An intervention target marker identification system for brain multi-dimensional characteristic signals

Abstract

The application relates to the technical field of biological feature recognition, and discloses an intervention target point marker recognition system for brain multi-dimensional feature signals, which comprises the following modules: a signal sampling module, which is used for converting multi-channel brain original potential signals into high-dimensional feature sequences; a topological manifold mapping module, which is used for projecting the sequences to a physiological reference manifold space to extract feature quantities; an active interactive detection module, which is used for applying sub-threshold physical stimulation and acquiring response signals; a response characteristic judgment module, which is used for determining response correlation strength according to a geometric deviation value; a marker locking module, which is used for outputting a recognition instruction; and a self-calibration module, which is used for updating a center coordinate according to a manifold residual error. The application realizes conversion from correlation to causality, eliminates neural fluctuation interference, guarantees target point recognition specificity, and compensates for feature drift caused by physiological state switching.

Description

Technical Field

[0001] This invention belongs to the field of biometric identification technology, and in particular relates to an intervention target marker identification system for multidimensional feature signals of the brain. Background Technology

[0002] The acquisition and identification of multi-channel brain potential signals is the foundation for implementing precise neurointervention. Current technologies typically employ threshold detection methods based on time-frequency domain amplitude, which determine the target intervention biomarker by monitoring amplitude jumps or energy distribution within a specific frequency band. However, since brain potential signals are physically non-stationary random processes, environmental noise, spontaneous neural activity, and real target biomarkers have overlapping features in their amplitude envelopes. This makes it difficult for conventional amplitude discrimination logic to extract specific recognition features, resulting in a large number of false biomarker trigger commands.

[0003] As monitoring duration increases, the periodic migration of the organism's internal environment or drug metabolism causes a slow shift in the signal baseline, leading to the failure of static recognition criteria in long-term operation. Although the industry has tried to optimize performance by increasing the filter order or building higher-order computational models, the complex computational sequences involved in such paths often introduce processing delays and adversely affect the real-time power consumption of computing hardware. To improve the system's recognition capabilities, the industry has tried multi-dimensional feature fusion strategies. For example, Chinese invention patent CN109222906B discloses a method for constructing a pain state prediction model based on brain electrical signals. This scheme extracts features in the time domain, frequency domain, and wavelet domain to establish an integrated pain state prediction model. Although such schemes increase feature dimensions, their logical essence is still limited to the passive monitoring mode. The recognition process relies on the statistical correlation of historical data and cannot verify the causal attributes of markers and intervention paths at the physical level. Discrete feature extraction methods are difficult to decouple non-stationary signal morphological features from background noise. Faced with global spectral drift caused by physiological steady-state migration, static recognition criteria show poor anti-drift capabilities and cannot guarantee the continuity and specificity of recognition decisions in complex physiological environments.

[0004] Therefore, how to construct an identification architecture that can suppress non-stationary interference, adapt to physiological drift, and have causal discrimination capability is the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a system for identifying intervention target markers of multidimensional feature signals in the brain, comprising: The signal sampling module is used to acquire multi-channel raw brain potential signals and convert each channel signal into a high-dimensional feature time series. The topological manifold mapping module is used to project high-dimensional feature time series onto the physiological baseline manifold space to generate manifold space feature quantities; The active interactive detection module is used to monitor the distribution trajectory of manifold space features in the physiological reference manifold space, and when the distribution trajectory enters the target area in the physiological reference manifold space, it applies a subthreshold physical stimulus with an energy intensity lower than the neuronal action potential burst threshold to the target point to be identified in the brain, and simultaneously acquires the transient response signal induced by the subthreshold physical stimulus. The response characteristic determination module is used to map the transient response signal to the physiological reference manifold space using the topological manifold mapping module to generate the response trajectory. By calculating the geometric deviation value of the response trajectory relative to the reference center point of the physiological reference manifold space, the response correlation strength between the manifold space feature quantity and the target point to be identified is determined. When the geometric deviation value is within the preset deviation threshold range, the response correlation strength is determined to be valid. The marker locking module is used to lock the manifold space features as intervention target markers when the response correlation strength determination is valid, and output the recognition command. The reference drift self-calibration module is used to monitor the manifold residual between the physiological reference manifold space and the background signal, and calculate the reference offset index, which characterizes the migration rate of the physiological background. When the reference offset index is within the preset steady-state offset range, the coordinates of the reference center point are updated based on the asymmetric step size adjustment parameter.

[0006] Preferably, when the reference drift self-calibration module performs coordinate updates based on the asymmetric step size adjustment parameters, it sets the asymmetric step size gain according to the polarity direction of the reference offset index, and sends a manifold reconstruction command to the topology manifold mapping module when the short-time mean of the manifold residual shows an increasing trend and the reference offset index exceeds the preset abrupt change threshold.

[0007] Preferably, the marker locking module performs a stability check before outputting the recognition command. The stability check includes: statistically analyzing the determination results of the response correlation strength in three consecutive sampling periods, and calculating the deviation variance of the geometric deviation value in each sampling period. When the determination results remain valid in three consecutive sampling periods and the deviation variance is lower than the preset stability threshold, the recognition command is confirmed to be valid.

[0008] Preferably, the reference drift self-calibration module calculates the reference offset index according to the following rules. : ,in, This represents the total number of channels in the multichannel brain's primitive potential signal. For the first The real-time coordinate components of the channel's manifold space features in the physiological reference manifold space. These are the preset weighting coefficients for each channel. The sampling time; the reference drift self-calibration module calculates... The first derivative is used to identify the switching between resting and active states in electrophysiological states.

[0009] Preferably, the topological manifold mapping module preserves the local topological correlation of high-dimensional signals by constructing a nearest neighbor weight matrix between each sampling point; the Euclidean distance of the manifold space features in the physiological baseline manifold space is used to characterize the physical coherence between signals from different brain regions.

[0010] Preferably, the subthreshold physical stimulation includes electrical pulse signals or modulated ultrasound beams.

[0011] Preferably, the response characteristic determination module uses the Hausdorff distance algorithm to measure the manifold distance between the response trajectory and the reference center point to obtain the geometric deviation value; wherein, the preset deviation threshold is calculated based on the patient's historical reference data through the maximum likelihood estimation method.

[0012] Preferably, the system also includes a sensing interface monitoring module for analyzing the power spectrum slope of the manifold residual; when the rate of change of the power spectrum slope exceeds a preset threshold, the sensing interface monitoring module generates an aging warning signal of the physical coupling state of the sensing interface.

[0013] Preferably, the signal sampling module has an automatic channel filtering logic, which is used to identify and remove channel signals that are affected by power frequency interference or electromyographic noise before extracting the features of the manifold space. The automatic channel filtering logic makes a judgment based on the mutual information between each channel signal and the physiological reference manifold space.

[0014] Preferably, the manifold spatial feature quantity includes a cyclic phase that reflects the periodic characteristics of the signal; the marker locking module also adjusts the output parameters of the subsequently connected neural intervention device according to the recognition instruction, so that the intervention phase of the neural intervention device is synchronously aligned with the cyclic phase.

[0015] Compared with existing technologies, the brain multidimensional feature signal intervention target marker recognition system of the present invention has the following advantages: 1. In the multidimensional feature signals of the brain, the original potential signals of the brain in multiple channels are processed by the topological space mapping mechanism. The high-dimensional non-stationary signals are converted into topological feature vectors with morphological stability by using the preset projection operator. This eliminates the feature overlap between pathological markers and spontaneous neural background fluctuations in the signal envelope, ensuring that the system still has the ability to accurately lock the intervention target in a strong noise interference environment.

[0016] 2. The sub-threshold excitation-induced response trajectory discrimination logic is adopted. By actively applying weak physical perturbations and analyzing the transient induced response characteristics generated in the brain region, the simple waveform correlation recognition is improved to the causal attribute judgment based on the feedback trajectory. This avoids metastable false triggering caused by changes in patient position or environmental electromagnetic interference, and ensures that the output of intervention instructions has clear pathological targeting.

[0017] 3. By introducing the instantaneous phase angle of the low-frequency phase component to perform a geometric rotation transformation on the target feature manifold, the real-time decoupling of the recognition features and physiological benchmark is achieved by utilizing the fluctuation law of the physiological background. This offsets the global spectral drift caused by sleep stage switching or drug metabolism fluctuations, ensuring the consistency of feature anchoring of the recognition system when running across physiological states. Attached Figure Description

[0018] Figure 1 This is a flowchart of the system logic control and signal processing of the present invention; Figure 2 This is a schematic diagram of the hardware architecture and data interaction link of the system of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0021] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal communication between two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0022] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0023] A system for identifying intervention target markers from multidimensional brain feature signals includes a signal sampling module, a topological manifold mapping module, an active interaction detection module, a response characteristic determination module, a marker locking module, and a reference drift self-calibration module. It achieves synchronous information flow and logical feedback via a data bus. The signal sampling module acquires multi-channel raw potential signals and converts them into high-dimensional feature sequences. The topological manifold mapping module maps the high-dimensional feature sequences to a physiological reference manifold space using a projection operator to extract manifold space features. The active interaction detection module applies subthreshold physical stimulation to the target region within the physiological reference manifold space based on the distribution trajectory. The response characteristic determination module calculates the geometric deviation of the response trajectory relative to the reference center point to determine the response correlation strength. The marker locking module outputs a recognition command after meeting stability verification conditions. The reference drift self-calibration module monitors the manifold residual and updates the coordinates of the reference center point based on the reference offset index. Addressing the non-stationary random characteristics of brain potential signals and the feature overlap between environmental noise and target markers in the amplitude envelope, the signal sampling module acquires... The system collects raw brain potential signals from multiple channels. The signal sampling module has an automatic channel filtering logic, which is used to identify and remove channel signals that are affected by power frequency interference or electromyographic noise before extracting manifold space features. The automatic filtering logic judges based on the mutual information between each channel signal and the physiological reference manifold space. When the mutual information of a certain channel is lower than the preset correlation threshold, the channel signal is determined to be a source of disturbance. The system converts the filtered raw potential signals into a high-dimensional feature time series according to a preset sampling rate. The preset sampling rate is set to 1024Hz to ensure that the high-dimensional feature series can completely represent the changes in the brain's electrophysiological state.

[0024] Signal sampling module acquisition Converting raw potential signals from a single channel into high-dimensional feature time series The processor is based on high-dimensional feature time series Constructing the nearest neighbor weight matrix eigenvalue decomposition to extract the projection operator Calculate high-dimensional feature time series The Euclidean distance between the reconstructed signal and the manifold residual determines the manifold residual. manifold residuals The calculation formula is as follows: ;In the formula, Represents the manifold residual. Represents a high-dimensional feature vector. Represents the projection operator. Represents the projection operator Transpose matrix; manifold residual If the root mean square value remains within the preset convergence threshold range and the variance of fluctuation is less than 2% during a continuous 60-second observation period, it is determined that the physiological reference manifold space has completed steady-state calibration and locking of the projection operator. Since time-frequency domain amplitude detection methods are difficult to decouple signal morphology from background noise, the topological manifold mapping module preserves the local topological correlation of high-dimensional signals by constructing a nearest neighbor weight matrix between each sampling point. The topological manifold mapping module projects the high-dimensional feature time series onto the physiological reference manifold space to generate manifold space feature quantities containing cyclic phases. These cyclic phases reflect the periodicity of the signal. The Euclidean distance of the manifold space feature quantities in the physiological reference manifold space is used to characterize the physical coherence between signals from different brain regions. The topological manifold mapping module uses a linear dimensionality reduction operator to perform weighted projection on the multidimensional signal feature sequence. This mapping process reduces the original dimension to three-dimensional space, thereby forming an observable topological feature vector in the physiological reference manifold space.

[0025] Since passive monitoring logic cannot verify the causal attributes of biomarkers and intervention paths online, the active interactive detection module monitors the distribution trajectory of manifold spatial features in the physiological baseline manifold space. When the distribution trajectory enters the target region within the physiological baseline manifold space, the active interactive detection module applies a subthreshold physical stimulus with an energy intensity lower than the neuronal action potential burst threshold to the target point in the brain and simultaneously acquires the transient response signal induced by the stimulus. This subthreshold physical stimulus uses a 50Hz electrical pulse signal with an intensity set to 70% to 85% of the neuronal firing threshold, or a modulated ultrasound beam. By applying a weak physical perturbation, the system will... Correlation identification is transformed into causal attribute determination based on feedback trajectory to avoid false triggering caused by changes in body position. The response characteristic determination module uses the topological manifold mapping module to map the transient response signal to the physiological reference manifold space to generate the response trajectory. The response characteristic determination module uses the Hausdorff distance algorithm to measure the manifold distance between the response trajectory and the reference center point to obtain the geometric deviation value. The preset deviation threshold is calculated based on the patient's historical reference data through the maximum likelihood estimation method. When the geometric deviation value is within the preset deviation threshold range, the response correlation strength is determined to be valid. If the geometric deviation value exceeds the range, the current signal is determined to be transient interference and the deviation determination logic is reset.

[0026] When performing Hausdorff distance calculation, the response characteristic determination module expands the reference center point of the physiological reference manifold space into a reference point set consisting of 50 historical steady-state sampling points. The algorithm calculates the Euclidean distance from each discrete sampling point in the response trajectory to the nearest point in the reference point set, and extracts the largest value from this series of distance values ​​as the geometric deviation value, which is used to quantify the farthest physical boundary of the response trajectory deviating from the steady-state region. If the value of the farthest boundary is continuously lower than the preset threshold of ten coordinate units, it is confirmed as a valid association. Before outputting the recognition command, the marker locking module performs stability verification, including statistically analyzing the determination results of the response association strength in three consecutive sampling periods and calculating the deviation variance of the geometric deviation value in each sampling period. When the determination results remain valid in three consecutive sampling periods and the deviation variance is lower than the preset stability threshold, the recognition command is confirmed to be valid. The preset stability threshold is set to 0.05. After confirming validity, the marker locking module locks the manifold space feature quantity as the intervention target marker and outputs the recognition command, which is used to adjust the output parameters of the subsequently connected neural intervention device so that the intervention phase of the neural intervention device is synchronously aligned with the cyclic phase.

[0027] The reference drift self-calibration module is used to compensate for feature drift caused by physiological state switching. It monitors the manifold residual between the physiological reference manifold space and the background signal, and calculates the reference offset index, which characterizes the migration rate of the physiological background, according to the following rules. : ,in: The baseline offset index; This represents the total number of channels for multichannel primitive potential signals in the brain; For the first Preset weighting coefficients for the corresponding channels; For the first Real-time coordinate components of the channel's manifold space features in the physiological reference manifold space; The sampling time; the reference drift self-calibration module calculates... The first derivative is used to identify the switching between resting and active states in electrophysiological states. When the reference offset index is within the preset steady-state offset range, the coordinates of the reference center point are updated based on the asymmetric step size adjustment parameter. When the short-time mean of the manifold residual shows an increasing trend and the reference offset index exceeds the preset abrupt change threshold, a manifold reconstruction command is sent to the topology manifold mapping module. During the asymmetric update of the reference center point coordinates, the reference drift self-calibration module monitors the reference offset index. Polarity direction adjustment coordinate update step size ; Baseline offset index When the value is positive and the brain's electrophysiological state shifts towards a more active state, a step size scaling factor of 0.01 is selected to slow down the drift rate of the reference center point coordinates; reference offset index When the value is negative and the regression is towards the resting state, a step size scaling factor of 0.05 is selected to accelerate the regression speed of the reference center point towards the steady-state region; coordinate update step size By reference offset index The step size is determined by multiplying the step size by a step size scaling factor, and the processor updates the step size of the coordinates. The coordinates of the reference center point in the physiological reference manifold space are compensated to complete the nonlinear tracking and dynamic correction of the physiological background migration.

[0028] The sensing interface monitoring module analyzes the power spectrum slope of the manifold residual. When the rate of change of the power spectrum slope exceeds a preset threshold, it generates an aging warning signal for the physical coupling state of the sensing interface, which characterizes changes in polarization layer thickness or tissue proliferation in the electrode system. When the physical coupling state deviates from a preset stable range, the system blocks the identification command output by the marker locking module. Dynamic sensing of the physical interface state is achieved through real-time analysis of the projected residual components. During the warning process of the sensing interface monitoring module, the processor extracts the manifold residual. The power spectrum slope is calculated for power spectrum components in the high-frequency band above 30Hz; when the rate of change of the power spectrum slope exceeds 20% within a 10-second sampling window, it is determined that the physical coupling state between the sensing electrode and brain tissue has aged or failed polarization; based on the determination result, a physical coupling aging warning signal of the sensing interface is generated, and the logic circuit is driven to block the identification command output of the marker locking module until the physical impedance of the sensor returns to the preset normal operating range, so as to avoid the risk of false triggering of markers due to sensor performance degradation.

[0029] Example 1: When the system faces the condition of electrode micro-displacement caused by the patient's daily activities, and the background intensity of spontaneous neural activity in the brain shifting with the sleep cycle, the signal sampling module acquires... The system collects raw brain potential signals from multiple channels and converts these signals into high-dimensional feature sequences at a sampling rate of 1024 Hz. Because the amplitude envelopes of pathological markers and spontaneous neural background fluctuations physically overlap in the time domain, energy threshold-based recognition logic cannot distinguish between target features and random oscillations. A topological manifold mapping module projects the high-dimensional feature sequences onto a physiological baseline manifold space. By extracting manifold space features, the system spatially decouples the intertwined features on the manifold geometry. The system then transforms the recognition task into a process of analyzing the distribution trajectory of these manifold space features within the physiological baseline manifold space. The stability assessment of the manifold geometric distribution involves the active interactive detection module applying a subthreshold physical stimulus with an energy intensity lower than the neuronal action potential burst threshold to the target point in the brain when the distribution trajectory enters the target region within the physiological reference manifold space. An electrical pulse signal with a frequency of 50 Hz is selected. The response characteristic determination module calculates the geometric deviation value of the response trajectory induced by the physical stimulus relative to the reference center point. The trigger point of the active interactive detection module is determined using the spatial positioning reference provided by the topological manifold mapping module, and the correlation between the manifold spatial features and the target point is verified based on the geometric deviation value.

[0030] The marker locking module outputs a recognition command when the judgment result is valid for three consecutive sampling periods and the variance of the deviation is less than 0.05. This command drives the subsequently connected neural intervention device to synchronize the intervention phase with the cyclic phase in the manifold space feature quantity. The reference drift self-calibration module simultaneously calculates the reference offset index. Baseline offset index The calculation formula is as follows: ,in: The baseline offset index; Total number of channels; For the first Preset weighting coefficients for the corresponding channels; For the first Real-time coordinate components of the channel's manifold space features in the physiological reference manifold space; The sampling time is used to detect the physical coupling state of the sensing interface by monitoring the power spectrum slope of the manifold residual. When the rate of change of the power spectrum slope exceeds 20% within 10 seconds, an aging warning signal is triggered and the identification command output is blocked. This process uses the topological characteristics of the data stream to compensate for physical interface failure and physiological benchmark drift.

[0031] Example 2: To verify the stability of the brain multidimensional feature signal intervention target marker recognition system under physiological background migration and electrode displacement conditions, the experimental platform adopted a multi-channel digital signal acquisition unit with a resolution of 24 bits. Its input end was connected to the potential signals of simulated pathological features and neural fluctuations. The experimental data came from the standard neurophysiological recording library and was converted into physical voltage waveforms by the signal generator. The signal acquisition frequency was set to 1024Hz. The mutual information threshold was set to balance the sensitivity of singular channel elimination and the retention of useful physiological features. According to the physical law that the signal bandwidth covers the main frequency oscillation range, the sampling rate was set to 1024Hz to correspond to the Nyquist sampling theorem requirement. The mutual information threshold was selected as 0.2 to cope with the typical working condition where the amplitude disturbance accounts for 20%. The sample group of this invention implemented the active interactive detection and response characteristic judgment procedure. The control group 1 adopted the time-domain energy threshold discrimination method, and the control group 2 adopted the passive manifold monitoring method. The recognition specificity was tested in a mixed signal field containing 50Hz power frequency interference and variable power Gaussian white noise.

[0032] Table 1: Comparison of recognition specificity data for each sample group under different signal-to-noise ratio gradients

[0033] Observing the data recorded in Table 1, when the signal-to-noise ratio (SNR) decreased from 20dB to -5dB, the recognition specificity of control group 1 deteriorated from 88.5% to 5.2%, confirming the physical limitations of the time-domain energy discrimination logic in environments with strong background overlap. In this invention's sample group, when the SNR decreased to 0dB, its recognition specificity remained at 91.5% and the variance of the geometric deviation was 0.046. When the SNR dropped to -5dB, the indicators showed a non-linear decline and the variance exceeded the safety boundary of 0.05. At this point, the marker locking module confirmed insufficient stability and executed logic locking. The performance inflection point data in Table 1 confirmed that the system operates at extremely low... The reliability of the automatic triggering protection mechanism under signal-to-noise ratio conditions was demonstrated. Experimental results confirmed that the local stability of manifold spatial characteristic quantities provided a geometric positioning benchmark for active detection. The subthreshold physical stimulation applied by the active interactive detection module eliminated non-pathological topological deviations by inducing transient response signals. The synergistic effect of the two improved the certainty of pathological marker extraction. The system sensed the coupling state of the sensing interface by analyzing the slope of the projection residual power spectrum generated by the manifold mapping process. When the rate of change of the power spectrum slope exceeded 20% within 10s, an aging warning signal was triggered, realizing online identification of physical interface failure without introducing additional electrical impedance measurement hardware.

[0034] Example 3: This example combines Figures 1 to 2 This describes a system for identifying intervention target biomarkers based on multidimensional brain feature signals, such as... Figure 1As shown, the system mainly consists of six logic modules. These modules work collaboratively through data flow and control signals. The signal sampling module acquires the raw potential signal and converts it into a high-dimensional feature time series, which is then transmitted to the topological manifold mapping module. The topological manifold mapping module projects the sequence onto the physiological baseline manifold space to generate manifold space feature quantities. Simultaneously, it outputs these manifold space feature quantities to the response characteristic determination module and allows the active interaction detection module to monitor the distribution trajectory of the manifold space feature quantities. Based on the monitored distribution trajectory, the active interaction detection module applies subthreshold physical stimulation, acquires the transient response signal induced by the stimulation, and transmits it to the response module. The characteristic determination module utilizes the mapping relationship provided by the topological manifold mapping module to map the transient response signal to generate a response trajectory. By calculating the geometric deviation value of the response trajectory relative to the reference center point, the response correlation strength is determined, and the determination result is transmitted to the marker locking module. After confirming that the response correlation strength is valid, the marker locking module locks the intervention target marker and outputs the recognition command. At the same time, the reference drift self-calibration module continuously monitors the manifold residual generated by the topological manifold mapping, calculates the reference offset index, and performs reference center point coordinate update accordingly. The updated reference center point is fed back to the response characteristic determination module to correct the determination reference.

[0035] like Figure 2 As shown, the system architecture comprises four main parts: front-end acquisition and control equipment, core computing workstation, physical sensing and interaction environment, and external medical linkage equipment. The physical sensing and interaction environment, located at the bottom layer, includes a multi-channel sensing electrode array for acquiring raw potential signals, a subthreshold physical stimulation probe for applying electrical pulses or ultrasound beams, and a sensing interface coupling medium serving as the monitored object. This medium transmits multi-channel analog signals to the front-end acquisition and control equipment and receives subthreshold physical stimulation from the front end. The front-end acquisition and control equipment integrates a high-precision analog-to-digital converter as the signal sampling module hardware, a programmable stimulation generator as the active interaction detection hardware, and an anti-interference channel screening circuit. This device utilizes high-speed data... The bus transmits digitized feature sequences to the core computing workstation and receives detection trigger commands from the workstation. The core computing workstation, as the processing hub of the system, internally runs a topology manifold mapping engine to perform high-dimensional projection and space construction, response characteristics and causal judgment logic to perform geometric deviation analysis, marker locking and command system to perform stability verification and command output, benchmark drift self-calibration service to be responsible for manifold residual monitoring and coordinate updates, sensor interface status monitoring module to provide physical coupling aging warning, and stores a patient historical benchmark database. After processing, the core computing workstation outputs recognition commands and cyclic phase synchronization parameters to external medical linkage devices, driving the included neural intervention devices to perform phase synchronization alignment operations.

[0036] Example 4: When the system is used for intervention in essential tremor and faces the situation where there are differences in the electrophysiological baseline manifold morphology among different patients, the system executes an initialization calibration procedure to determine the characteristic parameters of the physiological baseline manifold space; the signal sampling module acquires 60 seconds of multi-channel raw potential signals in a non-interventional state and determines them as the baseline observation sequence; the topology manifold mapping module constructs a nearest neighbor graph by calculating the Euclidean distance between any two sampling points in the baseline observation sequence, and based on each sampling point... Construct a local weight matrix using nearest neighbor points; select Projection operator The construction procedure involves performing eigenvalue decomposition on the local weight matrix and selecting the eigenvectors corresponding to the three smallest non-zero eigenvalues ​​to form the projection operator. It is used to linearly transform high-dimensional feature sequences generated in real time to a three-dimensional physiological reference manifold space, realizing the transformation from the original features to stable manifold coordinates.

[0037] In the specific construction of the projection operator, the topological manifold mapping module searches for the 12 nearest neighbors of each sampling point in the original high-dimensional space. By solving the equation for minimizing local reconstruction error, it calculates the reconstruction contribution ratio of these 12 nearest neighbors to the center point and ensures that the sum of the contribution ratios is always equal to one. The system performs eigenvalue decomposition on the symmetric matrix formed by the reconstruction contribution ratios, discards trivial solutions with zero eigenvalues, and selects the three non-zero eigenvectors with the smallest eigenvalues ​​as the projection reference. The original sampling voltage values ​​of the multi-channel are weighted and summed with these three eigenvectors to directly output the three-dimensional coordinate values ​​in the physiological reference manifold space. During the calibration phase, the system monitors the distribution density of the manifold space features during the 60-second observation period to calibrate the boundary of the target region in the physiological reference manifold space. The system enters a 60-second pre-observation phase, statistically analyzing the signal energy distribution within each 100-millisecond window, and uses the three channel mapping points with the highest energy as the initial target geometry. The initial geometric center is used as the center, and the manifold coordinate spacing corresponding to a 50 mV amplitude is used as the initial radius to define a temporary search area. The active interactive detection module applies test stimuli with a step increment of 0.5 mA within this area, monitoring the convergence direction of the response trajectory. If the response trajectory continues to move closer to the center, the radius remains unchanged; if the response trajectory shows a divergent trend, the initial radius is gradually reduced by 5% until the trajectory deviation variance stabilizes within 0.03 for five consecutive detection cycles. This completes the dynamic calibration from the initial search area to the precise pathological target area, resolving the logical conflict of the unpredictable target area before marker identification. The active interactive detection module performs kernel density estimation on the sampling points within the manifold space to identify the center of the characteristic point cluster corresponding to the pathological tremor frequency band. This center point is determined as the target geometric center, and the minimum convex hull region surrounding this center point and containing 95% of the sampling point weights is determined as the target area. The radius of the target area... Based on the standard deviation of the sampling point distribution Confirmed, the calculation formula is as follows: When the real-time distribution trajectory enters this radius Within a defined range, the system determines that the physical stimulus triggering condition is met. This discrimination logic based on distribution density makes the delineation of the target area controlled by the current electrophysiological state distribution.

[0038] For individualized calibration of subthreshold physical stimulus intensity, the active interactive detection module applies a 50Hz test electrical pulse starting from zero intensity, increasing in increments of 0.5mA, while simultaneously monitoring the geometric deviation value output by the response characteristic judgment module. When the first derivative reaches an inflection point and the system does not detect any physical twitching of the limb, the system simultaneously analyzes weak voltage signals with frequencies between 100Hz and 200Hz through the signal sampling module. If the peak-to-peak value of this signal exceeds 30 microvolts within 10 milliseconds, it is determined that a minor twitching of the limb has occurred. Without detecting such high-frequency twitching signals, the system continuously records the rate of decrease in the geometric deviation value. For every 0.5mA increase in stimulation current and the decrease in the geometric deviation value... When the slope decreases to within 20% of the initial slope, the system automatically stops the current step increase and locks 80% of the current intensity as the final subthreshold physical stimulation parameter. The high-frequency artifact monitoring of the EEG channel replaces the external EMG sensor, realizing closed-loop calibration under the existing hardware architecture. The system determines 80% of this critical intensity as the subthreshold physical stimulation intensity in the running state. In the running example of this embodiment, the calibrated stimulation intensity is 2.4mA. After the system enters the running state, the marker locking module performs active interactive detection based on the calibrated intensity. By executing a closed-loop procedure that includes projection matrix construction, geometric boundary delineation, and stimulation intensity calibration, the system achieves anchoring of specific brain feature signals.

[0039] Example 5: Under long-term operation and when faced with nonlinear shifts in the signal reference due to changes in the patient's physiological rhythm, the reference drift self-calibration module calculates the migration potential by monitoring the fluctuation rate of the manifold residual in the time dimension. Migration potential energy The absolute value of the second-order rate of change of the manifold residual with respect to the sampling time within a sliding window of length 512 sampling points is used by the system based on the migration potential. Dynamically adjust coordinate update step size Specifically, coordinate update step size The calculation formula is as follows: ,in: Update the step size for the coordinates of the reference center point; This is the step size scaling factor, with a value of 0.02. ; The migration potential energy is used as the basis for the reference drift self-calibration module. The system adjusts the following speed of the center coordinate in real time. When the migration potential energy exceeds the preset mutation threshold, the system determines that the physiological background has entered the state switching interval and starts the coordinate update process.

[0040] To ensure the stability of the benchmark offset index calculation, the benchmark drift self-calibration module first inputs the manifold residual sequence into a moving average filter with a length of 128 sampling points before calculation to filter out signal spikes caused by random electromagnetic interference. The system calculates the difference between the residual at the current sampling time and the previous time as the first-order rate of change, and then performs a difference operation on two adjacent first-order rates of change and takes the absolute value as the migration potential. If it exceeds 0.2 coordinate units per square second for 300 consecutive milliseconds, the system confirms that a trend migration of physiological state has occurred rather than random fluctuation, thereby triggering an asymmetric update of the benchmark center point coordinates. When the system monitors the impedance fluctuations caused by changes in tissue fluid composition in the signal generated by the sensing interface, the sensing interface monitoring module analyzes the projection residual distribution generated by the manifold mapping process and calculates the reconstruction gain of the manifold projection. Reconstruction gain Used to characterize the energy conservation before and after the projection process, reconstruction gain The calculation formula is as follows: ,in, For reconstruction gain; The total energy of the feature components after projecting the high-dimensional feature sequence onto the physiological baseline manifold space; The total energy of the high-dimensional feature sequence is input before projection; the benchmark drift self-calibration module reconstructs the gain. The coordinate update is confirmed to be effective when the value is within the range of 0.95 to 1.05, and the reconstruction gain is... If the values ​​deviate from the above range, it is determined that the current coordinate update has caused topological distortion and triggered a logic latch. This is achieved by monitoring the reconstruction gain. The changing trajectory enables feature alignment between the recognition system and the target marker.

[0041] Example 6: When the system is deployed on-site and faces the situation where the spectral distribution characteristics of brain potential signals differ among different patients, the system executes a sliding sampling window. With frequency resolution The joint calibration procedure involves the signal sampling module acquiring a 10-second resting-state raw potential signal and calculating the power spectral density curves under different window values, with a frequency resolution of [missing information]. The calculation formula is as follows: ,in: Frequency resolution; The sampling rate is fixed at 1024Hz; The number of sample points contained in the sliding sampling window; the system compares different The parameter for determining the energy concentration of the pathological tremor frequency band under the given value is when the sliding sampling window... When the value is selected as 512 and Hanning window weighting is used, the clustering of the resulting manifold space features in the physiological reference manifold space reaches its maximum value. This calibration procedure provides a preset discretized input length for the topological manifold mapping module.

[0042] When the system experiences signal quality mismatch due to inconsistent physical impedance of the sensing electrodes, the reference drift self-calibration module performs multi-channel weighting coefficient adjustments. The system employs an adaptive initialization procedure to calculate the signal-to-noise ratio of each channel during the calibration observation period. And assign weights according to the signal-to-noise ratio distribution ratio, the first Channel weighting coefficient The calculation formula is as follows: ,in: For the first Channel weighting coefficients; For the first The signal-to-noise ratio of the channel; This represents the total number of channels. In this embodiment, the signal-to-noise ratio (SNR) is determined based on the ratio of the signal's main frequency energy to the high-frequency residual energy. The system uses the weighting coefficients obtained from calibration. Adjusting the reference offset index To assess sensitivity, the system calculates the signal-to-noise ratio (SNR) for each channel. It defines the frequency band from 0.5 Hz to 40 Hz as the signal passband and the remaining bands as noise bands. The SNR value for each channel is obtained by calculating the ratio of the average voltage in the passband to the average voltage in the noise band. If the SNR of a channel is below 1.5 dB, the system automatically sets its corresponding weighting coefficient to zero to shield the interfering channel. If the SNR is above 10 dB, its weighting coefficient is increased to 1.2. This weighting mechanism based on the SNR gradient ensures that the calculation result of the reference offset index is controlled only by high-reliability channels with excellent signal quality, and that the coordinate update direction is controlled by specific channels with signal quality better than a preset threshold. The differences in SNR between channels are used to physically shield the risk of interface degradation.

[0043] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A system for identifying intervention target markers based on multidimensional feature signals of the brain, characterized in that, include: The signal sampling module is used to acquire multi-channel raw brain potential signals and convert each channel signal into a high-dimensional feature time series. The topological manifold mapping module is used to project high-dimensional feature time series onto the physiological baseline manifold space to generate manifold space feature quantities; The active interactive detection module is used to monitor the distribution trajectory of manifold space features in the physiological reference manifold space, and when the distribution trajectory enters the target area in the physiological reference manifold space, it applies a subthreshold physical stimulus with an energy intensity lower than the neuronal action potential burst threshold to the target point to be identified in the brain, and simultaneously acquires the transient response signal induced by the subthreshold physical stimulus. The response characteristic determination module is used to map the transient response signal to the physiological reference manifold space using the topological manifold mapping module to generate the response trajectory. By calculating the geometric deviation value of the response trajectory relative to the reference center point of the physiological reference manifold space, the response correlation strength between the manifold space feature quantity and the target point to be identified is determined. When the geometric deviation value is within the preset deviation threshold range, the response correlation strength is determined to be valid. The marker locking module is used to lock the manifold space features as intervention target markers when the response correlation strength determination is valid, and output the recognition command. The reference drift self-calibration module is used to monitor the manifold residual between the physiological reference manifold space and the background signal, and calculate the reference offset index, which characterizes the migration rate of the physiological background. When the reference offset index is within the preset steady-state offset range, the coordinates of the reference center point are updated based on the asymmetric step size adjustment parameter.

2. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, When the reference drift self-calibration module performs coordinate updates based on the asymmetric step size adjustment parameters, it sets the asymmetric step size gain according to the polarity direction of the reference offset index. When the short-time mean of the manifold residual shows an increasing trend and the reference offset index exceeds the preset abrupt change threshold, it sends a manifold reconstruction command to the topology manifold mapping module.

3. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, Before outputting the recognition command, the marker locking module performs a stability check. The stability check includes: statistically analyzing the judgment results of the response correlation strength in three consecutive sampling periods, and calculating the variance of the geometric deviation value in each sampling period. When the judgment results remain valid in three consecutive sampling periods and the variance of the deviation is lower than the preset stability threshold, the recognition command is confirmed to be valid.

4. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The reference drift self-calibration module calculates the reference offset index according to the following rules. : ,in, This represents the total number of channels in the multichannel brain's primitive potential signal. For the first The real-time coordinate components of the channel's manifold space characteristics in the physiological reference manifold space. These are the preset weighting coefficients for each channel. The sampling time; the reference drift self-calibration module calculates... The first derivative is used to identify the switching between resting and active states in electrophysiological states.

5. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The topological manifold mapping module preserves the local topological correlation of high-dimensional signals by constructing a nearest neighbor weight matrix between each sampling point; the Euclidean distance of manifold space features in the physiological baseline manifold space is used to characterize the physical coherence between signals from different brain regions.

6. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, Subthreshold physical stimuli include electrical pulse signals or modulated ultrasound beams.

7. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The response characteristic determination module uses the Hausdorff distance algorithm to measure the manifold distance between the response trajectory and the baseline center point to obtain the geometric deviation value; wherein, the preset deviation threshold is calculated based on the patient's historical baseline data through the maximum likelihood estimation method.

8. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The system also includes a sensor interface monitoring module, which is used to analyze the power spectrum slope of the manifold residual; when the rate of change of the power spectrum slope exceeds a preset threshold, the sensor interface monitoring module generates an aging warning signal of the physical coupling state of the sensor interface.

9. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The signal sampling module has an automatic channel filtering logic, which is used to identify and remove channel signals that are affected by power frequency interference or electromyographic noise before extracting manifold space features. The automatic channel filtering logic makes a judgment based on the mutual information between each channel signal and the physiological reference manifold space.

10. The brain multidimensional feature signal intervention target marker identification system according to claim 1, characterized in that, The manifold space features include a cyclic phase that reflects the periodic characteristics of the signal; the marker locking module also adjusts the output parameters of the subsequently connected neural intervention device according to the recognition instructions, so that the intervention phase of the neural intervention device is synchronously aligned with the cyclic phase.

Images