Intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system

Abstract

This invention discloses a system for synchronous acquisition and noise reduction of intracranial multi-channel neural signals, belonging to the field of neural signal processing technology. It synchronously acquires neural electrical signals and interface impedance spectrum data through a high-density microelectrode array; constructs a dynamic impedance distribution map and extracts the spatiotemporal propagation characteristics of physiological artifacts using spatial coordinates; projects the signals into the artifact and neural signal subspaces, dynamically determining the common-mode noise contribution weight of each channel at the current moment based on the energy distribution ratio and impedance change rate; and constructs a dynamically weighted common average reference layer, adaptively subtracting the reference layer signal to achieve common-mode suppression. This invention effectively eliminates physiological artifacts and crosstalk in free-movement scenarios, greatly improving the acquisition quality of high-fidelity neural signals from subjects such as the human brain.

Description

Technical Field

[0001] This invention relates to the field of neural signal processing technology, specifically to a system for synchronous acquisition and noise reduction of intracranial multi-channel neural signals. Background Technology

[0002] Brain-computer interfaces and high-density neurophysiological acquisition technology play a crucial role in basic neuroscience research and the diagnosis and treatment of major clinical brain diseases. In these applications, it is usually necessary to implant high-density microelectrode arrays into specific brain regions of human brains or experimental animals to obtain neural signals with extremely high spatiotemporal resolution, such as local field potentials and single-cell action potentials.

[0003] However, in real-world cognitive behavioral experiments or long-term clinical monitoring, subjects inevitably engage in free activities such as eating, walking, turning their heads, and chewing. In these dynamic scenarios, the acquisition of intracranial neural electrical signals faces extremely severe challenges. On the one hand, due to the local flow of cerebrospinal fluid, the pulsation of microvessels, and the microscopic displacement of brain tissue caused by muscle traction, the electrical characteristics of the interface between the microelectrode and the brain tissue experience dramatic dynamic fluctuations. On the other hand, macroscopic physiological activities such as heartbeat, respiration, and chewing electromyography generate low-frequency common-mode artifacts with amplitudes far exceeding the background neural signal. These multi-source physiological artifacts exhibit complex spatiotemporal propagation characteristics within the heterogeneous brain tissue, resulting in extremely uneven distribution among different electrode channels.

[0004] To address the aforementioned common-mode noise, traditional denoising methods primarily rely on fixed reference electrodes in hardware or conventional common average reference algorithms in software. However, existing technologies have the following inherent drawbacks:

[0005] First, the impedance state of the fixed reference electrode is static, which cannot track and reflect the dynamic changes in the micro-interface impedance and physiological artifacts at the recording electrode array in time and space. This causes the traditional hardware common-mode suppression mechanism to fail directly when the subject moves.

[0006] Secondly, traditional CAR algorithms perform indiscriminate arithmetic averaging of signals from all electrode channels during software processing. When a few electrode channels generate explosive artifacts with large values ​​due to local tissue deformation or violent movement, simple averaging not only fails to eliminate noise, but also forcibly spreads these extremely strong local noise components into the originally uncontaminated clean channels, causing serious "secondary contamination".

[0007] Finally, real neural electrical signals and multi-source physiological artifacts severely overlap in the frequency domain. Traditional bandpass filtering or independent component analysis often removes or distorts valuable neurophysiological indicators while filtering out noise. During high-density acquisition, parasitic capacitance crosstalk between electrode channels and complex background noise are superimposed, making effective separation difficult. Summary of the Invention

[0008] The purpose of this invention is to provide a system for synchronous acquisition and noise reduction of intracranial multi-channel neural signals to solve the above-mentioned problems.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system, comprising:

[0010] The data acquisition module is used to acquire the raw neural electrical signals of each channel of the human brain or experimental animals and the corresponding electrode-tissue interface impedance spectrum data through a high-density microelectrode array implanted in the brain in a multi-channel synchronous sampling manner.

[0011] The data processing module is used to construct a dynamic impedance distribution map of the electrode-tissue interface based on the interface impedance spectrum data acquired in real time from each channel, and to extract the spatiotemporal propagation characteristics of physiological artifacts by weighted principal component analysis in combination with the spatial coordinates of each channel signal.

[0012] The data determination module is used to project the original neural electrical signals of each channel to the artifact subspace and the neural signal subspace, and dynamically determine the common-mode noise contribution weight of each channel at the current moment based on the energy distribution ratio and impedance change rate of the signal in the artifact subspace and the neural signal subspace.

[0013] The result generation module is used to generate time-varying weighted coefficients based on the judgment result, construct a dynamically weighted common average reference layer (DWCAR), and adaptively subtract the dynamic reference layer signal from each channel signal to achieve common mode suppression, while retaining the neural signal subspace components to obtain the denoised target neural electrical signal.

[0014] Furthermore, the data acquisition module is specifically used for:

[0015] A weak test current of known frequency is injected into the high-density microelectrode array, and a mixed electrical signal containing neural activity information of the test subject, such as the human brain, and the response of the weak test current is acquired simultaneously.

[0016] The test current response is separated from the mixed electrical signal using a bandpass filter and lock-in amplification to calculate the electrode-tissue interface impedance spectrum data, while the original neural electrical signal is extracted.

[0017] Furthermore, the specific process of constructing the dynamic impedance distribution spectrum of the electrode-tissue interface in the data processing module includes:

[0018] Acquire the impedance amplitude and phase data of each channel within a continuous sliding time window;

[0019] Combined with the pre-calibrated high-density microelectrode array implanted in the brain and other subjects' cranial cavity, the three-dimensional spatial coordinates are determined.

[0020] Using a spatial interpolation algorithm, a three-dimensional dynamic distribution map representing the impedance change of each channel electrode-tissue interface at the current moment under free-moving conditions is generated.

[0021] Furthermore, in the data processing module, the specific process of extracting the spatiotemporal propagation characteristics of physiological artifacts using the weighted principal component analysis method includes:

[0022] The original neural electrical signals of each channel are spatially weighted using the electrode-tissue interface impedance spectrum data of each channel as weighting factors.

[0023] A spatial covariance matrix is ​​constructed from the weighted multi-channel signal and eigenvalue decomposition is performed to extract the large-amplitude low-frequency components corresponding to physiological artifacts such as heartbeat, respiration or electromyography, thereby obtaining the principal component eigenvectors to characterize the spatiotemporal propagation characteristics of physiological artifacts.

[0024] Furthermore, the step of projecting the signal onto the subspace in the data determination module specifically includes:

[0025] Based on the spatiotemporal propagation characteristics, a pseudo-trace projection matrix is ​​constructed. The original neural electrical signals of each channel are multiplied by the pseudo-trace projection matrix to obtain the pseudo-trace components projected onto the pseudo-trace subspace.

[0026] By projecting the original neural electrical signals of each channel into the neural signal subspace through orthogonal complementary space, neural activity components with major physiological artifacts filtered out are obtained.

[0027] Furthermore, in the data determination module, the calculation logic for dynamically determining the common-mode noise contribution weight includes:

[0028] The energy distribution ratio is obtained by calculating the ratio of the energy of the artifact component to the energy of the neural activity component in each single channel at the same analysis time.

[0029] Calculate the rate of change of interface impedance of the single channel at the current moment compared to the previous time window;

[0030] The energy distribution ratio and the impedance change rate are fused together to calculate and output the common-mode noise contribution weight of each channel at the current moment; wherein, the channel with the larger energy distribution ratio and the higher impedance change rate has a higher common-mode noise contribution weight.

[0031] Furthermore, the specific process of constructing the dynamically weighted common average reference layer DWCAR in the result generation module includes:

[0032] The common-mode noise contribution weights of all channels are globally normalized to calculate the time-varying weighting coefficients of each channel at the current time.

[0033] The signal amplitude of each channel at the current moment is multiplied by its corresponding time-varying weighting coefficient and then summed to generate the dynamic weighted common average reference layer signal that characterizes the global common-mode noise background.

[0034] Furthermore, the specific process of constructing the dynamically weighted common average reference layer DWCAR in the result generation module includes:

[0035] The common-mode noise contribution weights of all channels are globally normalized to calculate the time-varying weighting coefficients of each channel at the current time.

[0036] The signal amplitude of each channel at the current moment is multiplied by its corresponding time-varying weighting coefficient and then summed to generate the dynamic weighted common average reference layer signal that characterizes the global common-mode noise background.

[0037] The system also includes a channel crosstalk correction module, used for:

[0038] Based on the local interface abrupt change state reflected by the dynamic impedance distribution spectrum, when subtracting the dynamic reference layer signal from each channel signal, the phase and amplitude compensation factors of the subtraction operation are dynamically adjusted.

[0039] To separate inter-channel crosstalk and background noise during high-density acquisition, reduce the crosstalk ratio in the signal coherence spectrum, and preserve the true high-frequency local field potential and action potential characteristics of the human brain.

[0040] Compared with existing technologies, the intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system provided by this invention has the following beneficial effects:

[0041] 1. By constructing a three-dimensional dynamic impedance distribution map, the system can accurately capture the drastic fluctuations in interface impedance caused by micro-movements of brain tissue during free movement of the subject. Based on this, a dynamically weighted common average reference layer can adaptively follow the shift of the physical noise center of gravity of the entire array, improving the common-mode noise suppression efficiency by more than 50% compared with traditional methods under free movement conditions, and achieving high-quality signal extraction under extremely low signal-to-noise ratio.

[0042] 2. By fusing the signal energy distribution ratio within the subspace with the current interface impedance change rate, the common-mode noise contribution weight of each channel is quantified with extreme precision. Channels with severe contamination dominate when synthesizing the virtual reference layer, while the weight of clean channels is compressed to the extreme. This mathematically prevents secondary cross-contamination of noise between channels, ensuring the absolute purity of the clean channel signal.

[0043] 3. By abandoning the traditional frequency slicing filtering approach, a geometrical isolation method using weighted principal component analysis and orthogonal complementary space projection is adopted instead. This forces the high-dimensional mixed signal to be projected onto mutually perpendicular artifact subspaces and neural subspaces. While completely eliminating interference from ECG, EMG, and respiration, this method perfectly avoids the phase distortion and loss of low-frequency characteristic components inherent in traditional filtering, and faithfully preserves the original form of extremely fragile action potentials and low-frequency local field potentials.

[0044] 4. In the final stage of the noise reduction subtraction operation, this system fully utilizes the local interface abrupt changes reflected in the dynamic impedance distribution spectrum to proactively perform picosecond / nanosecond-level microscopic phase delay compensation and amplitude fine-tuning on the synthesized dynamic reference layer signal. This mechanism enables extremely precise gear-like meshing between the reference waveform and the target channel noise waveform on the time and amplitude axes, completely eliminating incomplete subtraction caused by local tissue deformation, significantly reducing signal coherence and inter-channel crosstalk caused by the high-density array, and providing an irreplaceable clear data source for precise localization of clinical epileptic foci and advanced cognitive neuroscience research. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0046] Figure 1 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0048] Please see Figure 1 A multi-channel intracranial neural signal synchronous acquisition and noise reduction processing system, including:

[0049] The data acquisition module is used to acquire the raw neural electrical signals of each channel of the human brain or experimental animals and the corresponding electrode-tissue interface impedance spectrum data through a high-density microelectrode array implanted in the brain in a multi-channel synchronous sampling manner.

[0050] The data processing module is used to construct a dynamic impedance distribution map of the electrode-tissue interface based on the interface impedance spectrum data acquired in real time from each channel, and to extract the spatiotemporal propagation characteristics of physiological artifacts by combining the spatial coordinates of the signals from each channel with the weighted principal component analysis method.

[0051] The data determination module projects the original neural electrical signals of each channel to the artifact subspace and the neural signal subspace. Based on the energy distribution ratio and impedance change rate of the signal in the artifact subspace and the neural signal subspace, it dynamically determines the common-mode noise contribution weight of each channel at the current moment.

[0052] The result generation module is used to generate time-varying weighted coefficients based on the judgment results, construct a dynamically weighted common average reference layer (DWCAR), and adaptively subtract the dynamic reference layer signal from each channel signal to achieve common mode suppression, while retaining the neural signal subspace components to obtain the denoised target neural electrical signal.

[0053] Specifically, high-density microelectrode arrays refer to implantable electrodes with a dense number of channels and miniaturized design. Dynamic impedance mapping is a real-time spatial topological mapping of the microscopic interface state of the electrode array during subject movement. The artifact subspace and neural signal subspace are separated into different independent dimensions through mathematical orthogonal transformations.

[0054] This system breaks through the limitations of traditional pure digital signal processing at the underlying architecture level, and establishes a closed-loop working mechanism that couples physical state and digital signal in two directions.

[0055] First, the system synchronously captures the subject's neurophysiological signals and physical impedance characteristics reflecting the microscopic contact state through a hardware front-end. Then, the algorithm transforms the discrete physical impedance into visualized three-dimensional spatial topological constraints, guiding the principal component analysis model to accurately locate physiological common-mode artifacts propagating across channels. Based on this, the system uses spatial geometric projection rules to forcibly tear the originally aliased signals into "pure neural dimensions" and "contaminated artifact dimensions," and precisely quantifies the "noise contamination index" of each physical channel at that moment by cross-comparing the energy proportion of the internal dimensions with the intensity of the external physical impedance. Finally, the system uses these indices to proportionally extract signals from each channel, dynamically piecing together a virtual reference layer that perfectly characterizes the current global common-mode noise pattern, and then performs subtraction stripping. This mechanism ensures that the noise reduction process can adapt to every moment of the subject's movement in real time.

[0056] It effectively solves the industry problem of common-mode noise suppression in free-moving states. The dynamic weighted reference layer greatly improves the common-mode noise suppression efficiency and significantly reduces signal crosstalk between channels, achieving high-fidelity neural signal acquisition under extremely low signal-to-noise ratio.

[0057] The data acquisition module is specifically used for:

[0058] A weak test current of known frequency is injected into a high-density microelectrode array, and a mixed electrical signal containing neural activity information of the test subjects, such as the human brain, and the response of the weak test current is acquired simultaneously.

[0059] The test current response was separated from the mixed electrical signal using a bandpass filter and lock-in amplification to calculate the electrode-tissue interface impedance spectrum data, while the original neural electrical signal was extracted.

[0060] Specifically, the weak test current is a carrier current of a specific high frequency with its amplitude controlled below the biosafety threshold, which will not induce abnormal neural discharges in the test subject. Phase-locked amplification (PLA) is a phase-sensitive detection technique capable of extracting weak signals of a specific frequency from strong background noise.

[0061] At the hardware acquisition front end, the system continuously injects a specific high-frequency microampere-level probe current into the tissue through a constant current source network. The impedance characteristics of the biological tissue modulate the amplitude and phase of this current, forming physical feedback. At this time, the probe electrodes simultaneously pick up low-frequency EEG signals and high-frequency impedance feedback signals, which naturally superimpose in the time domain to form a composite signal. After entering the digital domain, the system sets up a split processing link: one path directly filters out the high-frequency probe carrier through a low-pass / band-pass filter to restore the pure original neural electrical signal; the other path uses a lock-in amplifier, with the injected known high-frequency signal as a reference local oscillator, to perform mixing and low-pass filtering, accurately demodulating the response component that is in phase and frequency with the probe carrier, and then calculating the real and imaginary parts of the impedance at this moment using Ohm's law. This mechanism ensures sub-millisecond time alignment of two types of data with different attributes.

[0062] It achieves strict time-synchronous acquisition of the original neural signals and the physical impedance of the interface, completely eliminating relay switching noise and time delay errors caused by time-division switching acquisition, and providing an absolutely reliable real-time data source for subsequent dynamic calculation of weights.

[0063] In the data processing module, the specific process of constructing the dynamic impedance distribution spectrum of the electrode-tissue interface includes:

[0064] Acquire the impedance amplitude and phase data of each channel within a continuous sliding time window;

[0065] The three-dimensional spatial coordinates of implanted in the brain and other subjects' skulls, combined with a pre-calibrated high-density microelectrode array;

[0066] Using a spatial interpolation algorithm, a three-dimensional dynamic distribution map representing the impedance change of each channel electrode-tissue interface at the current moment under free-moving conditions is generated.

[0067] Specifically, a continuous sliding time window divides an infinitely long continuous time axis into equal-length computational periods with overlapping beginnings and ends. Spatial interpolation algorithms use impedance data from known discrete points to estimate data in unmeasured regions, thereby constructing a smooth global distribution surface.

[0068] This step is crucial for establishing the physical constraint model. During operation, as the sliding window advances, the system continuously captures snapshots of the impedance amplitude and phase of all channels within each time slice. The system pre-stores a "three-dimensional coordinate mapping table" reflecting the probe's physical structure. When a frame of impedance data is input, the system uses the impedance value of each channel as a Z-axis variable, attaching it to the corresponding XYZ physical coordinate point in the mapping table, forming a spatial scatter cloud. Subsequently, the spatial interpolation engine is activated, calculating and filling the physical gaps between the scatter points based on the correlation of impedances of adjacent electrodes, ultimately rendering a "three-dimensional impedance heatmap" covering the entire implantation target area in digital space. When the subject performs actions such as turning their head or chewing, causing local tissue displacement, this heatmap will instantly produce bulges or depressions in the corresponding area, thus transforming passive impedance values ​​into dynamic tissue deformation geometric features.

[0069] By visually transforming discrete and abstract electrical impedance data into a dynamic feature model with spatial continuity and topological correlation, it can accurately capture the spatiotemporal abrupt change in local impedance caused by the motion of the test object, providing a precise physical boundary for subsequent spatial filtering.

[0070] In the data processing module, the specific process of extracting the spatiotemporal propagation characteristics of physiological artifacts using weighted principal component analysis includes:

[0071] The original neural electrical signals of each channel were spatially weighted using the electrode-tissue interface impedance spectrum data of each channel as weighting factors.

[0072] A spatial covariance matrix is ​​constructed from the weighted multi-channel signal and eigenvalue decomposition is performed to extract the large-amplitude low-frequency components corresponding to physiological artifacts such as heartbeat, respiration or electromyography, thereby obtaining the principal component eigenvectors to characterize the spatiotemporal propagation characteristics of physiological artifacts.

[0073] Specifically, the spatial covariance matrix is ​​a square matrix whose diagonal elements represent the variance of a single channel signal and whose off-diagonal elements represent the correlation between channels. Eigenvalue decomposition is used to transform highly coupled mixed signals into several principal components that are orthogonal to each other and arranged in order of variance magnitude.

[0074] In traditional principal component analysis, all channels are treated equally. However, this system introduces the concept of "physical confidence," meaning that channels with more abnormal impedance and more drastic fluctuations are more likely to be affected by physiological motion interference. Therefore, when constructing the multi-channel data matrix, the system first uses data extracted from real-time impedance spectra as scaling factors to pre-multiply and weight the original signal. Channels with extremely high impedance have their signal amplitude mathematically amplified, while channels with stable impedance have their signal weights suppressed. After this highly biased preprocessing, the system calculates the covariance matrix between each channel. When performing eigenvalue decomposition on this matrix, due to the enormous energy of artifacts such as heartbeats and respiration, coupled with the mathematical amplification effect of high-impedance channels, these artifact components are forcibly pushed into the eigenvectors corresponding to the first few largest eigenvalues. These eigenvectors essentially depict the spatial path model of how interference noise propagates from its source through brain tissue with different impedances to each electrode.

[0075] By introducing physical impedance as the core weighting constraint, the blindness of pure mathematical statistics is broken, so that the extracted pseudo-trace principal component features are no longer simply the maximization of mathematical variance, but truly reflect the physical law of the propagation and attenuation of physiological noise in inhomogeneous tissues.

[0076] In the data determination module, the step of projecting the signal into the subspace specifically includes:

[0077] Based on the spatiotemporal propagation characteristics, a pseudo-trace projection matrix is ​​constructed. The original neural electrical signals of each channel are multiplied by the pseudo-trace projection matrix to obtain the pseudo-trace components projected onto the pseudo-trace subspace.

[0078] By projecting the original neural electrical signals of each channel into the neural signal subspace through orthogonal complementary space, the neural activity components that filter out major physiological artifacts are obtained.

[0079] Specifically, the artifact projection matrix is ​​a spatial mapping operator constructed from the eigenvectors of the extracted artifact principal components. The orthogonal complement space refers to a data space in a multidimensional linear space that is completely perpendicular to and does not intersect with the artifact subspace.

[0080] This module's operational logic departs from the traditional "frequency slicing" approach, instead employing "geometric dimensional isolation." The system treats the synchronously acquired high-density multi-channel signals as a complex set of vectors in a high-dimensional space. Since the spatial propagation path of the noise has been determined in the previous step, the system uses these vectors to construct a dedicated "artifact projection matrix." When the original high-dimensional signal vector is multiplied by this matrix, it's equivalent to light shining on an object and creating a shadow. The components in the mixed signal that match the noise pattern are precisely "projected" and extracted, forming pure artifact components. Simultaneously, based on the orthogonality principle of linear algebra, the system constructs the null space of this artifact matrix and forces the original signal to be projected into this space. Because the orthogonal complement space and the artifact space are absolutely perpendicular at 90 degrees in multidimensional geometry, this means that the signal components projected into the neural subspace have been mathematically stripped of any artifact-related features, leaving only pure activity components that conform to the random discharge characteristics of neurophysiology.

[0081] Rigid isolation between neural electrical signals and physiological artifacts is achieved in the multidimensional geometric dimension, completely avoiding the fatal defects inherent in traditional frequency filtering, such as phase distortion, passband ripple interference, and loss of low-frequency EEG characteristic components.

[0082] In the data determination module, the calculation logic for dynamically determining the common-mode noise contribution weight includes:

[0083] The energy distribution ratio is obtained by calculating the ratio of the artifact component energy to the neural activity component energy of each single channel at the same analysis time.

[0084] Calculate the rate of change of interface impedance of the single channel at the current moment compared to the previous time window;

[0085] The common-mode noise contribution weight of each channel is calculated by fusing the energy distribution ratio and the impedance change rate. The channel with the larger energy distribution ratio and the higher impedance change rate has a higher common-mode noise contribution weight.

[0086] In this embodiment, the specific logic of the fusion calculation can be implemented by defining a nonlinear activation function or a weighted summation formula. For example, the weighted output model can be set as follows:

[0087]

[0088] In the formula, For the first Common-mode noise contribution weights for each channel; This represents the energy distribution ratio. The rate of change of interfacial impedance; and The sensitivity adjustment constant set for the system; This is the normalized activation function.

[0089] This mechanism is equivalent to equipping each electrode channel with an intelligent evaluator. At each calculation instant, the evaluator simultaneously extracts two core indicators: the first is an internal signal attribute indicator, which is the sum of squares of the energy of the "artifact component" separated in the previous step, divided by the sum of squares of the energy of the "neural activity component". If this ratio is extremely large, it indicates that the signal captured by the channel at this moment is almost entirely noise. The second is an external physical attribute indicator, which is the impedance derivative of the current impedance of the channel compared to the previous window. If the rate of change is extremely high, it indicates that the channel has encountered a severe physical impact or is at risk of electrode dislocation. The system substitutes these two indicators, reflecting different physical natures, into the fusion formula for cross-verification. Only when a channel exhibits both an extremely high proportion of artifact energy at the digital level and an extremely unstable contact state at the physical level will the system output an extremely high "common-mode noise contribution weight", definitively identifying the channel as a primary source of contamination.

[0090] By deeply integrating the intrinsic digital statistical parameters of electrical signals with the external physical state parameters, a robust multimodal cross-verification barrier is constructed, which significantly improves the accuracy of common-mode noise weight allocation and prevents occasional single-channel abnormal high-frequency discharges from being misjudged as artifact noise.

[0091] In the results generation module, the specific process of constructing the dynamically weighted common average reference layer DWCAR includes:

[0092] The common-mode noise contribution weights of all channels are globally normalized to calculate the time-varying weighting coefficients of each channel at the current time.

[0093] The signal amplitude of each channel at the current moment is multiplied by its corresponding time-varying weighting coefficient and then summed to generate a dynamically weighted common average reference layer signal that represents the global common-mode noise background.

[0094] Specifically, global normalization maps all calculated independent weight values ​​to the [0,1] interval and ensures that the sum of the coefficients of all channels is strictly equal to 1. In this embodiment, the generation logic of the DWCAR signal is executed according to the following mathematical model:

[0095]

[0096] in, The amplitude of the synthesized dynamically weighted common average reference layer instantaneous signal; This represents the total number of channels for hardware sampling. After normalization, the first Time-varying weighting coefficients for each channel; For the first The amplitude of the original neural electrical signal at the current moment for each channel.

[0097] In the computation pipeline for generating the final reference layer, the system first performs a stringent normalization operation to constrain the boundaries. Assuming there are 100 channels, the system uses normalization to ensure that the sum of these 100 time-varying weighted coefficients is 1. At this point, based on the conclusion of the aforementioned decision module, the more heavily contaminated the channel, the more weighted it is assigned. The coefficient approaches 1 infinitely; and the purer the signal in a channel, the more its... The coefficients are forcibly compressed to near zero. Subsequently, the system performs a weighted summation operation. Under the influence of this formula, the background EEG signals from a massive number of pure channels are eliminated during the accumulation due to their extremely small coefficients; conversely, the signals from channels containing typical common-mode physiological noise are completely preserved and amplified. Through this extreme "sublation" operation, the system artificially creates a virtual signal curve—the DWCAR layer—that does not physically exist but highly fits and condenses the common-mode noise profile of the entire implanted brain region at the current moment.

[0098] The dynamic reference layer can spontaneously and instantaneously shift with the physical noise center of gravity of the entire array, fundamentally blocking the "large-area noise diffusion and secondary cross-contamination" phenomenon caused by the traditional arithmetic averaging algorithm, and ensuring the absolute purity of the clean channel.

[0099] In the final subtraction step of removing the common reference layer from the target signal, if a small deformation of local brain tissue causes a sudden change in impedance of a local channel, the noise captured by that channel will be slightly misaligned in time and phase with the globally synthesized reference layer signal. In this case, performing a simple algebraic subtraction directly will not only fail to eliminate the noise but will also generate new high-frequency harmonic artifacts due to the phase mismatch, resulting in incomplete subtraction.

[0100] The system also includes a channel crosstalk correction module, used for:

[0101] Based on the local interface abrupt change state reflected by the dynamic impedance distribution spectrum, the phase and amplitude compensation factors of the subtraction operation are dynamically adjusted when subtracting the dynamic reference layer signal from the signals of each channel.

[0102] To separate inter-channel crosstalk and background noise during high-density acquisition, reduce the crosstalk ratio in the signal coherence spectrum, and preserve the true high-frequency local field potential and action potential characteristics of the human brain.

[0103] Specifically, a local interface abrupt change refers to an abnormal impedance gradient appearing at individual electrode nodes within a short period of time in the three-dimensional impedance distribution spectrum. Phase and amplitude compensation factors are fine-tuning parameters applied to the synthesized reference signal before matrix subtraction by introducing a small time delay window and a scaling multiplier.

[0104] This crosstalk correction module acts as the final, precise alignment valve in the noise reduction process. It is used to perform the final purification subtraction. Previously, the system would retrieve a three-dimensional impedance spectrum and scan the current process to be processed. Whether each channel is in a state of interface abrupt change. If the deformation of this local tissue causes a slight physical delay in signal transmission, the correction module will calculate this delay and, before performing the subtraction, specifically adjust the signal transmission path. The signal is compensated for with a picosecond or nanosecond-level phase lag or lead, and its amplitude is microscaled and matched. The system aligns the peaks of the reference signal with the peaks of noise in the target channel with extreme precision on both the time and amplitude axes, much like aligning the meshing points of two gears under a microscope. After this microscopic alignment, a subtraction operation is performed.

[0105] It completely eliminates residual incomplete common-mode inhibition caused by physical differences at the micro-electrode-brain tissue interface, effectively removes signal crosstalk caused by parasitic capacitance between high-density channels, and perfectly preserves the extremely fragile action potential and high-frequency local field potential characteristics that carry core neural coding information.

[0106] It includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the data acquisition and noise reduction processing logic performed by the modules mentioned above, so as to achieve high-fidelity acquisition of intracranial neural signals from subjects such as the human brain.

[0107] Specifically, the memory can be read-only memory, random access memory, or high-speed solid-state drive, used to store static machine instructions for the processing algorithm and dynamic high-dimensional data matrices during runtime. The processor can be a high-performance multi-core central processing unit, a graphics processing unit, or a field-programmable gate array (FPGA) with high-concurrency, low-latency computing capabilities.

[0108] After the computer device is powered on and initialized, the processor retrieves pre-compiled noise reduction program instructions from the memory cache. During operation, the hardware bus, acting as the main artery for data transmission, continuously transports the native EEG and impedance data collected by the high-density electrode array to the processor's registers. Based on the instruction logic, the processor schedules its internal floating-point arithmetic unit and matrix multiplication acceleration engine to intensively perform a series of complex logical branches on the input data stream within extremely short clock cycles, including low-pass filtering, three-dimensional kriging interpolation, covariance matrix decomposition, spatial geometric projection, multi-parameter weight fusion, and array subtraction operations with phase compensation. After the calculations are completed, the processor packages the pure background neural signal data, stripped of common-mode artifacts, and outputs it to the device's display for real-time waveform rendering, or sends it to a medical database for archiving via a network interface.

[0109] This provides a solid and stable underlying computing environment and hardware platform support for the aforementioned innovative adaptive noise reduction methodology, ensuring low latency and real-time performance for synchronous processing of extremely high-density data streams with hundreds of channels or more.

[0110] A computer program is stored on a computer-readable storage medium. When the computer program is executed by a processor, it implements the signal processing logic performed by the modules described above.

[0111] Specifically, computer-readable storage media include, but are not limited to, USB flash drives, portable hard drives, read-only optical discs, cloud server object storage nodes, or various data-storing carriers such as non-volatile flash memory chip arrays deeply embedded in the motherboard of portable medical monitoring devices.

[0112] The R&D engineers compiled the complete source code project, which includes impedance mapping, weighted principal component extraction, and dynamic reference layer adaptive synthesis algorithms, into executable binary machine code. They then used magnetic, optical, or charge injection techniques to physically record and save these binary code sequences into the microscopic storage units of the storage medium. When clinical staff connect this storage medium to any neurophysiological monitoring terminal with basic general-purpose processing capabilities, the terminal's underlying operating system actively mounts the medium, reads the encapsulated program instruction stream, and quickly decompresses and deploys the noise reduction algorithm engine framework in the processor's memory space, thereby taking over and modifying the original device's signal processing data link.

[0113] This significantly reduces the dependence of high-order noise reduction algorithms on specific dedicated hardware motherboards, greatly improving the cross-platform portability of medical software systems, the commercial distribution efficiency of algorithm firmware, and the convenience of subsequent technical maintenance.

[0114] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A system for synchronous acquisition and noise reduction of intracranial multi-channel neural signals, characterized in that, include: The data acquisition module is used to acquire the raw neural electrical signals of each channel of the human brain or experimental animals and the corresponding electrode-tissue interface impedance spectrum data through a high-density microelectrode array implanted in the brain in a multi-channel synchronous sampling manner. The data processing module is used to construct a dynamic impedance distribution map of the electrode-tissue interface based on the interface impedance spectrum data acquired in real time from each channel, and to extract the spatiotemporal propagation characteristics of physiological artifacts by weighted principal component analysis in combination with the spatial coordinates of each channel signal. The data determination module is used to project the original neural electrical signals of each channel to the artifact subspace and the neural signal subspace, and dynamically determine the common-mode noise contribution weight of each channel at the current moment based on the energy distribution ratio and impedance change rate of the signal in the artifact subspace and the neural signal subspace. The result generation module is used to generate time-varying weighted coefficients based on the judgment result, construct a dynamically weighted common average reference layer (DWCAR), and adaptively subtract the dynamic reference layer signal from each channel signal to achieve common mode suppression, while retaining the neural signal subspace components to obtain the denoised target neural electrical signal.

2. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The data acquisition module is specifically used for: A weak test current of known frequency is injected into the high-density microelectrode array, and a mixed electrical signal containing neural activity information of the test subject, such as the human brain, and the response of the weak test current is collected simultaneously. The test current response is separated from the mixed electrical signal using a bandpass filter and lock-in amplification to calculate the electrode-tissue interface impedance spectrum data, while the original neural electrical signal is extracted.

3. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The specific process of constructing the dynamic impedance distribution spectrum of the electrode-tissue interface in the data processing module includes: Acquire the impedance amplitude and phase data of each channel within a continuous sliding time window; Combined with the pre-calibrated high-density microelectrode array implanted in the brain and other subjects' cranial cavity, the three-dimensional spatial coordinates are determined. Using a spatial interpolation algorithm, a three-dimensional dynamic distribution map representing the impedance change of each channel electrode-tissue interface at the current moment under free-moving conditions is generated.

4. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The specific process of extracting the spatiotemporal propagation characteristics of physiological artifacts using the weighted principal component analysis method in the data processing module includes: The original neural electrical signals of each channel are spatially weighted using the electrode-tissue interface impedance spectrum data of each channel as weighting factors. A spatial covariance matrix is ​​constructed from the weighted multi-channel signal and eigenvalue decomposition is performed to extract the large-amplitude low-frequency components corresponding to physiological artifacts such as heartbeat, respiration or electromyography, thereby obtaining the principal component eigenvectors to characterize the spatiotemporal propagation characteristics of physiological artifacts.

5. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The data determination module includes the following steps for projecting the signal into the subspace: Based on the spatiotemporal propagation characteristics, a pseudo-trace projection matrix is ​​constructed. The original neural electrical signals of each channel are multiplied by the pseudo-trace projection matrix to obtain the pseudo-trace components projected onto the pseudo-trace subspace. By projecting the original neural electrical signals of each channel into the neural signal subspace through orthogonal complementary space, neural activity components with major physiological artifacts filtered out are obtained.

6. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 5, characterized in that, The calculation logic for dynamically determining the common-mode noise contribution weight in the data determination module includes: The energy distribution ratio is obtained by calculating the ratio of the energy of the artifact component to the energy of the neural activity component in each single channel at the same analysis time. Calculate the rate of change of interface impedance of the single channel at the current moment compared to the previous time window; The energy distribution ratio and the impedance change rate are fused together to calculate and output the common-mode noise contribution weight of each channel at the current moment; wherein, the channel with the larger energy distribution ratio and the higher impedance change rate has a higher common-mode noise contribution weight.

7. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The specific process of constructing the dynamically weighted common average reference layer DWCAR in the result generation module includes: The common-mode noise contribution weights of all channels are globally normalized to calculate the time-varying weighting coefficients of each channel at the current time. The signal amplitude of each channel at the current moment is multiplied by its corresponding time-varying weighting coefficient and then summed to generate the dynamic weighted common average reference layer signal that characterizes the global common-mode noise background.

8. The intracranial multi-channel neural signal synchronous acquisition and noise reduction processing system according to claim 1, characterized in that, The system also includes a channel crosstalk correction module, used for: Based on the local interface abrupt change state reflected by the dynamic impedance distribution spectrum, when subtracting the dynamic reference layer signal from each channel signal, the phase and amplitude compensation factors of the subtraction operation are dynamically adjusted. To separate inter-channel crosstalk and background noise during high-density acquisition, reduce the crosstalk ratio in the signal coherence spectrum, and preserve the true high-frequency local field potential and action potential characteristics of the human brain.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the data acquisition and noise reduction processing logic performed by each module as described in any one of claims 1 to 8, so as to achieve high-fidelity acquisition of intracranial neural signals of subjects such as human brains.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the signal processing logic performed by each of the modules as described in any one of claims 1 to 8.

Images