Method and device for evaluating anesthesia depth based on joint disturbance and electroencephalogram signals

Abstract

The invention discloses a method and a device for evaluating anesthesia depth based on joint disturbance and electroencephalogram signals. According to the method, transcranial magnetic stimulation and transcranial ultrasonic stimulation are sequentially applied to a cortex target spot and a subcortex target spot; collecting first multi-channel electroencephalogram data and second multi-channel electroencephalogram data; based on the first multi-channel electroencephalogram data and the second multi-channel electroencephalogram data, reconstructing a cortex source activity time sequence and a subcortex source activity time sequence; calculating a disturbance complexity index according to the first multichannel electroencephalogram data and a source positioning algorithm; calculating a cortex-to-subcortex directional coherence value and a subcortex-to-cortex directional coherence value according to the cortex source activity time sequence and the subcortex source activity time sequence; and generating an anesthesia depth index based on the disturbance complexity index, the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value. By implementing the technical scheme provided by the invention, the problem of insufficient accuracy between different anesthetics and individuals is solved, and more stable and reliable anesthesia depth monitoring is realized.

Description

Technical Field

[0001] This application relates to the technical field of medical testing, specifically to a method and apparatus for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals. Background Technology

[0002] Monitoring the depth of anesthesia is a crucial aspect of clinical anesthesia management. Current technologies primarily rely on analyzing the spectral characteristics of spontaneous electroencephalogram (EEG) signals to assess a patient's level of consciousness. Typical commercial monitoring devices acquire EEG signals from the forehead, perform spectral decomposition to extract statistical parameters such as power ratios or entropy values ​​for specific frequency bands, and then convert these parameters into a numerical anesthesia depth index using a pre-defined algorithm. The core of this approach lies in the passive recording and statistical analysis of spontaneous electrical activity in the cerebral cortex, using the observation of patterns in the EEG spectrum changes with the depth of anesthesia to infer the patient's current level of consciousness.

[0003] However, the aforementioned monitoring methods based on spontaneous EEG spectral analysis have inherent technical limitations. Because different anesthetic drugs have different mechanisms of action on the brain, the resulting EEG spectral patterns also vary, making it difficult to guarantee the accuracy and reliability of the same monitoring algorithm under different drug conditions. Furthermore, spontaneous EEG signals only reflect the surface activity state of the cerebral cortex and cannot directly assess the functional integrity of the thalamic-cortical neural circuits closely related to the generation and maintenance of consciousness. This circuit is precisely the common neural substrate for the consciousness-inhibiting effects of various anesthetic drugs. Therefore, existing technologies struggle to accurately and stably reflect the degree of inhibition of core consciousness-related pathways by anesthesia at the neural mechanism level, exhibiting significant limitations in cross-drug and cross-individual application scenarios. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and apparatus for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals.

[0005] The first aspect of this application provides a method for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals, employing the following technical solution: According to the preset timing sequence, transcranial magnetic stimulation is applied to preset cortical target points and transcranial ultrasound stimulation is applied to preset subcortical target points in sequence. Acquire first multichannel EEG data in response to the transcranial magnetic stimulation and second multichannel EEG data in response to the transcranial ultrasound stimulation; Based on the first multi-channel EEG data and the second multi-channel EEG data, the cortical source activity time sequence of the cortical target and the subcortical source activity time sequence of the subcortical target are reconstructed through a preset source localization algorithm. Based on the first multi-channel EEG data and the source localization algorithm, a perturbation complexity index for characterizing the functional state of the cortex is calculated. Based on the cortical source activity time series and the subcortical source activity time series, the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value are calculated respectively. An anesthesia depth index is generated based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, using a preset weighted model.

[0006] By employing the aforementioned technical approach, combining transcranial magnetic stimulation (TMS) and ultrasound perturbation, and actively stimulating and simultaneously acquiring neural responses at cortical and subcortical target points, this method overcomes the limitations of traditional methods that rely solely on spontaneous EEG spectral analysis. By reconstructing cortical and subcortical source activities using source localization technology, and combining perturbation complexity and bidirectional directional coherence values, the functional integrity and information exchange efficiency of the thalamus-cortex circuit can be directly and quantitatively assessed. Starting from the common neural mechanisms of consciousness inhibition, this method significantly improves the neural specificity and mechanistic consistency of anesthesia depth assessment, effectively overcoming the accuracy limitations of existing technologies across different anesthetic drugs and individuals, and achieving more stable and reliable anesthesia depth monitoring.

[0007] Optionally, the step of reconstructing the cortical source activity timeline of the cortical target point and the subcortical source activity timeline of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data using a preset source localization algorithm includes: The pre-defined individual magnetic resonance imaging data is segmented into tissues, and a three-dimensional mesh head model is constructed based on the segmentation results; Based on the three-dimensional mesh head model and the preset conductivity parameters, the lead field matrix is ​​calculated using the boundary element method; The first multi-channel EEG data and the second multi-channel EEG data are segmented to obtain multiple data segments. The multiple data segments are subjected to a preset bandpass filtering process. Based on the multiple data segments after filtering, the data covariance matrix is ​​calculated. By combining the lead field matrix and the data covariance matrix, a spatial filter is constructed using a linearly constrained minimum variance beamforming algorithm. The spatial filter is applied to the first multi-channel EEG data and the second multi-channel EEG data to obtain the timing sequence of cortical source activity and the timing sequence of subcortical source activity.

[0008] By employing the above technical solutions, a high-precision positive model can be constructed based on the subject's own head anatomy, significantly improving the accuracy of source localization. The linearly constrained minimum variance beamforming algorithm maximizes the target source signal and suppresses background noise and interference, thereby separating and reconstructing the neural activity temporal sequence of specific cortical and subcortical target points from multi-channel EEG data with high spatiotemporal resolution.

[0009] Optionally, the step of calculating the perturbation complexity index for characterizing the cortical functional state based on the first multi-channel EEG data and the source localization algorithm includes: Extract data segments within a preset time window from the first multi-channel EEG data; By applying the spatial filter to the data segment, a corresponding activity intensity time series is generated for multiple preset cortical source points, and the multiple activity intensity time series are combined to form a source data matrix; By assigning the positions in the source data matrix whose values ​​are greater than a preset statistical threshold to a first preset state, and assigning the positions whose values ​​are not greater than the preset statistical threshold to a second preset state, a binarized spatiotemporal activity matrix is ​​generated based on the first preset state and the second preset state. The binarized spatiotemporal activity matrix is ​​compressed using a preset lossless compression algorithm to obtain compressed data, and the compressed data is then normalized to generate the perturbation complexity index.

[0010] By employing the aforementioned technical solution, combining source localization technology with spatiotemporal binarization and lossless compression algorithms, the dynamics of multi-source cortical activity induced by transcranial magnetic stimulation (TMS) are transformed into a single-dimensional index capable of quantitatively characterizing the complexity of cortical information processing. This index, based on individually reconstructed source activity temporal sequences, extracts significant neural responses and encodes them into spatiotemporal patterns, then uses the compression ratio to quantify the pattern complexity. This method provides a more sensitive and direct quantitative measure for assessing cortical functional status.

[0011] Optionally, the step of calculating the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value based on the cortical source activity time series and the subcortical source activity time series, respectively, includes: The cortical source activity time series and the subcortical source activity time series are jointly resampled to generate a time-aligned hybrid activity time series; The time series of the mixed activities was fitted using a pre-defined multivariate autoregressive model, and the model coefficient matrix was calculated. The transfer function matrix is ​​obtained by performing a Fourier transform on the model coefficient matrix, and the spectral matrix is ​​obtained based on the transfer function matrix. Extract the first cross-spectral density component representing the time sequence of cortical source activity to the time sequence of subcortical source activity from the spectral matrix, and perform integration and normalization on the first cross-spectral density component within a preset frequency band to generate the directional coherence value from the cortex to the subcortex. The second cross-spectral density component representing the time sequence from the subcortical source activity to the cortical source activity is extracted from the spectral matrix, and the second cross-spectral density component is integrated and normalized within the preset frequency band to generate the subcortical-to-cortical directional coherence value.

[0012] By employing the aforementioned technical solution, a multivariate autoregressive model is established, and the fitted model coefficients are transformed in the frequency domain. This allows for the precise decoupling and quantification of the directional causal interaction of neural information between the cortex and subcortical target points from the reconstructed source activity time series. This method overcomes the limitation of traditional coherence analysis in distinguishing the direction of information flow. By calculating the directional coherence values ​​in two directions, it assesses the independent changes in the efficiency of top-down and bottom-up information transmission in the thalamic-cortical circuit.

[0013] Optionally, the step of generating an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value through a preset weighted model includes: Obtain the baseline perturbation complexity index, baseline cortical-to-subcortical directional coherence value, and baseline subcortical-to-cortical directional coherence value corresponding to the preset awake state; Divide the perturbation complexity index by the baseline perturbation complexity index to obtain the complexity ratio; The first directional coherence ratio is obtained by dividing the cortical-to-subcortical directional coherence value by the baseline cortical-to-subcortical directional coherence value, and the second directional coherence ratio is obtained by dividing the subcortical-to-cortical directional coherence value by the baseline subcortical-to-cortical directional coherence value. The anesthesia depth index is generated by performing a preset weighted summation on the complexity ratio, the first directional coherence ratio, and the second directional coherence ratio.

[0014] By adopting the above technical solution, the influence of differences in basic neural activity among individuals on the assessment results is effectively overcome, and personalized calibration of the depth of anesthesia is achieved. By proportionally integrating the complexity index reflecting the functional state of the cortex with the directional coherence index reflecting the efficiency of bidirectional information flow in the thalamus-cortex circuit, the multi-level inhibitory effect of anesthetic drugs on consciousness-related neural circuits can be comprehensively and quantitatively characterized from two complementary dimensions: "node function" and "network connectivity".

[0015] Optionally, the method further includes: According to a preset evaluation cycle, a preset number of time-continuous target anesthesia depth indices are obtained, and the median of the multiple target anesthesia depth indices is calculated. Data quality checks are performed on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results; When the data quality check results meet the preset conditions, the anesthesia depth index is updated using the median.

[0016] By adopting the above technical solution, the stability and anti-interference ability of the anesthesia depth index output are effectively improved. Periodically collecting multiple continuous indices and taking the median can smooth out instantaneous fluctuations and suppress interference from random noise or artifacts. Combined with real-time data quality checks, low-confidence data periods caused by factors such as unstable signal acquisition can be automatically identified and eliminated. This dual-protection mechanism ensures that the final output anesthesia depth index is always calculated based on reliable, high-quality neural signals, thereby maintaining the continuity and robustness of monitoring results in complex clinical environments.

[0017] Optionally, the step of performing data quality checks on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results includes: For each channel of the first multi-channel EEG data and the second multi-channel EEG data, it is determined whether the signal amplitude of each channel exceeds a preset amplitude range, and the number of EEG channels that do not exceed the preset amplitude range is counted to calculate the proportion of the number to the total number of channels. Based on the first multi-channel EEG data and the second multi-channel EEG data, a pre-set artifact residue score is calculated using an artifact recognition algorithm. The channel ratio is compared with a first preset threshold, and the artifact residue score is compared with a second preset threshold to generate the data quality check result.

[0018] By employing the above technical solution, objective and quantitative real-time assessment of the quality of multi-channel EEG data is achieved. This solution can effectively identify signal distortion or excessive contamination caused by poor electrode contact, patient movement, or environmental interference, ensuring that the data used for source localization and index calculation are within a reliable physiological range. Automated judgment by setting clear thresholds provides a precise and repeatable quality control basis for subsequent adoption and updating of the anesthesia depth index.

[0019] Secondly, a system for assessing the depth of anesthesia based on combined perturbations and electroencephalogram (EEG) signals is provided, the system comprising: The combined perturbation application module is used to sequentially apply transcranial magnetic stimulation to preset cortical target points and transcranial ultrasound stimulation to preset subcortical target points according to a preset timing sequence. The EEG signal acquisition module is used to acquire first multi-channel EEG data in response to the transcranial magnetic stimulation and second multi-channel EEG data in response to the transcranial ultrasound stimulation. The brain source reconstruction module is used to reconstruct the cortical source activity time sequence of the cortical target point and the subcortical source activity time sequence of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data, using a preset source localization algorithm. The complexity index calculation module is used to calculate the perturbation complexity index used to characterize the functional state of the cortex based on the first multi-channel EEG data and the source localization algorithm. The directional connectivity analysis module is used to calculate the directional coherence values ​​from the cortex to the subcortex and from the subcortex to the cortex, respectively, based on the cortical source activity time series and the subcortical source activity time series. The anesthesia index generation module is used to generate an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, through a preset weighted model.

[0020] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the foregoing.

[0021] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any of the preceding descriptions.

[0022] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By utilizing personalized source localization technology to accurately reconstruct cortical and subcortical target activity, and combining this with perturbation complexity indices to quantify cortical information integration capabilities, and by assessing the bidirectional information flow efficiency of the thalamic-cortical circuit through directional coherence analysis, a multi-level, direct quantitative assessment of the functional integrity and network connectivity of consciousness-related neural circuits is achieved. Combined with baseline-normalized weighted fusion, real-time data quality checks, and periodic median smoothing, the resulting anesthesia depth index possesses high neurospecificity, strong anti-interference ability, and good individual and cross-drug adaptability, providing a more reliable and universally applicable method for monitoring anesthesia depth in clinical practice. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system architecture of an embodiment of a method or system for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals according to this application. Figure 2This is a flowchart illustrating a method for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals disclosed in an embodiment of this application. Figure 3 This is a schematic diagram of a module of a system for assessing the depth of anesthesia based on combined perturbation and electroencephalogram signals disclosed in this application; Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0024] Figure reference numerals: 100, System architecture; 101, First terminal device; 102, Second terminal device; 103, Third terminal device; 104, Network; 105, Server; 301, Joint perturbation application module; 302, EEG signal acquisition module; 303, Brain source reconstruction module; 304, Complexity index calculation module; 305, Directional connectivity analysis module; 306, Anesthesia index generation module; 401, Processor; 402, Communication bus; 403, User interface; 404, Network interface; 405, Memory. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0027] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, social platform software, etc.

[0028] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP3 (Moving Picture Experts Group Audio Layer IV) players, laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.

[0029] This embodiment discloses a method for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals. Figure 2 This is a flowchart illustrating a method for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals disclosed in an embodiment of this application. Figure 2 As shown, the method includes the following steps: S201. According to the preset timing sequence, transcranial magnetic stimulation is applied to the preset cortical target points and transcranial ultrasound stimulation is applied to the preset subcortical target points in sequence. Specifically, within a fixed assessment cycle (e.g., 10 seconds), firstly, at the start of the cycle (e.g., t=0 seconds), transcranial magnetic stimulation (TMS) is applied to a pre-defined cortical target. This target is preferably the left dorsolateral prefrontal cortex, and a single pulse with an intensity of 110% of the patient's resting motor threshold can be applied using a figure-eight coil to ensure physiological equivalence across individuals. Subsequently, after a pre-defined time interval (e.g., 3 seconds, i.e., at t=3 seconds), transcranial ultrasound stimulation is applied to a pre-defined subcortical target. This target is preferably the left anterior thalamic nucleus, and a low-intensity focused ultrasound pulse sequence with a duration of 100 milliseconds, a center frequency of 500 kHz, a pulse repetition frequency of 250 Hz, and a duty cycle of 5% can be applied via the temporal acoustic window. The acoustic energy intensity is strictly controlled within a pre-defined safety standard (e.g., spatial peak time mean intensity Ispta ≤ 500 mW / cm²) to achieve precise, non-invasive neuromodulation of deep nuclei. As an alternative implementation of the fixed timing scheme described above, the time interval between two stimuli and between each evaluation cycle can also be randomly varied within a preset range (e.g., 1.5-3.5 seconds) to reduce the adaptive effects that the brain may produce on periodic stimulation.

[0030] S202. Acquire first multichannel EEG data in response to the transcranial magnetic stimulation and second multichannel EEG data in response to the transcranial ultrasound stimulation; Specifically, throughout the evaluation period, a 64-channel high-density EEG acquisition system compatible with a transcranial magnetic stimulation (TMS) environment was used for continuous data recording. The system's electrodes were positioned according to the international 10-10 standard, with the reference electrode placed at the Cz point, and data acquisition was performed at a sampling rate of at least 5 kHz. To avoid signal saturation caused by large amplitude artifacts from TMS, the amplifier's input range was set to a wide range, for example, at least ±20 mV. During acquisition, the trigger signals from TMS and TMS ultrasound were synchronously input to a dedicated channel of the EEG acquisition system (e.g., channel 0), thus embedding microsecond-accurate time stamps within the EEG data stream. In subsequent processing, the first multichannel EEG data was defined as a 64-channel EEG data segment within a preset time window (e.g., from 100 milliseconds before stimulation to 500 milliseconds after stimulation) centered on each magnetic stimulation time marker; similarly, the second multichannel EEG data was defined as a 64-channel EEG data segment within the same time window centered on each ultrasound stimulation time marker.

[0031] S203. Based on the first multi-channel EEG data and the second multi-channel EEG data, the cortical source activity time sequence of the cortical target point and the subcortical source activity time sequence of the subcortical target point are reconstructed through a preset source localization algorithm. Specifically, the acquired first and second multi-channel EEG data are preprocessed. This preprocessing preferably includes filtering using a 1-45Hz bandpass filter and employing independent component analysis (ICA) to identify and remove physiological artifacts such as eye movements and electromyography (EMG). Next, based on the individual subject's T1-weighted magnetic resonance imaging (MRI) data, a high-precision three-layer boundary element model of the head is constructed. This model accurately delineates the boundaries of tissues such as the scalp, skull, and brain, and precisely registers the EEG electrodes with the individual's head anatomical space. Then, a pre-defined source localization algorithm is used to solve the EEG inverse problem. As a preferred implementation, this algorithm can be standardized low-resolution electromagnetic tomography (sLORETA), which reconstructs brain electrical activity by calculating the current density distribution at each location in a three-dimensional source space (e.g., containing approximately 8000 voxel points distributed in the cortex and subcortical structures). By inputting the preprocessed first multichannel EEG data into the algorithm, the temporal sequence of source activities in response to transcranial magnetic stimulation (TMS) across the entire brain can be obtained. Similarly, by inputting the second multichannel EEG data into the algorithm, the temporal sequence of source activities in response to transcranial ultrasound stimulation can be obtained. Finally, based on the preset anatomical coordinates of cortical target points (e.g., the left dorsolateral prefrontal cortex) and subcortical target points (e.g., the left anterior thalamic nucleus), corresponding regions of interest (ROIs) are defined in the reconstructed whole-brain source space. The source activities of all voxel points within each ROI are averaged to obtain the cortical source activity temporal sequence of the cortical target points and the subcortical source activity temporal sequence of the subcortical target points, respectively. As an optional or supplementary implementation of the sLORETA algorithm, the source localization algorithm can also be other algorithms known in the art, such as dynamic statistical parametric mapping or precise low-resolution electromagnetic tomography, to adapt to different signal-to-noise ratio and spatial resolution requirements.

[0032] Optionally, the step of reconstructing the cortical source activity timeline of the cortical target point and the subcortical source activity timeline of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data using a preset source localization algorithm includes: performing tissue segmentation on preset individual magnetic resonance imaging data, constructing a three-dimensional mesh head model based on the segmentation results; calculating the lead field matrix using the boundary element method based on the three-dimensional mesh head model and preset conductivity parameters; segmenting the first multi-channel EEG data and the second multi-channel EEG data to obtain multiple data segments, performing preset bandpass filtering on the multiple data segments, calculating the data covariance matrix based on the filtered multiple data segments; constructing a spatial filter using a linearly constrained minimum variance beamforming algorithm by combining the lead field matrix and the data covariance matrix; and applying the spatial filter to the first multi-channel EEG data and the second multi-channel EEG data to obtain the cortical source activity timeline and the subcortical source activity timeline.

[0033] Specifically, the pre-defined individual magnetic resonance imaging (MRI) data is segmented into tissues, and a three-dimensional mesh head model is constructed based on the segmentation results. Specifically, the individual MRI data is preferably high-resolution T1-weighted structural image data. A well-known neuroimaging software toolkit can be used to automatically segment the T1 data, thereby accurately extracting surface mesh models of different tissues such as the scalp, skull, and cerebral cortex. Based on these segmented surface meshes, an anatomically accurate three- or four-layer boundary element model (BEM) head model is constructed. This head model provides a precise geometric and physical basis for subsequent forward problem calculations, i.e., simulating how endogenous electrical currents propagate to the scalp surface.

[0034] Furthermore, based on the aforementioned three-dimensional mesh head model and preset conductivity parameters, the lead field matrix is ​​calculated using the boundary element method. The lead field matrix, also known as the positive problem matrix or gain matrix, mathematically describes the linear mapping between the activity of dipole sources at any given location within the brain and the potential signals that can be recorded by all EEG electrodes on the scalp surface. During the calculation, each layer in the BEM model (such as the scalp, skull, and brain tissue) needs to be assigned a preset conductivity parameter that conforms to physiological characteristics. For example, the conductivity of the scalp, skull, and brain can be set to 0.33 S / m, 0.0042 S / m, and 0.33 S / m, respectively. By solving the Laplace equation, a lead field matrix with dimensions of [number of electrodes × number of sources] is finally obtained. Each element of this matrix represents the contribution of a specific source point to the signal of a specific electrode.

[0035] Furthermore, the acquired first and second multi-channel EEG data are segmented and preprocessed, and a data covariance matrix is ​​calculated. Specifically, the continuously recorded raw EEG data (including the first and second parts) are segmented around the stimulus event, forming multiple data segments. These data segments are then subjected to a preset bandpass filter, such as a 1-45Hz bandpass filter, to remove high-frequency noise and low-frequency drift. Based on all the filtered data segments, or data segments within a specific time window (e.g., the pre-stimulation baseline period), a data covariance matrix with dimensions of [number of electrodes × number of electrodes] is calculated. This covariance matrix reflects the statistical correlation between signals from different electrode channels and is a key input for constructing an adaptive spatial filter.

[0036] Furthermore, by combining the lead field matrix and the data covariance matrix, a spatial filter is constructed using the Linearly Constrained Minimum Variance (LCMV) beamforming algorithm. LCMV beamforming is an adaptive spatial filtering technique whose core idea is to design an optimal spatial filter for each location (voxel or vertex) in the brain's source space. This filter can pass signals from the target location with unity gain and without distortion, while minimizing the contributions from signals from all other brain regions and external noise, thereby maximizing the signal-to-noise ratio of the output signal. This process is achieved by solving an optimization problem with the data covariance matrix and the lead field vector corresponding to a specific source point as input, ultimately generating a set of optimal filter weights for each location in the source space.

[0037] Furthermore, the constructed spatial filters are applied to the first and second multi-channel EEG data respectively to obtain the temporal sequences of cortical source activity and subcortical source activity. The spatial filter weights constructed for each location in the source space are linearly weighted and summed (i.e., matrix multiplication) with the preprocessed first multi-channel EEG data (i.e., data responding to magnetic stimulation) to reconstruct the activity time sequence of each source point in the brain after responding to magnetic stimulation. The temporal sequence of cortical source activity is obtained by extracting the activity sequence of the source point corresponding to a preset cortical target point (e.g., DLPFC). Using the same method, the spatial filters are applied to the second multi-channel EEG data (i.e., data responding to ultrasound stimulation) to reconstruct the source activity in response to ultrasound stimulation, and then the activity sequence of the source point corresponding to a preset subcortical target point (e.g., anterior thalamic nucleus) is extracted, which is the temporal sequence of subcortical source activity.

[0038] S204. Based on the first multi-channel EEG data and the source localization algorithm, calculate the perturbation complexity index used to characterize the functional state of the cortex. In this embodiment, the functional state of the cortex is quantitatively characterized by calculating a perturbation complexity index. This calculation process is preferably based on the cortical source activity timeline reconstructed by the source localization algorithm in the previous step in response to transcranial magnetic stimulation (TMS). Specifically, firstly, the first multichannel EEG data corresponding to multiple TMS trials are averaged at the source space level to obtain a TMS-evoked source response with a high signal-to-noise ratio. Next, to identify significant neural activity truly evoked by TMS, statistical testing of the source response data is required. A preferred implementation is to compare the activity amplitude of each cortical source point within the post-stimulation time window (e.g., 0-300 ms) with its activity during the pre-stimulation baseline period (e.g., -500 ms to -10 ms), calculate the Z-score, and set a significance threshold (e.g., |Z|>3.0). When the Z-score of the activity amplitude of a source point at a certain time point exceeds this threshold, the point is considered to have generated significant activation at that time. Based on this, a binary spatiotemporal activation matrix is ​​constructed, where the rows represent all cortical source points, and the columns represent time points after stimulation. A matrix element of 1 indicates significant activation of the corresponding source point at that time, while 0 indicates no significant activation. Finally, this binary matrix is ​​unfolded into a one-dimensional binary sequence in chronological order, and the complexity of this sequence is calculated using a Lempel-Ziv complexity algorithm (e.g., a variant of the LZ76 algorithm). To ensure comparability, this complexity value is typically normalized, for example, by dividing it by the average complexity calculated after multiple random permutations of the original binary matrix. This final normalized value is the perturbation complexity index, whose magnitude directly reflects the richness of the spatiotemporal dynamic patterns that the cerebral cortex can generate when responding to external perturbations, thus effectively characterizing the current cortical functional state. As an alternative, the statistical method used to determine significant activations can also be a nonparametric method based on permutation tests, and the algorithm used to calculate complexity can be other information-theoretic metrics, such as Shannon entropy or multiscale entropy, as long as they can quantify the unpredictability of spatiotemporal patterns.

[0039] Optionally, the step of calculating the perturbation complexity index for characterizing the cortical functional state based on the first multi-channel EEG data and the source localization algorithm includes: extracting data segments within a preset time window from the first multi-channel EEG data; applying the spatial filter to the data segments to generate corresponding activity intensity time series for multiple preset cortical source points, and combining the multiple activity intensity time series to form a source data matrix; assigning positions in the source data matrix with values ​​greater than a preset statistical threshold as a first preset state, and assigning positions with values ​​not greater than the preset statistical threshold as a second preset state, and generating a binarized spatiotemporal activity matrix based on the first preset state and the second preset state; compressing the binarized spatiotemporal activity matrix using a preset lossless compression algorithm to obtain compressed data, and normalizing the compressed data to generate the perturbation complexity index.

[0040] Specifically, a data segment within a predetermined time window is extracted from the first multichannel EEG data. This step aims to focus the analysis on the causal chain directly triggered by the TMS perturbation. Specifically, this predetermined time window typically refers to a period of time after the TMS pulse is applied, for example, starting from the stimulation application moment (0 ms) and continuing until 300 ms or 500 ms later, to ensure that the entire dynamic process of cortical activity induction, propagation, and attenuation can be fully captured. For each independent TMS stimulation trial, such a data segment is extracted from its corresponding first multichannel EEG data for subsequent source space analysis.

[0041] Furthermore, by applying a constructed spatial filter (e.g., an LCMV spatial filter) to the data segment, activity intensity time series corresponding to each of the multiple preset cortical source points are generated, and these time series are combined to form a source data matrix. Here, a cortical source point can refer to all grid vertices covering the entire surface of the cerebral cortex, for example, approximately 8000 to 15000 points, to achieve whole-brain analysis. Applying the spatial filter corresponding to each source point to the data segment of each experiment yields the activity intensity time series of that source point in that experiment. Organizing the activity intensity values ​​of all cortical source points at all time points constitutes a source data matrix. The preferred dimension of this matrix is ​​[number of source points × number of time points], where each element represents the neural activity intensity at a specific cortical location at a specific moment after stimulation.

[0042] Furthermore, a binarized spatiotemporal activity matrix is ​​generated by thresholding the source data matrix. The purpose of this step is to convert continuous activity intensity signals into discrete active or inactive states, thereby identifying significant spatiotemporal activity patterns truly induced by TMS. Specifically, a preset statistical threshold needs to be set first. A preferred approach is to calculate the Z-score by comparing the distribution of activity intensity at each time point after stimulation with the distribution of activity intensity during the pre-stimulation baseline period (e.g., -500ms to -10ms), and using, for example, |Z|>2.5 or |Z|>3.0 as the threshold. Then, each element in the source data matrix is ​​iterated: if its value is greater than the preset statistical threshold, it is assigned a first preset state (e.g., value 1, representing significant activation) at the corresponding position in the binarized spatiotemporal activity matrix; if its value is not greater than the threshold, it is assigned a second preset state (e.g., value 0, representing inactivation). The resulting binarized matrix clearly depicts the propagation path and range of significant neural activity in the cortical spatial and temporal dimensions.

[0043] Furthermore, the binarized spatiotemporal activity matrix is ​​compressed using a preset lossless compression algorithm, and the compression result is normalized to generate the perturbation complexity index. The core idea of ​​this step is that the more complex and irregular the content of a matrix (representing rich and non-repetitive activity patterns generated by the brain), the lower its compression efficiency. Specifically, the Lempel-Ziv series of lossless compression algorithms (e.g., LZ76 or LZ77), well-known in the art, can be used to compress the binarized spatiotemporal activity matrix generated in the previous step, obtaining a compressed data size. To eliminate the influence of the matrix size itself on the result and make it comparable, normalization is required. A preferred normalization method is to randomly permutate the positions of 1s and 0s in the original binarized matrix multiple times, generating multiple (e.g., 100) random surrogate matrices. These surrogate matrices are compressed separately, and the average size of the compressed matrices is calculated. Finally, the compressed size of the original matrix is ​​divided by this average. The resulting dimensionless ratio is the perturbation complexity index. This indicator can robustly quantify the certainty and integrative nature of the brain’s response to external disturbances, thus providing an objective measure of cortical functional status.

[0044] S205. Based on the cortical source activity time series and the subcortical source activity time series, calculate the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value, respectively. Specifically, using the obtained cortical source activity time series (e.g., signals representing activity in the dorsolateral prefrontal cortex) and subcortical source activity time series (e.g., signals representing activity in the anterior thalamic nucleus), the connection strength in both directions is calculated using a frequency domain analysis method based on Granger causality. A preferred implementation is to employ the Partial Directed Coherence (PDC) algorithm. First, the cortical source activity time series and the subcortical source activity time series are treated as a bivariate system and fitted with a Multivariable Autoregressive (MVAR) model. This model aims to predict the current values ​​of the two time series using their past values; the model order (i.e., the length of time to revisit) can be automatically determined using criteria such as the Akaike Information Criterion or the Bayesian Information Criterion. After successfully fitting the MVAR model and obtaining the model coefficients, the model can be transformed into the frequency domain. The cortical-to-subcortical directional coherence value is obtained by calculating the average or peak value of the PDC spectrum of the connection path from cortical signals to subcortical signals in the model over a specific frequency band (e.g., the Theta band 4-8 Hz or the Alpha band 8-12 Hz, which is relevant to cognitive function). Similarly, the subcortical-to-cortical directional coherence value is obtained by calculating the average or peak value of the PDC spectrum of the connection path from subcortical signals to cortical signals in the same frequency band. As alternative technical solutions, in addition to the PDC algorithm, other equivalent directional connectivity analysis methods can be used, such as Directed Transfer Function (DTF), spectral Granger causality analysis, or phase-based methods. These methods can also infer the directional influence between two neural signals from MVAR models or other time-frequency representations, thereby achieving accurate quantification of the bidirectional communication efficiency of the cortical-to-subcortical loop.

[0045] Optionally, the step of calculating the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value based on the cortical source activity time series and the subcortical source activity time series respectively includes: jointly resampling the cortical source activity time series and the subcortical source activity time series to generate a time-aligned hybrid activity time series; fitting the hybrid activity time series to a preset multivariate autoregressive model to calculate the model coefficient matrix; performing a Fourier transform on the model coefficient matrix to calculate the transfer function matrix, and calculating the... The spectral matrix is ​​used to extract a first cross-spectral density component representing the time sequence from the cortical source activity to the subcortical source activity, and the first cross-spectral density component is integrated and normalized within a preset frequency band to generate the cortical-to-subcortical directional coherence value; the spectral matrix is ​​also used to extract a second cross-spectral density component representing the time sequence from the subcortical source activity to the cortical source activity, and the second cross-spectral density component is integrated and normalized within the preset frequency band to generate the subcortical-to-cortical directional coherence value.

[0046] Specifically, the cortical source activity time series and the subcortical source activity time series are jointly resampled to generate a temporally precisely aligned mixed activity time series. This step is fundamental to the accuracy of subsequent causal analysis, aiming to eliminate minor temporal deviations that may be caused by different processing procedures or differences in the original sampling rate. Specifically, a uniform target sampling rate can be set, such as 500Hz or 1000Hz, and then linear interpolation, cubic spline interpolation, or a Fourier transform-based method can be used to simultaneously resample both time series to this target frequency. This process ensures that each time point in the two activity time series has a strict one-to-one correspondence, thereby constructing a dual-channel (or multi-channel, if more brain regions are involved) mixed activity time series data structure, preparing for subsequent multivariate model fitting.

[0047] Furthermore, a pre-defined MVAR model is used to fit the mixed activity time series generated in the previous step, and the model coefficient matrix is ​​calculated. The core idea of ​​the MVAR model is that the state of the system at the current moment (i.e., the activity intensity of each of the two brain regions) can be predicted by a linear combination of the system's states at several past moments. The key parameter here is the order of the model, which determines how far back in time the model can be traced. To objectively determine the optimal order p, well-known criteria in the field are preferred, such as the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC). The criterion values ​​at different orders are calculated, and the order corresponding to the minimum point is selected as the optimal choice. After the model is fitted, the result is a set of model coefficient matrices, which quantify the predictive weight of the past values ​​of each time series on the current values ​​of other time series (including itself).

[0048] Furthermore, a Fourier transform is performed on the model coefficient matrix to calculate the transfer function matrix, and then the spectral matrix is ​​calculated based on this transfer function matrix. This step transforms the analysis from the time domain to the frequency domain, allowing the examination of brain region interactions across different rhythmic frequency bands. Specifically, a frequency-dependent transfer function matrix H(f) is obtained by performing a discrete-time Fourier transform on the coefficient matrix of the MVAR model. This matrix describes the linear relationship between the system input and output. Subsequently, the spectral matrix S(f) is calculated using the formula S(f) = H(f)·ΣH(f)·H (where Σ is the covariance matrix of the model residuals, and H represents the conjugate transpose). This spectral matrix is ​​a 2x2 matrix (in the bivariate case), where the diagonal elements represent the power spectrum of each signal itself, and the off-diagonal elements represent the cross-spectral density between the two signals, containing the coupling information of their phase and amplitude.

[0049] Further, a first cross-spectral density component representing the temporal information flow from cortical source activity to subcortical source activity is extracted from the spectral matrix S(f), and integrated and normalized within a preset frequency band to generate the cortical-to-subcortical directional coherence value. This step is crucial for quantifying the intensity of information transmission in a specific direction. In the spectral matrix S(f), the off-diagonal element S12(f) (assuming index 1 represents the cortex and index 2 represents the subcortical region) is the desired first cross-spectral density component. Next, the magnitude (or power) of this cross-spectral density component needs to be integrated or averaged within one or more physiologically significant preset frequency bands (e.g., the Theta band 4-8 Hz or the Alpha band 8-12 Hz, which are related to specific cognitive functions). To obtain a standardized coherence value between 0 and 1, it is usually normalized, for example, by dividing by the total power of the subcortical signal within that frequency band. The final value obtained is the cortical-to-subcortical directional coherence value. The larger the value, the stronger the causal driving effect of cortical activity on subcortical activity.

[0050] Furthermore, in a similar manner, a second cross-spectral density component representing the time sequence from subcortical source activity to cortical source activity is extracted from the spectral matrix S(f), and integrated and normalized within the preset frequency band to generate the subcortical-to-cortical directional coherence value. Specifically, this step mirrors the previous step. The cross-spectral density component S21(f) is extracted from the spectral matrix; this component reflects the influence of subcortical activity on cortical activity. Similarly, the magnitude of S21(f) is integrated or averaged within the same preset frequency band, and appropriately normalized (e.g., divided by the total power of the cortical signal within that frequency band). The resulting value is the subcortical-to-cortical directional coherence value, which quantifies the intensity of the uplink information flow and reflects the degree of modulation or driving of cortical activity by the subcortical structure. By calculating the directional coherence values ​​in these two directions respectively, this application can comprehensively evaluate the bidirectional dynamic communication mode of the cortical-subcortical functional loop.

[0051] S206. Based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, an anesthesia depth index is generated through a preset weighted model.

[0052] In this embodiment, step S206 aims to effectively fuse multiple feature indices calculated in the preceding steps, each reflecting different neurophysiological dimensions—namely, the perturbation complexity index (PCI), the cortical-to-subcortical orientational coherence value, and the subcortical-to-cortical orientational coherence value—to generate a single, intuitive, and robust Anesthesia Depth Index (ADI). This process is achieved through a pre-defined weighted model. A preferred implementation is to use a linear weighted summation model, whose mathematical expression can be: ADI = w1·f1(PCI) + w2·f2(PDCCS) + w3·f3(PDCSC) + C. In this model, w1, w2, and w3 are preset weighting coefficients, C is a constant bias term, PDCCS represents the cortical-to-subcortical directional coherence value, used to quantify the intensity of the downward information flow from the cortex to the subcortex, and PDCSC represents the subcortical-to-cortical directional coherence value, used to quantify the intensity of the upward information flow from the subcortical to the cortex. f1, f2, and f3 are optional normalization or transformation functions used to adjust input indicators of different dimensions and ranges to a comparable scale before combination. These weighting coefficients are not arbitrarily set, but are obtained based on a large amount of clinical trial data through multivariate regression analysis or machine learning algorithms (such as support vector regression, ridge regression, or neural networks). Specifically, the training process uses the above three indicators collected at different depths of anesthesia (e.g., calibrated by clinical rating scales such as OAA / S or simultaneously collected BIS indexes) as input features, with the known depth of anesthesia as the target output. An optimization algorithm iteratively calculates the optimal weight combination that minimizes the error between the model's predicted values ​​and the actual values. As an alternative technical solution, the pre-defined weighted model can also be a more complex nonlinear model, such as a multinomial regression model, a decision tree model, or a pre-trained shallow neural network. These models can capture the nonlinear interactions that may exist between the three indicators, thus potentially providing a more accurate assessment of the depth of anesthesia. Regardless of the model used, the final generated ADI will be mapped to a standardized, clinically interpretable numerical range, such as 0 to 100, where 100 represents a fully conscious state, while lower values ​​(such as below 40) represent deep anesthesia or even loss of consciousness. This provides anesthesiologists with a comprehensive quantitative assessment tool for the patient's level of consciousness, integrating two core mechanisms: cortical information processing capacity and cortical-subcortical circuit connectivity.

[0053] Optionally, the step of generating an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value through a preset weighted model includes: obtaining a baseline perturbation complexity index, a baseline cortical-to-subcortical directional coherence value, and a baseline subcortical-to-cortical directional coherence value corresponding to a preset conscious state; dividing the perturbation complexity index by the baseline perturbation complexity index to obtain a complexity ratio; dividing the cortical-to-subcortical directional coherence value by the baseline cortical-to-subcortical directional coherence value to obtain a first directional coherence ratio, and dividing the subcortical-to-cortical directional coherence value by the baseline subcortical-to-cortical directional coherence value to obtain a second directional coherence ratio; and performing a preset weighted summation on the complexity ratio, the first directional coherence ratio, and the second directional coherence ratio to generate the anesthesia depth index.

[0054] As a more specific and preferred implementation of the aforementioned weighted model, this application provides a method for generating an anesthesia depth index based on a normalized individualized baseline. The first step of this method is to obtain the baseline perturbation complexity index, baseline cortical-to-subcortical orientation coherence value, and baseline subcortical-to-cortical orientation coherence value corresponding to a preset conscious state. The core purpose of this step is to establish a personalized reference standard for each patient, representing their fully conscious brain function, thereby eliminating the influence of individual differences on subsequent index calculations. In clinical practice, a typical implementation involves collecting EEG signal data from the patient for a period of time (e.g., 2 to 5 minutes) in a quiet, closed-eye, and relaxed environment during the preparation stage before anesthesia induction. Subsequently, using the same method as in the aforementioned embodiments, this data representing the conscious state is processed to calculate the patient-specific baseline perturbation complexity index (PCIbaseline), baseline cortical-to-subcortical orientation coherence value (PDCCS_baseline), and baseline subcortical-to-cortical orientation coherence value (PDCSC_baseline). These baseline values ​​will be stored and used as a benchmark for subsequent calculations.

[0055] After obtaining the aforementioned baseline values, the next step in this embodiment is to compare the real-time calculated indicators with their corresponding baseline values ​​at any point during the anesthesia process, thereby generating standardized ratio values. Specifically, this step includes: dividing the real-time calculated perturbation complexity index (PCIcurrent) by the baseline perturbation complexity index (PCIbaseline) to obtain a dimensionless complexity ratio (RatioPCI=PCIcurrent / PCIbaseline). Similarly, dividing the real-time calculated cortical-to-subcortical directional coherence value (PDCCS_current) by its baseline value (PDCCS_baseline) to obtain a first directional coherence ratio (RatioCS=PDCCS_current / PDCCS_baseline); and dividing the real-time calculated subcortical-to-cortical directional coherence value (PDCSC_current) by its baseline value (PDCSC_baseline) to obtain a second directional coherence ratio (RatioSC=PDCSC_current / PDCSC_baseline). This normalization process has significant technical advantages: it transforms absolute values ​​into relative changes, making the interpretation of indicators more intuitive. For example, a complexity ratio of 0.5 clearly indicates that the brain's information processing capacity has decreased to half of its waking state. Simultaneously, this also makes indicators comparable across different individuals and time points. To prevent calculation instability caused by a zero or excessively small denominator, a technical improvement is to add a very small positive constant (epsilon) to the denominator.

[0056] Furthermore, the three standardized ratio values ​​generated in the previous step—namely, the complexity ratio, the first directional coherence ratio, and the second directional coherence ratio—are summed using a pre-defined weighted average to generate the final Anesthesia Depth Index (ADI). The mathematical expression for this step can be exemplarily represented as: ADI = WPCI·RatioPCI + WCS·RatioCS + WSC·RatioSC. Here, WPCI, WCS, and WSC are pre-defined weighting coefficients that collectively determine the relative importance of the three physiological dimensions in the final index composition. These weights can also be determined based on extensive clinical data, trained and optimized using machine learning regression models to ensure the final ADI has the highest fit with the gold standard (such as clinical scoring). For example, research might find that a decrease in the perturbation complexity index is most closely associated with the level of consciousness; therefore, it can be assigned a relatively high weight (e.g., WPCI = 0.6), while the coherence ratios in the two directions are assigned relatively low weights (e.g., WCS = 0.2, WSC = 0.2). In this way, this embodiment not only integrates neurophysiological information from multiple dimensions, but also outputs a highly reliable, easy-to-interpret quantitative indicator that accurately reflects the patient's depth of anesthesia through individualized baseline normalization and refined weighted fusion.

[0057] Optionally, the method further includes: acquiring a preset number of time-continuous target anesthesia depth indices according to a preset evaluation cycle, and calculating the median of the target anesthesia depth indices; performing data quality checks on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results; and updating the anesthesia depth index using the median when the data quality check results meet preset conditions.

[0058] To further enhance the stability and anti-interference capability of ADI in clinical applications, this application also provides an optional real-time update and smoothing mechanism based on data quality checks. This method continuously processes the anesthesia depth index over a recent period using a sliding time window. Specifically, the mechanism acquires a preset number of time-continuous target anesthesia depth indices according to a preset evaluation cycle. The preset evaluation cycle can be configured according to clinical real-time requirements, for example, performing data acquisition and calculation every 1 second or every 2 seconds. The preset number determines the time scale of the smoothing process; for example, it can be set to collect the most recent 5 to 10 consecutively calculated ADI values, forming a data sequence for subsequent analysis.

[0059] After obtaining the data sequence containing multiple target depth of anesthesia indices, the next step is to smooth the sequence to filter out potential noise and transient fluctuations. This embodiment preferably calculates the median of these multiple target depth of anesthesia indices. The median is chosen as the statistical indicator, rather than a simple arithmetic mean, based on its statistical robustness. The median is insensitive to extreme outliers in the data sequence, meaning that even if one or two abnormally high or low ADI values ​​appear in the data sequence due to accidental signal artifacts or physiological mutations, the calculated median can still stably reflect the central trend of the data sequence. This processing method can effectively smooth the ADI output curve, avoid unnecessary jumps, and provide clinicians with a value that better represents the recent stable trend of the patient's anesthetic state.

[0060] In parallel with the aforementioned exponential smoothing process, the system also performs a rigorous quality assessment on the raw signals used to calculate these anesthesia depth indices. Specifically, this step involves performing a data quality check on the first multichannel EEG data (e.g., from cortical regions) and the second multichannel EEG data (e.g., from subcortical regions) to generate a data quality check result. This check process can integrate multiple algorithms and strategies. A first option is amplitude detection, which checks whether the amplitude of the EEG signal exceeds a preset physiologically reasonable range (e.g., ±100 μV). Signal segments exceeding this range may be contaminated by artifacts such as electromyography, electrooculography, or poor electrode contact. A second option is spectral analysis, which uses Fast Fourier Transform to detect the presence of significant power frequency interference (e.g., 50 Hz or 60 Hz) and its harmonic components in the signal. A third option is real-time monitoring of the contact impedance between each electrode and the scalp to ensure it is below a clinically acceptable threshold (e.g., 5 kΩ or 10 kΩ). The results of these checks can be combined into a comprehensive data quality indicator, such as a simple Boolean value (good or poor) or a more refined continuous quality score (such as a score from 0 to 1).

[0061] Furthermore, the system uses the results of data quality checks as a decision-making basis to determine whether to adopt the smoothed index value. Specifically, the system only uses the previously calculated median to update and display the final anesthesia depth index when the data quality check results meet preset conditions. These preset conditions can be flexibly defined; for example, they could be that the signal quality of all channels is good, or the overall quality score is higher than a preset threshold (e.g., greater than 0.8). If the data quality check results do not meet these preset conditions, the system will implement a protective strategy, such as temporarily freezing and maintaining the previous valid ADI value, while simultaneously issuing a visual or auditory alarm on the monitoring device's user interface to alert medical staff to the current poor signal quality. Through this conditional update mechanism, this embodiment ensures that every anesthesia depth index value ultimately presented to clinicians is based on high-quality raw EEG signals and generated through a robust smoothing algorithm, thereby greatly enhancing the reliability and clinical guidance value of the index.

[0062] Optionally, the step of performing data quality checks on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results includes: for each channel of the first multi-channel EEG data and the second multi-channel EEG data, determining whether the signal amplitude of each channel exceeds a preset amplitude range, and counting the number of EEG channels that do not exceed the preset amplitude range to calculate the channel ratio of the number to the total number of channels; calculating a artifact residue score based on the first multi-channel EEG data and the second multi-channel EEG data using a preset artifact detection algorithm; comparing the channel ratio with a first preset threshold, and comparing the artifact residue score with a second preset threshold to generate the data quality check results.

[0063] As a specific and preferred implementation, this application first introduces a channel-level evaluation method based on signal amplitude. This method independently determines whether the signal amplitude of each channel in the first and second multi-channel EEG data exceeds a preset amplitude range within the current analysis time window. This preset amplitude range, for example, can be set to ±100 microvolts or ±150 microvolts, and is a physiological range within which normal EEG signals should be located, derived from extensive clinical experience. Any signal point exceeding this range is generally considered to be due to severe artifacts, such as electrode detachment, significant patient movement, or equipment saturation. After evaluating all channels, the system counts the number of valid EEG channels that do not exceed the preset amplitude range and further calculates the proportion of this number to the total number of channels. This channel proportion, or channel integrity rate, provides an intuitive and quantitative indicator for assessing the spatial extent of artifact contamination.

[0064] While performing the aforementioned amplitude-based channel-level checks, or as a supplement, this embodiment also introduces a more refined, algorithm-based global artifact assessment method. Specifically, this step involves calculating an artifact residue score based on the first and second multi-channel EEG data as a whole, using a pre-defined artifact detection algorithm. This step aims to identify more subtle artifacts with normal amplitudes but abnormal waveforms, such as electrooculography (EOG) or electromyography (EMG) interference. The pre-defined artifact detection algorithm can have several specific implementation schemes. One scheme can employ an independent component analysis (ICA) method, which decomposes multi-channel EEG signals into multiple statistically independent source signals. By identifying independent components with typical artifact characteristics and calculating the energy proportion of these artifact components in the original signal, an artifact residue score is generated. A second scheme can employ a machine learning or deep learning model, such as a convolutional neural network pre-trained on a large amount of labeled EEG data. This model can directly input a segment of multi-channel EEG data and output a score between zero and one, representing the probability or severity of artifacts in the data segment. This artifact persistence score can provide a more profound assessment of data quality from a signal morphology perspective.

[0065] Furthermore, the system integrates the evaluation results from the two dimensions mentioned above to generate a final, comprehensive data quality check result. This step specifically includes comparing the calculated channel ratio with a first preset threshold and the calculated artifact residue score with a second preset threshold. These two thresholds are key parameters for ensuring data quality and can be set by clinical experts based on practical experience. For example, the first preset threshold can be set to 80%, meaning that at least 80% of the channels must be intact; the second preset threshold can be set according to the definition of the score—if a higher score indicates more severe artifacts, it can be set to 0.3. The system generates the final data quality check result based on the comparison results. A simple implementation uses an AND logic: the final data quality check result is judged as good or meets the preset conditions only if the channel ratio is greater than the first preset threshold and the artifact residue score is less than the second preset threshold. If either condition is not met, the result is judged as poor or does not meet the preset conditions. This dual verification mechanism ensures that the data segment is considered high quality only when the spatial distribution and morphological severity of signal artifacts are at an acceptable level, thus providing a solid and reliable foundation for the subsequent calculation and updating of the anesthesia depth index.

[0066] Figure 3 This is a schematic diagram of a module of a system for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals, as described in an embodiment of this application. This system can be implemented as all or part of a system through software, hardware, or a combination of both. For example... Figure 3 As shown, the system includes: The combined perturbation application module 301 is used to sequentially apply transcranial magnetic stimulation to preset cortical target points and transcranial ultrasound stimulation to preset subcortical target points according to a preset timing sequence. The EEG signal acquisition module 302 is used to acquire first multi-channel EEG data in response to the transcranial magnetic stimulation and second multi-channel EEG data in response to the transcranial ultrasound stimulation; The brain source reconstruction module 303 is used to reconstruct the cortical source activity time sequence of the cortical target point and the subcortical source activity time sequence of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data through a preset source localization algorithm. The complexity index calculation module 304 is used to calculate the perturbation complexity index for characterizing the functional state of the cortex based on the first multi-channel EEG data and the source localization algorithm. The directional connectivity analysis module 305 is used to calculate the directional coherence values ​​from the cortex to the subcortex and from the subcortex to the cortex based on the cortical source activity time series and the subcortical source activity time series, respectively. The anesthesia index generation module 306 is used to generate an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, through a preset weighted model.

[0067] Optionally, the brain source reconstruction module 303 is specifically used for: performing tissue segmentation on preset individual magnetic resonance imaging data, and constructing a three-dimensional mesh head model based on the segmentation results; calculating the lead field matrix using the boundary element method based on the three-dimensional mesh head model and preset conductivity parameters; segmenting the first multi-channel EEG data and the second multi-channel EEG data to obtain multiple data segments, performing preset bandpass filtering on the multiple data segments, and calculating the data covariance matrix based on the filtered multiple data segments; constructing a spatial filter using a linearly constrained minimum variance beamforming algorithm by combining the lead field matrix and the data covariance matrix; and applying the spatial filter to the first multi-channel EEG data and the second multi-channel EEG data to obtain the cortical source activity time series and the subcortical source activity time series.

[0068] Optionally, the complexity index calculation module 304 is specifically used for: extracting data segments within a preset time window from the first multi-channel EEG data; applying the spatial filter to the data segments to generate corresponding activity intensity time series for multiple preset cortical source points, and combining the multiple activity intensity time series to form a source data matrix; assigning positions in the source data matrix with values ​​greater than a preset statistical threshold to a first preset state, and assigning positions with values ​​not greater than the preset statistical threshold to a second preset state, generating a binarized spatiotemporal activity matrix based on the first preset state and the second preset state; compressing the binarized spatiotemporal activity matrix using a preset lossless compression algorithm to obtain compressed data, and normalizing the compressed data to generate the perturbation complexity index.

[0069] Optionally, the directional connectivity analysis module 305 is specifically used for: jointly resampling the cortical source activity time series and the subcortical source activity time series to generate a time-aligned hybrid activity time series; fitting the hybrid activity time series to a preset multivariate autoregressive model to calculate a model coefficient matrix; performing a Fourier transform on the model coefficient matrix to calculate a transfer function matrix, and calculating a spectral matrix based on the transfer function matrix; extracting a first cross-spectral density component representing the transition from the cortical source activity time series to the subcortical source activity time series from the spectral matrix, and integrating and normalizing the first cross-spectral density component within a preset frequency band to generate the cortical-to-subcortical directional coherence value; extracting a second cross-spectral density component representing the transition from the subcortical source activity time series to the cortical source activity time series from the spectral matrix, and integrating and normalizing the second cross-spectral density component within the preset frequency band to generate the subcortical-to-cortical directional coherence value.

[0070] Optionally, the anesthesia index generation module 306 is specifically used for: acquiring a baseline perturbation complexity index, a baseline cortical-to-subcortical directional coherence value, and a baseline subcortical-to-cortical directional coherence value corresponding to a preset conscious state; dividing the perturbation complexity index by the baseline perturbation complexity index to obtain a complexity ratio; dividing the cortical-to-subcortical directional coherence value by the baseline cortical-to-subcortical directional coherence value to obtain a first directional coherence ratio, and dividing the subcortical-to-cortical directional coherence value by the baseline subcortical-to-cortical directional coherence value to obtain a second directional coherence ratio; and performing a preset weighted summation on the complexity ratio, the first directional coherence ratio, and the second directional coherence ratio to generate the anesthesia depth index.

[0071] Optionally, the anesthesia index generation module 306 is specifically used to: acquire a preset number of time-continuous target anesthesia depth indices according to a preset evaluation cycle, and calculate the median of the target anesthesia depth indices; perform data quality checks on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results; and update the anesthesia depth index using the median when the data quality check results meet preset conditions.

[0072] Optionally, the anesthesia index generation module 306 is specifically used for: determining whether the signal amplitude of each channel of the first multi-channel EEG data and the second multi-channel EEG data exceeds a preset amplitude range, and counting the number of EEG channels that do not exceed the preset amplitude range to calculate the channel ratio of the number to the total number of channels; calculating a artifact residue score based on the first multi-channel EEG data and the second multi-channel EEG data using a preset artifact recognition algorithm; comparing the channel ratio with a first preset threshold, and comparing the artifact residue score with a second preset threshold to generate the data quality check result.

[0073] This embodiment also discloses an electronic device, as shown in the reference. Figure 4 The electronic device may include: at least one processor 401, at least one communication bus 402, a user interface 403, a network interface 404, and at least one memory 405. The communication bus 402 is used to enable communication between these components. The user interface 403 may include a display screen or a camera; optionally, the user interface 403 may also include a standard wired interface or a wireless interface. The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0074] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0075] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0076] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0077] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 405 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory 405 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

Claims

1. A method for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals, characterized in that, Applied to a server, the method includes: According to the preset timing sequence, transcranial magnetic stimulation is applied to preset cortical target points and transcranial ultrasound stimulation is applied to preset subcortical target points in sequence. Acquire first multichannel EEG data in response to the transcranial magnetic stimulation and second multichannel EEG data in response to the transcranial ultrasound stimulation; Based on the first multi-channel EEG data and the second multi-channel EEG data, the cortical source activity time sequence of the cortical target and the subcortical source activity time sequence of the subcortical target are reconstructed through a preset source localization algorithm. Based on the first multi-channel EEG data and the source localization algorithm, a perturbation complexity index for characterizing the functional state of the cortex is calculated. Based on the cortical source activity time series and the subcortical source activity time series, the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value are calculated respectively. An anesthesia depth index is generated based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, using a preset weighted model.

2. The method according to claim 1, characterized in that, The step of reconstructing the cortical source activity timeline of the cortical target point and the subcortical source activity timeline of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data using a preset source localization algorithm includes: The pre-defined individual magnetic resonance imaging data is segmented into tissues, and a three-dimensional mesh head model is constructed based on the segmentation results; Based on the three-dimensional mesh head model and the preset conductivity parameters, the lead field matrix is ​​calculated using the boundary element method; The first multi-channel EEG data and the second multi-channel EEG data are segmented to obtain multiple data segments. The multiple data segments are subjected to a preset bandpass filtering process. Based on the multiple data segments after filtering, the data covariance matrix is ​​calculated. By combining the lead field matrix and the data covariance matrix, a spatial filter is constructed using a linearly constrained minimum variance beamforming algorithm. The spatial filter is applied to the first multi-channel EEG data and the second multi-channel EEG data to obtain the timing sequence of cortical source activity and the timing sequence of subcortical source activity.

3. The method according to claim 2, characterized in that, The step of calculating the perturbation complexity index to characterize the cortical functional state based on the first multi-channel EEG data and the source localization algorithm includes: Extract data segments within a preset time window from the first multi-channel EEG data; By applying the spatial filter to the data segment, a corresponding activity intensity time series is generated for multiple preset cortical source points, and the multiple activity intensity time series are combined to form a source data matrix; By assigning the positions in the source data matrix whose values ​​are greater than a preset statistical threshold to a first preset state, and assigning the positions whose values ​​are not greater than the preset statistical threshold to a second preset state, a binarized spatiotemporal activity matrix is ​​generated based on the first preset state and the second preset state. The binarized spatiotemporal activity matrix is ​​compressed using a preset lossless compression algorithm to obtain compressed data, and the compressed data is then normalized to generate the perturbation complexity index.

4. The method according to claim 1, characterized in that, The step of calculating the cortical-to-subcortical directional coherence value and the subcortical-to-cortical directional coherence value based on the cortical source activity time series and the subcortical source activity time series, respectively, includes: The cortical source activity time series and the subcortical source activity time series are jointly resampled to generate a time-aligned hybrid activity time series; The time series of the mixed activities was fitted using a pre-defined multivariate autoregressive model, and the model coefficient matrix was calculated. The transfer function matrix is ​​obtained by performing a Fourier transform on the model coefficient matrix, and the spectral matrix is ​​obtained based on the transfer function matrix. Extract the first cross-spectral density component representing the time sequence of cortical source activity to the time sequence of subcortical source activity from the spectral matrix, and perform integration and normalization on the first cross-spectral density component within a preset frequency band to generate the directional coherence value from the cortex to the subcortex. The second cross-spectral density component representing the time sequence from the subcortical source activity to the cortical source activity is extracted from the spectral matrix, and the second cross-spectral density component is integrated and normalized within the preset frequency band to generate the subcortical-to-cortical directional coherence value.

5. The method according to claim 1, characterized in that, The process of generating an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value through a preset weighted model includes: Obtain the baseline perturbation complexity index, baseline cortical-to-subcortical directional coherence value, and baseline subcortical-to-cortical directional coherence value corresponding to the preset awake state; Divide the perturbation complexity index by the baseline perturbation complexity index to obtain the complexity ratio; The first directional coherence ratio is obtained by dividing the cortical-to-subcortical directional coherence value by the baseline cortical-to-subcortical directional coherence value, and the second directional coherence ratio is obtained by dividing the subcortical-to-cortical directional coherence value by the baseline subcortical-to-cortical directional coherence value. The anesthesia depth index is generated by performing a preset weighted summation on the complexity ratio, the first directional coherence ratio, and the second directional coherence ratio.

6. The method according to claim 5, characterized in that, The method further includes: According to a preset evaluation cycle, a preset number of time-continuous target anesthesia depth indices are obtained, and the median of the multiple target anesthesia depth indices is calculated. Data quality checks are performed on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results; When the data quality check results meet the preset conditions, the anesthesia depth index is updated using the median.

7. The method according to claim 6, characterized in that, The step of performing data quality checks on the first multi-channel EEG data and the second multi-channel EEG data to generate data quality check results includes: For each channel of the first multi-channel EEG data and the second multi-channel EEG data, it is determined whether the signal amplitude of each channel exceeds a preset amplitude range, and the number of EEG channels that do not exceed the preset amplitude range is counted to calculate the proportion of the number to the total number of channels. Based on the first multi-channel EEG data and the second multi-channel EEG data, a pre-set artifact residue score is calculated using an artifact recognition algorithm. The channel ratio is compared with a first preset threshold, and the artifact residue score is compared with a second preset threshold to generate the data quality check result.

8. A system for assessing the depth of anesthesia based on combined perturbation and electroencephalogram (EEG) signals, characterized in that, The system includes: The combined perturbation application module is used to sequentially apply transcranial magnetic stimulation to preset cortical target points and transcranial ultrasound stimulation to preset subcortical target points according to a preset timing sequence. The EEG signal acquisition module is used to acquire first multi-channel EEG data in response to the transcranial magnetic stimulation and second multi-channel EEG data in response to the transcranial ultrasound stimulation. The brain source reconstruction module is used to reconstruct the cortical source activity time sequence of the cortical target point and the subcortical source activity time sequence of the subcortical target point based on the first multi-channel EEG data and the second multi-channel EEG data, using a preset source localization algorithm. The complexity index calculation module is used to calculate the perturbation complexity index used to characterize the functional state of the cortex based on the first multi-channel EEG data and the source localization algorithm. The directional connectivity analysis module is used to calculate the directional coherence values ​​from the cortex to the subcortex and from the subcortex to the cortex, respectively, based on the cortical source activity time series and the subcortical source activity time series. The anesthesia index generation module is used to generate an anesthesia depth index based on the perturbation complexity index, the cortical-to-subcortical directional coherence value, and the subcortical-to-cortical directional coherence value, through a preset weighted model.

9. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images