Intracranial multimodal data monitoring system based on FBG grating technology

Abstract

This invention proposes an intracranial multimodal data monitoring system based on FBG grating technology, including a monitoring device, a three-channel fiber Bragg grating, and a three-channel fiber Bragg grating probe. Before monitoring, the monitoring device implants the three-channel fiber Bragg grating probe into the skull for decoupling calibration. The center wavelength is aligned at a uniform sampling time to form a three-dimensional observation sequence. A decoupling drift factorization interface is constructed using a drift mode basis set to restrict the decoupling drift to the subspace spanned by the basis set. A score diffusion latent variable inference network is used to generate a posterior distribution of the drift latent variable. The drift confidence is calculated based on the consistency between the posterior dispersion and continuous evolution. The drift coefficient is determined under confidence gating, range pruning, and continuity fusion to correct the initial decoupling matrix; otherwise, a baseline drift coefficient is used. Based on the current decoupling matrix, three types of monitoring values ​​are decoupled and output in real time, and a drift judgment flag is generated and sent to the monitoring device.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing and neural monitoring technology, and in particular to an intracranial multimodal data monitoring system based on FBG grating technology. Background Technology

[0002] In intracranial intensive care often requires simultaneous acquisition of multimodal information such as intracranial pressure, local temperature, and cerebral oxygenation to support treatment decisions and risk warnings. Fiber Bragg gratings (FBGs) are small, implantable, resistant to electromagnetic interference, and reusable at multiple points, making them suitable for integration with intracranial probes. However, the center wavelength of FBGs is cross-sensitive to various physical quantities, and the adhesion and stress state with tissues after implantation changes over time, easily causing a drift in the mapping relationship from wavelength to physiological quantity, thus affecting the continuous delivery of the monitoring chain.

[0003] Existing solutions typically employ multi-channel FBGs or combinations of FBGs with other sensing units. A fixed linear decoupling matrix is ​​established through a single calibration, mapping the changes in the three center wavelengths to parameters such as intracranial pressure, temperature, and oxygenation. In engineering implementation, this is combined with grating demodulation equipment to locate the peak of the reflectance spectrum and extract the center wavelength. Stable readings are obtained through signal processing such as channel alignment, filtering, and threshold alarms. Some solutions reduce the impact of temperature drift / strain coupling by adding a reference grating, improving the packaging structure, or periodically recalibrating. Other solutions use regression models to directly output multi-parameter estimation results from the multi-channel wavelength sequence to simplify the decoupling process.

[0004] The main shortcomings of the above schemes are: the decoupling matrix relies heavily on one-time calibration during ex vivo or early implantation, making it difficult to characterize the decoupling relationship drift caused by the slow change in the fit after implantation; clinical settings often lack repeatable calibration loads or reliable reference quantities, making it difficult to determine "whether the decoupling relationship should be updated" based solely on online three-channel wavelength observations when drift occurs; when using direct regression or empirical thresholds, there is usually a lack of expression and gating mechanisms for the uncertainty of drift estimation, which can easily lead to erroneous corrections or the inability to provide interpretable fit drift status markers, thereby affecting the availability and deliverability of monitoring data.

[0005] Therefore, a multimodal intracranial data monitoring system that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intracranial multimodal data monitoring system based on FBG grating technology. The core technical problem to be solved by this application is: under the condition that only the online three-channel FBG center wavelength timing and one-time calibration information (initial decoupling matrix and prior constraints of drift mode) after implantation can be obtained, how to reliably identify and estimate the decoupling relationship drift caused by the change of probe fit state, and thereby realize closed-loop online decoupling calibration and drift state output.

[0007] This invention provides an intracranial multimodal data monitoring system based on FBG grating technology, including a monitoring device, a three-channel fiber Bragg grating, and a three-channel fiber Bragg grating probe. One end of the three-channel fiber Bragg grating is connected to the monitoring device via a monitoring port, and the other end is fitted with the three-channel fiber Bragg grating probe. Before monitoring, the three-channel fiber Bragg grating probe is implanted intracranially for decoupling calibration, as detailed below:

[0008] S1. Obtain the center wavelength timing of the three-channel fiber Bragg grating implanted in the brain, form the three-channel center wavelength timing, and receive the initial decoupling matrix obtained from one calibration.

[0009] S2. Read the drift mode basis set corresponding to the three-channel fiber Bragg grating probe, construct the decoupling drift factorization interface in combination with the initial decoupling matrix, and limit the dimension and range of drift coefficient values.

[0010] S3. Input the three-channel center wavelength time sequence into the score diffusion latent variable inference network. The scorediffusion latent variable inference network characterizes the probe fitting state change with cross-channel shared drift drive, and generates the fitting drift latent variable posterior distribution under continuous evolution constraint. Calculate the fitting drift confidence based on the fitting drift latent variable posterior distribution to obtain the fitting drift confidence.

[0011] S4. Determine the fit drift confidence level and the fit drift confidence level threshold. If the condition is met, map the posterior distribution of the latent variable of fit drift to the drift coefficient, and modify the initial decoupling matrix through the decoupling drift factorization interface to obtain the current decoupling matrix. If the condition is not met, use the preset benchmark drift coefficient to obtain the current decoupling matrix through the decoupling drift factorization interface.

[0012] S5. Based on the current decoupling matrix, decouple the timing of the three-channel center wavelength to obtain the intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value.

[0013] S6. Generate a fit drift judgment mark based on the drift coefficient, and send the fit drift judgment mark along with the intracranial pressure monitoring value, intracranial temperature monitoring value, and cerebral oxygenation monitoring value to the monitoring equipment.

[0014] Optionally, step S1 specifically includes:

[0015] The three-channel fiber Bragg grating probe implanted in the brain is connected to the grating demodulation device to continuously demodulate the three-channel reflection spectrum and obtain the original timing sequence of the center wavelength of the three-channel fiber Bragg grating.

[0016] The original timing sequence of the center wavelength of the three-channel fiber Bragg grating is time-aligned according to a unified sampling time. Three center wavelength scalars are extracted at each sampling time and formed into a three-dimensional observation vector in a fixed channel order. The three-dimensional observation vector is then arranged in time to form the three-channel center wavelength timing sequence.

[0017] The system receives the initial decoupling matrix obtained from a one-time calibration, performs a consistency check between the row and column indices of the initial decoupling matrix and the fixed channel order, as well as the parameter order of intracranial pressure monitoring values, intracranial temperature monitoring values, and brain oxygenation monitoring values. If the check is inconsistent, the initial decoupling matrix is ​​rearranged to meet the consistency constraints.

[0018] Optionally, step S2 specifically includes:

[0019] Read the drift mode basis set corresponding to the three-channel fiber Bragg grating probe, represent each drift mode as a three-row, three-column drift mode matrix consistent with the fixed channel order, and count the number of drift mode matrices to determine the dimension of the drift coefficient values;

[0020] The initial decoupling matrix and the drift mode matrix are connected to the decoupling drift factorization interface. In the decoupling drift factorization interface, the drift coefficients are weighted and combined according to the order of the drift mode matrix to generate a controlled perturbation matrix. The controlled perturbation matrix is ​​then added to the initial decoupling matrix to form a calibrable expression of the decoupling matrix.

[0021] Based on the range of drift coefficient values ​​obtained from a one-time calibration, upper and lower limits are set for each dimension of the drift coefficient, and these upper and lower limits are used as constraints on the range of drift coefficient values ​​when determining the drift coefficient.

[0022] Optionally, step S3 specifically includes:

[0023] A random noise vector is generated based on the dimension of the drift coefficient, and a noise level index is generated according to the preset diffusion noise level sequence. The noise level index is then input into the noise level coding layer to obtain the noise level coding vector.

[0024] The three-channel center wavelength time sequence is input into the cross-channel shared drift driving layer. Multi-layer one-dimensional convolutional neurons are used to extract the cross-channel common slowly varying features, and the convolutional features are converged in time to obtain the shared drift representation vector.

[0025] The time series of the three center wavelengths are input into the three channel branch layers respectively, and the channel difference features are extracted by stacking independent one-dimensional convolutional neurons to obtain the three channel difference representation vectors.

[0026] The shared drift representation vector and the three channel difference representation vectors are concatenated and fused in channel order, and the conditional feature vector is obtained through the fusion mapping layer. The conditional feature vector is only used as the inference condition for the latent variable of drift and is not used as the regression target for intracranial pressure monitoring value, intracranial temperature monitoring value and brain oxygenation monitoring value.

[0027] The conditional feature vector is input into the continuous evolution constraint layer. Evolutionary consistency features are generated based on the change amplitude of the conditional feature vector at adjacent time points. The evolutionary consistency features are used as constraint information to fit the temporal continuity of the drift latent variables.

[0028] The random noise vector, conditional feature vector, and noise level encoding vector are input into the scoring layer. Multi-layer fully connected neurons and residual connections are used to generate the score vector. Based on the score vector-driven diffusion-type posterior generation mechanism, the random noise vector is subjected to multi-level denoising iteration to obtain a sample sequence of drifting latent variables.

[0029] The posterior center vector and posterior dispersion vector are obtained by statistically aggregating the latent variable sample sequences of the fit drift. The fit drift confidence is obtained by confidence mapping based on the posterior dispersion vector and the evolutionary consistency feature.

[0030] Optionally, when obtaining the fit drift confidence score by mapping the confidence score between the posterior discreteness vector and the evolutionary consistency feature, a confidence function is used to map the confidence score between the posterior discreteness vector and the evolutionary consistency feature. Specifically, the confidence function is:

[0031] ;

[0032] in, Sampling time The reliability of the fit drift It is an exponential function. The bias parameters for the credibility parameter layer. The weight parameters are the features corresponding to evolutionary consistency. For the posterior discreteness vector, the first... The corresponding weight parameters for each dimension Sampling time The evolutionary consistency characteristic scalar, Sampling time posterior discreteness vector The dimensional components, The dimension for the drift coefficient values. This is the index of the components of the posterior discreteness vector.

[0033] Optionally, step S4 specifically includes:

[0034] The confidence level of the bonding drift is compared with the confidence threshold of the bonding drift to obtain the confidence level judgment result;

[0035] When the credibility determination result meets the condition of fitting the drift credibility threshold, the posterior center vector and the posterior dispersion vector are read, the posterior dispersion vector is mapped to the uncertainty weight vector, and the posterior center vector is weighted dimension by dimension to obtain the drift candidate vector.

[0036] The drift candidate vectors are checked for consistency in the dimensions of their values, and the drift candidate vectors are pruned dimension by dimension according to the upper and lower limits of each dimension of the drift coefficient to obtain range-constrained drift candidate vectors.

[0037] Read the evolutionary consistency features, map the evolutionary consistency features to continuous weights, and perform weighted fusion of the range-constrained drift candidate vectors and the preset benchmark drift coefficients based on the continuous weights to determine the drift coefficients;

[0038] When the confidence determination result does not meet the confidence threshold condition for the fit drift, the drift coefficient is determined by using a preset benchmark drift coefficient;

[0039] The drift coefficients and drift mode matrix are input into the decoupled drift factorization interface. The controlled perturbation matrix is ​​generated by weighted combination of each dimension of the drift coefficients and the corresponding drift mode matrix. The controlled perturbation matrix is ​​then restricted to the subspace spanned by the drift mode matrix.

[0040] The controlled disturbance matrix is ​​added to the initial decoupling matrix to form the current decoupling matrix.

[0041] Optionally, after reading evolutionary consistency features and mapping them to continuous weights, and when the confidence determination result meets the condition of fitting the drift confidence threshold, the posterior discreteness vector is mapped to an uncertainty weight vector, and the posterior center vector is weighted dimension-wise. Based on obtaining range-constrained drift candidate vectors by dimension-wise pruning of the drift candidate vectors according to the upper and lower limits of each dimension of the drift coefficient, a coefficient function is used to weight and fuse the range-constrained drift candidate vectors with a preset benchmark drift coefficient to determine the drift coefficient. When the confidence determination result does not meet the condition of fitting the drift confidence threshold, the preset benchmark drift coefficient is used to determine the drift coefficient. The coefficient function is specifically as follows:

[0042] ;

[0043] in, Sampling time The drift coefficient vector, To standardize sampling times, Sampling time The reliability of the fit drift To match the drift confidence threshold, Sampling time Continuous weight scalar, Sampling time The posterior center vector, Sampling time Uncertainty weight vector, for The dimensional components, The scalar normalization factor is obtained by summing the components of the uncertainty weight vector. For dimension-wise multiplication operations, This is the lower bound vector for the drift coefficient. This is the upper limit vector of the drift coefficients. As a preset baseline drift coefficient vector, To perform the operation by taking the smaller value in each dimension, To perform the operation by taking the larger value in each dimension, The dimension for the drift coefficient values. This is the dimension index for the drift coefficient.

[0044] Optionally, step S5 specifically includes:

[0045] At each sampling moment, the three center wavelength scalars of the three-channel center wavelength timing are combined into a center wavelength observation vector in a fixed channel order, and the reference center wavelength vector determined in the one-time calibration stage is read. The center wavelength change vector is obtained by performing reference difference on the center wavelength observation vector.

[0046] The center wavelength change vector is mapped to the current decoupling matrix to obtain a three-parameter monitoring vector arranged in the order of intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value.

[0047] The three-parameter monitoring vectors at each sampling time are arranged in chronological order, and the three components of the three-parameter monitoring vectors are respectively used to form the time series of intracranial pressure monitoring values, intracranial temperature monitoring values, and cerebral oxygenation monitoring values.

[0048] Optionally, step S6 specifically includes:

[0049] Read the drift coefficient, drift mode matrix sensitivity weight, upper threshold of fit drift, lower threshold of fit drift and fit drift judgment flag of the previous sampling time. Take the absolute value of the drift coefficient dimension by dimension and aggregate it according to the sensitivity weight to obtain the fit drift amplitude index. When the fit drift amplitude index is greater than the upper threshold of fit drift, update the fit drift judgment flag as fit drift abnormal. When the fit drift amplitude index is less than the lower threshold of fit drift, update the fit drift judgment flag as fit drift normal.

[0050] The sampling time, the fit drift judgment mark, the value of the intracranial pressure monitoring value corresponding to the sampling time, the value of the intracranial temperature monitoring value corresponding to the sampling time, and the value of the brain oxygenation monitoring value corresponding to the sampling time are concatenated in a fixed field order to generate the monitoring transmission frame timing.

[0051] The monitoring equipment communication interface is used to send the monitoring transmission frame timing sequence to the monitoring equipment in the order of sampling time.

[0052] The beneficial effects of this invention are:

[0053] (1) This proposal presents an improved online calibration method for multi-parameter decoupling of intracranial three-channel FBG. In the one-time calibration phase, a drift mode basis set is extracted from multiple sets of decoupling matrix deviation samples under controlled bonding / loading states. In the online phase, a decoupling drift factorization interface is constructed in the form of "initial decoupling matrix + weighted combination of drift modes," restricting the correction of the decoupling relationship to the subspace spanned by the drift modes. Simultaneously, consistency checks and upper and lower bound constraints are implemented on the dimension and dimensional value range of the drift coefficients. The controlled perturbation matrix can be projected back into this subspace and then added to the initial decoupling matrix to form the current decoupling matrix. Compared to treating the decoupling matrix as a fixed constant or arbitrarily adaptively updated, this design constrains the decoupling drift caused by changes in bonding state as an "interpretable and controllable" low-dimensional perturbation, reducing the risk of non-physical corrections, channel mismatches, or uncontrollable matrix drift during online updates. This maintains an executable decoupling calculation path under the condition of relying solely on online center wavelength observation and one-time calibration priors.

[0054] (2) This proposal proposes a novel posterior inference and confidence gating mechanism for fitting drift latent variables. Based on the score diffusion latent variable inference network, a cross-channel shared drift driving representation is established for the three-channel center wavelength time series. Combined with the channel branch difference characteristics, a fitting drift latent variable sample sequence is generated under continuous evolution constraints, thereby obtaining the posterior center vector and the posterior dispersion vector. Then, the posterior dispersion and evolution consistency are mapped to the fitting drift confidence, which is used to trigger or suppress the drift coefficient update. When the confidence meets the threshold, the uncertainty weight is constructed using the dispersion to generate drift candidates by weighting the posterior center, and then fused with the continuous weight and the benchmark drift coefficient. When the confidence is insufficient, it directly regresses to the benchmark drift coefficient. Compared to direct regression of multiple parameters or the use of empirical thresholds to judge drift, this mechanism introduces a computable uncertainty representation and gated update path in scenarios where reliable reference quantities are not available online. It supports interpretable judgment on "whether the decoupling relationship should be corrected", reduces erroneous updates when noise, transient disturbances or unclear fitting status are present, and provides traceable posterior statistical basis for the determination of drift coefficients.

[0055] (3) This proposal proposes a closed-loop method for monitoring delivery. The original time sequence of the center wavelength obtained by demodulating the three-channel reflection spectrum is aligned according to a unified sampling time. Combined with the consistency verification of channel order and parameter order, the monitoring values ​​are formed by mapping the reference center wavelength difference with the current decoupling matrix and arranged in a fixed order of intracranial pressure, intracranial temperature, and brain oxygenation. When the consistency of matrix and vector dimensions is not satisfied, protective skipping is performed to avoid generating non-corresponding data. At the same time, the fit drift amplitude index is constructed based on the drift coefficient and drift mode sensitivity weight. The fit drift judgment mark is generated by combining the upper and lower thresholds and state maintenance logic. The timestamp, judgment mark and three types of monitoring values ​​are encapsulated into a transmission frame according to fixed fields and sent to the monitoring device after verification. This overall design not only outputs physiological monitoring results, but also synchronously outputs fit drift status, so that the monitoring end can distinguish the business consequences brought about by "physiological changes" and "decoupling changes caused by fit drift", forming a deliverable data chain including calibration, decoupling and status marking. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0057] Figure 1 This is a structural diagram of an intracranial multimodal data monitoring system based on FBG grating technology proposed in this invention;

[0058] Figure 2 This is a flowchart of the decoupling calibration method in this invention;

[0059] Figure 3 This is a flowchart of the drift mode basis set and the decoupling drift factorization interface in the decoupling calibration method of the present invention;

[0060] Figure 4 This is a flowchart of the score diffusion latent variable inference and bonding drift confidence process in the decoupling calibration method of this invention.

[0061] Figure 5 This is a flowchart of the bonding drift confidence gating and current decoupling matrix update process in the decoupling calibration method of the present invention.

[0062] Figure 6 This is a flowchart of the decoupling calculation based on the current decoupling matrix in the decoupling calibration method of the present invention;

[0063] Figure 7 This is a comparative schematic diagram showing the decoupling misalignment caused by probe bonding drift and the recovery of stability through self-calibration in the decoupling calibration method of the present invention.

[0064] Figure 8 This is a schematic diagram of the decoupling matrix correction mechanism in the confined drift subspace of the decoupling calibration method of the present invention. Detailed Implementation

[0065] In Example 1, reference Figures 1 to 8 A multimodal intracranial data monitoring system based on FBG grating technology includes a monitoring device 1, a three-channel fiber Bragg grating 2, and a three-channel fiber Bragg grating probe 3. One end of the three-channel fiber Bragg grating 2 is connected to the monitoring device 1 via a monitoring port, and the other end is fitted with the three-channel fiber Bragg grating probe 3. Before monitoring, the monitoring device 1 implants the three-channel fiber Bragg grating probe 3 into the cranium for decoupling calibration, the specific method of which is as follows:

[0066] S1. Obtain the center wavelength timing of the three-channel fiber Bragg grating 2 implanted in the cranium, form the three-channel center wavelength timing, and receive the initial decoupling matrix obtained by one-time calibration;

[0067] S2. Read the drift mode basis set corresponding to the three-channel fiber Bragg grating probe 3, construct the decoupling drift factorization interface in combination with the initial decoupling matrix, and limit the dimension and range of drift coefficient values.

[0068] S3. Input the three-channel center wavelength time sequence into the score diffusion latent variable inference network. The scorediffusion latent variable inference network characterizes the probe fitting state change with cross-channel shared drift drive, and generates the fitting drift latent variable posterior distribution under continuous evolution constraint. Calculate the fitting drift confidence based on the fitting drift latent variable posterior distribution to obtain the fitting drift confidence.

[0069] S4. Determine the fit drift confidence level and the fit drift confidence level threshold. If the condition is met, map the posterior distribution of the latent variable of fit drift to the drift coefficient, and modify the initial decoupling matrix through the decoupling drift factorization interface to obtain the current decoupling matrix. If the condition is not met, use the preset benchmark drift coefficient to obtain the current decoupling matrix through the decoupling drift factorization interface.

[0070] S5. Based on the current decoupling matrix, decouple the timing of the three-channel center wavelength to obtain the intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value.

[0071] S6. Generate a fitting drift judgment mark based on the drift coefficient, and send the fitting drift judgment mark along with the intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value to the monitoring device 1.

[0072] In this embodiment, step S1 specifically includes:

[0073] After the three-channel fiber Bragg grating probe 3 is implanted intracranially, the three fiber optic ports of the probe 3 are connected to the corresponding channels of the grating demodulation device. Channel identifiers consistent with the fixed channel sequence are configured on the grating demodulation device side. The fixed channel sequence is denoted as follows: , , The grating demodulation device continuously scans and demodulates the three-channel reflection spectrum of each channel. Within each scan cycle, it acquires a discrete sampling sequence showing the amplitude variation of the reflection spectrum with wavelength. The sampling point with the maximum amplitude in the discrete sampling sequence is located as the peak point. A preset proportional threshold for the peak point's amplitude is used as a boundary criterion. The process extends sequentially from the peak point towards the short-wavelength and long-wavelength directions until a sampling point with an amplitude lower than the boundary criterion is encountered, thus obtaining the main reflection peak interval. Within the main reflection peak interval, a wavelength- and amplitude-weighted convergence is performed on each sampling point, with the weight taken as the amplitude. The convergence result is divided by the sum of the amplitude weights to obtain the center wavelength value. The center wavelength of the channel is denoted as ,Will The center wavelength of the channel is denoted as ,Will The center wavelength of the channel is denoted as ,in The timestamps corresponding to the scanning cycles are used to form the original timing sequence of the center wavelength of the three-channel fiber Bragg grating 2;

[0074] To meet the synchronous observation input requirements of the score diffusion latent variable inference network for cross-channel shared drift-driven synchronous observation, the original time series of the center wavelength of the three-channel fiber Bragg grating 2 were time-aligned according to a unified sampling time, denoted as _____. ,in For each channel, the sampling sequence number is incremented by time. ,exist The search timestamps on both sides satisfy A pair of adjacent demodulation results and and in When the preset maximum interval is not exceeded, adjust according to the time distance ratio. and Perform linear interpolation to obtain When only one-sided demodulation results exist or When the preset maximum interval is exceeded, the timestamp closest to the set timestamp is selected. The demodulation results are used as ,Will , , Three-dimensional observation vectors are formed according to a fixed channel sequence. , The three components are as follows: , , and will Arranged chronologically to form the center wavelength sequence of the three channels. ;

[0075] The initial decoupling matrix is ​​given in the one-time calibration phase, and the initial decoupling matrix is ​​denoted as . To ensure the decoupling drift factorization interface is compatible with The corrections are consistent with the fixed channel order of the three-channel center wavelength timing, and ensure that the parameter order of intracranial pressure monitoring values, intracranial temperature monitoring values, and cerebral oxygenation monitoring values ​​remains consistent in the decoupling calculation. The row and column indexes are subjected to consistency checks, with the parameter order fixed as intracranial pressure monitoring value, intracranial temperature monitoring value, and cerebral oxygenation monitoring value, and denoted as... , , The consistency verification reads the index mapping information generated during the one-time calibration phase. This index mapping information is used in the construction... The time is written by the calibration process, including Three columns and , , The correspondence, and Three lines and , , The correspondence, when any correspondence does not satisfy the fixed channel order and parameter order, for Perform row and column rearrangement, The column order is adjusted to , , and will The row order is adjusted to , , ,make It maintains consistency with the three-channel center wavelength timing, decoupling drift factorization interface, and decoupling calculation module under a unified indexing system.

[0076] In this embodiment, step S2 specifically includes:

[0077] In the one-time calibration phase, a drift mode basis set corresponding to the three-channel fiber Bragg grating probe 3 is constructed and stored as probe configuration parameters. During the one-time calibration phase, multiple sets of controlled bonding states or controlled loading states are set, and the corresponding calibration decoupling matrix is ​​obtained for each state. The calibration decoupling matrix is ​​denoted as... ,in To determine the state sequence number and read the initial decoupling matrix. For each set of states, calculate the matrix deviation, denoted as . , Each element is composed of Subtract corresponding element The corresponding elements are obtained, and multiple groups are obtained. A bias sample set is formed by superimposing state numbers. Principal component decomposition or singular value decomposition is performed on the bias sample set to obtain several principal component matrices. The principal component matrices are then sorted from high to low contribution, and the top ones are selected. The principal component matrices serve as the basis set for the drift modes, where To represent the number of drift mode matrices, the first... The drift patterns are denoted as the drift pattern matrix. ,in This is the drift mode number. It is a 3x3 matrix. Column order and fixed channel order , , Consistent, The order of the data is consistent with the parameter order of intracranial pressure monitoring values, intracranial temperature monitoring values, and cerebral oxygenation monitoring values, for each... Amplitude normalization is performed to ensure that different drift mode matrices participate in weighted combination at a uniform scale. When reading the drift mode basis set, it is loaded in the order of storage. to and will as drift coefficient vector The value dimensions, where For length is The vector, for The dimensional components, and One-to-one correspondence enables the score diffusion latent variable inference network to generate random noise vectors with... Determine the dimension of the random noise vector and ensure that the posterior distribution of the fitted drift latent variable has a consistent parameterization entry in the dimension of the drift coefficient values;

[0078] The initial decoupling matrix With drift mode matrix When accessing the decoupling drift factorization interface, the decoupling drift factorization interface reads the drift coefficient vector. Combine with the drift pattern matrix set and verify. Is the dimension equal to When the dimensions are consistent, the drift coefficients of each dimension are processed sequentially according to the order of the drift mode matrix. As the weight multiplied into the corresponding The weighted matrices are obtained, and the elements of all weighted matrices are added together one by one according to their positions to form the controlled perturbation matrix. ,in It is a 3x3 matrix. The Line 1 The column elements are obtained by summing the corresponding elements of all weighted matrices, and the controlled perturbation matrix is ​​then used. With the initial decoupling matrix By adding each element sequentially according to its position, a calibrable representation of the decoupling matrix is ​​obtained. This allows changes in probe contact state to be controlled and corrected only through the subspace spanned by the drift mode basis set, and enables the drift coefficients determined by S4 to act on the decoupling matrix with a clear matrix operation path.

[0079] To establish a range constraint for determining the drift coefficient, during the one-time calibration phase, based on each set of calibration states... Solve for the corresponding drift coefficient vector ,in For length is The vector is solved by minimizing the sum of squared differences between matrix elements. With each Expanded into a vector of length nine in a fixed order, constructing... The design matrix is ​​obtained by using the vector of drift mode elements as columns, and then solving it using normal equations or generalized inverse equations. The results obtained from all calibration states Calculate the maximum and minimum values ​​by dimension, and denote them as follows: and ,in For the first Upper limit of dimensional drift coefficient, For the first The lower limit of the dimension drift coefficient, and will and Write the drift coefficient value range configuration so that S4 can apply upper and lower bound constraints to the drift coefficient for each dimension when determining the drift coefficient from the posterior distribution of the drift latent variable.

[0080] In this embodiment, step S3 specifically includes:

[0081] Timing of the center wavelength of the three channels As an online observation of the score diffusion latent variable inference network, Three-dimensional observation vectors arranged according to uniform sampling time constitute, The three components are the center wavelength scalars of the three channels, and the values ​​of the drift coefficients are read from the dimension. , Consistent with the number of drift pattern matrices, a pseudo-random number generator is used to generate a dimension of [missing value] at the start of each inference. random noise vector , Each dimension is sampled from a zero-mean, unit-variance distribution to read a preset diffuse noise level sequence. , Arranged from highest to lowest noise intensity, the noise level index is denoted as follows: , Taken from The valid set of indexes, indexing noise levels. The input noise level coding layer uses a trainable embedding table to... Mapped to noise level coding vector The noise level encoding vector is used to characterize the noise intensity level of a random noise vector in the diffusion denoising iteration;

[0082] Timing of the center wavelength of the three channels The input is a cross-channel shared drift driving layer, which uses multiple one-dimensional convolutional neurons to extract cross-channel common slowly varying features along the time axis. Each convolutional layer uses three channels simultaneously as input channels, mapping the joint changes of the three channels at multiple adjacent sampling times to a convolutional feature sequence. Non-linear activation neurons are set between convolutional layers, and a convolution with a stride of one is used to maintain temporal resolution. Temporal convergence is performed on the feature sequence of the last convolutional layer. Temporal convergence is achieved by averaging the features across the entire time period or by performing a moving average according to a preset window length to obtain the shared drift representation vector. The shared drift representation vector is used to represent the components that appear together in the three channels and change slowly over time;

[0083] Timing of the center wavelength of the three channels Each of the three input channels is a branch layer, and each branch layer corresponds to a fixed channel order. , , Each channel branch layer reads only the one-dimensional center wavelength time sequence of the corresponding channel, and uses stacked independent one-dimensional convolutional neurons to extract channel difference features, thus obtaining the channel difference representation vector. , , The convolution parameters of the channel branch layers are not shared, allowing the three channels to exhibit different responses through... , , Reserved;

[0084] Shared drift representation vector The difference representation vectors from the three channels are concatenated in a fixed channel order, and the resulting fused vector is denoted as... , will fuse vector The input is a fusion mapping layer, which uses fully connected neurons to map the fusion vector into a conditional feature vector. The conditional feature vector is aligned with the sampling time. It serves only as a condition for fitting the drift latent variable inference and is not used as a regression target for intracranial pressure, intracranial temperature, and cerebral oxygenation monitoring values. The conditional feature vector is input into the continuous evolution constraint layer. This layer calculates the change amplitude of the conditional feature vector at adjacent sampling times. The change amplitude is calculated by taking the absolute value of the difference between each dimension of the conditional feature vector at two adjacent times and summing them to obtain a scalar change. Then, the average of multiple consecutive scalar changes over a preset window length is used to obtain the evolutionary consistency feature. Evolutionary consistency features serve as constraint information to fit the temporal continuity of drift latent variables;

[0085] The random noise vector, conditional feature vector, and noise level encoding vector are input into the scoring layer. The random noise vector is encoded in terms of noise level. The corresponding iteration step is denoted as Scoring layer , and The concatenated vector is used as input, and a score vector is generated using multilayer fully connected neurons and residual connections. The dimension of the score vector and Consistent, diffuse posterior generation mechanism according to noise level sequence Perform multi-level denoising iterations from high noise to low noise, and in each level, base the results on the score vector. Denoising and updating are performed, and a random perturbation matching the current noise level is injected after the update according to a preset noise schedule. This process is repeated a preset number of times at each sampling time to obtain a sample vector that fits the drift latent variable. ,in The sampling sequence number is used to assign the sampling time to each sampling time. Arrange the samples according to time to form a sequence of samples that fit the drift latent variables;

[0086] Statistical aggregation is performed on the latent variable sample vectors of the fit drift at each sampling time, and the arithmetic mean of the sample vectors in each dimension is calculated to obtain the posterior center vector. The posterior dispersion vector is obtained by calculating the mean absolute deviation of the sample vector relative to the posterior center vector in each dimension. The dimensions of the posterior center vector and the posterior discreteness vector are both... The posterior discreteness vector and evolutionary consistency features are input into the confidence parameter layer. The confidence parameter layer uses a linear combination of dimension-wise monotonically compressed posterior discreteness vector and monotonically compressed evolutionary consistency features, and maps it to fit drift confidence through a logistic function. Fit drift confidence is calculated according to the confidence function.

[0087] ;

[0088] in, Sampling time The reliability of the fit drift It is an exponential function. The bias parameters for the credibility parameter layer. The weight parameters are the features corresponding to evolutionary consistency. For the posterior discreteness vector, the first... The corresponding weight parameters for each dimension Sampling time The evolutionary consistency characteristic scalar, Sampling time posterior discreteness vector The dimensional components, The dimension for the drift coefficient values. This is the index of the components of the posterior discreteness vector.

[0089] In this embodiment, step S4 specifically includes:

[0090] At each sampling time Read the bonding drift confidence With Adhesion Drift Confidence Threshold ,right and The confidence level is determined by comparing the magnitudes. The confidence level is used to control the triggering constraints of the drift coefficient determination path, so that the drift coefficient is determined only by the posterior distribution of the latent variable of the drift when the confidence level of the drift meets the threshold condition.

[0091] When the confidence determination result meets the confidence threshold condition for fit drift, the posterior center vector is read. With the posterior discreteness vector ,in and The dimensions are all the dimensions of the drift coefficient values. Mapping the posterior discreteness vector to an uncertainty weight vector The first uncertain weight vector dimension It is obtained by monotonically decreasing compression, and the compression method is to... Adding this to a constant and taking the reciprocal gives a larger weight to the dimension with smaller dispersion. Used for the posterior central vector By performing dimension-wise weighting, we obtain the drift candidate vector. Read the upper limit of each dimension of the drift coefficient. and lower limit ,in and All are of length The vector, based on and right Dimensional pruning yields range-constrained drift candidate vectors. The cropping rule is to take the upper limit when the corresponding dimension is greater than the upper limit, and to take the lower limit when the corresponding dimension is less than the lower limit;

[0092] Reading evolutionary consistency features Evolutionary consistency features are scalar values ​​output by the continuous evolution constraint layer, and these features are mapped to continuous weights. The continuity weights are obtained by monotonically decreasing compression, and the compression method is to... Add the result to a constant and take the reciprocal, then divide the compressed result into intervals. Apply range restrictions, the range restriction rule is that the value is less than Time to take The value is greater than Time to take Read the preset baseline drift coefficient The preset baseline drift coefficient is the length of The vector, based on continuous weights and Determine the drift coefficient by performing dimension-wise weighted fusion. When the credibility determination result does not meet the credibility threshold condition for fit drift, the following is adopted: Determine the drift coefficient Drift coefficient Calculations are performed using coefficient functions:

[0093] ;

[0094] in, Sampling time The drift coefficient vector, To standardize sampling times, Sampling time The reliability of the fit drift To match the drift confidence threshold, Sampling time Continuous weight scalar, Sampling time The posterior center vector, Sampling time Uncertainty weight vector, for The dimensional components, The scalar normalization factor is obtained by summing the components of the uncertainty weight vector. For dimension-wise multiplication operations, This is the lower bound vector for the drift coefficient. This is the upper limit vector of the drift coefficients. As a preset baseline drift coefficient vector, To perform the operation by taking the smaller value in each dimension, To perform the operation by taking the larger value in each dimension, The dimension for the drift coefficient values. Index for the drift coefficient dimension;

[0095] drift coefficient With drift mode matrix to The input decoupled drift factorization interface generates a controlled perturbation matrix by weighting and combining each dimension of the drift coefficient with the corresponding drift mode matrix. The weighted combination adopts the first Dimensional drift coefficient and the 1st Each drift mode matrix undergoes element-wise matrix multiplication to obtain a weighted matrix. Then, all weighted matrices are summed element-wise according to their positions to obtain the controlled perturbation matrix. To confine the controlled perturbation matrix within the subspace spanned by the drift mode matrices, during the one-time calibration phase... to Expand the matrix element by element into a basis vector of length nine and perform orthogonal normalization to obtain the subspace basis matrix. During the online computation phase, the controlled perturbation matrix is ​​expanded into a vector of length nine in the order of identical elements and projected onto the subspace basis matrix. Within Zhang Cheng's space, the projection results are then reconstructed into a controlled perturbation matrix in element-wise order, and the initial decoupling matrix is ​​read. The controlled perturbation matrix The current decoupling matrix is ​​formed by adding the elements of the initial decoupling matrix element by element. .

[0096] In this embodiment, step S5 specifically includes:

[0097] At each sampling time Read the center wavelength observation vector of the three channels The three-channel center wavelength observation vectors are arranged in a fixed channel order. , , Arrangement, in which For the first channel sign, For the second channel sign, This is the identifier for the third channel. The three components of the center wavelength observation vector for the three channels are as follows: , , , , , For the grating demodulation equipment at the sampling time The output center wavelength scalar value is used to read the reference center wavelength vector determined during the one-time calibration phase. The three components of the reference center wavelength vector are respectively , , The reference center wavelength vector is obtained by time convergence of the three channel center wavelengths during the one-time calibration stage under the reference bonding state, and is bound and stored with the three-channel fiber Bragg grating probe 3. Reference difference is performed on the three-channel center wavelength observation vector, and calculation is performed separately for each channel. ,in The channel number has a value of , , ,Will , , The center wavelength variation vector is formed according to a fixed channel sequence. The center wavelength change vector is used to characterize the amount of center wavelength change relative to the reference fitting state;

[0098] Read the current decoupling matrix The current decoupling matrix is ​​a 3x3 matrix. The column order of the current decoupling matrix is ​​consistent with the fixed channel order, and the row order of the current decoupling matrix is ​​consistent with the parameter order of intracranial pressure monitoring values, intracranial temperature monitoring values, and cerebral oxygenation monitoring values. A decoupling matrix mapping operation is performed between the center wavelength change vector and the current decoupling matrix, using matrix-vector multiplication. Mapped to a three-parameter monitoring vector To clarify the multiplication-addition calculation process, the current decoupling matrix is... Line 1 Column elements are denoted as ,in The parameter row index and its value is , , , The channel column is indexed and its value is , , The third parameter of the three-parameter monitoring vector Each component is composed of , , The summation of the three products yields the three-parameter monitoring vector, which is a three-dimensional column vector. The third parameter of the three-parameter monitoring vector... Wei Jiwei And corresponding to the intracranial pressure monitoring value, the first Wei Jiwei And corresponding to the intracranial temperature monitoring value, the first Wei Jiwei And corresponding to the brain oxygenation monitoring values, so that the three-parameter monitoring vectors are arranged strictly in the order of the parameters;

[0099] The three-parameter monitoring vectors at each sampling time are written into the three-parameter monitoring sequence in chronological order. And extract three monitoring time series from the three-parameter monitoring sequence according to the component position, and then extract them from all the time series. The sequence formed by arranging values ​​according to time is defined as the intracranial pressure monitoring value time series, which will consist of all The sequence formed by arranging values ​​according to time is defined as the time series of intracranial temperature monitoring values, which will consist of all The sequence formed by arranging values ​​according to time is defined as the brain oxygenation monitoring time series. To ensure that the monitoring time series is consistent with the decoupling relationship, the decoupling matrix mapping operation is performed before... Matrix dimensions and Consistency checks are performed on the vector dimension, and the consistency checks include The number of columns is three and The three-dimensional verification, and The number of rows is three and The consistency check is performed in three dimensions, and the sampling time is skipped if the consistency check fails. The three-parameter monitoring vector is written to avoid generating monitoring values ​​that do not correspond to the parameter order.

[0100] In this embodiment, step S6 specifically includes:

[0101] At sampling time Read the drift coefficient Drift mode matrix sensitivity weight vector Adhesion drift upper threshold , Adhesion drift threshold With sampling time Adhesion drift judgment mark Drift coefficient For length is The vector, For the first dimensional drift coefficient scalar, The dimension index for the drift coefficient and its value is... to Drift mode matrix sensitivity weight vector For length is non-negative vectors, For the first Sensitivity weight scalar, sensitivity weight vector The data is obtained offline during the one-time calibration phase. The offline acquisition process is as follows: For the first... A drift pattern matrix The absolute values ​​of the nine matrix elements are taken and summed to obtain the contribution scalar. Then, normalization is performed on all contribution scalars to obtain... ,in It is a 3x3 matrix;

[0102] For drift coefficient Taking absolute value from one dimension and according to the sensitivity weight vector Weighted convergence yields the fit drift amplitude index. Fit the drift amplitude index The calculation process is as follows: calculate the weighting magnitude for each dimension. ,right to The scalar is obtained by summing the weighted magnitudes. The drift amplitude index will be aligned. With the adhesion drift upper threshold , Adhesion drift threshold Compare and update the fit drift judgment markers Adhesion drift detection mark Values This indicates an abnormal bonding drift; bonding drift detection flag. Values This indicates that the fit and drift are normal. When Set as ,when When Set as ,when When Set as This ensures that the determination within the threshold range remains stable.

[0103] At sampling time Generate monitoring transmission frame At that time, the sampling time corresponding to the intracranial pressure monitoring value is read. numerical value The time sequence of intracranial temperature monitoring values ​​corresponds to the sampling time. numerical value Brain oxygenation monitoring values ​​time sequence corresponding to sampling time numerical value Concatenate monitoring and sending frames according to a fixed field order. The fixed field order is as follows: frame header field, sampling time field, fit drift judgment mark field, intracranial pressure monitoring value field, intracranial temperature monitoring value field, cerebral oxygenation monitoring value field, and verification field. The sampling time field uses an integer millisecond timestamp encoding, the fit drift judgment mark field uses a single-byte encoding, and the three monitoring value fields use... Encoding and verification fields use Calculations show that The calculation process is as follows: Set the initial register to hexadecimal. The byte stream from the frame header field to the brain oxygenation monitoring value field is XORed byte by byte and then right-shifted eight times in a loop with a polynomial hexadecimal. The bitwise XOR update outputs the low and high bytes of the register as a check field, and arranges the monitoring transmission frames at all sampling times in the order of sampling time to form the monitoring transmission frame timing. , A time-ordered sequence of frames;

[0104] The timing of the monitoring transmission frame is sent using the communication interface of monitoring device 1. At that time, the monitoring device 1 communication interface will send monitoring frames. The data is converted into a byte stream and written to the transmission buffer in the order of sampling time. The monitoring device 1's communication interface sequentially transmits the byte stream in the transmission buffer according to the transmission rate configured for the communication link, ensuring the sampling time... Adhesion drift judgment mark Intracranial pressure monitoring values Intracranial temperature monitoring values Brain oxygenation monitoring values The data is received in a fixed field order in the monitoring device 1's data receiving channel.

[0105] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multimodal intracranial data monitoring system based on FBG grating technology, characterized in that, The system includes a monitoring device, a three-channel fiber Bragg grating, and a three-channel fiber Bragg grating probe. One end of the three-channel fiber Bragg grating is connected to the monitoring device via a monitoring port, and the other end is fitted with the three-channel fiber Bragg grating probe. Before monitoring, the three-channel fiber Bragg grating probe is implanted intracranially for decoupling calibration, as detailed below: S1. Obtain the center wavelength timing of the three-channel fiber Bragg grating implanted in the brain, form the three-channel center wavelength timing, and receive the initial decoupling matrix obtained from one calibration. S2. Read the drift mode basis set corresponding to the three-channel fiber Bragg grating probe, construct the decoupling drift factorization interface in combination with the initial decoupling matrix, and limit the dimension and range of drift coefficient values. S3. Input the three-channel center wavelength time sequence into the score diffusion latent variable inference network. The score diffusion latent variable inference network characterizes the change of probe fitting state by cross-channel shared drift drive, and generates the posterior distribution of fitting drift latent variable under continuous evolution constraint. Calculate the fitting drift confidence based on the posterior distribution of fitting drift latent variable to obtain the fitting drift confidence. S4. Determine the fit drift confidence level and the fit drift confidence level threshold. If the condition is met, map the posterior distribution of the latent variable of fit drift to the drift coefficient, and modify the initial decoupling matrix through the decoupling drift factorization interface to obtain the current decoupling matrix. If the condition is not met, use the preset benchmark drift coefficient to obtain the current decoupling matrix through the decoupling drift factorization interface. S5. Based on the current decoupling matrix, decouple the timing of the three-channel center wavelength to obtain the intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value. S6. Generate a fit drift judgment mark based on the drift coefficient, and send the fit drift judgment mark along with the intracranial pressure monitoring value, intracranial temperature monitoring value, and cerebral oxygenation monitoring value to the monitoring equipment.

2. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S1 specifically involves: The three-channel fiber Bragg grating probe implanted in the skull is connected to the grating demodulation device to continuously demodulate the three-channel reflection spectrum and obtain the original timing sequence of the center wavelength of the three-channel fiber Bragg grating. The original timing sequence of the center wavelength of the three-channel fiber Bragg grating is time-aligned according to a unified sampling time. Three center wavelength scalars are extracted at each sampling time and formed into a three-dimensional observation vector in a fixed channel order. The three-dimensional observation vector is then arranged in time to form the three-channel center wavelength timing sequence. The system receives the initial decoupling matrix obtained from a one-time calibration, performs a consistency check between the row and column indices of the initial decoupling matrix and the fixed channel order, as well as the parameter order of intracranial pressure monitoring values, intracranial temperature monitoring values, and brain oxygenation monitoring values. If the check is inconsistent, the initial decoupling matrix is ​​rearranged to meet the consistency constraints.

3. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S2 specifically involves: Read the drift mode basis set corresponding to the three-channel fiber Bragg grating probe, represent each drift mode as a three-row, three-column drift mode matrix consistent with the fixed channel order, and count the number of drift mode matrices to determine the dimension of the drift coefficient values; The initial decoupling matrix and the drift mode matrix are connected to the decoupling drift factorization interface. In the decoupling drift factorization interface, the drift coefficients are weighted and combined according to the order of the drift mode matrix to generate a controlled perturbation matrix. The controlled perturbation matrix is ​​then added to the initial decoupling matrix to form a calibrable expression of the decoupling matrix. Based on the range of drift coefficient values ​​obtained from a one-time calibration, upper and lower limits are set for each dimension of the drift coefficient, and these upper and lower limits are used as constraints on the range of drift coefficient values ​​when determining the drift coefficient.

4. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S3 specifically involves: A random noise vector is generated based on the dimension of the drift coefficient, and a noise level index is generated according to the preset diffusion noise level sequence. The noise level index is then input into the noise level coding layer to obtain the noise level coding vector. The three-channel center wavelength time sequence is input into the cross-channel shared drift driving layer. Multi-layer one-dimensional convolutional neurons are used to extract the cross-channel common slowly varying features, and the convolutional features are converged in time to obtain the shared drift representation vector. The time series of the three center wavelengths are input into the three channel branch layers respectively, and the channel difference features are extracted by stacking independent one-dimensional convolutional neurons to obtain the three channel difference representation vectors. The shared drift representation vector and the three channel difference representation vectors are concatenated and fused in channel order, and the conditional feature vector is obtained through the fusion mapping layer. The conditional feature vector is only used as the inference condition for the latent variable of drift and is not used as the regression target for intracranial pressure monitoring value, intracranial temperature monitoring value and brain oxygenation monitoring value. The conditional feature vector is input into the continuous evolution constraint layer. Evolutionary consistency features are generated based on the change amplitude of the conditional feature vector at adjacent time points. The evolutionary consistency features are used as constraint information to fit the temporal continuity of the drift latent variables. The random noise vector, conditional feature vector, and noise level encoding vector are input into the scoring layer. Multi-layer fully connected neurons and residual connections are used to generate the score vector. Based on the score vector-driven diffusion-type posterior generation mechanism, the random noise vector is subjected to multi-level denoising iteration to obtain a sample sequence of drifting latent variables. The posterior center vector and posterior dispersion vector are obtained by statistically aggregating the latent variable sample sequences of the fit drift. The fit drift confidence is obtained by confidence mapping based on the posterior dispersion vector and the evolutionary consistency feature.

5. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 4, characterized in that, When obtaining the fit drift confidence score by mapping confidence score based on the posterior discreteness vector and evolutionary consistency features, a confidence function is used to map the posterior discreteness vector and evolutionary consistency features. Specifically, the confidence function is: ； in, Sampling time The reliability of the fit drift It is an exponential function. The bias parameters for the credibility parameter layer. The weight parameters are the features corresponding to evolutionary consistency. For the posterior discreteness vector, the first... The corresponding weight parameters of the dimension Sampling time The evolutionary consistency characteristic scalar, Sampling time posterior discreteness vector The dimensional components, The dimension for the drift coefficient values. This is the index of the components of the posterior discreteness vector.

6. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S4 specifically involves: The confidence level of the bonding drift is compared with the confidence threshold of the bonding drift to obtain the confidence level judgment result; When the credibility determination result meets the condition of fitting the drift credibility threshold, the posterior center vector and the posterior dispersion vector are read, the posterior dispersion vector is mapped to the uncertainty weight vector, and the posterior center vector is weighted dimension by dimension to obtain the drift candidate vector. The drift candidate vectors are checked for consistency in the dimensions of their values, and the drift candidate vectors are pruned dimension by dimension according to the upper and lower limits of each dimension of the drift coefficient to obtain range-constrained drift candidate vectors. Read the evolutionary consistency features, map the evolutionary consistency features to continuous weights, and perform weighted fusion of the range-constrained drift candidate vectors and the preset benchmark drift coefficients based on the continuous weights to determine the drift coefficients; When the confidence determination result does not meet the confidence threshold condition for the fit drift, the drift coefficient is determined by using a preset benchmark drift coefficient; The drift coefficients and drift mode matrix are input into the decoupled drift factorization interface. The controlled perturbation matrix is ​​generated by weighted combination of each dimension of the drift coefficients and the corresponding drift mode matrix. The controlled perturbation matrix is ​​then restricted to the subspace spanned by the drift mode matrix. The controlled disturbance matrix is ​​added to the initial decoupling matrix to form the current decoupling matrix.

7. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 6, characterized in that, After reading evolutionary consistency features and mapping them to continuous weights, and when the confidence determination result meets the drift confidence threshold condition, the posterior discreteness vector is mapped to an uncertainty weight vector, and the posterior center vector is weighted dimension-wise. Based on obtaining range-constrained drift candidate vectors by dimension-wise pruning of the drift candidate vectors according to the upper and lower limits of each dimension of the drift coefficient, a coefficient function is used to weight and fuse the range-constrained drift candidate vectors with a preset benchmark drift coefficient to determine the drift coefficient. When the confidence determination result does not meet the drift confidence threshold condition, the preset benchmark drift coefficient is used to determine the drift coefficient. The specific coefficient function is as follows: ； in, Sampling time The drift coefficient vector, To standardize sampling times, Sampling time The reliability of the fit drift To match the drift confidence threshold, Sampling time Continuous weight scalar, Sampling time The posterior center vector, Sampling time Uncertainty weight vector, for The dimensional components, The scalar normalization factor is obtained by summing the components of the uncertainty weight vector. For dimension-wise multiplication operations, This is the lower bound vector for the drift coefficient. This is the upper limit vector of the drift coefficients. As a preset baseline drift coefficient vector, To perform the operation by taking the smaller value in each dimension, To perform the operation by taking the larger value in each dimension, The dimension for the drift coefficient values. This is the dimension index for the drift coefficient.

8. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S5 specifically involves: At each sampling moment, the three center wavelength scalars of the three-channel center wavelength timing are combined into a center wavelength observation vector in a fixed channel order, and the reference center wavelength vector determined in the one-time calibration stage is read. The center wavelength change vector is obtained by performing reference difference on the center wavelength observation vector. The center wavelength change vector is mapped to the current decoupling matrix to obtain a three-parameter monitoring vector arranged in the order of intracranial pressure monitoring value, intracranial temperature monitoring value, and brain oxygenation monitoring value. The three-parameter monitoring vectors at each sampling time are arranged in chronological order, and the three components of the three-parameter monitoring vectors are respectively used to form the time series of intracranial pressure monitoring values, intracranial temperature monitoring values, and cerebral oxygenation monitoring values.

9. The intracranial multimodal data monitoring system based on FBG grating technology according to claim 1, characterized in that, Step S6 specifically involves: Read the drift coefficient, drift mode matrix sensitivity weight, upper threshold of fit drift, lower threshold of fit drift and fit drift judgment flag of the previous sampling time. Take the absolute value of the drift coefficient dimension by dimension and aggregate it according to the sensitivity weight to obtain the fit drift amplitude index. When the fit drift amplitude index is greater than the upper threshold of fit drift, update the fit drift judgment flag as fit drift abnormal. When the fit drift amplitude index is less than the lower threshold of fit drift, update the fit drift judgment flag as fit drift normal. The sampling time, the fit drift judgment mark, the value of the intracranial pressure monitoring value corresponding to the sampling time, the value of the intracranial temperature monitoring value corresponding to the sampling time, and the value of the brain oxygenation monitoring value corresponding to the sampling time are concatenated in a fixed field order to generate the monitoring transmission frame timing. The monitoring equipment communication interface is used to send the monitoring transmission frame timing sequence to the monitoring equipment in the order of sampling time.

Images