A method for detecting local brain tissue blood oxygen saturation

Abstract

The present invention relates to the technical field of blood oxygen saturation, and specifically to a method for detecting local brain tissue blood oxygen saturation, comprising the following steps: S1, based on the original spectral data of the brain tissue, using high-pass and low-pass filter methods to perform signal preprocessing, remove low-frequency and high-frequency noise in the signal, and use discrete Fourier transform and wavelet transform to perform frequency domain analysis and filtering processing to generate preprocessed spectral data. Based on the original spectral data of the brain tissue, the present invention uses high-pass and low-pass filters to perform signal preprocessing, effectively removes low-frequency and high-frequency noise, and uses discrete Fourier transform and wavelet transform to perform frequency domain analysis and filtering processing to generate high-quality preprocessed spectral data.

Description

Technical Field

[0001] The present invention relates to the technical field of blood oxygen saturation, and in particular to a method for detecting local brain tissue blood oxygen saturation. Background Art

[0002] The field of blood oxygen saturation technology involves the measurement and monitoring of oxygen content in the blood, mainly used to evaluate the individual's respiratory function and blood circulation status. This technology is widely used in the medical field, especially in critical care, anesthesia management and chronic disease monitoring. Through non-invasive means such as pulse oximeters or spectrometers, blood oxygen levels are monitored in real time to provide important physiological parameters. The technical field covers sensor design, data acquisition and processing, signal filtering and correction, multimodal data fusion and other aspects, aiming to improve the accuracy and reliability of blood oxygen saturation measurement and provide reliable data support for clinical diagnosis and treatment.

[0003] However, traditional methods have obvious deficiencies in signal processing and feature extraction. First, traditional signal preprocessing methods have limited filtering effects on low-frequency and high-frequency noise, resulting in more residual noise in the signal, affecting the accuracy of subsequent analysis. Secondly, baseline drift correction is insufficient, and fixed algorithms are often used, which cannot be adjusted dynamically, resulting in baseline drift affecting the reliability of the results. In the process of signal denoising, traditional methods are insufficient in processing complex noise, and the signal-to-noise ratio is not fully optimized. These shortcomings lead to poor results of traditional methods in practical applications, limiting their application in high-precision medical testing. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for detecting local brain tissue blood oxygen saturation.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme, a method for detecting local brain tissue blood oxygen saturation, comprising the following steps:

[0006] S1, based on the original spectral data of brain tissue, high-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering to generate preprocessed spectral data;

[0007] S2, based on the preprocessed spectral data, using an adaptive baseline correction algorithm to perform baseline drift correction, dynamically adjust the signal baseline level, and generate corrected spectral data;

[0008] S3, based on the corrected spectral data, using a Kalman filter to perform signal denoising processing, optimize the signal-to-noise ratio of the signal, and generate denoised spectral data;

[0009] S4, based on the denoised spectral data, extracting blood oxygen saturation correlation features by using principal component analysis and independent component analysis methods to generate blood oxygen saturation feature data;

[0010] S5, based on the blood oxygen saturation characteristic data, using support vector regression and random forest method to predict and estimate the rScO2 value to generate the rScO2 value;

[0011] S6, based on the rScO2 value and in combination with the electrical impedance data, a finite element analysis method is used to calculate the current distribution, reconstruct the electrical impedance image of the brain tissue, and generate an electrical impedance image;

[0012] S7, based on the electrical impedance image and the denoised spectral data, Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion, calculate the blood oxygen saturation of local brain tissue, and generate a local brain tissue blood oxygen saturation image.

[0013] Furthermore, the preprocessed spectral data includes spectral data from which low-frequency noise and high-frequency noise have been removed, the corrected spectral data specifically refers to spectral signals from which baseline has been corrected, the denoised spectral data includes spectral data from which random noise and system noise have been removed, the blood oxygen saturation characteristic data includes principal component characteristics and independent component characteristics, the rScO2 value includes a blood oxygen saturation value predicted and estimated by regression analysis and a machine learning algorithm, the electrical impedance image includes a current distribution map and an electrical impedance value map, and the local brain tissue blood oxygen saturation image includes a fused blood oxygen saturation distribution map and a numerical map.

[0014] Furthermore, high-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering. The steps for generating preprocessed spectral data are as follows:

[0015] S101, based on the original spectral data of the brain tissue, use a high-pass filter to process the signal, set a high-frequency threshold, extract and retain the high-frequency components of the signal, remove the frequency components below the threshold, and reconstruct the signal to generate spectral data after high-pass filtering;

[0016] S102, based on the spectral data after high-pass filtering, use a low-pass filter to process the signal, set a low-frequency threshold, extract and retain the low-frequency components of the signal, remove the frequency components higher than the threshold, and reconstruct the signal to generate spectral data after low-pass filtering;

[0017] S103, based on the spectral data after low-pass filtering, use a frequency domain analysis tool to perform spectrum transformation, decompose the various frequency components of the signal, and perform wavelet transformation to identify and separate noise components, reconstruct the signal, and generate pre-processed spectral data.

[0018] Furthermore, an adaptive baseline correction algorithm is used to correct the baseline drift and dynamically adjust the signal baseline level. The steps for generating corrected spectral data are as follows:

[0019] S201, based on the preprocessed spectral data, extract the baseline part of the spectrum, perform data segmentation, divide the spectral data into multiple interval sections, calculate the minimum value of each section, obtain the minimum value data of multiple intervals, merge the minimum value data of multiple intervals, and generate initial baseline data;

[0020] S202, adjusting the baseline level of the spectral data based on the initial baseline data, calculating the baseline offset of each section by matching the minimum value data of multiple intervals, and performing baseline adjustment, performing offset correction on the spectral data of multiple intervals, and generating adjusted baseline data;

[0021] S203 , based on the adjusted baseline data, applying the adjusted baseline level to the entire spectral data, correcting the baseline drift of the segment, merging the adjusted data of the segment, and generating corrected spectral data.

[0022] Furthermore, a Kalman filter is used to perform signal denoising and optimize the signal-to-noise ratio of the signal. The steps for generating denoised spectral data are as follows:

[0023] S301, based on the corrected spectral data, using a Kalman filter to process the signal, initialize filter parameters, perform state estimation and measurement update, recursively process data points, and perform noise reduction to generate spectral data after Kalman filtering;

[0024] S302, based on the spectral data after Kalman filtering, using adaptive noise reduction to process the signal, setting adaptive parameters, analyzing the noise characteristics in the signal, and adjusting the parameters to match the noise changes, gradually reducing the noise components, and generating spectral data after adaptive noise reduction;

[0025] S303, based on the spectral data after the adaptive noise reduction, optimizing the signal-to-noise ratio of the signal, performing signal smoothing processing, eliminating residual noise, and generating denoised spectral data.

[0026] Furthermore, the principal component analysis and independent component analysis methods are used to extract the blood oxygen saturation correlation features, and the steps of generating blood oxygen saturation feature data are as follows:

[0027] S401, performing data centering processing based on the denoised spectral data, subtracting the average value of the data from each data point to obtain the centralized data, performing orthogonal transformation, and performing orthogonal projection on the centralized data to generate principal component data;

[0028] S402, performing data independence processing based on the principal component data, using signal separation technology to separate each principal component to obtain independent signal data, extracting signals associated with blood oxygen saturation, and generating independent component data;

[0029] S403, performing a feature extraction operation based on the independent component data, analyzing the characteristic values ​​of each independent signal, calculating characteristic parameters highly correlated with the blood oxygen saturation, selecting correlated features, and generating blood oxygen saturation characteristic data.

[0030] Furthermore, support vector regression and random forest methods were used to predict and estimate the rScO2 value. The steps for generating the rScO2 value were as follows:

[0031] S501, based on the blood oxygen saturation characteristic data, using a data preprocessing step, performing data cleaning and normalization processing, then removing outliers and performing linear interpolation, and performing format conversion to generate preprocessed characteristic data;

[0032] S502, based on the preprocessed feature data, training a support vector regression model, setting model parameters, dividing the data set into a training set and a test set, training the model and optimizing the parameters, and performing cross-validation to generate a support vector regression model;

[0033] S503, based on the support vector regression model, training the random forest model, setting the number and depth of trees, evaluating the importance of features and adjusting model parameters, combining the support vector regression model to predict and estimate the rScO2 value, and verifying the results to generate the rScO2 value.

[0034] Furthermore, the finite element analysis method is used to calculate the current distribution and reconstruct the electrical impedance image of the brain tissue in combination with the electrical impedance data. The specific steps of generating the electrical impedance image are as follows:

[0035] S601, based on the rScO2 value, combined with the electrical impedance data, data integration is performed, the rScO2 value is used to match the electrical impedance data, the rScO2 value and the electrical impedance data are matched according to the position, and a preliminary model of brain tissue electrical impedance is constructed to generate a preliminary electrical impedance model;

[0036] S602, based on the preliminary electrical impedance model, current distribution calculation is performed, current distribution of each point of the model is determined by finite element analysis, the positions of the current source and the measurement point are adjusted by point-by-point analysis, data is collected, and current distribution data is generated;

[0037] S603, performing electrical impedance image reconstruction based on the current distribution data, using the current distribution data of each point to perform image reconstruction, applying the current distribution data to the brain tissue model, calculating the electrical impedance value of each area, and generating an electrical impedance image.

[0038] Furthermore, Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion to calculate the blood oxygen saturation of local brain tissue. The steps for generating a local brain tissue blood oxygen saturation image are as follows:

[0039] S701, based on the electrical impedance image and the denoised spectral data, Kalman filtering is used to perform data fusion, and each data point is recursively processed and state estimation and measurement update are performed, and noise reduction is performed to generate Kalman filtering fusion data;

[0040] S702, based on the Kalman filter fusion data, using the Bayesian estimation method to perform probability distribution calculation, perform data fusion by updating prior and posterior probabilities, and perform uncertainty assessment to generate Bayesian estimation fusion data;

[0041] S703, based on the Bayesian estimation fusion data, calculate the local brain tissue blood oxygen saturation, generate a two-dimensional image through interpolation and mapping, and perform pseudo-color coding for data visualization to generate a local brain tissue blood oxygen saturation image.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] The present invention is based on the original spectral data of brain tissue, uses high-pass and low-pass filters to perform signal preprocessing, effectively removes low-frequency and high-frequency noise, and performs frequency domain analysis and filtering processing through discrete Fourier transform and wavelet transform to generate high-quality preprocessed spectral data. Further, an adaptive baseline correction algorithm is used to dynamically adjust the signal baseline level, correct baseline drift, and ensure the consistency and accuracy of the data. The Kalman filter is used for signal denoising, optimizes the signal-to-noise ratio of the signal, and ensures the purity of the data. In the feature extraction stage, principal component analysis and independent component analysis are combined to extract features associated with blood oxygen saturation to generate efficient feature data. Support vector regression and random forest methods are used to predict and estimate rScO2 values ​​to obtain accurate rScO2 values. Combined with electrical impedance data, the finite element analysis method is used to calculate the current distribution, reconstruct the electrical impedance image of brain tissue, and achieve fine imaging of brain tissue. Finally, multimodal data fusion is performed through Kalman filtering and Bayesian estimation methods, the blood oxygen saturation of local brain tissue is calculated, and an accurate blood oxygen saturation image is generated. The above improvements make data processing more accurate, feature extraction more effective, blood oxygen saturation monitoring more reliable and intuitive, and improve the overall detection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] In the description of the present invention, it should be understood that the terms length, width, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc. indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of multiple is two or more, unless otherwise clearly and specifically defined. Example

[0047] See also Figure 1 The present invention provides a technical solution, a method for detecting local brain tissue blood oxygen saturation, comprising the following steps:

[0048] S1, based on the original spectral data of brain tissue, high-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering to generate preprocessed spectral data;

[0049] In the signal preprocessing step, based on the original spectral data of brain tissue, high-pass and low-pass filter methods are used for signal preprocessing. First, the signal is processed by a high-pass filter to remove low-frequency noise and retain high-frequency components. The specific operations include setting the filter threshold, applying the recursive filter equation, and reconstructing the signal. Next, the signal is processed by a low-pass filter to remove high-frequency noise and retain low-frequency components. The specific operations include setting the filter threshold, applying the convolution operation, and reconstructing the signal. After that, the discrete Fourier transform is used for frequency domain analysis to decompose the various frequency components of the signal, and the noise components are identified and separated by spectrum analysis. Finally, the wavelet transform is used for multi-scale analysis to further filter out the residual noise, and the signal is reconstructed and smoothed to generate the preprocessed spectral data.

[0050] S2, based on the preprocessed spectral data, an adaptive baseline correction algorithm is used to perform baseline drift correction, dynamically adjust the signal baseline level, and generate corrected spectral data;

[0051] In the baseline drift correction step, an adaptive baseline correction algorithm is used to correct the baseline drift based on the preprocessed spectral data. The baseline level of the signal is dynamically adjusted by real-time analysis of the signal. The specific operations include calculating the trend component of the signal, adjusting the baseline position through an adaptive filter, and applying polynomial fitting technology to eliminate the drift effect. First, the low-frequency component of the signal is analyzed to identify the baseline drift trend. Then, an adaptive filter is applied to adjust the signal baseline to match the actual data and dynamically update it. Finally, the signal is smoothed by a polynomial fitting method to eliminate the effect of baseline drift and generate corrected spectral data.

[0052] S3, based on the corrected spectral data, using a Kalman filter to perform signal denoising, optimize the signal-to-noise ratio of the signal, and generate denoised spectral data;

[0053] In the signal denoising step, a Kalman filter is used to perform signal denoising based on the corrected spectral data. The signal is recursively estimated by the Kalman filter to optimize the signal-to-noise ratio. The specific operations include initializing the filter parameters, performing state estimation and measurement updates, and reducing noise through prediction-correction steps. First, the state variables and covariance matrix of the Kalman filter are initialized. Then, the state estimate is updated using the measurement data, the predicted value is adjusted by the Kalman gain, and error correction is performed. Finally, each data point of the signal is recursively processed to gradually reduce the noise component and generate denoised spectral data.

[0054] S4, based on the denoised spectral data, principal component analysis and independent component analysis methods are used to extract blood oxygen saturation correlation features and generate blood oxygen saturation feature data;

[0055] In the feature extraction step, principal component analysis and independent component analysis are used to extract blood oxygen saturation related features based on the denoised spectral data. First, principal component analysis is performed to extract the main eigenvectors and eigenvalues ​​of the data by calculating the covariance matrix, reduce the data dimension, and retain the main information. Next, independent component analysis is applied to separate independent signal sources, and components with independent statistical characteristics are extracted by maximizing non-Gaussianity. Specific operations include calculating the covariance matrix, performing eigenvalue decomposition, extracting independent components through the rotation matrix, and reconstructing the characteristic signal. Generate blood oxygen saturation feature data. The signal is an infrared signal and the spectrum is an infrared spectrum.

[0056] S5, based on the blood oxygen saturation characteristic data, the rScO2 value is predicted and estimated by using support vector regression and random forest method to generate the rScO2 value;

[0057] In the rScO2 value prediction step, support vector regression and random forest methods are used to predict and estimate rScO2 values ​​based on the blood oxygen saturation feature data. Through the support vector regression model, the kernel function and penalty parameters are set, the feature data is fitted, and the rScO2 value is predicted. The specific operations include dividing the training set and the test set, and performing model training and parameter optimization. Then, the random forest method is applied to further optimize the prediction results by constructing multiple decision trees, performing ensemble learning and feature importance evaluation. The specific operations include setting the number and depth of trees, and performing model training and cross-validation. The rScO2 value is generated. This step combines the advantages of support vector regression and random forest methods to accurately predict blood oxygen saturation.

[0058] S6, based on the rScO2 value and combined with the electrical impedance data, the finite element analysis method is used to calculate the current distribution, reconstruct the electrical impedance image of the brain tissue, and generate the electrical impedance image;

[0059] In the electrical impedance image reconstruction step, based on the rScO2 value and combined with the electrical impedance data, the finite element analysis method is used to calculate the current distribution and reconstruct the electrical impedance image of the brain tissue. By establishing a finite element model, the numerical simulation of the electrical impedance data is performed. The specific operations include setting boundary conditions, meshing, and solving the current distribution equation. First, based on the rScO2 value and the electrical impedance data, a finite element model of the brain tissue is constructed. Then, the current injection and measurement positions are set, meshing is performed, and the current distribution is solved by the finite element method. Finally, the electrical impedance image of the brain tissue is reconstructed to generate an electrical impedance image.

[0060] S7, based on the electrical impedance image and the denoised spectral data, uses Kalman filtering and Bayesian estimation methods to perform multimodal data fusion, calculates the blood oxygen saturation of local brain tissue, and generates a local brain tissue blood oxygen saturation image.

[0061] In the multimodal data fusion step, Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion based on the electrical impedance image and denoised spectral data. The multimodal data is recursively processed by Kalman filtering to optimize the fusion results. The specific operations include initializing the filter parameters, performing state estimation and measurement updates, and reducing noise and uncertainty through prediction-correction steps. Next, the Bayesian estimation method is applied to calculate the probability distribution, and data fusion is achieved by updating the prior and posterior probabilities. The specific operations include setting the prior distribution, calculating the conditional probability of the data, and updating the posterior probability. Finally, the blood oxygen saturation of the local brain tissue is calculated through multimodal data fusion, and the local brain tissue blood oxygen saturation image is generated.

[0062] The preprocessed spectral data include spectral data from which low-frequency noise and high-frequency noise have been removed. The corrected spectral data specifically refers to spectral signals from which baseline has been corrected. The denoised spectral data include spectral data from which random noise and system noise have been removed. The blood oxygen saturation feature data include principal component features and independent component features. The rScO2 value includes the blood oxygen saturation value predicted and estimated by regression analysis and machine learning algorithm. The electrical impedance image includes a current distribution map and an electrical impedance value map. The local brain tissue blood oxygen saturation image includes a fused blood oxygen saturation distribution map and a numerical map.

[0063] The high-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering. The steps to generate preprocessed spectral data are as follows:

[0064] S101, based on the original spectral data of the brain tissue, use a high-pass filter to process the signal, set a high-frequency threshold, extract and retain the high-frequency components of the signal, remove the frequency components below the threshold, and reconstruct the signal to generate spectral data after high-pass filtering;

[0065] In sub-step S101, based on the original spectral data of brain tissue, the signal is processed with a high-pass filter. First, a high-frequency threshold is set to ensure that only signal components above this threshold are retained. The specific operation includes inputting the original spectral data into a high-pass filter, calculating the spectral components of the signal, and removing the frequency components below the high-frequency threshold through frequency domain filtering technology. When reconstructing the signal, the frequency domain signal is converted back to the time domain signal through an inverse Fourier transform to ensure the retention of high-frequency components and the removal of low-frequency noise. The format of the spectral data after high-pass filtering is a time domain signal sequence, and each data point represents the signal intensity at a specific time point. Through this method, the high-frequency components of the signal are successfully extracted and retained, the low-frequency noise is removed, and the spectral data after high-pass filtering is generated. The generated high-pass filtered spectral data has clearer high-frequency components, reduces low-frequency interference, and is convenient for further analysis.

[0066] S102, based on the spectral data after high-pass filtering, use a low-pass filter to process the signal, set a low-frequency threshold, extract and retain the low-frequency components of the signal, remove the frequency components higher than the threshold, and reconstruct the signal to generate spectral data after low-pass filtering;

[0067] In sub-step S102, based on the spectral data after high-pass filtering, the signal is processed with a low-pass filter. A low-frequency threshold is set to ensure that only signal components below this threshold are retained. The specific operation includes inputting the spectral data after high-pass filtering into a low-pass filter, calculating the spectral components of the signal, and removing the frequency components above the low-frequency threshold through frequency domain filtering technology. When reconstructing the signal, the frequency domain signal is converted back to the time domain signal through inverse Fourier transform to ensure the retention of low-frequency components and the removal of high-frequency noise. The format of the spectral data after low-pass filtering is a time domain signal sequence, and each data point represents the signal intensity at a specific time point. Through this method, the low-frequency components of the signal are successfully extracted and retained, the high-frequency noise is removed, and the spectral data after low-pass filtering is generated. The generated spectral data after low-pass filtering has clearer low-frequency components, reduces high-frequency interference, and is convenient for further analysis.

[0068] S103, based on the spectral data after low-pass filtering, use frequency domain analysis tools to perform spectrum transformation, decompose the various frequency components of the signal, and perform wavelet transformation to identify and separate the noise components, reconstruct the signal, and generate pre-processed spectral data.

[0069] In the S103 sub-step, based on the spectrum data after low-pass filtering, a frequency domain analysis tool is used to perform spectrum transformation. First, the time domain signal is converted into a frequency domain signal by Fourier transform, and each frequency component of the signal is decomposed. The specific operation includes Fourier transforming the spectrum data after low-pass filtering, obtaining a spectrum diagram, and identifying the amplitude and phase of each frequency component. Next, wavelet transform is applied for multi-scale analysis, and the signal is decomposed by selecting a suitable wavelet basis function, identifying and separating the noise component. The specific operation includes selecting a wavelet basis function, performing wavelet decomposition, decomposing the signal into wavelet coefficients of different scales, and identifying and removing the noise coefficient. Finally, the wavelet coefficients are reconstructed into a time domain signal by an inverse wavelet transform to generate pre-processed spectrum data. The format of the pre-processed spectrum data is a time domain signal sequence, and each data point represents the signal intensity at a specific time point. Through this method, the signal is successfully decomposed and reconstructed, the noise component is removed, and the generated pre-processed spectrum data has a higher signal-to-noise ratio, which is convenient for subsequent analysis.

[0070] Adopting the adaptive baseline correction algorithm method, baseline drift correction is performed to dynamically adjust the signal baseline level. The specific steps for generating corrected spectral data are as follows:

[0071] S201, based on the preprocessed spectral data, extract the baseline part of the spectrum, perform data segmentation, divide the spectral data into multiple interval sections, calculate the minimum value of each section, obtain the minimum value data of multiple intervals, merge the minimum value data of multiple intervals, and generate initial baseline data;

[0072] In sub-step S201, the baseline part of the spectrum is extracted through the pre-processed spectral data. First, the spectral data is segmented, and the entire spectral data is divided into multiple equally spaced segments according to time or wavelength intervals. The data in each segment is stored in an array or matrix form to ensure the convenience of subsequent processing. Next, the minimum value of each segment is calculated. Specifically, for each segment, its data points are traversed, the lowest value is found and recorded, and a new minimum value array is formed. The array contains the minimum value data of all segments. The generated initial baseline data is a set of values ​​reflecting the baseline level of the spectrum, which can provide a reference for subsequent baseline adjustment. The role of this initial baseline data is to provide a standard baseline for subsequent steps to correct the baseline drift problem in the spectral data.

[0073] S202, adjusting the baseline level of the spectral data based on the initial baseline data, calculating the baseline offset of each section by matching the minimum value data of multiple intervals, and performing baseline adjustment, performing offset correction on the spectral data of multiple intervals, and generating adjusted baseline data;

[0074] In sub-step S202, the baseline level of the spectral data is adjusted through the initial baseline data. First, the baseline offset of each paragraph is calculated by matching the minimum value data of multiple interval paragraphs. The specific operation is to compare the spectral data of each paragraph with the corresponding initial baseline data, calculate the offset of each data point, and obtain the offset array by using the difference calculation method of the data point. Next, the baseline adjustment is performed, the offset is applied to the spectral data, and the baseline level of the spectral data is adjusted by subtracting the value in the offset array, so that the spectral data of each paragraph returns to a unified baseline level. The generated adjusted baseline data is the spectral data after baseline correction, which can reflect the real spectral signal and eliminate the influence of baseline drift. The adjusted baseline data lays the foundation for the subsequent global correction, making the spectral data more accurate and reliable.

[0075] S203 , based on the adjusted baseline data, applying the adjusted baseline level to the entire spectral data, correcting the baseline drift of the segment, merging the adjusted data of the segment, and generating corrected spectral data.

[0076] In sub-step S203, the adjusted baseline level is applied to the entire spectral data through the adjusted baseline data. First, a global correction is performed, and the adjusted baseline level is applied to all segments to ensure that the baseline drift of each segment is corrected. The specific operation is to extend the application of the adjusted baseline data to the entire spectral data, and by reusing the adjusted baseline data, ensure that the baselines of all segments are uniformly adjusted. Then, the adjusted data of all paragraphs are merged, and the corrected data of each paragraph are integrated to generate a complete spectral data set. The merging operation can be achieved through data connection or matrix splicing to ensure the continuity and integrity of the data. Finally, the corrected spectral data is generated, and the data set reflects the spectral signal after comprehensive correction. The corrected spectral data has high accuracy and consistency, can be used for subsequent spectral analysis and feature extraction, and provide a reliable data basis. Through the above steps, the corrected spectral data is successfully generated, which provides high-quality data support for scientific research and practical applications.

[0077] The steps of using Kalman filter to perform signal denoising and optimize the signal-to-noise ratio of the signal to generate denoised spectral data are as follows:

[0078] S301, based on the corrected spectral data, using a Kalman filter to process the signal, initialize the filter parameters, perform state estimation and measurement update, recursively process data points, and perform noise reduction to generate spectral data after Kalman filtering;

[0079] In the S301 sub-step, the initial state of the filter is ensured to be accurate by initializing the Kalman filter parameters, including state variables, covariance matrix, process noise and measurement noise. In the specific implementation, the corrected spectral data is first converted into the input data of the state space model, the state vector and measurement vector are defined, the initial state covariance matrix is ​​set to the unit matrix, and the process noise covariance and measurement noise covariance are set based on prior knowledge and experimental data. The prediction and update formula of the Kalman filter are used to recursively process the data points of each time step. The prediction step updates the state estimate through the state transfer equation and calculates the prediction covariance matrix; the update step uses the actual measurement value to correct the predicted value, calculates the Kalman gain, updates the state estimate and covariance matrix, and finally obtains the filtered state estimate. The noise reduction effect is evaluated by the mean square error of the measurement residual (i.e., the difference between the measured value and the estimated value), and the spectral data file after the Kalman filter is generated. This process improves the signal-to-noise ratio of the spectral data, reduces the influence of measurement noise, and lays the foundation for subsequent data processing.

[0080] S302, based on the spectral data after Kalman filtering, using adaptive noise reduction to process the signal, setting adaptive parameters, analyzing the noise characteristics in the signal, and adjusting the parameters to match the noise changes, gradually reducing the noise components, and generating spectral data after adaptive noise reduction;

[0081] In sub-step S302, the spectral data after Kalman filtering is processed by an adaptive noise reduction algorithm, adaptive parameters are set, the noise characteristics in the signal are analyzed, the parameters are adjusted to match the noise changes, and the noise components are gradually reduced. In the specific implementation, the noise statistical characteristics of the spectral data are first calculated, including the noise mean and variance, and the parameters of the adaptive filter are initialized based on the statistical characteristics. The adaptive filter adopts the LMS (least mean square) algorithm, updates the filter weights according to the input signal and the expected output signal, and adjusts the step size parameters to achieve the noise reduction effect. For each data point, the error between the current filter output and the actual signal is calculated, and the filter weight is updated to gradually reduce the error, and finally obtain spectral data with lower noise. The noise reduction effect is evaluated by calculating the signal-to-noise ratio of the filtered signal to generate a spectral data file after adaptive noise reduction. This process effectively reduces the random noise component in the spectral data and improves the quality of the signal.

[0082] S303, based on the spectral data after adaptive noise reduction, optimizing the signal-to-noise ratio of the signal, performing signal smoothing processing, eliminating residual noise, and generating denoised spectral data.

[0083] In sub-step S303, the signal smoothing process is performed on the spectral data after adaptive noise reduction to optimize the signal-to-noise ratio and eliminate residual noise. In the specific implementation, a moving average filter is used to smooth the spectral data, the size of the smoothing window is set, and the appropriate window length is selected according to the characteristics of the data to ensure the smoothing effect. For each data point, the average value of the data in the window is calculated to replace the value of the original data point, thereby achieving a smoothing effect. After smoothing, the standard deviation and mean of the signal are calculated to evaluate the smoothing effect, ensure that the high-frequency noise in the signal is effectively eliminated, and generate the final denoised spectral data file. This process further improves the signal-to-noise ratio of the spectral data, makes the signal more stable, and is conducive to subsequent spectral analysis and application.

[0084] The principal component analysis and independent component analysis methods are used to extract the blood oxygen saturation correlation features and generate the blood oxygen saturation feature data as follows:

[0085] S401, performing data centering processing based on the denoised spectral data, subtracting the average value of the data from each data point to obtain the centralized data, performing orthogonal transformation, and performing orthogonal projection on the centralized data to generate principal component data;

[0086] In sub-step S401, the spectral data after denoising is centrally processed, and the average value of the data is subtracted from each data point to obtain the centralized data. The specific operation includes calculating the mean of the spectral data after denoising, and subtracting the mean from each data point to obtain the centralized data. Then, an orthogonal transformation is performed, and the centralized data is orthogonally projected to generate principal component data. The orthogonal transformation calculates the covariance matrix and performs eigenvalue decomposition to find the principal component direction of the data, and projects the data onto the principal component direction, thereby achieving data dimensionality reduction and feature extraction. The format of the generated principal component data is a principal component score matrix, and each element represents the projection value of the data in a certain principal component direction.

[0087] S402, based on the principal component data, performing data independence processing, using signal separation technology to separate each principal component, obtaining independent signal data, extracting the signal associated with the blood oxygen saturation, and generating independent component data;

[0088] In sub-step S402, the principal component data is processed for data independence, and the signal separation technology is used to separate the principal components to obtain independent signal data. The specific operation includes selecting a suitable signal separation method, such as independent component analysis, taking the principal component data as input, and separating the signal components with independent statistical characteristics by maximizing non-Gaussianity or minimizing mutual information. Each independent component represents a signal that has no linear relationship between different sources. In this way, the mixed signal can be effectively decomposed into independent signals. Then, the signal associated with the blood oxygen saturation is extracted, and the degree of correlation is determined by comparing each independent signal with the known blood oxygen saturation reference signal, and the signal with high correlation is selected. The format of the generated independent component data is an independent signal matrix, and each element represents the intensity of the independent signal at a certain time point. This step decomposes the mixed signal into independent components through signal separation technology, thereby extracting signals that are highly correlated with blood oxygen saturation, generating independent component data, and providing a basis for subsequent feature extraction.

[0089] S403, based on the independent component data, perform feature extraction operations, analyze the characteristic values ​​of each independent signal, calculate characteristic parameters that are highly correlated with the blood oxygen saturation, select related features, and generate blood oxygen saturation characteristic data.

[0090] In sub-step S403, feature extraction is performed on the independent component data, the characteristic values ​​of each independent signal are analyzed, and characteristic parameters highly correlated with blood oxygen saturation are calculated. Specific operations include calculating the statistical characteristics of each independent signal, such as mean, variance, kurtosis, skewness, etc. The main characteristics of the signal can be described by the characteristic parameters. Then, the correlation analysis method is used to calculate the correlation between the characteristic parameters and the blood oxygen saturation, and the features with high correlation are selected as the final features. Finally, the selected features are combined to generate blood oxygen saturation feature data. The format of the generated blood oxygen saturation feature data is a feature matrix, and each element represents the value of a certain feature at a certain time point. This step extracts the characteristic parameters that are highly correlated with blood oxygen saturation by analyzing the characteristic values ​​of independent signals. The generated blood oxygen saturation feature data provides key input for subsequent prediction and estimation.

[0091] Support vector regression and random forest methods were used to predict and estimate rScO2 values. The steps for generating rScO2 values ​​were as follows:

[0092] S501, based on the blood oxygen saturation characteristic data, using the data preprocessing step, performing data cleaning and normalization processing, then removing outliers and performing linear interpolation, and performing format conversion to generate preprocessed characteristic data;

[0093] In sub-step S501, data preprocessing is performed using the blood oxygen saturation feature data. First, data cleaning is performed to remove missing values ​​and duplicate values ​​in the data set to ensure data integrity and consistency. Next, data normalization is performed to scale all feature data to a range of 0 to 1. Then, outliers are detected and removed, and by calculating the mean and standard deviation of each feature, those abnormal data points that deviate from the mean by more than 3 standard deviations are marked and deleted to avoid the impact of outliers on model training. After that, linear interpolation is performed to fill in the missing data caused by the removal of outliers, and the average value of adjacent data points is used for interpolation to maintain data continuity. Finally, format conversion is performed to convert the processed data into the format required for model training. The preprocessed feature data file is used for subsequent model training to ensure data quality and consistency.

[0094] S502, based on the preprocessed feature data, training a support vector regression model, setting model parameters, dividing the data set into a training set and a test set, training the model and optimizing the parameters, and performing cross-validation to generate a support vector regression model;

[0095] In sub-step S502, the support vector regression model is trained through the pre-processed feature data. First, the model parameters are set, including the kernel function type, regularization parameter and error tolerance, etc., to ensure the basic architecture of the model. Next, the pre-processed feature data is divided into a training set and a test set. Then, the training set data is used for model training, and the model parameters are adjusted by minimizing the loss function so that the model can accurately fit the data. In order to avoid overfitting, the cross-validation technique is used to further divide the training set into several subsets, and different subsets are used cyclically for training and verification, to optimize the model parameters and improve the generalization ability of the model. After the training is completed, a support vector regression model is generated, and the model file is used for subsequent prediction tasks with high prediction accuracy and stability.

[0096] S503, based on the support vector regression model, the random forest model is trained, the number and depth of trees are set, the rScO2 value is predicted and estimated by combining the support vector regression model by evaluating the feature importance and adjusting the model parameters, and the result is verified to generate the rScO2 value.

[0097] In sub-step S503, the random forest model is trained through the support vector regression model. First, the parameters of the random forest model, including the number of trees and the depth of the trees, are set to ensure the complexity of the model and the training efficiency. Next, the importance of each feature is evaluated, the contribution of each feature in the model is calculated, and the most important features are selected for training to reduce the complexity of the model. Then, the pre-processed feature data is predicted in combination with the support vector regression model to obtain a preliminary rScO2 value prediction result. Next, the prediction results are used as input data to train the random forest model, and the prediction performance of the model is improved by gradually increasing the number of trees and adjusting the depth of the trees. Finally, the rScO2 value prediction file is generated, and the result verification is performed. By comparing the actual value and the predicted value, the accuracy and stability of the model are evaluated. The generated rScO2 value file is used for practical applications to provide reliable blood oxygen saturation estimation and support clinical decision-making and health monitoring.

[0098] Combined with the electrical impedance data, the finite element analysis method is used to calculate the current distribution and reconstruct the electrical impedance image of the brain tissue. The steps for generating the electrical impedance image are as follows:

[0099] S601, based on the rScO2 value and the electrical impedance data, data integration is performed, the rScO2 value is used to match the electrical impedance data, the rScO2 value and the electrical impedance data are matched according to the position, a preliminary model of brain tissue electrical impedance is constructed, and a preliminary electrical impedance model is generated;

[0100] In the S601 sub-step, data integration is performed based on rScO2 values ​​combined with electrical impedance data, and a matching algorithm is used to correspond rScO2 values ​​and electrical impedance data according to spatial positions to construct a preliminary model of brain tissue electrical impedance. First, the rScO2 values ​​and electrical impedance data are converted into matrix form, wherein the rScO2 values ​​are in time sequence and the electrical impedance data are in spatial coordinate sequence. The two sets of data are matched using the least squares method, and the electrical impedance data are adjusted according to the rScO2 values ​​of the corresponding positions to ensure that the two data sets are consistent in the same spatial position. After the matching is completed, the data is integrated through matrix operations to generate a preliminary model of brain tissue electrical impedance. The model provides a preliminary brain tissue electrical impedance distribution map, and the generated preliminary electrical impedance model file contains the electrical impedance values ​​of each region of the brain tissue, which provides a basis for further current distribution calculations.

[0101] S602, based on the preliminary electrical impedance model, current distribution calculation is performed, the current distribution of each point of the model is determined by finite element analysis, the positions of the current source and the measurement point are adjusted by point-by-point analysis, data is collected, and current distribution data is generated;

[0102] In sub-step S602, the current distribution is calculated based on the preliminary electrical impedance model, and the current distribution of each point in the model is determined by the finite element analysis method. First, the preliminary electrical impedance model is imported into the finite element analysis software, a finite element mesh model of the brain tissue is established, and the positions of the current source and the measurement points are set. The current density and voltage distribution of each point in the grid are calculated one by one through point-by-point analysis. The position of the current source is adjusted to obtain the best current distribution to ensure the accuracy of the measurement. During the data acquisition process, the current density value of each grid point is recorded to generate current distribution data. This process uses accurate current distribution calculations to generate a current distribution data file that records in detail the current density distribution of each point in the brain tissue, providing the necessary data support for subsequent electrical impedance image reconstruction.

[0103] S603, performing electrical impedance image reconstruction based on the current distribution data, using the current distribution data of each point to perform image reconstruction, applying the current distribution data to the brain tissue model, calculating the electrical impedance value of each area, and generating an electrical impedance image.

[0104] In the S603 sub-step, the impedance image is reconstructed based on the current distribution data, and the current distribution data of each point is used to generate the impedance image of the brain tissue. First, the current distribution data is imported into the image reconstruction algorithm, and the current distribution data is processed using the back projection algorithm or the iterative reconstruction algorithm. The specific operation includes converting the current density data of each measuring point into an image pixel value and constructing an initial model of the impedance image. Through iterative calculation, the impedance value of each point is gradually adjusted to optimize the resolution and contrast of the image. During the reconstruction process, the algorithm parameters are adjusted according to the error of the reconstruction result until the optimal image effect is obtained. The impedance image file finally generated shows the distribution of impedance values ​​in various regions of the brain tissue, providing an intuitive image reference for medical diagnosis and research, and can help doctors more accurately judge the health status of brain tissue.

[0105] Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion and calculate the blood oxygen saturation of local brain tissue. The steps for generating a local brain tissue blood oxygen saturation image are as follows:

[0106] S701, based on the electrical impedance image and the denoised spectral data, Kalman filtering is used to perform data fusion, and Kalman filtering fusion data is generated by recursively processing each data point and performing state estimation and measurement update, and performing noise reduction;

[0107] In sub-step S701, data fusion is performed by using Kalman filtering on the electrical impedance image and the denoised spectral data. First, the electrical impedance image data and the denoised spectral data are preliminarily matched and aligned to ensure that the data are consistent in time and space. Next, the state variables and covariance matrix of the Kalman filter are initialized, and the estimated values ​​of the initial state and measurement noise are set. The Kalman filtering process includes two main steps: prediction and update. In the prediction step, the state estimate and state transfer matrix of the previous moment are used to predict the state at the current moment. Then, in the update step, the state estimate is updated using the current measurement data, and the predicted value is adjusted by calculating the Kalman gain to make it closer to the actual measurement value. All data points are recursively processed to gradually reduce noise and optimize the fusion results to generate Kalman filter fusion data. The generated Kalman filter fusion data format is a multidimensional spatiotemporal sequence, and each data point contains the fused electrical impedance value and spectral value.

[0108] S702, based on the Kalman filter fusion data, using the Bayesian estimation method, performing probability distribution calculation, performing data fusion by updating the prior and posterior probabilities, and performing uncertainty assessment to generate Bayesian estimation fusion data;

[0109] In sub-step S702, the probability distribution is calculated by using the Bayesian estimation method for the Kalman filter fusion data. First, the prior probability distribution is set, and the initial distribution is estimated based on the existing knowledge and prior data. Then, the prior probability is updated using the Bayesian formula, the likelihood function is calculated by the observed data, and the posterior probability is calculated in combination with the prior distribution. The specific operations include: sampling the Kalman filter fusion data, calculating the conditional probability of each data point, and obtaining the posterior distribution by cumulative calculation. In this process, the Markov chain Monte Carlo method or the particle filter method is used to sample and estimate the high-dimensional data to ensure the accuracy and stability of the calculation. When performing uncertainty assessment, the variance and confidence interval of the posterior distribution are analyzed to determine the reliability of data fusion. The generated Bayesian estimation fusion data format is a probability distribution sequence, and each data point contains its corresponding probability value and uncertainty parameter. Through this method, the information of multi-source data is effectively integrated, the accuracy and reliability of data fusion are improved, and the generated data provides a solid foundation for the subsequent blood oxygen saturation calculation.

[0110] S703, based on the Bayesian estimation fusion data, the local brain tissue blood oxygen saturation is calculated, a two-dimensional image is generated by interpolation and mapping, and pseudo-color coding is performed for data visualization to generate a local brain tissue blood oxygen saturation image.

[0111] In sub-step S703, the local brain tissue blood oxygen saturation is calculated by using the Bayesian estimation fusion data. First, the Bayesian estimation fusion data is interpolated spatially and temporally using an interpolation method to convert discrete data points into a continuous data field. The specific operation includes selecting a suitable interpolation method, such as bilinear interpolation or spline interpolation, performing interpolation calculations on each data point, and generating a continuous two-dimensional data matrix. Next, the interpolated data is mapped to a two-dimensional image plane using a mapping method to generate a preliminary blood oxygen saturation image. Then, the pseudo-color coding technology is applied to visualize the image data, and different blood oxygen saturation values ​​are corresponding to different colors by setting a color mapping table to generate a pseudo-color image with an intuitive display effect. The generated local brain tissue blood oxygen saturation image format is a two-dimensional color image file, and each pixel represents the blood oxygen saturation value of the corresponding position. Through this method, the complex multimodal fusion data is successfully converted into an intuitive image form, and the generated blood oxygen saturation image can clearly display the blood oxygen situation of the local brain tissue, providing an important reference for medical diagnosis and research.

[0112] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for detecting local brain tissue blood oxygen saturation, characterized in that: The following steps are involved: Based on the original spectral data of brain tissue, high-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering to generate preprocessed spectral data; Based on the preprocessed spectral data, an adaptive baseline correction algorithm is used to perform baseline drift correction, dynamically adjust the signal baseline level, and generate corrected spectral data; Based on the corrected spectral data, a Kalman filter is used to perform signal denoising processing, optimize the signal-to-noise ratio of the signal, and generate denoised spectral data; Based on the denoised spectral data, principal component analysis and independent component analysis methods are used to extract blood oxygen saturation correlation features and generate blood oxygen saturation feature data; Based on the blood oxygen saturation characteristic data, support vector regression and random forest method are used to predict and estimate the rScO2 value to generate the rScO2 value; Based on the rScO2 value and in combination with the electrical impedance data, a finite element analysis method is used to calculate the current distribution, reconstruct the electrical impedance image of the brain tissue, and generate the electrical impedance image; Based on the electrical impedance image and the denoised spectral data, Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion, calculate the blood oxygen saturation of local brain tissue, and generate a local brain tissue blood oxygen saturation image.

2. The method for detecting local brain tissue oxygen saturation according to claim 1, characterized in that: The preprocessed spectral data includes spectral data from which low-frequency noise and high-frequency noise have been removed, the corrected spectral data specifically refers to spectral signals from which baseline has been corrected, the denoised spectral data includes spectral data from which random noise and system noise have been removed, the blood oxygen saturation characteristic data includes principal component characteristics and independent component characteristics, the rScO2 value includes a blood oxygen saturation value predicted and estimated by regression analysis and a machine learning algorithm, the electrical impedance image includes a current distribution map and an electrical impedance value map, and the local brain tissue blood oxygen saturation image includes a fused blood oxygen saturation distribution map and a numerical map.

3. The method for detecting local brain tissue blood oxygen saturation according to claim 1, characterized in that: High-pass and low-pass filter methods are used for signal preprocessing to remove low-frequency and high-frequency noise in the signal, and discrete Fourier transform and wavelet transform are used for frequency domain analysis and filtering. The steps to generate preprocessed spectral data are as follows: Based on the original spectral data of the brain tissue, the signal is processed with a high-pass filter, a high-frequency threshold is set, high-frequency components of the signal are extracted and retained, frequency components below the threshold are removed, and the signal is reconstructed to generate spectral data after high-pass filtering; Based on the spectral data after high-pass filtering, the signal is processed with a low-pass filter, a low-frequency threshold is set, the low-frequency component of the signal is extracted and retained, the frequency components higher than the threshold are removed, and the signal is reconstructed to generate the spectral data after low-pass filtering; Based on the spectral data after low-pass filtering, a frequency domain analysis tool is used to perform spectrum transformation, decompose the various frequency components of the signal, and perform wavelet transformation to identify and separate the noise components, reconstruct the signal, and generate pre-processed spectral data.

4. The method for detecting local brain tissue blood oxygen saturation according to claim 1, characterized in that: Adopting the adaptive baseline correction algorithm method, baseline drift correction is performed to dynamically adjust the signal baseline level. The specific steps for generating corrected spectral data are as follows: Based on the pre-processed spectral data, extract the baseline part of the spectrum, perform data segmentation, divide the spectral data into multiple interval sections, calculate the minimum value of each section, obtain the minimum value data of multiple intervals, merge the minimum value data of multiple intervals, and generate initial baseline data; Based on the initial baseline data, adjusting the baseline level of the spectral data, by matching the lowest value data of multiple intervals, calculating the baseline offset of each paragraph, and performing baseline adjustment, performing offset correction on the spectral data of multiple intervals, and generating adjusted baseline data; Based on the adjusted baseline data, the adjusted baseline level is applied to the entire spectral data, the baseline drift of the segment is corrected, and the adjusted data of the segment are merged to generate the corrected spectral data.

5. The method for detecting local brain tissue oxygen saturation according to claim 1, characterized in that: The steps of using Kalman filter to perform signal denoising and optimize the signal-to-noise ratio of the signal to generate denoised spectral data are as follows: Based on the corrected spectral data, a Kalman filter is used to process the signal, filter parameters are initialized, state estimation and measurement update are performed, data points are recursively processed, and noise reduction is performed to generate spectral data after Kalman filtering; Based on the spectral data after Kalman filtering, adaptive noise reduction is used to process the signal, adaptive parameters are set, noise characteristics in the signal are analyzed, and parameters are adjusted to match noise changes, so as to gradually reduce noise components and generate spectral data after adaptive noise reduction; Based on the spectral data after the adaptive noise reduction, the signal-to-noise ratio of the signal is optimized, the signal smoothing process is performed, the residual noise is eliminated, and the denoised spectral data is generated.

6. The method for detecting local brain tissue blood oxygen saturation according to claim 1, characterized in that: The principal component analysis and independent component analysis methods are used to extract the blood oxygen saturation correlation features and generate the blood oxygen saturation feature data as follows: Based on the denoised spectral data, data centering is performed, the average value of the data is subtracted from each data point to obtain the centralized data, an orthogonal transformation is performed, and the centralized data is orthogonally projected to generate principal component data; Based on the principal component data, data independence processing is performed, and signal separation technology is used to separate each principal component to obtain independent signal data, and the signal associated with the blood oxygen saturation is extracted to generate independent component data; Based on the independent component data, a feature extraction operation is performed to analyze the characteristic values ​​of each independent signal, calculate characteristic parameters highly correlated with the blood oxygen saturation, select related features, and generate blood oxygen saturation characteristic data.

7. The method for detecting local brain tissue blood oxygen saturation according to claim 1, characterized in that: Support vector regression and random forest methods were used to predict and estimate rScO2 values. The steps for generating rScO2 values ​​were as follows: Based on the blood oxygen saturation characteristic data, a data preprocessing step is adopted to perform data cleaning and normalization processing, then outliers are removed and linear interpolation is performed, and format conversion is performed to generate preprocessed characteristic data; Based on the preprocessed feature data, training a support vector regression model is performed, model parameters are set, and the data set is divided into a training set and a test set, model training is performed and parameters are optimized, and cross-validation is performed to generate a support vector regression model; Based on the support vector regression model, the random forest model is trained, the number and depth of trees are set, and the rScO2 value is predicted and estimated by combining the support vector regression model by evaluating the feature importance and adjusting the model parameters, and the result is verified to generate the rScO2 value.

8. The method for detecting local brain tissue oxygen saturation according to claim 1, characterized in that: Combined with the electrical impedance data, the finite element analysis method is used to calculate the current distribution and reconstruct the electrical impedance image of the brain tissue. The steps for generating the electrical impedance image are as follows: Based on the rScO2 value, combined with the electrical impedance data, data integration is performed, the rScO2 value is used to match the electrical impedance data, the rScO2 value and the electrical impedance data are matched according to the position, and a preliminary model of brain tissue electrical impedance is constructed to generate a preliminary electrical impedance model; Based on the preliminary electrical impedance model, current distribution calculation is performed, and the current distribution of each point of the model is determined by finite element analysis. The positions of the current source and the measurement point are adjusted by point-by-point analysis, and data collection is performed to generate current distribution data; Based on the current distribution data, electrical impedance image reconstruction is performed, the current distribution data of each point is used to perform image reconstruction, the current distribution data is applied to the brain tissue model, the electrical impedance value of each area is calculated, and the electrical impedance image is generated.

9. The method for detecting local brain tissue blood oxygen saturation according to claim 1, characterized in that: Kalman filtering and Bayesian estimation methods are used to perform multimodal data fusion and calculate the blood oxygen saturation of local brain tissue. The steps for generating a local brain tissue blood oxygen saturation image are as follows: Based on the electrical impedance image and the denoised spectral data, Kalman filtering is used to perform data fusion, and Kalman filtering fusion data is generated by recursively processing each data point and performing state estimation and measurement update, and performing noise reduction; Based on the Kalman filter fusion data, a Bayesian estimation method is used to perform probability distribution calculation, data fusion is performed by updating prior and posterior probabilities, and uncertainty assessment is performed to generate Bayesian estimation fusion data; Based on the Bayesian estimation fusion data, local brain tissue blood oxygen saturation is calculated, a two-dimensional image is generated by interpolation and mapping, and pseudo-color coding is performed for data visualization to generate a local brain tissue blood oxygen saturation image.

Images