

[0201] 1. General surgical operations: such as laparoscopic cholecystectomy, appendectomy, etc. In these operations, this system can help anesthesiologists accurately control the depth of anesthesia and reduce common complications such as postoperative nausea and vomiting.

[0202] 2. Heart surgery: such as coronary artery bypass grafting, valve replacement, etc. This type of surgery requires extremely high control of anesthesia depth, and the high-precision evaluation of this system can significantly improve surgical safety.

[0203] 3. Neurosurgery: such as brain tumor resection, spinal cord surgery, etc. In these operations, precise control of anesthesia depth is crucial to protect neurological function.

[0204] 4. Day surgery: such as cataract surgery, arthroscopy, etc. This system can help achieve rapid anesthesia induction and awakening, and improve the efficiency of day surgery centers.

[0205] 5. Intensive Care Unit (ICU): For patients who require long-term sedation, this system can help medical staff better manage the depth of sedation and reduce complications such as ventilator-associated pneumonia.

[0206] 6. Obstetric anesthesia: During caesarean section, this system can help anesthesiologists minimize the impact on the fetus while ensuring the safety of the mother.

[0207] 7. Pediatric surgery: The physiological characteristics of pediatric patients make anesthesia management more challenging. The personalized adaptability of this system can provide more precise guidance for pediatric anesthesia.

[0208] 8. Surgery for elderly patients: Elderly patients often have multiple underlying diseases and have different sensitivities to anesthetic drugs. This system can help achieve safer anesthesia management for elderly patients.

[0209] 9. Robot-assisted surgery: such as prostatectomy assisted by the da Vinci surgical system. In such high-precision surgeries, the system can provide stable anesthesia depth control.

[0210] 10. Organ transplantation: such as liver transplantation, kidney transplantation, etc. This type of surgery takes a long time and has high risks. This system can help maintain a long-term stable anesthesia state.

[0211] Preferably, in one embodiment of the present invention, the parameter settings of the system in different types of surgery can be fine-tuned. For example, in cardiac surgery, the weight of the autonomic nervous system balance index can be increased; in neurosurgery, more attention can be paid to changes in specific frequency bands of EEG.

[0212] In summary, the real-time anesthesia depth assessment method and system provided by the present invention effectively solves the problems faced by the existing technology through innovative algorithm design and system integration, and makes important contributions to improving anesthesia safety, optimizing anesthesia management processes, and promoting the development of personalized precision anesthesia. It can not only improve surgical safety, but also optimize resource utilization and reduce medical costs, which is of great significance to promoting the development of modern anesthesiology.

[0213] In order to verify the superiority of the present invention, we designed an embodiment and three comparative examples, and conducted detailed comparative tests. The test environment was a general surgical operating room in a tertiary hospital, and the test subjects were 100 patients (aged 18-65 years old, ASA grade I-II) receiving general anesthesia.

[0214] Embodiment 1: The method for real-time assessment of anesthesia depth of the present invention;

[0215] Comparative Example 1: Traditional BIS index monitoring method;

[0216] Comparative Example 2: A single EEG signal based on deep learning method;

[0217] Comparative Example 3: Multimodal signal fusion method without attention mechanism and dynamic time window.

[0218] Test indicators and methods:

[0219] 1. Accuracy of anesthesia depth assessment: Three experienced anesthesiologists assessed the anesthesia depth based on the patient's clinical manifestations, physiological indicators and postoperative interview results. The expert assessment results were used as the gold standard to calculate the assessment accuracy of each method.

[0220] 2. System response time: Use a high-precision timer to record the time from signal input to output of anesthesia depth assessment results.

[0221] 3. Individual difference adaptability: Calculate the standard deviation of the assessment accuracy of patients in different age groups (18-40 years old, 41-65 years old) and different ASA grades. The smaller the standard deviation, the better the adaptability.

[0222] 4. Anti-interference ability: Test with and without electric knife interference, and calculate the percentage decrease in evaluation accuracy.

[0223] 5. Prevention of intraoperative awareness: Record the number of intraoperative awareness events reported by patients after surgery.

[0224] 6. Anesthetic drug dosage: Record the total amount of propofol and remifentanil used during the operation and calculate the dosage per kilogram of body weight.

[0225] The test results are shown in the following table:

[0226] index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 Anesthesia depth assessment accuracy 95.2% 82.5% 88.7% 90.1% System response time 85ms 1500ms 210ms 150ms Individual difference adaptability (standard deviation) % 2.1% 7.8% 5.2% 4.5% Anti-interference ability (accuracy decrease) % 3.2% 15.6% 9.8% 7.5% Number of intraoperative awareness events 0 3 1 1 Propofol dosage (mg / kg / h) 4.2 5.8 5.1 4.8 Remifentanil dosage (μg / kg / h) 0.15 0.22 0.18 0.17

[0227] Combined with the above test results, the analysis and discussion are as follows:

[0228] 1. Anesthesia depth assessment accuracy: The accuracy of Example 1 is significantly higher than other methods, reaching 95.2%. This proves the effectiveness of the multimodal signal fusion, hierarchical attention mechanism and dynamic time window technology of the present invention. High accuracy means that the anesthesia depth can be controlled more accurately and the risk of complications can be reduced.

[0229] 2. System response time: The response time of Example 1 is only 85ms, which is much better than other methods. Rapid response is crucial to timely capture sudden changes in the anesthesia state, which can help anesthesiologists make quick adjustments and improve surgical safety.

[0230] 3. Adaptability to individual differences: The standard deviation of Example 1 is the smallest, which is 2.1%, indicating that the method of the present invention can better adapt to the individual differences of different patients. This adaptability is of great significance for achieving precise and personalized anesthesia management.

[0231] 4. Anti-interference ability: In the case of electric knife interference, the accuracy of Example 1 decreases the least, only 3.2%. This proves that the present invention has stronger robustness in complex operating room environments and can provide more reliable anesthesia depth assessment. Please refer to Fig.10 Through this line graph, it can be clearly seen that the present invention (red line) maintains the highest accuracy under all interference levels, and the slope of the line is the smallest, indicating that it has the strongest anti-interference ability. The lines of comparative examples 1-3 decrease significantly with the increase of interference level, especially under moderate and severe interference, the accuracy decreases significantly.

[0232] 5. Prevention of intraoperative awareness: No intraoperative awareness event occurred in Example 1, while other methods all had varying degrees of occurrence, which indicates that the present invention can more effectively prevent intraoperative awareness and improve the patient's surgical experience and satisfaction.

[0233] 6. Dosage of anesthetic drugs: The dosage of propofol and remifentanil in Example 1 is significantly lower than that of other methods. This shows that the present invention can help anesthesiologists to more accurately control the depth of anesthesia and avoid over-anesthesia, thereby reducing the dosage of drugs, reducing the risk of drug-related adverse reactions, and possibly accelerating the patient's postoperative recovery.

[0234] Reference Fig.11 Through this 3D surface graph, it can be intuitively seen that the present invention provides a more stable and accurate anesthesia depth assessment during the entire surgical process, which is manifested as a smoother surface and a more reasonable frequency distribution. Based on the above test results, Example 1 is determined to be the best embodiment. It performs best in all test indicators, especially in terms of accuracy, response time and anti-interference ability. This improvement in comprehensive performance is due to the unique technical innovations of the present invention, including:

[0235] 1. The hierarchical attention fusion network (HAFN) effectively integrates the information of multimodal physiological signals and improves the comprehensiveness and accuracy of the evaluation.

[0236] 2. Dynamic time window (DTW) technology enhances the system's ability to adapt to rapid changes in anesthetic status and improves the real-time nature of the assessment.

[0237] 3. The multi-task learning framework not only improves the performance of the main task (anesthesia depth assessment), but also provides additional valuable information (such as EEG frequency band energy ratio, autonomic nervous system balance index, etc.), providing comprehensive support for anesthesia management.

[0238] In summary, these test results fully demonstrate the significant advantages of the present invention in the field of real-time assessment of anesthesia depth. It not only improves the accuracy and real-time performance of the assessment, but also enhances the adaptability and anti-interference ability of the system. These improvements are of great significance for improving surgical safety, optimizing anesthesia management, reducing drug use and preventing complications. The present invention provides strong technical support for promoting the development of personalized precision anesthesia and is expected to play an important role in clinical practice. It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time assessment method for anesthesia depth based on deep learning, the method comprising the following steps: Acquire multimodal physiological signals of the patient, wherein the multimodal physiological signals include electroencephalogram (EEG), electrocardiogram (ECG), photoplethysmogram (PPG), blood pressure (BP) and respiratory signal (RESP); Preprocessing the multimodal physiological signal; Characterized in that the method further comprises: The pre-processed multimodal physiological signals are fused using the hierarchical attention fusion network HAFN to obtain fusion features; The hierarchical attention fusion network includes a signal-level attention layer, a time-scale attention layer, and a frequency component attention layer; Adjusting the time scale of the fusion feature based on the dynamic time window DTW to obtain an adjusted feature; The dynamic time window uses a reinforcement learning method to adaptively adjust the size of the observation window according to the current signal characteristics and historical status; The adjusted features are input into a multi-task learning framework to simultaneously predict anesthesia depth classification, EEG frequency band energy ratio, autonomic nervous system balance index, and drug concentration; Based on the output of the multi-task learning framework, the patient's anesthesia depth is determined and an anesthesia depth staging map is generated.

2. The method according to claim 1, characterized in that The step of preprocessing the multimodal physiological signal includes: performing adaptive wavelet denoising on the multimodal physiological signal; wherein, the optimal wavelet basis function and threshold are automatically selected through an unsupervised learning method; compressed sensing sampling is performed on the denoised signal; wherein, the sparsity of the signal is utilized to perform dimensionality reduction sampling through a random projection matrix; online anomaly detection and correction are performed on the sampled signal; wherein, a local anomaly factor LOF algorithm is used to detect and correct signal anomalies in real time.

3. The method according to claim 1, characterized in that: The processing steps of the hierarchical attention fusion network HAFN include: applying signal-level attention to the input multimodal physiological signal to obtain signal-level fusion features; applying time-scale attention to the signal-level fusion features to obtain time-scale fusion features; applying frequency component attention to the time-scale fusion features to obtain final fusion features.

4. The method according to claim 1, characterized in that: The implementation steps of the dynamic time window DTW include: defining a state space, including current signal characteristics and historical window size; defining an action space, including the amplitude of adjusting the window size; defining a reward function based on signal stability and information volume; using a reinforcement learning model of an Actor-Critic architecture to learn the optimal window adjustment strategy by interacting with the environment.

5. The method according to claim 1, characterized in that: The processing steps of the multi-task learning framework include: setting the main task as anesthesia depth classification; setting auxiliary tasks including EEG frequency band energy ratio prediction, autonomic nervous system balance index prediction and drug concentration estimation; building a shared feature extraction network and task-specific output layers; defining a multi-task loss function, comprehensively considering the loss and weight of each task; and optimizing the performance of all tasks simultaneously through a back-propagation algorithm.

6. The method according to claim 1, characterized in that The following feature extraction steps are also included: The phase synchronization index PSI is calculated to quantify the synchronization between different brain regions or different physiological signals; Calculate the multi-scale entropy MSE to evaluate the complexity of the signal at different time scales; Extract symbolic dynamic features SDF, convert continuous signals into symbol sequences, and extract the dynamic characteristics of the system.

7. The method according to claim 1, characterized in that The multi-task learning framework adopts the following optimized model architecture: Use dynamic depth-wise separable convolution to dynamically adjust the size and number of convolution kernels according to the complexity of input features; Introducing the attention mechanism into the residual connection to realize the attention-guided jump connection; Adaptive activation function is used to automatically adjust the parameters of the activation function according to the needs of different layers.

8. The method according to claim 1, characterized in that The following post-processing steps are also included: Calculate the probability distribution of anesthesia depth; Detect abnormal conditions and generate alerts; Conduct interpretability analysis to quantify the contribution of different input features to the prediction results; Generate anesthetic adjustment recommendations based on prediction results and interpretability analysis.

9. The method according to claim 1, characterized in that: The steps of generating the anesthesia depth staging diagram include: The depth of anesthesia is divided into six stages: awake, mild sedation, moderate sedation, deep sedation, general anesthesia and too deep anesthesia; Different colors are used to represent different stages of anesthesia, and the depth of the color indicates the reliability of the stage; With time as the horizontal axis and the depth of anesthesia as the vertical axis, a bar graph of dynamic changes is drawn.

10. A real-time evaluation system for depth of anesthesia for implementing the method according to any one of claims 1 to 9, characterized in that: The system includes: A multi-source signal acquisition module, used to acquire multi-modal physiological signals of patients; The signal preprocessing module is used to perform denoising, compressed sensing sampling and anomaly detection on the collected signals; Feature extraction module, used to calculate time-frequency features, phase synchronization index, multi-scale entropy and symbolic dynamics features; Deep learning inference module, including hierarchical attention fusion network, dynamic time window adjustment unit and multi-task learning framework; Results analysis and presentation module, used to generate anesthesia depth staging diagrams and decision-making recommendations; Among them, the deep learning reasoning module is implemented using edge computing equipment to ensure real-time performance; the output of the result analysis and display module is used to guide the multi-source signal acquisition module to adjust the acquisition parameters to form a closed-loop control.

Citation Information

Patent Citations

  • Anesthesia depth monitoring method and apparatus

    CN112006657A

  • Method of treating the effects of stroke

    CN112891374A

  • Electroencephalogram signal classification method and device, equipment and storage medium

    CN116392148A

  • Perioperative anesthesia depth monitoring system based on deep learning

    CN116982937A

  • Anesthesia depth monitoring system and method based on artificial intelligence

    CN117457229A

Cited By

  • Electroencephalogram state real-time monitoring method and device and electronic equipment

    CN119498866A

  • Anesthesia depth evaluation method and system based on physical sign data of patient

    CN119791609A

  • Anesthesia depth monitoring system and method based on skin electric signal analysis

    CN119908668A

  • Intraoperative patient anesthesia state evaluation system based on electroencephalogram signal and ear temperature monitoring

    CN120304780A

  • Anesthesia early warning method based on multi-source signal fusion

    CN120753633A