Method and system for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid ultrasound

Abstract

The application provides a cervical artery ultrasound dynamic evolution-based hypotension prediction method and system after spinal anesthesia, relates to the technical field of medical monitoring and prediction, and comprises the following steps: acquiring a cervical artery blood flow signal before a parturient woman is subjected to spinal anesthesia, and preprocessing to obtain blood flow velocity time series data; modeling the blood flow velocity time series data, and generating a three-dimensional ultrasound hemodynamic diagram; constructing a multi-modal feature vector from the three-dimensional ultrasound hemodynamic diagram and the original signal; calculating a scalar index quantifying the degree of hemodynamic evolution caused by body position change based on the multi-modal feature vector; and fusing the multi-modal feature vector and the scalar index, and outputting the risk probability of hypotension of the parturient woman after spinal anesthesia; the application dynamically learns and analyzes hemodynamic parameters and changes thereof of a spinal anesthesia parturient woman in a lateral decubitus position and a supine position, constructs a prediction model of intraoperative hypotension, realizes high-precision hypotension risk prediction, thereby reduces the incidence of intraoperative complications, and guarantees the safety of the patient.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring and prediction technology, specifically to a method and system for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound. Background Technology

[0002] Spinal anesthesia is a commonly used anesthesia method for cesarean section, but the incidence of supine hypotension syndrome (SHS) caused by it is as high as 50%, which is a major cause of adverse outcomes such as maternal nausea, vomiting and even fetal distress. At present, there is no effective means of actively warning of spinal anesthesia hypotension in clinical practice.

[0003] Existing technologies mainly rely on invasive hemodynamic monitoring, such as artificial intelligence tools based on arterial waveforms (e.g., hypotension prediction index). Although other methods obtain carotid artery blood flow parameters through ultrasound (e.g., corrected carotid flow time FTc, velocity-time integral VTI), they focus on general signal analysis and parameter calculation, without addressing the specific physiological scenario of spinal anesthesia in postpartum women, nor do they involve dynamic feature evolution and risk prediction modeling under positional changes.

[0004] Current research on predicting SHS mostly relies on static blood flow parameters (such as heart rate and blood pressure baseline) at a single moment, which has limited predictive efficacy. Although some studies have applied RBF (radial basis function) neural network modeling to signals such as respiration and electrocardiogram, its application field (adenoid hypertrophy), signal source (respiratory flow rate), and modeling purpose (morphological recognition) are completely different from this application scenario, and it has not revealed how to quantify the changes in dynamic processes for risk prediction.

[0005] Prenatal positional changes in parturients before spinal anesthesia (such as changing from supine to lateral decubitus) cause complex and nonlinear hemodynamic redistribution. The information contained in this dynamic transition process is crucial for predicting the vascular response to anesthetic drugs, but current technologies have not captured or analyzed this information.

[0006] Therefore, existing technologies mainly rely on static parameter monitoring, which fails to effectively capture and quantify the continuous and nonlinear dynamic evolution characteristics of carotid artery hemodynamics during changes in the mother's position before anesthesia, resulting in insufficient accuracy in predicting hypotension after spinal anesthesia during cesarean section. Summary of the Invention

[0007] To address the aforementioned issues, this invention proposes a method and system for predicting postoperative hypotension after spinal anesthesia based on dynamic evolution of carotid ultrasound. By dynamically learning and analyzing the hemodynamic parameters and their changes in the lateral and supine positions of spinal anesthetized mothers, a predictive model for intraoperative hypotension is constructed, achieving high-precision prediction of hypotension risk, thereby reducing the incidence of intraoperative complications and ensuring patient safety.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid ultrasound includes: Carotid artery blood flow signals were acquired before spinal anesthesia was administered to the mother. The signals covered two body position stages: stable supine position and stable lateral position. The acquired raw signals were preprocessed into time-series blood flow velocity data. Using an adaptive radial basis function neural network, local dynamic system identification and modeling of blood flow velocity time-series data at each body position stage are performed to generate a three-dimensional ultrasound hemodynamic map; Features in the time domain, frequency domain, and nonlinear dynamic domain are extracted from the three-dimensional ultrasound hemodynamic maps and raw signals at each body position stage to construct multimodal feature vectors for each body position stage. Based on multimodal feature vectors, a scalar index is calculated to comprehensively quantify the degree of hemodynamic evolution caused by changes in body position by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals. The extracted multimodal feature vectors are fused with a scalar exponent and input together into a trained machine learning prediction model to output the probability of postpartum hypotension.

[0009] According to some embodiments, the present invention adopts the following technical solution: A post-spinal anesthesia hypotension prediction system based on dynamic evolution of carotid ultrasound includes: The signal acquisition module is configured to acquire carotid artery blood flow signals of the parturient before spinal anesthesia, the signals covering two body position stages: stable supine position and stable lateral position, and preprocess the acquired raw signals into blood flow velocity time-series data; The dynamic modeling module is configured to: use an adaptive radial basis function neural network to identify and model the local dynamic system of blood flow velocity time series data at each body position stage, and generate a three-dimensional ultrasound hemodynamic map; The feature extraction module is configured to extract features in the time domain, frequency domain, and nonlinear dynamic domain from the three-dimensional ultrasound hemodynamic map and raw signal at each body position stage, and construct multimodal feature vectors for each body position stage. The index calculation module is configured to: calculate a scalar index that comprehensively quantifies the degree of hemodynamic evolution caused by changes in body position based on multimodal feature vectors by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals; The risk prediction module is configured to fuse the extracted multimodal feature vectors with a scalar exponent and input them into a trained machine learning prediction model to output the probability of postpartum hypotension.

[0010] According to some embodiments, the present invention adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the method for predicting post-spinal hypotension based on dynamic evolution of carotid ultrasound.

[0011] According to some embodiments, the present invention adopts the following technical solution: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid ultrasound.

[0012] According to some embodiments, the present invention adopts the following technical solution: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the method for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid ultrasound.

[0013] 0. Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires carotid artery blood flow signals covering two body positions: stable supine and stable lateral decubitus. It then performs local dynamic system identification and modeling of the time-series blood flow velocity data for each position to generate a three-dimensional ultrasound hemodynamic map. This captures the continuous dynamic evolution of hemodynamics caused by positional changes, overcoming the limitations of monitoring single static parameters. Furthermore, it extracts multimodal feature vectors in the time domain, frequency domain, and nonlinear dynamic domain from the three-dimensional ultrasound hemodynamic map and the original signal. Combining the relative changes in feature vectors between the two stages with the divergence of the blood flow velocity signal probability distribution, a comprehensive scalar exponent is calculated. This effectively quantifies the nonlinear redistribution information of hemodynamics implicit in the positional transition into computable features. Finally, the multimodal feature vectors and the scalar exponent are fused and input into a machine learning prediction model, enabling a probabilistic output of the risk of hypotension after spinal anesthesia. This provides a high-precision proactive warning before anesthesia, reducing the risk of intraoperative hypotension and related complications. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figure 1 This is a flowchart of the method in Example 1.

[0016] Figure 2This is a time-series data diagram of blood flow velocity after preprocessing in Example 1.

[0017] Figure 3 This is an ultrasound hemodynamic diagram of Example 1. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0021] Example 1 One embodiment of the present invention provides a method for predicting postoperative hypotension after spinal anesthesia based on dynamic evolution of carotid ultrasound. By dynamically learning and analyzing the hemodynamic parameters and their changes in the lateral and supine positions of women undergoing spinal anesthesia, a predictive model for intraoperative hypotension is constructed, achieving high-precision prediction of hypotension risk, thereby reducing the incidence of intraoperative complications and ensuring patient safety. Figure 1 As shown, the specific steps are as follows: S1. Acquire carotid artery blood flow signals of the parturient before spinal anesthesia, the signals covering two body position stages: stable supine position and stable lateral position, and preprocess the acquired raw signals into blood flow velocity time series data; Specifically, before administering spinal anesthesia to the mother, carotid artery blood flow signals were continuously acquired using an ultrasound Doppler probe. The acquisition process covered both stable supine (T0 phase) and stable lateral decubitus (T1 phase) positions. The signals acquired in the supine position were used as the blood flow baseline. Morphological processing and noise reduction were performed on the acquired raw ultrasound images. The blood flow velocity signals were then uniformly resampled to 500Hz high-frequency time-series data through interpolation. Figure 2 The blood flow velocity time series data shown .

[0022] S2. Using an adaptive radial basis function neural network, local dynamic system identification and modeling are performed on the time-series data of blood flow velocity at each body position stage to generate a three-dimensional ultrasound hemodynamic map. Furthermore, the identification and modeling of the local dynamic system involves the following specific steps: Based on physiological priors during pregnancy, time delays were applied to the blood flow velocity time-series data. and embedding dimension The optimized selection reconstructs the high-dimensional phase space trajectory; An adaptive radial basis function neural network is used to locally approximate the dynamics of the hemodynamic system during pregnancy in the reconstructed phase space; The approximation results are visualized to generate a three-dimensional ultrasound hemodynamic map.

[0023] Furthermore, the adaptive radial basis function neural network employs an adaptive weight update rule, integrating the local Lyapunov index of carotid artery blood flow during pregnancy and a physiological state correction term during pregnancy, to dynamically adjust the neural network weights, as expressed by the formula:

[0024]

[0025] in, For local Lyapunov exponents and instantaneous speed Adaptive adjustment factor; For state The estimation error, This is a correction item for physiological conditions during pregnancy; It is the neural network weight update rate; and These are the weighting coefficients for the adaptive adjustment factor; Centered on the j-th neuron The Gaussian radial basis function with kernel.

[0026] Specifically, one of the core innovations of this embodiment lies in the customized dynamic modeling scheme for the unique physiological characteristics of cesarean section mothers during pregnancy, constructing a three-dimensional ultrasound hemodynamic map that reflects blood flow stability, breaking through the limitations of traditional general modeling in adapting to special physiological scenarios, and achieving accurate identification and visualization of blood flow dynamics.

[0027] The time-series signals of blood flow velocity at each stage obtained from S1 A radial basis function (RBF) neural network incorporating physiological prior constraints during pregnancy is used for the identification and modeling of local dynamic systems, specifically including: S201, Pregnancy-specific phase space reconstruction: Carotid artery blood flow during pregnancy exhibits a unique pattern of "high flow velocity fluctuations, low stability tolerance, and delayed compensatory response." These characteristics are fundamentally different from those in non-pregnant individuals and other physiological scenarios (such as respiration and electrocardiogram signals). Under the aforementioned physiological prior constraints of pregnancy, a time delay is specified. and embedding dimension The range.

[0028] Furthermore, the CC method is used to calculate the optimal delay time τ and embedding dimension m (e.g., τ=10, m=3). The CC method quantifies the nonlinear temporal correlation characteristics of the carotid artery blood flow signal during pregnancy by using correlation integrals and feature statistics to select the optimal time delay τ. Based on τ, the effectiveness of phase space reconstruction under different embedding dimensions is verified by the convergence test of the false nearest neighbor rate, thus achieving the selection of the optimal embedding dimension m. The entire calculation is performed within the parameter range limited by the physiological prior of pregnancy, which avoids the modeling bias and computational resource consumption of unconstrained search, and ensures that the optimal parameters are highly adapted to the unique signal characteristics of carotid artery blood flow during pregnancy.

[0029] Furthermore, the one-dimensional blood flow velocity time-series signal Phase space reconstruction is performed using the optimal delay time τ and embedding dimension m to restore the state of the carotid artery blood flow dynamic system, expressed by the formula:

[0030] in, To obtain the phase space sampling data after reconstruction, the phase space is sampled. , These are the sampling points.

[0031] S202, Adaptive Local RBF Network Modeling: The RBF network used for modeling system dynamics (i.e., local dynamics system identification) adopts an existing three-layer structure of input layer, hidden layer, and output layer. The input layer receives sampled data from the reconstructed phase space. The input is then fed into the hidden layer; each neuron in the hidden layer calculates the input and the neuron center. The Euclidean distance between them, expressed by the Gaussian function The activation values ​​are obtained, and all outputs are used to construct the regression subvector. ,in, The neuron number; the output layer will regress the subvectors. and corresponding connection weights Linear weighting yields an estimate of the system's dynamic characteristics, expressed by the formula:

[0032] in, , These are the hemodynamic characteristics of the carotid artery in various dimensions. It is the weight after the weight update is completed. It is a regressor subvector. It is the first A precise mathematical representation of 3D dynamic properties. To model residuals, visualizing the dynamic characteristics can generate results such as... Figure 3 The ultrasound hemodynamic diagram shown.

[0033] Within the reconstructed phase space, the dynamics of the carotid artery blood flow system are dynamically approximated locally. Another core innovation of this embodiment is the proposal of an adaptive weight update rule specific to carotid artery blood flow during pregnancy, which overcomes the defect of the traditional RBF network's "general parameter update" in being unable to adapt to special physiological states.

[0034] Center vector, width parameter, and connection weights of RBF network Instead of fixed or simple gradient updates, the weights are dynamically adjusted by incorporating the local Lyapunov index of carotid artery blood flow during pregnancy and a physiological state correction term during pregnancy. This allows the network to more accurately learn the nonlinear dynamic patterns of blood flow at different stages of pregnancy and in different individuals. This principle is expressed as:

[0035]

[0036] in, For local Lyapunov exponents and instantaneous speed An adaptive adjustment factor is used to enhance the learning rate and flexibility of the network. Specifically, the local Lyapunov exponent... Based on the reconstructed phase space trajectory x(k), a small-data-volume local orbit algorithm is used to solve the trajectory segment by segment within the constraints of physiological parameters during pregnancy; instantaneous velocity From the original pre-processed blood flow velocity timing Extracted point by point according to the sampling time sequence. For state The estimation error, The physiological state correction term during pregnancy is obtained by fitting the gestational age (G) and body mass index (BMI) of the mother. It is used to correct the influence of blood volume changes at different gestational ages and vascular elasticity differences at different BMIs on the modeling. It is the neural network weight update rate; and The weighting coefficients for the adaptive adjustment factor were obtained through cross-validation calibration using clinical samples. Centered on the j-th neuron The Gaussian radial basis function with kernel.

[0037] S3. Extract the features in the time domain, frequency domain, and nonlinear dynamic domain from the three-dimensional ultrasound hemodynamic maps and raw signals at each body position stage to construct the multimodal feature vectors for each body position stage. Furthermore, the multimodal feature vector includes time-domain parameters, frequency-domain parameters, and nonlinear dynamic parameters. The nonlinear dynamic parameters are extracted from the three-dimensional ultrasound hemodynamic map and include at least local attractor morphology indices, spectral fitting index, maximum Lyapunov index, and maximum rate of change.

[0038] Specifically, from the ultrasound hemodynamic map and raw signal generated by S2, multimodal feature sets in the time domain, frequency domain, and nonlinear domain are extracted, including: (1) Time-domain parameters: carotid corrected flow time (FTc), velocity-time integral (VTI), mean flow velocity (MFV), pulsatility index (PI), and resistance index (RI).

[0039] (2) Frequency domain parameters: The power spectrum is calculated by fast Fourier transform (FFT), and the spectrum fit index (SFI), spectrum centroid (SC), and low-frequency to high-frequency energy ratio (LF / HF) are extracted.

[0040] (3) Nonlinear dynamic parameters: Extract nonlinear features such as local attractor morphology index, spectral fit index (DSFI), maximum Lyapunov index (used to quantify the sensitivity of the carotid blood flow dynamic system to initial conditions), and maximum rate of change (Rate_max) from the three-dimensional dynamic diagram.

[0041] S4. Based on multimodal feature vectors, a scalar index is calculated to comprehensively quantify the degree of hemodynamic evolution caused by changes in body position by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals. Furthermore, the scalar index comprises two parts: the relative characteristic changes between the lateral and supine positions, and the asymmetric shift of the lateral position blood flow velocity signal probability distribution from the baseline distribution, expressed by the formula:

[0042] Wherein, FT0 is the multimodal feature vector for the supine position, and FT1 is the multimodal feature vector for the lateral position. The norm of a vector This represents the relative characteristic changes between the lateral and supine positions. The KL divergence of the probability distribution of blood flow velocity signals in the lateral and supine positions. These are the weighting coefficients.

[0043] Specifically, the third core innovation of this embodiment lies in quantifying the deviation of the hemodynamic index caused by lateral decubitus position relative to the blood flow baseline, namely the carotid artery hemodynamic evolution index (ΔCBDI), which is a comprehensive scalar and is calculated using the following formula:

[0044] Among them, FT0 is the blood flow baseline (i.e., the multimodal dynamic feature vector in the supine position), and FT1 is the multimodal dynamic feature vector of blood flow extracted in the lateral position, which comprehensively characterizes the hemodynamic state in the lateral position. The norm of a vector Calculate the relative characteristic changes between steady states to represent the deviation between the lateral decubitus position and the blood flow baseline; It is the probability of blood flow velocity signal in the lateral decubitus and supine positions; The KL divergence of blood flow velocity signal probability distribution in lateral and supine positions measures the asymmetric shift of the lateral position blood flow velocity signal probability distribution from the baseline distribution, captures "distribution morphology abnormalities" that cannot be covered by traditional parameter difference, and reflects the intrinsic stability changes of hemodynamics; α and β are weighting coefficients, which are determined through training and optimization using historical data and satisfy α+β=1.

[0045] The ΔCBDI index combines static bias and distribution shift, providing a highly condensed and physically meaningful input feature for machine learning models.

[0046] S5. The extracted multimodal feature vectors and scalar exponents are fused together and input into the trained machine learning prediction model to output the risk probability of postpartum hypotension caused by spinal anesthesia.

[0047] A clinically prior-driven synergistic prediction method combining "multimodal features + ΔCBDI index" is employed, specifically, a set of clinical risk factors is constructed. Each factor is assigned a value of 0 (none) or 1 (yes).

[0048] Furthermore, construct Clinical prior-feature association matrix ( To the total number of features to be merged, (Number of risk factors), matrix elements Indicates the first The first clinical risk factor and the multimodal feature vector. The association strength of each characteristic was determined based on clinical guidelines and statistical analysis of historical sample data. This was based on individual risk factors. Vector automatic adjustment of feature weights This ensures that features corresponding to high-risk clinical factors receive higher weighting.

[0049] Based on this, the multimodal dynamic feature set extracted by S3 will be... (As a high-dimensional fine-grained feature) is fused with the ΔCBDI index calculated by S4 (as a comprehensive evolution index) to form a weighted feature vector of the adapted individual, which is used as the input of the prediction model.

[0050] Specifically, the comprehensive evolution index ΔCBDI is denoted as... To adapt to high-dimensional feature fusion, it is expanded into a single-dimensional feature vector. Introducing individual fit weighting coefficients and Satisfying normalization constraints ,in Weights for multimodal fine-grained features, The weights are the comprehensive evolution index of ΔCBDI.

[0051] For high-dimensional feature vectors and the extended ΔCBDI eigenvector Normalization is performed separately (to eliminate dimensional differences) to obtain normalized eigenvectors. and Ultimately, the weighted fusion feature vector is adapted to the individual. The fusion formula is:

[0052] in, This indicates a feature dimension splicing operation, which achieves the dimensional fusion of high-dimensional fine-grained features and comprehensive evolution indicators.

[0053] In this embodiment, an improved interpretable ensemble tree model is used as the prediction model to predict risk, and the optimal prediction threshold is determined by the receiver operating characteristic (ROC) curve. The model output is the probability of postpartum hypotension in the mother.

[0054] The following is a specific example: Preparation stage: Connect the ultrasound equipment and ensure that the probe frequency and settings are suitable for carotid artery blood flow detection; the mother should lie supine and breathe calmly.

[0055] Data acquisition: Start continuously recording the Doppler spectrum of carotid artery blood flow, i.e., carotid artery blood flow signal; assist the mother to smoothly turn to the lateral decubitus position and continue recording until the lateral decubitus position is stable for 10 seconds; save video and spectrum data in both supine and lateral decubitus positions.

[0056] Signal preprocessing: Exported data, and performed median filtering and motion artifact removal on T0 (stable segment in supine position) and T1 (stable segment in lateral decubitus position) data respectively, and interpolated to generate 500Hz blood flow velocity time series data. .

[0057] Dynamic modeling and atlas construction: for each positional phase The optimal delay τ and embedding dimension m (e.g., τ=10, m=3) are calculated using the CC method to construct the phase space; the RBF network is initialized, and the learning rate η is increased by the adaptive factor Γ to quickly capture dynamic changes; the network is trained to approximate the dynamic system of carotid artery blood flow; the dynamic model is visualized in three-dimensional space to obtain a three-dimensional graph.

[0058] Feature extraction (S3): Calculate parameters such as FTc, VTI, PI, RI, SC, and DSFI for periods T0 and T1 from the signal and 3D graph to construct a multimodal feature vector. .

[0059] Calculate ΔCBDI (S4): Obtain the optimized weight coefficients from the database (e.g., α=0.7, β=0.3). Then, use the feature vector obtained in step 3... This yields a scalar value ΔCBDI (e.g., 0.85).

[0060] Risk prediction (S5): The multimodal feature vector from S3 and the scalar index obtained from S4 are fused and input into the trained improved ensemble model. After receiving the fused features, the ensemble model outputs a final risk probability of 0.75 and gives a decision rule. If this probability is higher than the preset threshold of 0.6, the system issues a "high risk" warning.

[0061] Clinical intervention: After receiving the warning, the anesthesiologist can prepare vasoactive drugs in advance and closely monitor the patient after spinal anesthesia, and deal with the condition in a timely manner, which effectively prevents the occurrence of severe hypotension.

[0062] The method provided in this embodiment has the following advantages: 1. Dynamic process capture: For the first time, multi-position ultrasound data was included in the analysis, capturing valuable information on hemodynamic changes, which surpasses the limitations of traditional static information (such as FTC).

[0063] 2. Innovative Modeling Methods: An adaptive RBF neural network modeling method under the physiological prior constraints of pregnancy is proposed. The individual physiological fingerprint is integrated into the phase space reconstruction parameter customization and weight update rules, which improves the model's recognition accuracy and generalization ability for complex physiological system dynamics.

[0064] 3. Index Quantification Innovation: The Comprehensive Evolution Index ΔCBDI is innovatively defined. Its calculation formula integrates static changes and probability distribution shifts, providing a powerful and novel feature engineering tool for risk prediction.

[0065] 4. Prediction model optimization: It effectively combines high-dimensional multimodal features and comprehensive evolution index, which significantly improves the accuracy and robustness of the prediction model (AUC reaches 0.80).

[0066] 5. Strong clinical applicability: This method can complete the assessment before anesthesia, providing clinicians with intuitive three-dimensional maps, quantitative indices, and clear risk probabilities, which helps to take preventive measures in advance and ensure maternal and infant safety.

[0067] Example 2 One embodiment of the present invention provides a post-spinal anesthesia hypotension prediction system based on dynamic evolution of carotid artery ultrasound, comprising: The signal acquisition module is configured to acquire carotid artery blood flow signals of the parturient before spinal anesthesia, the signals covering two body position stages: stable supine position and stable lateral position, and preprocess the acquired raw signals into blood flow velocity time-series data; The dynamic modeling module is configured to: use an adaptive radial basis function neural network to identify and model the local dynamic system of blood flow velocity time series data at each body position stage, and generate a three-dimensional ultrasound hemodynamic map; The feature extraction module is configured to extract features in the time domain, frequency domain, and nonlinear dynamic domain from the three-dimensional ultrasound hemodynamic map and raw signal at each body position stage, and construct multimodal feature vectors for each body position stage. The index calculation module is configured to: calculate a scalar index that comprehensively quantifies the degree of hemodynamic evolution caused by changes in body position based on multimodal feature vectors by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals; The risk prediction module is configured to fuse the extracted multimodal feature vectors with a scalar exponent and input them into a trained machine learning prediction model to output the probability of postpartum hypotension.

[0068] Example 3 One embodiment of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid artery ultrasound.

[0069] Example 4 In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, the method for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid artery ultrasound is implemented.

[0070] Example 5 One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the method for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid artery ultrasound.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0073] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound, characterized in that, include: Carotid artery blood flow signals were acquired before spinal anesthesia was administered to the mother. The signals covered two body position stages: stable supine position and stable lateral position. The acquired raw signals were preprocessed into time-series blood flow velocity data. Using an adaptive radial basis function neural network, local dynamic system identification and modeling of blood flow velocity time-series data at each body position stage are performed to generate a three-dimensional ultrasound hemodynamic map; Features in the time domain, frequency domain, and nonlinear dynamic domain are extracted from the three-dimensional ultrasound hemodynamic maps and raw signals at each body position stage to construct multimodal feature vectors for each body position stage. Based on multimodal feature vectors, a scalar index is calculated to comprehensively quantify the degree of hemodynamic evolution caused by changes in body position by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals. The extracted multimodal feature vectors are fused with a scalar exponent and input together into a trained machine learning prediction model to output the probability of postpartum hypotension.

2. The method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound as described in claim 1, characterized in that, The preprocessing involves morphological processing and filtering to reduce noise in the original signal, and then resampling the blood flow velocity signal into high-frequency time-series data through interpolation.

3. The method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound as described in claim 1, characterized in that, The specific steps for identifying and modeling the local dynamic system are as follows: Based on physiological priors during pregnancy, time delays were applied to the blood flow velocity time-series data. and embedding dimension The optimized selection reconstructs the high-dimensional phase space trajectory; An adaptive radial basis function neural network is used to locally approximate the dynamics of the hemodynamic system during pregnancy in the reconstructed phase space; The approximation results are visualized to generate a three-dimensional ultrasound hemodynamic map.

4. The method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound as described in claim 3, characterized in that, The adaptive radial basis function neural network employs an adaptive weight update rule, integrating the Lyapunov index of local carotid artery blood flow during pregnancy and a physiological state correction term during pregnancy, to dynamically adjust the neural network weights, as expressed by the formula: in, For local Lyapunov exponents and instantaneous speed Adaptive adjustment factor; For state The estimation error, This is a correction item for physiological conditions during pregnancy; It is the neural network weight update rate; and These are the weighting coefficients for the adaptive adjustment factor; Centered on the j-th neuron The Gaussian radial basis function with kernel.

5. The method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound as described in claim 1, characterized in that, The multimodal feature vector includes time-domain parameters, frequency-domain parameters, and nonlinear dynamic parameters. The nonlinear dynamic parameters are extracted from the three-dimensional ultrasound hemodynamic map and include at least local attractor morphology indices, spectral fitting index, maximum Lyapunov index, and maximum rate of change.

6. The method for predicting post-spinal anesthesia hypotension based on dynamic evolution of carotid artery ultrasound as described in claim 1, characterized in that, The scalar index consists of two parts: the relative characteristic changes between the lateral and supine positions, and the asymmetric shift of the lateral blood flow velocity signal probability distribution from the baseline distribution. It is expressed by the formula: in, The multimodal feature vector for the supine position. This is a multimodal feature vector for the lateral decubitus position. The norm of a vector This represents the relative characteristic changes between the lateral and supine positions. The KL divergence of the probability distribution of blood flow velocity signals in the lateral and supine positions. These are the weighting coefficients.

7. A system for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid artery ultrasound, characterized in that, include: The signal acquisition module is configured to acquire carotid artery blood flow signals of the parturient before spinal anesthesia, the signals covering two body position stages: stable supine position and stable lateral position, and preprocess the acquired raw signals into blood flow velocity time-series data; The dynamic modeling module is configured to: use an adaptive radial basis function neural network to identify and model the local dynamic system of blood flow velocity time series data at each body position stage, and generate a three-dimensional ultrasound hemodynamic map; The feature extraction module is configured to extract features in the time domain, frequency domain, and nonlinear dynamic domain from the three-dimensional ultrasound hemodynamic map and raw signal at each body position stage, and construct multimodal feature vectors for each body position stage. The index calculation module is configured to: calculate a scalar index that comprehensively quantifies the degree of hemodynamic evolution caused by changes in body position based on multimodal feature vectors by fusing the relative changes in feature vectors between two stages with the divergence of the probability distribution of blood flow velocity signals; The risk prediction module is configured to fuse the extracted multimodal feature vectors with a scalar exponent and input them into a trained machine learning prediction model to output the probability of postpartum hypotension.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for predicting post-spinal hypotension based on dynamic evolution of carotid ultrasound as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the method for predicting post-spinal hypotension based on dynamic evolution of carotid ultrasound as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the method for predicting hypotension after spinal anesthesia based on dynamic evolution of carotid ultrasound as described in any one of claims 1-6.

Images