Multi-site multi-wavelength parallel acquisition blood oxygen blood flow real-time measurement method and device

Abstract

The application discloses a kind of based on multi-site multi-wavelength parallel acquisition blood oxygen blood flow real-time measurement method, device, computer equipment and computer readable storage medium, it is related to biomedical optical measurement and computing technical field, wherein method includes: obtaining by multiple wavelength laser irradiation biological tissue same position, and in multiple different distance detection site parallel acquisition obtained multi-wavelength multi-site speckle intensity fluctuation sequence, and it is preprocessed, obtains the average autocorrelation curve and down-sampling intensity fluctuation curve corresponding to each group signal, input into multi-branch fusion neural network, by calculating and output blood oxygen saturation and blood flow velocity.The application realizes multi-site, multi-wavelength condition under high sensitivity blood oxygen and blood flow monitoring, overcomes the limitation that traditional single wavelength or single point detection is difficult to simultaneously quantify two indicators, and improves the real-time performance and accuracy of dynamic measurement by deep learning.

Description

Technical Field

[0001] This invention relates to the field of biomedical optical measurement and calculation technology, and in particular to a method and device for real-time measurement of blood oxygen and blood flow based on multi-site multi-wavelength parallel acquisition. Background Technology

[0002] Diffuse correlation spectroscopy is a non-invasive optical technique based on the principle of optical coherence. It is used to measure microscopic blood flow velocity and dynamic blood flow information within biological tissues, and is particularly suitable for monitoring blood flow in deep tissues such as the brain, muscles, and organs. When coherent light enters an turbid medium, photons propagate through diffusion within the tissue, randomly scattering and producing speckle patterns. The movement of scattering bodies within the tissue (such as red blood cells) causes fluctuations in light intensity, leading to changes in the speckle pattern. By analyzing the light intensity fluctuations in the speckle, the movement of red blood cells in the blood can be inferred, and thus the blood flow velocity and related hemodynamic parameters within the tissue can be calculated.

[0003] In existing technologies, calculating the blood flow index using diffuse correlation spectroscopy typically requires prior information on the absorption coefficient (μa) and reduced scattering coefficient (μs'). However, these parameters are often difficult to measure or obtain accurately and are significantly affected by individual differences, tissue composition, and environmental factors. Therefore, diffuse correlation spectroscopy is usually combined with near-infrared spectroscopy (NIRS) or diffuse optical spectroscopy (DOS) to obtain real-time information on tissue absorption and scattering characteristics using multiple wavelengths. Furthermore, it allows for the estimation of the concentrations of oxyhemoglobin (HbO2) and deoxyhemoglobin (HbR) within the tissue, thereby enabling simultaneous detection of blood oxygenation and blood volume, and providing more comprehensive data support for tissue oxygen metabolism.

[0004] Currently, non-invasive blood oxygenation and blood flow monitoring technologies have important applications in medical diagnosis and health monitoring, and are widely used to assess tissue oxygenation status and hemodynamics. However, traditional measurement methods typically rely on single-wavelength or single-point detection, resulting in limited spatial resolution and measurement accuracy, making it difficult to comprehensively reflect complex physiological changes. Traditional DCS systems usually rely on avalanche photodiodes to detect single-photon level light signals, but they are prone to high dark current and noise under weak light conditions, and can only detect speckle autocorrelation signals in a single channel. Random noise interference reduces the signal-to-noise ratio, thus limiting the accurate interpretation of weak dynamic signals. Since the theoretical derivation of blood flow velocity and blood oxygen saturation is related to the dynamic and static characteristics of tissues, respectively, diffuse correlation spectroscopy needs to be combined with NIRS or DOS to simultaneously detect these two types of information. The complexity of hybrid device systems limits their practical application in clinical practice. In addition, most existing methods are based on traditional fitting algorithms for calculation, which are insufficient for real-time, rapid, and accurate reflection of dynamic blood flow and blood oxygenation changes. Therefore, this invention proposes a multi-site, multi-wavelength parallel acquisition system, and calculates blood flow velocity and blood oxygen saturation simultaneously with high precision and high sensitivity based on deep learning algorithms. Summary of the Invention

[0005] The main objective of this invention is to provide a method for real-time measurement of blood oxygen and blood flow based on multi-site, multi-wavelength parallel acquisition.

[0006] Another objective of this invention is to propose a real-time blood oxygen and blood flow measurement device based on multi-site, multi-wavelength parallel acquisition.

[0007] The third objective of this invention is to provide an electronic device.

[0008] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, a first aspect of the present invention proposes a method for real-time measurement of blood oxygenation and blood flow based on multi-site, multi-wavelength parallel acquisition, comprising:

[0010] S1, obtain the multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and acquiring data in parallel at multiple detection sites at different distances; S2, preprocess the multi-wavelength multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals; S3, input the average autocorrelation curve and the downsampling intensity fluctuation curve into the multi-branch fusion neural network, and calculate and output the blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

[0011] Optionally, irradiating the same location of biological tissue with multiple wavelengths of laser light also includes: pass Lasers of different wavelengths generate a light source, among which The wavelength is between red and near-infrared light; This allows the laser to simultaneously irradiate the same location on the surface of biological tissue via multiple multimode optical fibers.

[0012] Optionally, the multi-wavelength, multi-site speckle intensity fluctuation sequence acquired in parallel at multiple detection sites at different distances also includes: Set on the surface of the biological tissue There are 10 detection points, each at a different distance from the laser illumination point. ; At each detection point A bundle of multimode optical fibers is used for signal transmission, and the signal is divided into... Each signal path passes through different filters to obtain a signal of a specific wavelength.

[0013] Optionally, the multi-wavelength, multi-site speckle intensity fluctuation sequence is preprocessed to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals, and the process further includes: For each region, a normalized autocorrelation curve is calculated for the speckle intensity fluctuation sequence corresponding to all detector pixels. The average of all curves within the region is then used as the final result for that region. For the average autocorrelation curve; The average speckle intensity fluctuation sequence corresponding to all detector pixels in each region is taken, and the averaged result is downsampled to have the same length as the average autocorrelation curve, thus obtaining... The downsampling intensity fluctuation curve.

[0014] Optionally, the average autocorrelation curve and the downsampled intensity fluctuation curve are input into a multi-branch fusion neural network, and the blood oxygen saturation and blood flow velocity are calculated and output by the neural network. The method further includes: Will The average autocorrelation curves are arranged in different ways to form two modes of data, including: data from the same probe site. By concatenating the data from beginning to end, we obtain a result containing information about different wavelengths. Multi-wavelength data for multiple channels; and data that will pass through the same filter. By concatenating the data from beginning to end, we can obtain data containing information about different loci. Channel multi-site data; Will contain information of different wavelengths Channel multi-wavelength data and data containing information at different locations Multi-point data from each channel is used as two input data streams and fed into a three-layer one-dimensional convolutional neural network for temporal feature extraction. The kernel size of each layer of the one-dimensional convolutional neural network decreases layer by layer to achieve step-by-step extraction of signal features from local to global time features. After each convolutional layer, batch normalization and ReLU activation are performed sequentially. Then, max pooling is used to reduce the dimensionality of the features to retain key information. The two time features extracted by the three-layer one-dimensional convolutional neural network are concatenated and fused, and then input into the fully connected layer to obtain features based on the average autocorrelation curve. The downsampled intensity fluctuation curve is input into a time-series feature extraction model containing a two-layer long short-term memory network to extract the time dependence and dynamic change characteristics of the signal and generate high-dimensional time features. The high-dimensional temporal features are fused with features extracted based on the average autocorrelation curve and input into a fully connected layer to output the estimated results of blood oxygen saturation and blood flow velocity.

[0015] To achieve the above objectives, a second aspect of the present invention provides a real-time blood oxygen and blood flow measurement device based on multi-site, multi-wavelength parallel acquisition, comprising: The collection module is used to acquire the multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and collecting data in parallel at multiple detection sites at different distances; The preprocessing module is used to preprocess the multi-wavelength, multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals. The calculation module is used to input the average autocorrelation curve and the downsampling intensity fluctuation curve into a multi-branch fusion neural network, and calculate and output blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

[0016] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0017] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing the real-time blood oxygen and blood flow measurement method based on multi-site multi-wavelength parallel acquisition as described in the first aspect embodiment.

[0018] To achieve the above objectives, the fourth aspect of this application proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the real-time measurement method for blood oxygenation and blood flow based on multi-site multi-wavelength parallel acquisition as described in the first aspect embodiment.

[0019] The embodiments of the present invention have the following beneficial effects: This invention achieves highly sensitive blood oxygenation and blood flow monitoring under multi-site and multi-wavelength conditions, overcoming the limitations of traditional single-wavelength or single-point detection in simultaneously quantifying two indicators. Furthermore, it improves the real-time performance and accuracy of dynamic measurements through deep learning, and has broad application prospects. Attached Figure Description

[0020] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of a method for real-time measurement of blood oxygen and blood flow based on multi-site multi-wavelength parallel acquisition provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a multi-site, multi-wavelength parallel measurement system provided in an embodiment of the present invention; Figure 3 A schematic diagram of a data acquisition model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a high-speed blood oxygenation and blood flow measurement method based on multimodal data, provided in an embodiment of the present invention. Figure 5 A schematic diagram of a neural network architecture provided in an embodiment of the present invention; Figure 6 A schematic diagram of a convolution model provided in an embodiment of the present invention; Figure 7 This is a structural diagram of a real-time blood oxygen and blood flow measurement device based on multi-site multi-wavelength parallel acquisition, provided in an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] The following description, with reference to the accompanying drawings, describes a method and apparatus for real-time measurement of blood oxygen and blood flow based on multi-site, multi-wavelength parallel acquisition according to embodiments of the present invention.

[0024] Example 1 This invention provides a method for real-time measurement of blood oxygen and blood flow based on multi-site, multi-wavelength parallel acquisition. Figure 1 This is a flowchart illustrating a real-time blood oxygen and blood flow measurement method based on multi-site, multi-wavelength parallel acquisition, provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps: S1, acquire the multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and collecting data in parallel at multiple detection sites at different distances.

[0025] To achieve high-sensitivity acquisition of multimodal and multichannel data and support high-speed blood oxygenation and blood flow measurement methods, this application designs a multi-site, multi-wavelength parallel measurement method consisting of a light source and an acquisition model, and completes hardware parameter optimization.

[0026] In this embodiment, the multi-site multi-wavelength parallel measurement method is divided into two core parts: a light source model and an acquisition model. The two models have clear division of labor and work together to achieve high-sensitivity acquisition of multimodal and multi-channel data. The light source model is responsible for providing multi-wavelength illumination signals, while the acquisition model is responsible for receiving and processing echo signals from biological tissues. The combination of the two ensures that multispectral information related to the internal components and structure of biological tissues can be obtained, laying the data foundation for high-speed blood oxygenation and blood flow measurement methods.

[0027] In this embodiment, the specific composition and working principle of the light source model are detailed. The light source model consists of three long-coherent continuous-wave lasers with different wavelengths and multimode optical fibers. The wavelengths of the three lasers are 785 nm, 808 nm, and 830 nm, respectively, and the output power of each is 100 mW. This wavelength selection can effectively penetrate biological tissue and specifically interact with components such as hemoglobin within the tissue to obtain compositional information within the tissue. In this application, the three lasers are simultaneously irradiated onto the surface of the biological tissue through three multimode optical fibers bundled together. This parallel irradiation method can simultaneously provide illumination signals of multiple wavelengths, avoiding the time delay problem caused by the traditional alternating irradiation method, and providing a foundation for subsequent parallel acquisition to improve temporal resolution.

[0028] In this embodiment, the specific composition and working principle of the acquisition model are detailed. The acquisition model consists of multimode fiber, filters, and a 32×32 SPAD array. Three detection points are set at different distances from the light source, with distances of 10mm, 15mm, and 20mm respectively. These detection points at different distances can acquire echo signals from different depths of biological tissue, enabling layered detection of the internal structure of the tissue. After the echo signal transmitted through the multimode fiber passes through the filters, the SPAD array collects high-resolution, high-sensitivity speckle data from multiple detection sites at multiple wavelengths. The entire acquisition model, with the multimode fiber, filters, and SPAD array working collaboratively, ensures efficient and accurate signal transmission and processing.

[0029] In this embodiment, the signal transmission and processing flow of the acquisition model is refined. The signal from each detection point is transmitted through an optical fiber bundle composed of three multimode optical fibers, and the signal is divided into three paths. Each path is filtered by a different filter. There are three filters in total, and their center wavelengths correspond to the three laser wavelengths of the light source model, namely 785 nm, 808 nm, and 830 nm, respectively, thus obtaining signals of different wavelengths. In this application, a total of nine signals are obtained from the three detection points, which are received by the SPAD array. Specifically, the SPAD array is a 32×32 pixel array, and the nine circular detection areas A1-A9 represent the projection areas of the SPAD array corresponding to the nine detection optical fibers on the left. This design enables parallel reception and processing of nine signals, significantly improving the time resolution.

[0030] In this embodiment, by optimizing the spatial spacing between the detection fiber and the SPAD array, effective amplification and matching of the speckle pattern are achieved, ensuring improved speckle contrast and data signal-to-noise ratio. Specifically, during spatial propagation between the detection fiber and the SPAD array, the speckle pattern is amplified. If the distance is too close, each SPAD pixel will contain multiple independent speckles, reducing speckle contrast; if the distance is too far, it will reduce the signal-to-noise ratio. To match the speckle size with the effective area of ​​the SPAD array, this application adopts the following relationship:

[0031] Determine the distance between the optical fiber and the SPAD array. ,in The size of the speckle is... It is the wavelength of the illumination laser. This measures the core diameter of the optical fiber. By adjusting this distance, the speckle size can be made close to the effective area size of the SPAD pixel, ensuring that each pixel contains only one independent speckle, thereby improving speckle contrast and data signal-to-noise ratio.

[0032] In this embodiment, the distance between the detection fiber and the SPAD array is accurately calculated based on specific hardware parameters. The multimode detection fiber used in this application has a core diameter D=1mm, and the effective pixel area size of the 32×32 SPAD array is... Based on the above formula, the distance between the detection fiber and the SPAD array should be adjusted to about 8.5mm. This distance setting can ensure the best match between the speckle size and the effective area of ​​the SPAD pixels, maximizing speckle contrast and data signal-to-noise ratio.

[0033] In this embodiment, a multi-site, multi-wavelength parallel measurement method composed of a light source and a data acquisition model is designed. The spacing between the optical fiber and the SPAD array is optimized and parallel acquisition is achieved, enabling high-sensitivity and high-temporal-resolution acquisition of multi-modal, multi-channel speckle data, providing data support for high-speed blood oxygenation and blood flow measurement methods.

[0034] S2, preprocess the multi-wavelength, multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals.

[0035] In order to provide high signal-to-noise ratio and structured unified input data for subsequent neural network models, this application performs grouping, autocorrelation calculation, curve averaging and downsampling normalization processing on the speckle intensity fluctuation sequence acquired by the SPAD array.

[0036] In this embodiment, the raw data collected by the SPAD array is first processed by grouping. The SPAD array collects a speckle intensity fluctuation sequence of 32×32 pixels. According to the region division method set above, this application divides these 32×32 speckle intensity fluctuation sequences into 9 groups. Each group of data corresponds to the speckle data collected from different detection sites at different wavelengths. Each group of data contains the speckle intensity fluctuation sequence corresponding to each pixel in the region. By grouping, the data attribution of different wavelengths and different detection sites can be clearly defined, laying the foundation for subsequent regional processing.

[0037] In this embodiment, for each set of data, i.e., the speckle intensity fluctuation sequence corresponding to all pixels in each region, a normalized autocorrelation curve is calculated to obtain the average autocorrelation curve for each region. Specifically, to quantify the fluctuation of the dynamic scattering field within biological tissue, this application uses a specific equation to calculate the normalized time-intensity autocorrelation function of each pixel within the region. The equation is as follows:

[0038] in This represents the speckle intensity fluctuation sequence corresponding to the j-th pixel in the i-th SPAD region of the array at a given time t. Representing the delay time or correlation time, this equation can accurately quantify the temporal correlation of speckle intensity for each pixel, reflecting the dynamic scattering characteristics of biological tissues.

[0039] In this embodiment, the calculation parameters of the autocorrelation function are specifically set to ensure the rationality and practicality of the calculation results. In this embodiment, each autocorrelation function... All calculations were performed at P=10000 time points, combined with the time sampling rate of SPAD. Each This will generate a new autocorrelation curve, which this application refers to as the "autocorrelation curve rate". This rate is the effective measurement rate of this method, which can meet the real-time requirements of subsequent data processing.

[0040] In this embodiment, based on past experience and experimental results, the range of the delay time τ was optimized. It is known that when the delay time... When the time exceeds 200 μs, the autocorrelation curve has nearly decayed to 1, and subsequent correlation data can no longer provide important information about tissue dynamics. Therefore, this application sets the delay time τ to a range of 0 to 300 μs, and the interval is set to... It contains 101 delayed time points. This setting can both preserve key information about tissue dynamics and avoid redundant calculations of invalid data, thereby improving processing efficiency.

[0041] In this embodiment of the application, all SPAD pixels within each region are completed. After calculation, this application averages the autocorrelation curves within the region to obtain the average autocorrelation curve for each region. The calculation equation is as follows:

[0042] in This represents the number of SPAD pixels in the i-th region. This application, through this averaging operation, effectively reduces the impact of random noise on the data, filters out invalid interference, generates an average autocorrelation curve with a higher signal-to-noise ratio, significantly improves data quality, and provides a more reliable input basis for subsequent neural network model training.

[0043] In this embodiment, for the speckle intensity fluctuation sequence corresponding to all pixels within each region, the same averaging operation as used in calculating the average autocorrelation curve is performed to average these time series, resulting in the average intensity fluctuation curve for each region. Subsequently, this application downsamples the average intensity fluctuation curve of each region to 101 points, making it correlated with the average autocorrelation curve of that region. With the same sequence length, the downsampled curve is called the downsampling intensity fluctuation curve.

[0044] In this embodiment, by grouping, calculating the autocorrelation function, averaging, and downsampling the speckle intensity fluctuation sequence acquired by the SPAD array, a high signal-to-noise ratio, equal-length average autocorrelation curve and downsampled intensity fluctuation curve are obtained, laying the foundation for the unified structured input data of the subsequent neural network model.

[0045] S3, input the average autocorrelation curve and the downsampling intensity fluctuation curve into the multi-branch fusion neural network, and calculate and output the blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

[0046] To achieve high-precision and high-sensitivity blood oxygen and blood flow measurement, this application inputs the preprocessed two types of curves into a multi-branch fusion neural network to complete feature extraction, fusion, and index calculation.

[0047] In this embodiment, the core working logic of the multi-branch fusion neural network is clearly defined. The average autocorrelation curve and the downsampling intensity fluctuation curve obtained after preprocessing are used as the input of the neural network. By designing different branch networks, independent feature information is extracted from the two types of curves respectively. Then, the features extracted by multiple branches are fused and processed. Finally, the blood oxygen and blood flow measurement results are output through subsequent calculations, so as to achieve complementary advantages of the two types of data and improve the measurement performance.

[0048] In this embodiment, the structure and data processing flow of the autocorrelation curve branches are designed in detail. For the nine average autocorrelation curves, this application first constructs two modes of data through different arrangements to fully extract wavelength and location information from the data. Mode one is constructed by concatenating three data points (e.g., data from regions A1, A4, and A7) from the same detection site, resulting in three-channel multi-wavelength data containing different wavelength information. This mode can effectively capture the coordinated variation characteristics of different wavelengths at the same detection site. Mode two is constructed by concatenating three data points (e.g., data from regions A1, A2, and A3) that pass through the same filter, resulting in three-channel multi-site data containing different location information. This mode can effectively capture the spatial distribution characteristics of different detection sites at the same wavelength.

[0049] In this embodiment, data from two modalities are input into a convolutional model, which extracts their respective internal temporal features. Specifically, this application sets up three consecutive one-dimensional convolutional layers in the convolutional model to extract multi-scale temporal features. The kernel size decreases layer by layer, set to 7, 5, and 3 respectively. This gradient-based kernel design achieves gradual capture from local subtle features to global overall features, ensuring comprehensive feature extraction. After each convolutional operation, batch normalization and ReLU activation functions are added. Batch normalization accelerates model convergence, improves model training stability, and avoids gradient vanishing or exploding problems. The ReLU activation function enhances the model's non-linear expressive ability, allowing the model to better fit complex data distributions.

[0050] In this embodiment, a max pooling layer of size 2 is set at the end of the convolutional model. Max pooling can downsample the features output by the convolutional layer, reducing the feature dimension and the computational cost of the model while effectively retaining the most significant and critical feature information, preventing overfitting, and further improving the generalization ability of the model.

[0051] In this embodiment, after the data of the two modalities are processed by their respective convolutional models to extract local temporal features, this application achieves the fusion of the two types of features by splicing. The fused features are then input into a fully connected layer, which further integrates and filters the features, and finally outputs the 64-dimensional feature vector of the autocorrelation curve branch, providing high-quality feature support for subsequent overall feature fusion.

[0052] In this embodiment, the structure and data processing flow of the intensity fluctuation curve branches are designed in detail. Nine downsampled intensity fluctuation curves are directly input into a two-layer LSTM network for feature extraction. The shape of the input data is set to (9, 101), indicating that the data contains 101 time steps, each containing nine light intensity values, consistent with the length of the downsampled data. The hidden layer size of the LSTM network is set to 64, meaning that each time step of the sequence is encoded as a 64-dimensional feature vector. Through the gating mechanism and time memory capability of the LSTM network, the time dependence and dynamic change characteristics of the intensity fluctuation curves are accurately captured, revealing the intrinsic correlation between light intensity fluctuations and changes in blood oxygenation and blood flow.

[0053] In this embodiment, after the nine downsampled intensity fluctuation curves are processed by two layers of LSTM network, a 64-dimensional global feature representation is finally generated. This feature vector integrates the temporal dynamic features of all intensity fluctuation curves and serves as the final feature output of the intensity fluctuation curve branch. It corresponds to the 64-dimensional feature vector of the autocorrelation curve branch, thus preparing for subsequent feature fusion.

[0054] In this embodiment, the two types of preprocessed curves are input into a multi-branch fusion neural network, which extracts multi-scale and time-dynamic features by branch and completes feature fusion calculation, thereby achieving high-precision and high-sensitivity measurement of blood oxygen saturation and blood flow velocity.

[0055] Example 2 This invention relates to a real-time blood oxygen and blood flow measurement system based on multi-site, multi-wavelength parallel acquisition, such as... Figure 2-6 As shown: Figure 2 The schematic diagram of the present invention illustrates that the multi-site, multi-wavelength parallel measurement system comprises two parts: a light source module and an acquisition module. It performs high-sensitivity acquisition of multimodal, multi-channel data, providing data for high-speed blood oxygenation and blood flow analysis methods based on multimodal data. The light source module consists of three long-coherent continuous-wave lasers with different wavelengths (785 nm, 808 nm, 830 nm, 100 mW) and multimode optical fibers. The three lasers are simultaneously irradiated onto the surface of biological tissue through three bundled multimode optical fibers to provide multispectral information related to the tissue's internal composition and structure. The acquisition module consists of multimode optical fibers, filters, and... The system consists of a SPAD array with three detection points at different distances from the light source, specifically 10mm, 15mm, and 20mm. After transmission through multimode fiber and a filter, the SPAD array collects high-resolution, high-sensitivity speckle data from multiple detection points at multiple wavelengths. The collected multi-wavelength, multi-site speckle data is then transmitted to a computer for data preprocessing. A multi-branch fusion neural network is used to calculate blood flow velocity and blood oxygen saturation, ultimately achieving high-precision, high-sensitivity, high-speed blood oxygen and blood flow measurement.

[0056] Figure 3 This is a schematic diagram of the acquisition module according to an embodiment of the present invention. The entire acquisition module consists of multimode optical fiber, filters, and a SPAD array working together. Specifically, as shown... Figure 3 As shown in Figure a, the signal at each detection point is transmitted via an optical fiber bundle consisting of three multimode fibers, splitting the signal into three paths. Each path passes through different filters (three filters in total, with center wavelengths of 785 nm, 808 nm, and 830 nm), resulting in signals of different wavelengths. A total of nine signals are obtained from the three detection points, which are received by the SPAD array. The specific reception method is as follows... Figure 3 As shown in Figure b, the SPAD array is a 32×32 pixel array. Nine circular detection areas A1-A9 represent the projection areas of the SPAD array corresponding to the nine detection fibers on the left. This design enables the system to acquire spatially and spectrally distributed signals with high sensitivity. Furthermore, compared to the method of acquiring signals after alternating laser illumination, this parallel acquisition design effectively improves the system's temporal resolution. There is a spatial gap between the detection fibers and the SPAD array. Propagation in this space amplifies the speckle pattern appearing in the detection fibers. Too close a distance results in multiple independent speckles per SPAD pixel, reducing speckle contrast; too far a distance reduces the signal-to-noise ratio. To match the speckle size with the effective area of ​​the SPAD array, the distance between the fiber and the SPAD array is... The following relationship is used to determine this:

[0057] The speckle size is where It is the wavelength of the illumination laser. This involves detecting the core diameter of the optical fiber. By adjusting the distance between the optical fiber and the SPAD array, the speckle size can be made close to the effective area size of the SPAD pixel, resulting in each pixel containing only one independent speckle, thereby improving speckle contrast and data signal-to-noise ratio. In this embodiment of the invention, the following method is used: The multimode detection fiber, the effective pixel area size of the 32×32 SPAD array is... Therefore, the distance between the detection fiber and the SPAD array should be adjusted to 8.5. The data collected by the SPAD array will be transmitted to a computer, and high-precision, high-sensitivity blood flow velocity and blood oxygen saturation measurements will be achieved based on multi-wavelength, multi-site data.

[0058] like Figure 4 The diagram shown is a flowchart of the high-speed blood oxygenation and blood flow measurement method based on multimodal data according to an embodiment of the present invention. The specific implementation steps are as follows: Step 1: The data acquired by the SPAD array in the above system is a speckle intensity fluctuation sequence of 32×32 pixels. This 32×32 sequence is then divided into... Each group represents speckle data collected from different detection sites at different wavelengths. Each group represents the speckle intensity fluctuation sequence for each pixel within a different region.

[0059] Step 2: Calculate the normalized autocorrelation curve for the speckle intensity fluctuation sequence corresponding to all pixels in each region, and take the average of the autocorrelation curves corresponding to all pixels in the region as the average autocorrelation curve for that region.

[0060] Furthermore, to quantify the fluctuations of the dynamic scattering field within biological tissue, the following equation was used to calculate the normalized time-intensity autocorrelation function of each pixel within the region. :

[0061] in It is the speckle intensity fluctuation sequence corresponding to the j-th pixel in the i-th SPAD region of the array at a given time t. It is a delay or correlation time. In this embodiment of the invention, each autocorrelation function... exist =Calculated over 10,000 time points, combined with SPAD's time sampling rate =3μs indicates that each A new autocorrelation curve is generated every 30 ms, called the "autocorrelation rate," which is the effective measurement rate of this system. Based on past experience and experimental results, it is known that when the delay time... When the time exceeds 200 μs, the autocorrelation curve has already decayed to nearly 1, and subsequent correlation data cannot provide important information about tissue dynamics. Therefore, the time delay is... The range is set to 0 to 300 μs, with intervals of . =3μs, with a total of 101 delay time points.

[0062] Furthermore, in calculating all SPAD pixels in each region Then, the autocorrelation curves within the region are averaged to obtain the average autocorrelation curve for each region. :

[0063] in, This represents the number of SPAD pixels in the i-th region. This averaging operation effectively reduces the impact of random noise, thereby generating a curve with a higher signal-to-noise ratio, providing a more reliable input for subsequent neural network model training.

[0064] Step 3: For the speckle intensity fluctuation sequence corresponding to all pixels in each region, perform the same averaging operation as in Step 2 to average these time series. Then, downsample the average intensity fluctuation curve of each region to 101 points to make it match the average autocorrelation curve. Curves with the same sequence length are called downsampling intensity fluctuation curves. Having curves of the same length provides unified and structured input data for subsequent neural network models, thus enabling more effective feature extraction and pattern recognition.

[0065] Step 4: Input the nine sets of average autocorrelation curves and downsampling intensity fluctuation curves corresponding to the nine regions into a multi-branch fusion neural network for calculation, and finally achieve high-precision and high-sensitivity blood oxygen and blood flow measurement.

[0066] like Figure 5 The diagram shown is a neural network architecture diagram in the algorithm module of this invention. The average autocorrelation curve and downsampling intensity fluctuation curve obtained by steps 1-3 are used as inputs to the multi-branch fusion neural network. Independent features are extracted through different branch networks, and the final result is calculated after fusing multiple features.

[0067] Furthermore, for the autocorrelation curve branches, the nine average autocorrelation curves are first arranged in different ways to form two modes of data: Mode 1 combines data from the same probe site. The first mode concatenates three data points (e.g., data from regions A1, A4, and A7) to obtain 3-channel multi-wavelength data containing information at different wavelengths. The second mode concatenates three data points (e.g., data from regions A1, A2, and A3) that have passed through the same filter to obtain 3-channel multi-site data containing information at different locations. Subsequently, both data streams are processed by a convolutional module to extract their respective internal temporal features, as detailed below. Figure 6 As shown, multi-scale features are extracted through three consecutive layers of one-dimensional convolutions, with the kernel size decreasing layer by layer (7, 5, and 3 respectively), thus achieving gradual capture of features from local to global. Batch normalization and ReLU activation functions are added after each convolution layer to accelerate model convergence, improve training stability, and enhance non-linear expressive power. The final convolutional module uses a max-pooling layer of size 2, which can retain important features while reducing their dimensionality. After local features are extracted from the two data streams through the convolutional module, feature fusion is achieved by concatenation, and finally, a fully connected layer outputs the 64-dimensional feature vector of this branch. For the intensity fluctuation curves, nine downsampled intensity fluctuation curves are input into a two-layer LSTM network for feature extraction. The input data shape is (9, 101), representing 101 time steps, each containing nine light intensity values. The hidden layer size of the LSTM is set to 64, meaning that each time step of the sequence will be encoded as a 64-dimensional feature vector to capture temporal dependence and dynamic characteristics. After processing through two layers of LSTM, a 64-dimensional global feature representation is generated, which serves as the feature output for this branch. Finally, the features from the two branches are concatenated and fused before being input into a fully connected layer, ultimately outputting blood oxygen saturation and blood flow velocity.

[0068] Example 3 This invention provides a real-time blood oxygen and blood flow measurement device based on multi-site, multi-wavelength parallel acquisition, such as... Figure 7 As shown, the device includes: The collection module 100 is used to acquire a multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and simultaneously acquiring data at multiple detection sites at different distances. The preprocessing module 200 is used to preprocess the multi-wavelength multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals. The calculation module 300 is used to input the average autocorrelation curve and the downsampling intensity fluctuation curve into a multi-branch fusion neural network, and calculate and output blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

[0069] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0070] Example 4 To implement the methods of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0071] Example 5 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0073] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0074] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for real-time measurement of blood oxygenation and blood flow based on multi-site, multi-wavelength parallel acquisition, characterized in that, include: S1, obtain the multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and acquiring data in parallel at multiple detection sites at different distances; S2, preprocess the multi-wavelength multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals; S3, input the average autocorrelation curve and the downsampling intensity fluctuation curve into the multi-branch fusion neural network, and calculate and output the blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

2. The method according to claim 1, characterized in that, The irradiation of the same location of biological tissue by multiple wavelength lasers also includes: pass Lasers of different wavelengths generate a light source, among which The wavelength is between red and near-infrared light; This allows the laser to simultaneously irradiate the same location on the surface of biological tissue via multiple multimode optical fibers.

3. The method according to claim 1, characterized in that, The multi-wavelength, multi-site speckle intensity fluctuation sequence acquired in parallel from multiple detection sites at different distances also includes: Set on the surface of the biological tissue There are 10 detection points, each at a different distance from the laser illumination point. ; At each detection point A bundle of multimode optical fibers is used for signal transmission, and the signal is divided into... Each signal path passes through different filters to obtain a signal of a specific wavelength.

4. The method according to claim 1, characterized in that, The preprocessing of the multi-wavelength, multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each signal group further includes: For each region, a normalized autocorrelation curve is calculated for the speckle intensity fluctuation sequence corresponding to all detector pixels. The average of all curves within the region is then used as the final result for that region. For the average autocorrelation curve; The average speckle intensity fluctuation sequence corresponding to all detector pixels in each region is taken, and the averaged result is downsampled to have the same length as the average autocorrelation curve, thus obtaining... The downsampling intensity fluctuation curve.

5. The method according to claim 1, characterized in that, The step of inputting the average autocorrelation curve and the downsampling intensity fluctuation curve into a multi-branch fusion neural network, and calculating and outputting blood oxygen saturation and blood flow velocity through the neural network, further includes: Will The average autocorrelation curves are arranged in different ways to form two modes of data, including: data from the same probe site. By concatenating the data from beginning to end, we obtain a result containing information about different wavelengths. Multi-wavelength data for multiple channels; and data that will pass through the same filter. By concatenating the data from beginning to end, we can obtain data containing information about different loci. Channel multi-site data; Will contain information of different wavelengths Channel multi-wavelength data and data containing information at different locations Multi-point data from each channel is used as two input data streams and fed into a three-layer one-dimensional convolutional neural network for temporal feature extraction. The kernel size of each layer of the one-dimensional convolutional neural network decreases layer by layer to achieve step-by-step extraction of signal features from local to global time features. After each convolutional layer, batch normalization and ReLU activation are performed sequentially. Then, max pooling is used to reduce the dimensionality of the features to retain key information. The two time features extracted by the three-layer one-dimensional convolutional neural network are concatenated and fused, and then input into the fully connected layer to obtain features based on the average autocorrelation curve. The downsampled intensity fluctuation curve is input into a time-series feature extraction model containing a two-layer long short-term memory network to extract the time dependence and dynamic change characteristics of the signal and generate high-dimensional time features. The high-dimensional temporal features are fused with features extracted based on the average autocorrelation curve and input into a fully connected layer to output the estimated results of blood oxygen saturation and blood flow velocity.

6. A real-time blood oxygen and blood flow measurement device based on multi-site, multi-wavelength parallel acquisition, characterized in that, include: The collection module is used to acquire the multi-wavelength, multi-site speckle intensity fluctuation sequence obtained by irradiating the same location of biological tissue with multiple wavelength lasers and collecting data in parallel at multiple detection sites at different distances; The preprocessing module is used to preprocess the multi-wavelength, multi-site speckle intensity fluctuation sequence to obtain the average autocorrelation curve and downsampled intensity fluctuation curve corresponding to each group of signals. The calculation module is used to input the average autocorrelation curve and the downsampling intensity fluctuation curve into a multi-branch fusion neural network, and calculate and output blood oxygen saturation and blood flow velocity through the multi-branch fusion neural network.

7. An electronic device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images