Intelligent fusion analysis method and system for brain oxygen signals in high-altitude environment

Abstract

This invention relates to the field of high-altitude environmental health monitoring technology, and particularly to an intelligent fusion analysis method and system for brain oxygen signals in high-altitude environments. This method acquires and fuses multi-channel brain oxygen saturation, multi-dimensional physiological parameters, and time-varying environmental data to mine the synergistic characteristics of oxygen metabolism across brain regions and construct a spatiotemporal evolution tensor. Based on this, a multi-scale, time-dependent hypoxia susceptibility dynamic model is constructed, generating a prospective risk assessment benchmark. Furthermore, risk extrapolation and causal chain reconstruction of brain oxygen evolution are performed, outputting structured prediction results and dynamically recalibrating the model. This achieves accurate, prospective assessment and dynamic tracking of individual brain oxygen risk in high-altitude environments.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology in high-altitude environments, and in particular to a method and system for intelligent fusion analysis of brain oxygen signals in high-altitude environments. Background Technology

[0002] The unique low-pressure, low-oxygen environment of high-altitude regions poses a severe challenge to human physiological functions. Brain tissue is particularly sensitive to hypoxia, and changes in brain oxygen saturation are a key physiological indicator for assessing an individual's altitude acclimatization and risk of acute mountain sickness. Current technologies for monitoring and analyzing brain oxygenation status in high-altitude environments typically employ portable brain oxygenation monitoring devices based on near-infrared spectroscopy to continuously collect single-channel or limited-channel brain tissue blood oxygenation parameters.

[0003] Conventional analysis methods primarily focus on independent time-domain or frequency-domain analysis of the acquired raw brain oxygen saturation signals. This includes calculating the mean, trend, or power spectral density of specific frequency bands to assess the overall brain oxygenation level. Simultaneously, other common physiological parameters such as heart rate, blood oxygen saturation, and blood pressure are collected and analyzed in parallel with the brain oxygen data, performing simple linear correlation analysis or setting static thresholds for alarms. These methods treat the brain oxygen signal as an isolated, homogeneous whole.

[0004] However, such conventional approaches have significant limitations. Oxygen metabolism activities in different functional areas of the brain are not synchronous or homogeneous, but rather involve complex spatiotemporal coordination and compensatory mechanisms. Existing methods do not pay sufficient attention to the spatial heterogeneity of brain oxygen signals and lack in-depth exploration of the dynamic coordination patterns of oxygen metabolism among multiple brain regions, resulting in an inability to accurately characterize the networked features and reserve capacity of brain oxygen metabolism. Simply establishing a linear correlation between multidimensional physiological parameters and brain oxygen signals makes it difficult to reveal the complex nonlinear coupling relationships and multi-timescale dynamic feedback processes behind them. For example, it is difficult to effectively distinguish and provide early warning of chronic compensatory regulation and acute decompensation critical states. Summary of the Invention

[0005] This invention provides a method and system for intelligent fusion analysis of brain oxygen signals in high-altitude environments, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides an intelligent fusion analysis method for brain oxygen signals in high-altitude environments, comprising:

[0007] The system acquires multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. It performs spatiotemporal multi-scale separation and reconstruction on the multi-channel brain oxygen saturation data, mines the features of brain inter-brain oxygen metabolism synergistic patterns, and constructs a brain oxygen saturation spatiotemporal evolution tensor.

[0008] Based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensatory limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector.

[0009] The hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A forward-looking risk assessment benchmark is generated by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state.

[0010] Based on the aforementioned forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction. The triggering factors and propagation paths of hypoxia risk are traced and quantified in reverse, and structured prediction results are output.

[0011] Based on the structured prediction results, subsequent multi-channel brain oxygen saturation data are continuously tracked, and the multi-scale temporal dependence in the hypoxia susceptibility dynamic model is dynamically recalibrated in parameter space using the deviation between the actual observed data and the predicted trajectory.

[0012] The multi-channel brain oxygen saturation data were subjected to spatiotemporal multi-scale separation and reconstruction to mine the features of coordinated oxygen metabolism patterns between brain regions, and a spatiotemporal evolution tensor of brain oxygen saturation was constructed, including:

[0013] Variational mode decomposition was performed on multi-channel brain oxygen saturation data, and the signals of each channel were adaptively decomposed into a set of intrinsic mode components representing different physiological time scales. The intrinsic mode component sets corresponded to the rapid brain oxygen fluctuation layer, the respiratory heart rate coupling layer and the baseline metabolic drift layer in descending order of characteristic frequency. The instantaneous frequency envelope and instantaneous phase envelope of each intrinsic mode component were extracted as multi-scale time domain feature vectors.

[0014] Based on the cross-scale phase coupling relationship between the intrinsic modal component set and the channels of different brain functional areas, a brain inter-modal coupling tensor is constructed. The multi-scale temporal domain feature vector is spatiotemporally embedded with the brain inter-modal coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modal scale, and time evolution as each dimension, the spatiotemporal evolution tensor of brain oxygen saturation is obtained.

[0015] The multi-scale temporal feature vectors are spatiotemporally embedded with the brain inter-region modality coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modality scales, and temporal evolution as dimensions, the spatiotemporal evolution tensor of brain oxygen saturation is obtained, which includes:

[0016] Based on the phase-locking index and phase amplitude modulation index in the brain inter-modal coupling tensor, a cross-scale coupling propagation operator is constructed. The cross-scale coupling propagation operator maps the local representation of multi-scale temporal feature vectors in each intrinsic modal scale dimension to a global representation that integrates cross-scale synergistic effects by encoding the nonlinear interaction mechanism between different intrinsic modal scales.

[0017] The global representation is tensor-organized according to the spatial topological relationship of the brain functional area channels. Based on the anatomical location and functional connectivity strength of the brain functional area corresponding to each channel, a brain region spatial adjacency structure is constructed. The brain region spatial adjacency structure embeds the global representation of discrete channels into tensor slices that preserve the spatial dependence between brain regions by quantifying the topological distance and functional synergy between brain functional area nodes.

[0018] The tensor slices are stacked sequentially along the time evolution direction to construct a third-order tensor structure with brain functional area channels as the spatial dimension, intrinsic modality scale as the scale dimension, and time series as the temporal dimension.

[0019] The third-order tensor structure is rearranged dimensionally according to the topological centrality index of brain functional area nodes in the spatial dimension and the characteristic frequency of intrinsic modal components in the scale dimension. This eliminates redundant dependencies between dimensions and highlights the dominant cooperative mode, resulting in the spatiotemporal evolution tensor of brain oxygen saturation.

[0020] Based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. This hypoxia susceptibility dynamic model characterizes the dynamic boundary between an individual's brain oxygen metabolism reserve capacity and compensatory limit through multi-time-scale physiological feedback loops, resulting in a hypoxia susceptibility state vector including:

[0021] We quantify the cross-scale coupling strength of brain region oxygen metabolism coordination pattern characteristics and multidimensional physiological parameter data by constructing a conditional transfer entropy matrix under multiple time scale windows.

[0022] Based on the conditional transfer entropy matrix, closed-loop physiological feedback loops are identified. By detecting brain oxygen metabolism coordination patterns and physiological parameter pairs that exhibit bidirectional transfer entropy and fast-slow transfer delay characteristics, a set of feedback loops is extracted.

[0023] The set of feedback loops is constructed as a coupled set of dynamic differential equations. By defining the brain oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established.

[0024] Phase space trajectory analysis was performed on the hypoxia susceptibility dynamic model. By solving the phase space evolution trajectory under different initial reserve states and physiological disturbance intensities, the set of critical bifurcation points that lead to trajectory escape from the stable attraction domain were identified. The set of critical bifurcation points constitutes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit. The phase space coordinates of the dynamic boundary are represented as a hypoxia susceptibility state vector.

[0025] The feedback loop set is constructed as a coupled set of dynamic differential equations. By defining the cerebral oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established, including:

[0026] The feedback loops in the feedback loop set are decomposed into a structure. By identifying the brain oxygen metabolism coordination mode nodes and physiological parameter nodes in each feedback loop, the causal transmission paths and transmission delays between nodes are extracted, and a feedback loop topology diagram is constructed.

[0027] Based on the feedback loop topology diagram, the system phase variables and external driving variables are defined. The brain oxygen metabolism reserve state and its consumption rate and recovery rate at different time scales are defined as the system phase variable set, and the multidimensional physiological parameters and their fluctuation amplitude and change trend at different time scales are defined as the external driving variable set.

[0028] Based on the causal propagation path and propagation delay in the feedback loop topology diagram, each feedback loop is expressed as a differential equation, and by introducing delay terms and coupling terms, the differential equations of multiple feedback loops are coupled into a system of dynamic differential equations.

[0029] The dynamic differential equations are subjected to time-scale separation transformation. By solving the rapidly changing system phase variables and the slowly changing system phase variables at different time scales, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed.

[0030] The hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A forward-looking risk assessment benchmark is generated by constructing a nonlinear mapping function between environmental stress factors and hypoxia susceptibility states, including:

[0031] Cross-modal semantic projection is performed on the hypoxia susceptibility state vector and time-varying environmental parameter data. By calculating the mutual information transmission strength between each dimension component of the hypoxia susceptibility state vector and each dimension component of the time-varying environmental parameter data in the embedding space, a cross-modal correlation matrix is ​​constructed.

[0032] Based on the cross-modal correlation matrix, an environmental stress factor set is extracted. By identifying environmental parameter dimensions in the matrix whose influence weight exceeds a preset weight threshold, an environmental stress factor set is constructed.

[0033] The set of environmental stress factors and the hypoxia susceptibility state vector are constructed into a nonlinear mapping function. The nonlinear mapping function is then prospectively extrapolated. By inputting the predicted sequence of environmental parameters within a future time window into the nonlinear mapping function, the evolution trajectory of the hypoxia susceptibility state vector is solved. The minimum distance between the evolution trajectory and the dynamic boundary of the compensation limit and its corresponding arrival time are calculated. The minimum distance and the arrival time constitute a prospective risk assessment benchmark.

[0034] Based on the aforementioned forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction. The triggering factors and propagation paths of hypoxia risk are traced and quantified in reverse, and the structured prediction results are output, including:

[0035] Based on the arrival time in the forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is extrapolated in multiple forward steps. By taking the tensor at the current time as the initial state and combining it with the evolution trajectory predicted by the nonlinear mapping function, the tensor state at each future time step is iteratively calculated to construct the spatiotemporal evolution tensor sequence.

[0036] Risk pattern recognition is performed on the spatiotemporal evolution tensor sequence. By detecting abrupt changes in the spatial distribution of brain oxygen saturation and instability points in the brain inter-regional coordination patterns in the spatiotemporal evolution tensor sequence, time nodes and brain functional area nodes are marked.

[0037] Using the time nodes and brain functional area nodes as anchor points, reverse causal chain reconstruction is performed. By constructing a time-series dependency graph of the brain inter-brain oxygen metabolism coordination mode and performing backpropagation on the time-series dependency graph, the precursor brain functional area node sequence and precursor environmental stress factor sequence that cause the oxygen saturation mutation at the brain functional area node are identified.

[0038] The contribution of the precursor brain functional area node sequence and the precursor environmental stress factor sequence is quantified. By calculating the causal contribution intensity of each node in the precursor brain functional area node sequence to the oxygen saturation mutation and the driving contribution intensity of each factor in the precursor environmental stress factor sequence to the evolution of the hypoxia susceptibility state vector, a structured prediction result is constructed.

[0039] A second aspect of this invention provides an intelligent fusion analysis system for brain oxygen signals in high-altitude environments, comprising:

[0040] The brain oxygen data unit is used to acquire multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. The multi-channel brain oxygen saturation data is then subjected to spatiotemporal multi-scale separation and reconstruction to mine the features of brain inter-brain oxygen metabolism synergistic patterns and to construct a brain oxygen saturation spatiotemporal evolution tensor.

[0041] The hypoxia susceptibility unit is used to construct a multi-scale time-dependent hypoxia susceptibility dynamic model based on the nonlinear correlation between the characteristics of the brain region oxygen metabolism coordination mode and the multi-dimensional physiological parameter data. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector.

[0042] The risk assessment unit is used to perform cross-modal semantic alignment between the hypoxia susceptibility state vector and the time-varying environmental parameter data, and to generate a forward-looking risk assessment benchmark by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state.

[0043] The risk extrapolation unit is used to perform multi-step extrapolation and risk causal chain reconstruction on the spatiotemporal evolution tensor of brain oxygen saturation based on the aforementioned forward-looking risk assessment benchmark, reverse trace and quantify the triggering factors and propagation paths of hypoxia risk, and output structured prediction results.

[0044] The model dynamic unit is used to continuously track subsequent multi-channel brain oxygen saturation data based on the structured prediction results, and to dynamically recalibrate the multi-scale temporal dependencies in the hypoxia susceptibility dynamic model by using the deviation between the actual observed data and the predicted trajectory.

[0045] A third aspect of the present invention provides an electronic device, comprising:

[0046] processor;

[0047] Memory used to store processor-executable instructions;

[0048] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0050] This method can extract deep collaborative patterns of cerebral oxygen metabolism from complex high-altitude data, constructing a cerebral oxygen evolution tensor containing spatiotemporal information, thereby more comprehensively characterizing the dynamic correlation and overall evolution trend of oxygenation status in different brain regions. By establishing a multi-scale, time-dependent hypoxia susceptibility dynamic model, this method effectively simulates the feedback and compensation mechanisms of physiological systems at different time scales, accurately quantifying the dynamic boundaries and compensation limits of individual cerebral oxygen metabolism reserves, and providing an intrinsic physiological and dynamic basis for assessing hypoxia risk.

[0051] By employing cross-modal semantic alignment and nonlinear mapping, a deep fusion of time-varying environmental parameters and individual susceptibility states was achieved, generating a forward-looking risk assessment benchmark. Based on this benchmark, multi-step extrapolation and causal chain reconstruction can trace back to the specific triggering factors of hypoxia risk and its propagation path in brain networks, elevating risk prediction from a single indicator judgment to a structured analysis of causal mechanisms, significantly enhancing the accuracy and interpretability of early warnings.

[0052] This method introduces a dynamic recalibration mechanism based on the deviation between prediction and actual observation, enabling the core dynamic model to adaptively optimize based on real-time individual feedback, continuously improving the model's prediction accuracy and robustness in individualized scenarios. The entire analysis process forms a complete intelligent analysis system from data fusion, feature mining, model construction, risk extrapolation to closed-loop correction, providing reliable technical support for early, accurate, and personalized warning and intervention of cerebral hypoxia risk in high-altitude environments. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the intelligent fusion analysis method for brain oxygen signals in high-altitude environments according to an embodiment of the present invention.

[0054] Figure 2 This is a flowchart of the multi-step extrapolation and causal reconstruction of hypoxia risk based on a forward-looking risk assessment benchmark, as described in an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0056] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0057] Figure 1 This is a flowchart illustrating the intelligent fusion analysis method for brain oxygen signals in high-altitude environments according to an embodiment of the present invention. Figure 1 As shown, the intelligent fusion analysis method for brain oxygen signals in high-altitude environments includes:

[0058] The system acquires multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. It performs spatiotemporal multi-scale separation and reconstruction on the multi-channel brain oxygen saturation data, mines the features of brain inter-brain oxygen metabolism synergistic patterns, and constructs a brain oxygen saturation spatiotemporal evolution tensor.

[0059] Based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensatory limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector.

[0060] The hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A forward-looking risk assessment benchmark is generated by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state.

[0061] Based on the aforementioned forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction. The triggering factors and propagation paths of hypoxia risk are traced and quantified in reverse, and structured prediction results are output.

[0062] Based on the structured prediction results, subsequent multi-channel brain oxygen saturation data are continuously tracked, and the multi-scale temporal dependence in the hypoxia susceptibility dynamic model is dynamically recalibrated in parameter space using the deviation between the actual observed data and the predicted trajectory.

[0063] In one optional implementation, the multi-channel brain oxygen saturation data undergoes spatiotemporal multi-scale separation and reconstruction to mine features of coordinated oxygen metabolism patterns across brain regions, and a brain oxygen saturation spatiotemporal evolution tensor is constructed, including:

[0064] Variational mode decomposition was performed on multi-channel brain oxygen saturation data, and the signals of each channel were adaptively decomposed into a set of intrinsic mode components representing different physiological time scales. The intrinsic mode component sets corresponded to the rapid brain oxygen fluctuation layer, the respiratory heart rate coupling layer and the baseline metabolic drift layer in descending order of characteristic frequency. The instantaneous frequency envelope and instantaneous phase envelope of each intrinsic mode component were extracted as multi-scale time domain feature vectors.

[0065] Based on the cross-scale phase coupling relationship between the intrinsic modal component set and the channels of different brain functional areas, a brain inter-modal coupling tensor is constructed. The multi-scale temporal domain feature vector is spatiotemporally embedded with the brain inter-modal coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modal scale, and time evolution as each dimension, the spatiotemporal evolution tensor of brain oxygen saturation is obtained.

[0066] After acquiring multi-channel brain oxygen saturation data of the target individual, the raw signal first needs to be preprocessed to remove noise components such as power line interference and motion artifacts. The preprocessed multi-channel brain oxygen saturation data is denoted as... ,in Indicates the total number of channels in brain functional areas. Indicates the time sampling point. For the signal of each channel. An adaptive decomposition technique, Variational Mode Decomposition (VMD), is employed. VMD constructs a variational optimization problem, decomposing the signal into several intrinsic mode components with finite bandwidth characteristics, each corresponding to a specific center frequency. Specifically, the goal of VMD is to solve an optimization problem that minimizes the sum of the bandwidths of all mode components. By introducing a quadratic penalty term and Lagrange multipliers, the constrained optimization problem is transformed into an unconstrained saddle point problem for iterative solution.

[0067] For the The signals from each channel are obtained by variational mode decomposition. There are 1 intrinsic modal component, denoted as . ,in This represents the modal index. These intrinsic modal components are sorted from high to low center frequency. High-frequency components mainly reflect the rapid fluctuation characteristics of cerebral oxygen saturation, usually related to the local hemodynamic response caused by neurovascular coupling. Their frequency range is generally between 0.6 and 2 Hz. Signal fluctuations in this frequency band can capture the immediate response of brain regions to external stimuli or internal cognitive activities. The intrinsic modal components corresponding to this frequency band are classified as the rapid fluctuation layer of cerebral oxygen. Mid-frequency components reflect the periodic modulation effect of respiration and heart rate on cerebral blood flow. The respiratory rate is usually in the range of 0.2 to 0.4 Hz, and the heart rate rate is about 0.8 to 1.5 Hz. These two physiological rhythms have a coupled effect on cerebral oxygen saturation through blood pressure fluctuations and intracranial pressure changes. The corresponding intrinsic modal components are defined as the respiratory-heart rate coupling layer. Low-frequency components correspond to a center frequency below 0.1 Hz and mainly reflect the slow drift of the baseline level of cerebral oxygen metabolism over time. This drift is closely related to the overall metabolic state of the body, the oxygen-carrying capacity of blood, and the long-term high-altitude adaptation process, constituting the baseline metabolic drift layer.

[0068] For each intrinsic modal component The instantaneous amplitude envelope is obtained by extracting its analytic signal representation through Hilbert transform. and instantaneous phase envelope The instantaneous amplitude envelope characterizes the energy change characteristics of the modal component along the time axis, while the instantaneous phase envelope reflects the instantaneous frequency evolution of the signal. Combining the instantaneous amplitude and phase envelopes of all channels and all modes forms a multi-scale time-domain feature vector. The dimension of this feature vector is This study comprehensively characterized the dynamic characteristics of multi-channel brain oxygen saturation signals at different time scales.

[0069] After obtaining the intrinsic modal components of each channel, the phase coupling relationships between channels in different brain functional areas at the same or different modal scales were further analyzed. Phase coupling between brain regions reflects the synergy and synchronicity of different brain functional areas in oxygen metabolism processes, and is key to understanding the dynamics of brain oxygen metabolism networks. For channels... The Modal components and channels The For each modal component, its phase lock value is calculated as a measure of cross-scale phase coupling strength. The phase lock value is defined by calculating the modulus of the complex exponential average of the instantaneous phase difference between two signals, and its value ranges from 0 to 1. The closer the value is to 1, the stronger the phase coupling between the two signals. The phase lock values ​​between all channel pairs and all modal pairs are organized into a three-dimensional tensor structure, denoted as the brain inter-modal coupling tensor. Tensor elements Indicates channel modality With channel modality The phase coupling strength between them.

[0070] To unify the representation of temporal and spatially coupled features, a spatiotemporal joint embedding framework is constructed. Specifically, multi-scale temporal feature vectors are... Divide the time window into segments using a sliding window, with each time window having a length set to [value]. To ensure temporal continuity, adjacent windows overlap to a certain extent. For each time window, the average instantaneous amplitude and average instantaneous phase of each channel and modality within that window are calculated. Simultaneously, statistical features of the brain region modal coupling tensors corresponding to that time window are extracted, such as the mean and variance of coupling strength and network topology indices. These temporal features are then concatenated with the spatial coupling features to form a joint feature representation for each time window.

[0071] Based on this, a tensor structure with three dimensions—brain functional area channels, intrinsic modality scale, and temporal evolution—is constructed, namely the spatiotemporal evolution tensor of brain oxygen saturation. ,in This represents the total number of time windows. The first dimension of the tensor corresponds to... Each brain functional area channel, the second dimension corresponds to Each intrinsic modal scale, the third dimension corresponds to Each time evolution step. Each element in the tensor Indicates the first The first channel in the At the modal scale, the first The comprehensive feature value of each time window takes into account instantaneous amplitude, instantaneous phase, and coupling relationship with other channels. Through this tensor representation, the temporal evolution, spatial distribution pattern, and multi-scale dynamic characteristics of brain oxygen saturation signals can be simultaneously characterized within a unified mathematical framework, providing structured high-dimensional feature inputs for subsequent hypoxia susceptibility modeling and risk prediction.

[0072] In constructing the spatiotemporal evolution tensor of brain oxygen saturation, it is necessary to normalize the tensor to eliminate dimensional differences between different channels and modes. A statistical moment-based normalization method is employed, calculating the mean and standard deviation of the time series for each channel and mode, followed by standardization to ensure the normalized data has zero mean and unit variance. Furthermore, to enhance the robustness of the tensor representation, tensor decomposition techniques can be introduced to perform a low-rank approximation of the original tensor, extracting the dominant spatiotemporal modal components and filtering out noise and redundant information, thus obtaining a more compact and stable spatiotemporal evolution tensor representation of brain oxygen saturation.

[0073] In one optional implementation, the multi-scale temporal feature vector is spatiotemporally embedded with the brain inter-region modality coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modality scales, and temporal evolution as dimensions, the spatiotemporal evolution tensor of brain oxygen saturation is obtained, including:

[0074] Based on the phase-locking index and phase amplitude modulation index in the brain inter-modal coupling tensor, a cross-scale coupling propagation operator is constructed. The cross-scale coupling propagation operator maps the local representation of multi-scale temporal feature vectors in each intrinsic modal scale dimension to a global representation that integrates cross-scale synergistic effects by encoding the nonlinear interaction mechanism between different intrinsic modal scales.

[0075] The global representation is tensor-organized according to the spatial topological relationship of the brain functional area channels. Based on the anatomical location and functional connectivity strength of the brain functional area corresponding to each channel, a brain region spatial adjacency structure is constructed. The brain region spatial adjacency structure embeds the global representation of discrete channels into tensor slices that preserve the spatial dependence between brain regions by quantifying the topological distance and functional synergy between brain functional area nodes.

[0076] The tensor slices are stacked sequentially along the time evolution direction to construct a third-order tensor structure with brain functional area channels as the spatial dimension, intrinsic modality scale as the scale dimension, and time series as the temporal dimension.

[0077] The third-order tensor structure is rearranged dimensionally according to the topological centrality index of brain functional area nodes in the spatial dimension and the characteristic frequency of intrinsic modal components in the scale dimension. This eliminates redundant dependencies between dimensions and highlights the dominant cooperative mode, resulting in the spatiotemporal evolution tensor of brain oxygen saturation.

[0078] After obtaining the multi-scale temporal feature vectors and the brain intermodal coupling tensor, they need to be spatiotemporally coupled to construct a complete spatiotemporal evolution tensor of brain oxygen saturation. This process first extracts the phase-locking index and phase amplitude modulation index from the brain intermodal coupling tensor to characterize the nonlinear interaction mechanism between different intrinsic modal scales. The phase-locking index quantifies the phase synchronization degree between two intrinsic modal components; it is calculated by performing a Hilbert transform on the instantaneous phase difference between the two modal components and then taking the time average of the complex magnitudes. The phase amplitude modulation index characterizes the coupling strength between the amplitude envelope of the high-frequency modal component and the phase of the low-frequency modal component, quantified by calculating the mutual information between the instantaneous phase of the high-frequency component's amplitude envelope and the instantaneous phase of the low-frequency component.

[0079] Based on the extracted phase-locking exponent and phase amplitude modulation exponent, a cross-scale coupling propagation operator is constructed. This operator employs tensor decomposition to represent the interaction relationships between different intrinsic mode scales as a linear combination of a set of basis functions. Specifically, for the ... The intrinsic modal scale and the first The coupling relationship between intrinsic modal scales is defined by the propagation operator. Its function is to make the first Local eigenvectors at each scale Mapped to the fusion of the first Enhanced characterization of scale-based synergistic effects. The construction of the propagation operator requires consideration of the phase-locking exponent. With phase amplitude modulation index The combined effect of these two coupling modes is integrated into a unified propagation weight through a weighted fusion mechanism. For each scale component in the multi-scale temporal feature vector, the cross-scale coupling propagation operator is applied iteratively, enabling the local representation at each scale to absorb collaborative information from other scales. After multiple rounds of propagation iterations, the local representations at each scale gradually evolve into a global representation that integrates cross-scale collaborative effects. This global representation not only preserves the temporal dynamics within a single scale but also encodes the nonlinear interaction patterns between different scales.

[0080] After obtaining the global representation, tensor organization is required based on the spatial topological relationships of brain functional area channels. Brain functional area channels correspond to different anatomical locations within the brain, and specific spatial adjacency relationships and functional connectivity patterns exist between each channel. When constructing the brain region spatial adjacency structure, the topological distance between channels is first calculated using Euclidean distance or geodesic distance, based on the anatomical location of each channel's corresponding brain functional area. For channels in different brain regions such as the prefrontal, temporal, and parietal lobes, the topological distance reflects the degree of anatomical spatial proximity. Simultaneously, the functional connectivity strength between channels is calculated based on correlation analysis of brain oxygen saturation data. This strength is quantified by calculating the Pearson correlation coefficient or mutual information between the time series of brain oxygen saturation from different channels. After normalizing the topological distance and functional connectivity strength, a comprehensive brain region spatial adjacency matrix is ​​constructed using a weighted average. The elements of this matrix... Indicates the first The first channel and the first Spatial dependency strength between channels.

[0081] Based on the spatial adjacency structure of brain regions, the global representations of discrete channels are embedded into tensor slices that preserve the spatial dependencies between brain regions. At each time point, the global representations of all channels are arranged according to the topological order defined by the spatial adjacency matrix, forming a two-dimensional tensor slice. The row dimension corresponds to channels in brain functional areas, and the column dimension corresponds to the intrinsic modality scale. During the arrangement process, channels with strong spatial adjacency are placed in adjacent positions, resulting in a smooth gradient change in the tensor slice across the spatial dimensions, facilitating subsequent spatial convolution operations and region co-analysis. The element values ​​of each tensor slice are the global representation components of the corresponding channel at a specific intrinsic modality scale. These components comprehensively reflect the brain oxygen metabolism dynamics of that channel at that scale and its synergistic effects with other scales.

[0082] Tensor slices were sequentially stacked along the temporal evolution direction to construct a third-order tensor structure. Tensor slices from different time points were stacked in chronological order to form a three-dimensional tensor. Its first dimension is the spatial dimension of brain functional area channels, the second dimension is the scale dimension of intrinsic modality, and the third dimension is the temporal dimension of the time series. This third-order tensor structure completely encodes the evolutionary characteristics of brain oxygen saturation data in the spatial, scale, and temporal dimensions. Each element of the tensor... Indicates the first The brain functional area pathways in the first Under the intrinsic modal scale, at the th... Global representation values ​​at each time point. Through this tensor representation, brain oxygen saturation information, which was originally scattered across multiple channels and scales, is integrated into a unified mathematical structure, facilitating multi-dimensional joint analysis and pattern mining.

[0083] To further optimize the representational power of the third-order tensor structure, its dimensions need to be rearranged to eliminate redundant dependencies and highlight dominant collaborative patterns. Within the spatial dimension, channels are reordered based on the topological centrality indices of brain functional region nodes. Topological centrality is quantified using degree centrality or betweenness centrality. Degree centrality measures the importance of each node in the network by calculating the number of connections in the brain region's spatial adjacency structure, while betweenness centrality assesses its pivotal role in information propagation by calculating the number of shortest paths passing through that node. Channels with high topological centrality are placed at the front of the tensor, ensuring that core brain regions that contribute significantly to brain oxygen metabolism collaborative patterns are prioritized for processing and analysis.

[0084] Within the scale dimension, the intrinsic modal components are reordered according to their characteristic frequencies. These characteristic frequencies are determined by extracting the peak frequency of the power spectrum after performing a Fourier transform on each intrinsic modal component; this frequency reflects the dominant oscillation period of that scale component. Ordering the characteristic frequencies from low to high or high to low allows the dynamic processes at different time scales to exhibit an ordered hierarchical structure in the scale dimension of the tensor. Low-frequency components correspond to slow physiological regulatory processes, while high-frequency components correspond to rapid neurovascular coupling responses. Frequency ordering clearly distinguishes the physiological mechanisms at different time scales.

[0085] After rearranging the spatial and scale dimensions, the third-order tensor structure was decomposed to identify dominant cooperative patterns. Tucker decomposition or CANDECOMP / PARAFAC decomposition was used to decompose the tensor into the product of a core tensor and three factor matrices. The core tensor encodes the interaction relationships between dimensions, while the factor matrices represent spatial, scale, and temporal patterns, respectively. By retaining components with large singular values ​​in the core tensor and filtering out redundant secondary patterns, a dimensionality-reduced spatiotemporal evolution tensor of brain oxygen saturation was obtained. This tensor significantly reduced the data dimensionality while preserving the main cooperative patterns and eliminating redundant dependencies between dimensions, providing high-quality input features for subsequent hypoxia susceptibility dynamic modeling and risk prediction. The resulting spatiotemporal evolution tensor of brain oxygen saturation comprehensively characterizes the multi-scale spatiotemporal evolution of brain oxygen metabolism in target individuals in high-altitude environments, laying a data foundation for a deeper understanding of the brain's functional adaptation mechanisms under hypoxic stress.

[0086] In one optional implementation, based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. This hypoxia susceptibility dynamic model characterizes the dynamic boundary between an individual's brain oxygen metabolism reserve capacity and compensatory limit through multi-timescale physiological feedback loops, resulting in a hypoxia susceptibility state vector including:

[0087] We quantify the cross-scale coupling strength of brain region oxygen metabolism coordination pattern characteristics and multidimensional physiological parameter data by constructing a conditional transfer entropy matrix under multiple time scale windows.

[0088] Based on the conditional transfer entropy matrix, closed-loop physiological feedback loops are identified. By detecting brain oxygen metabolism coordination patterns and physiological parameter pairs that exhibit bidirectional transfer entropy and fast-slow transfer delay characteristics, a set of feedback loops is extracted.

[0089] The set of feedback loops is constructed as a coupled set of dynamic differential equations. By defining the brain oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established.

[0090] Phase space trajectory analysis was performed on the hypoxia susceptibility dynamic model. By solving the phase space evolution trajectory under different initial reserve states and physiological disturbance intensities, the set of critical bifurcation points that lead to trajectory escape from the stable attraction domain were identified. The set of critical bifurcation points constitutes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit. The phase space coordinates of the dynamic boundary are represented as a hypoxia susceptibility state vector.

[0091] After obtaining the characteristics of brain regions' oxygen metabolism coordination patterns and multidimensional physiological parameter data, it is necessary to delve into the nonlinear correlation between the two to construct a hypoxia susceptibility dynamic model. The multidimensional physiological parameter data specifically includes time-series data of physiological indicators such as heart rate, blood pressure, respiratory rate, blood oxygen saturation, and skin temperature. These parameters have complex cross-scale coupling relationships with the characteristics of brain oxygen metabolism coordination patterns.

[0092] In the cross-scale coupling strength quantification process, multiple timescale windows were set to capture physiological regulatory mechanisms at different frequencies. Short timescale windows (10-30 seconds) were used to capture rapid neurovascular coupling responses; medium timescale windows (1-5 minutes) reflected respiratory and circulatory regulation; and long timescale windows (10-30 minutes) characterized metabolic adaptation processes. Within each timescale window, the conditional transfer entropy between the brain region oxygen metabolism coordination pattern feature vector and various physiological parameters was calculated. The conditional transfer entropy was calculated by constructing a joint probability distribution using kernel density estimation. Quantification from physiological parameters Characteristics of brain oxygen metabolism In condition variables Information transmission volume under constraints, among which express of Rank history embedding vector, express of Rank history embedding vector, Representing condition variables of The historical embedding vector is calculated by traversing all combinations of brain region oxygen metabolism features and physiological parameters, and calculating the conditional transfer entropy at different time scale windows. A three-dimensional conditional transfer entropy matrix is ​​constructed, where the first dimension corresponds to the brain region oxygen metabolism feature index, the second dimension corresponds to the physiological parameter index, and the third dimension corresponds to the time scale index.

[0093] When identifying closed-loop physiological feedback loops based on the constructed conditional transfer entropy matrix, it is necessary to simultaneously detect the existence of bidirectional information transfer and the separation characteristics of the speed of transfer delay. This applies to any brain oxygen metabolism coordination pattern feature. With physiological parameters Pairing, calculating forward propagation entropy With reverse propagation entropy When the transfer entropy in both directions is significantly greater than zero, bidirectional information flow is considered to exist. Significance testing is performed using the substitution data method. The original time series is phase-randomized to generate 100 substitution sequences. The transfer entropy distribution of the substitution sequences is calculated. If the original transfer entropy exceeds the 95th percentile of the substitution distribution, it is considered significant. After confirming bidirectional transfer, the separation characteristics of fast and slow transfer delays are further analyzed. Using a time-delay embedding method, the transfer entropy is recalculated under different time-delay parameters, and a curve showing the change of transfer entropy with time delay is plotted. If the forward transfer entropy peaks at a short time delay while the reverse transfer entropy peaks at a long time delay, or vice versa, the pair is considered to exhibit fast-slow separation characteristics and is included in the feedback loop set. For example, regarding the relationship between prefrontal cortex oxygen metabolism characteristics and heart rate in a certain brain region, the transfer entropy from heart rate to brain oxygen metabolism peaks at a 2-second time delay, while the transfer entropy from brain oxygen metabolism to heart rate peaks at a 15-second time delay. This indicates that the rapid influence of heart rate on brain oxygen metabolism and the slow regulation of heart rate by brain oxygen metabolism form a closed-loop feedback.

[0094] When constructing the identified set of feedback loops into a coupled system of dynamic differential equations, the brain oxygen metabolism reserve state is defined as the phase variable of the system. Indicates the first Each brain region at any time The oxygen metabolism reserve capacity, with a value ranging from 0 to 1, where 0 indicates complete depletion of reserves and 1 indicates sufficient reserves. Multidimensional physiological parameters, as external driving variables, are denoted as vectors. ,in The number of physiological parameters. Based on the coupling relationships identified in the feedback loop set, a system of differential equations is established. ,in For the first The evolutionary function of the reserve state of each brain region. For the number of brain regions, This represents the coupling coefficient between brain regions. This represents the driving coefficient of physiological parameters on the reserve state of brain regions. Inter-brain coupling coefficient. This represents the influence strength of the j-th brain region on the i-th brain region, determined based on DTI white matter fiber tract density and fMRI functional connectivity matrix. For brain region pairs with direct connections, the value ranges from 0.2 to 0.5; otherwise, it is 0. (Drive coefficient) This represents the degree of influence of the k-th physiological parameter on the reserve of the i-th brain region, determined based on physiological mechanisms and sensitivity analysis of clinical data: the driving coefficient of MAP ranges from 0.25 to 0.40, ICP from -0.45 to -0.25, and SpO2 from 0.30 to 0.50. Evolutionary function. The specific form is designed as a superposition of multiple time-scale terms based on the fast and slow separation characteristics of the feedback loop. Fast feedback terms are in linear or quadratic form, while slow feedback terms are in integral or delayed form. For example, for the prefrontal cortex, its evolutionary function can be expressed as:

[0095] ;

[0096] in The self-decay coefficient represents the natural recovery rate of a brain region's reserve state in the absence of external input. Its value is determined based on the physiological metabolic characteristics of the brain region, with a typical range of 0.05-0.15, and a value of 0.08 for the prefrontal cortex. and These are reference values ​​for physiological parameters. This is the integration time window.

[0097] When performing phase space trajectory analysis on the constructed hypoxia susceptibility dynamic model, the first step is to determine the dimensions and coordinate axes of the phase space. The dimension of the phase space is equal to the number of brain regions. Each coordinate axis corresponds to the reserve state of a brain region. Different initial reserve state combinations were set; for example, a reserve state of 0.8 for all brain regions indicates a good initial state, while a reserve state of 0.5 for some brain regions indicates a partially damaged initial state. Simultaneously, different physiological perturbation intensities were set, and the physiological parameter vectors were changed accordingly. The amplitude of the disturbance is adjusted by applying perturbation modes such as step changes or periodic fluctuations. For each set of initial conditions and perturbation settings, the phase space evolution trajectory is obtained by numerically solving the system of differential equations. The numerical solution adopts the fourth-order Runge-Kutta method, with a time step of 0.1 seconds and a simulation duration of 3600 seconds to cover the typical time scale of high-altitude exposure.

[0098] In phase space, the stable attraction domain corresponds to the state region where an individual can maintain normal cerebral oxygen metabolism. By analyzing the convergence behavior of a large number of trajectories, the boundaries of the attraction domain are identified. When the initial state is within the attraction domain and the perturbation intensity is small, the trajectory eventually converges to a stable equilibrium point or a stable limit cycle. When the initial state approaches the boundary or the perturbation intensity increases, the trajectory escapes the attraction domain, manifested as a continuous decline in the reserve state until it falls below a safe threshold. The set of critical bifurcation points is defined as the combination of parameters that causes a qualitative change in trajectory behavior. Specifically, it is identified in the following ways: fixing the initial state, gradually increasing the perturbation intensity, and recording the perturbation intensity value and corresponding phase space coordinates when the trajectory first escapes the attraction domain; fixing the perturbation intensity, changing the initial state, and recording the initial state coordinates when the trajectory is exactly at the boundary of the attraction domain. These critical points form a bifurcation surface in phase space, which constitutes the dynamic boundary between an individual's cerebral oxygen metabolism reserve capacity and compensatory limit.

[0099] The phase space coordinates of the dynamic boundary are represented as a hypoxia susceptibility state vector. This vector contains multiple components, with the first component representing the reserve value of each brain region under critical conditions. The second component is the critical perturbation intensity that triggers escape. The third component serves as a bifurcation type identifier, used to distinguish different instability mechanisms such as saddle-node bifurcation and Hopf bifurcation. Principal component analysis is used to reduce the dimensionality of the high-dimensional hypoxia susceptibility state vector, extracting the first five principal components as a compact representation, retaining over 95% of the variance information. The final hypoxia susceptibility state vector comprehensively reflects an individual's vulnerability to hypoxia stress under current physiological conditions; the smaller the vector value, the closer to the compensation limit and the higher the risk of hypoxia.

[0100] In one optional implementation, the set of feedback loops is constructed as a coupled set of dynamic differential equations. By defining the cerebral oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established, including:

[0101] The feedback loops in the feedback loop set are decomposed into a structure. By identifying the brain oxygen metabolism coordination mode nodes and physiological parameter nodes in each feedback loop, the causal transmission paths and transmission delays between nodes are extracted, and a feedback loop topology diagram is constructed.

[0102] Based on the feedback loop topology diagram, the system phase variables and external driving variables are defined. The brain oxygen metabolism reserve state and its consumption rate and recovery rate at different time scales are defined as the system phase variable set, and the multidimensional physiological parameters and their fluctuation amplitude and change trend at different time scales are defined as the external driving variable set.

[0103] Based on the causal propagation path and propagation delay in the feedback loop topology diagram, each feedback loop is expressed as a differential equation, and by introducing delay terms and coupling terms, the differential equations of multiple feedback loops are coupled into a system of dynamic differential equations.

[0104] The dynamic differential equations are subjected to time-scale separation transformation. By solving the rapidly changing system phase variables and the slowly changing system phase variables at different time scales, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed.

[0105] After obtaining the set of feedback loops, it needs to be transformed into a computable mathematical model to achieve a quantitative assessment of hypoxia susceptibility. First, each feedback loop in the set is structurally decomposed. Specifically, all feedback loops in the set are traversed, and for each loop, the nodes involved in brain oxygen metabolism coordination patterns and physiological parameter nodes are identified. Brain oxygen metabolism coordination pattern nodes include regional coordination feature nodes such as the prefrontal lobe oxygen metabolism coordination index, temporal lobe oxygen metabolism coordination index, and parietal lobe oxygen metabolism coordination index, as well as global coordination feature nodes such as the strength of the whole-brain oxygen metabolism network and oxygen metabolism synchronicity. Physiological parameter nodes include multidimensional physiological indicator nodes such as heart rate, blood pressure, respiratory rate, blood oxygen saturation, and heart rate variability. Based on the identified nodes, the causal transmission paths between nodes are extracted.

[0106] For example, in a typical feedback loop, the heart rate node affects cerebral blood flow perfusion through cardiac output, which in turn affects the prefrontal cortex oxygen metabolism coordination index node. The prefrontal cortex oxygen metabolism coordination index node, in turn, affects the heart rate node through autonomic nervous regulation, forming a closed-loop causal transmission path. Simultaneously, the transmission delay along the causal transmission path is extracted; this delay reflects the time required for a physiological signal to propagate from one node to another. For example, the delay from heart rate change to cerebral blood flow perfusion change is approximately 2 to 5 seconds, the delay from cerebral blood flow perfusion change to cerebral oxygen metabolism coordination pattern change is approximately 3 to 8 seconds, and the delay from cerebral oxygen metabolism coordination pattern change to heart rate change through autonomic nervous regulation is approximately 5 to 15 seconds. By performing the above decomposition on all feedback loops, a feedback loop topology diagram is constructed. This topology diagram uses nodes to represent cerebral oxygen metabolism coordination pattern characteristics and physiological parameters, and directed edges to represent causal transmission paths, with transmission delays labeled on the edges, forming a multi-level network topology.

[0107] Based on the feedback loop topology diagram, system phase variables and external driving variables are defined. System phase variables describe the evolution of the system's internal state, while external driving variables describe the driving effect of external inputs on the system. The cerebral oxygen metabolic reserve is defined as the core system phase variable, quantifying an individual's remaining capacity to maintain normal cerebral oxygen metabolism under current environmental conditions. Furthermore, the depletion rate and recovery rate of the cerebral oxygen metabolic reserve at different time scales are also defined as system phase variables. The depletion rate reflects the rapid decline of cerebral oxygen metabolic reserves under stress conditions, while the recovery rate reflects the rebound trend of cerebral oxygen metabolic reserves after stress relief.

[0108] On a fast timescale, the consumption rate is mainly influenced by acute hypoxic stress, with a period of change ranging from seconds to minutes; on a slow timescale, the recovery rate is mainly influenced by the body's compensatory mechanisms and adaptive regulation, with a period of change ranging from tens of minutes to hours. These state variables at different timescales are combined into a set of system phase variables, denoted as […]. ,in Indicates the brain's oxygen metabolism reserve status. to Variables representing consumption rates on fast timescales. to Variables representing recovery rate over slow time scales The total number of phase variables in the system. Let time be the variable. The multidimensional physiological parameters and their fluctuation amplitudes and trends at different time scales are defined as the set of external driving variables. Multidimensional physiological parameters include real-time monitored physiological indicators such as heart rate, blood pressure, and respiratory rate. Fluctuation amplitude reflects the short-term variability of physiological parameters, and the trend reflects the long-term evolution direction of physiological parameters. The set of external driving variables is denoted as... ,in to Represents multidimensional physiological parameters, to This indicates the amplitude of fluctuations on a fast time scale. to Indicates the trend of change on a slow time scale. This represents the total number of external driving variables.

[0109] Based on the causal propagation paths and propagation delays in the feedback loop topology diagram, each feedback loop is expressed as a differential equation. For the ... The general form of the differential equation for a feedback loop is: ,in It is a nonlinear function that describes the phase variables of the system. The relationship between the rate of change of [the variable] and other system phase variables and external driving variables. The propagation delay between phase variables in the system. This represents the propagation delay from the external driving variable to the system phase variable. (Nonlinear function) The specific form is determined based on the physiological mechanism of the feedback loop. For example, for a feedback loop describing the depletion of the brain's oxygen metabolism reserve, its differential equation can be expressed as follows: ,in This is the consumption coefficient. This is a function representing the effect of external driving variables on the consumption rate. The specific expression uses the Sigmoid function form: ,in The steepness parameter is set to a value of 2-5. The threshold parameter is 20 mmHg for ICP. This function describes the nonlinear amplification effect on the rate of consumption when external driving variables (such as ICP) exceed the physiological threshold. The recovery coefficient represents the recovery rate of the reserve state. Its value is determined based on the brain's autoregulation ability and oxygen supply capacity, with a typical range of 0.05-0.2. For brain regions with strong autoregulation capabilities, the value is 0.15. This is a function of the influence of other system phase variables on the recovery rate. The specific expression takes the form of a linear or saturated function: ,in The recovery efficiency coefficient is set between 0.5 and 1.0. To represent the maximum recovery rate, this function describes the supporting role and saturation characteristics of the reserve states of other brain regions in the recovery of this brain region. By introducing time delay and coupling terms, the differential equations of multiple feedback loops are coupled into a system of dynamic differential equations. The time delay term reflects the hysteresis effect of physiological signal propagation, and the coupling term reflects the interaction between different feedback loops. The coupled system of dynamic differential equations can be expressed as follows: ,in For vector functions, the differential equations contain all feedback loops. Let the time delay vector be the time delay vector between the phase variables of the system. The time delay vector of the external driving variables.

[0110] A time-scale separation transformation is performed on the dynamic differential equations to handle the simultaneous occurrence of rapid and slow changes within the system. Based on singular perturbation theory, this transformation categorizes system phase variables into fast and slow variables. Fast variables correspond to dynamic processes on fast timescales, such as the rapid depletion of cerebral oxygen reserves under acute hypoxic stress; slow variables correspond to dynamic processes on slow timescales, such as the slow recovery of cerebral oxygen reserves driven by compensatory mechanisms. Small parameters are introduced. The ratio of the fast time scale to the slow time scale is used to decompose the system's phase variables into... ,in For fast variables, For slow variables. The system of dynamic differential equations is rewritten as follows: and On a fast timescale, let The quasi-steady-state equation for the fast variable is obtained. Solving this algebraic equation yields the expression for the fast variable in terms of the slow variable and the external driving variable. .

[0111] In one optional implementation, the hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A prospective risk assessment benchmark is generated by constructing a nonlinear mapping function between environmental stress factors and hypoxia susceptibility states, including:

[0112] Cross-modal semantic projection is performed on the hypoxia susceptibility state vector and time-varying environmental parameter data. By calculating the mutual information transmission strength between each dimension component of the hypoxia susceptibility state vector and each dimension component of the time-varying environmental parameter data in the embedding space, a cross-modal correlation matrix is ​​constructed.

[0113] Based on the cross-modal correlation matrix, an environmental stress factor set is extracted. By identifying environmental parameter dimensions in the matrix whose influence weight exceeds a preset weight threshold, an environmental stress factor set is constructed.

[0114] The set of environmental stress factors and the hypoxia susceptibility state vector are constructed into a nonlinear mapping function. The nonlinear mapping function is then prospectively extrapolated. By inputting the predicted sequence of environmental parameters within a future time window into the nonlinear mapping function, the evolution trajectory of the hypoxia susceptibility state vector is solved. The minimum distance between the evolution trajectory and the dynamic boundary of the compensation limit and its corresponding arrival time are calculated. The minimum distance and the arrival time constitute a prospective risk assessment benchmark.

[0115] After obtaining the hypoxia susceptibility state vector, it is necessary to perform deep fusion analysis with time-varying environmental parameter data to achieve accurate prediction of future hypoxia risk. The hypoxia susceptibility state vector typically contains multiple dimensions reflecting an individual's brain oxygen metabolism reserve capacity, such as the brain region oxygen supply-demand balance index, metabolic compensation rate, and hemodynamic response intensity. These components can have 15 to 30 dimensions. Time-varying environmental parameter data covers physical environmental factors such as altitude, atmospheric pressure, oxygen partial pressure, temperature, humidity, and wind speed, as well as behavioral environmental factors such as individual activity intensity, load conditions, and psychological stress levels. These parameters typically have 10 to 20 dimensions. Because these two types of data originate from different measurement systems and have heterogeneous physical dimensions and numerical scales, direct numerical calculations or correlation analyses are not possible.

[0116] To address the aforementioned heterogeneous data fusion problem, a cross-modal semantic projection technique is employed to map the two types of data into a unified embedding space. Specifically, each dimension component in the hypoxia susceptibility state vector is normalized, mapping its numerical range to the interval between 0 and 1 while preserving the relative relationships between components. The time-varying environmental parameter data are similarly normalized, and time-scale alignment is performed based on the time constants of each environmental parameter's impact on human physiology. For example, the effect of atmospheric pressure changes on blood oxygen saturation typically manifests within 5 to 10 minutes, while the effect of temperature changes requires 20 to 30 minutes to fully manifest; therefore, appropriate time delay corrections need to be applied to different environmental parameters. After normalization and time alignment, the hypoxia susceptibility state vector is denoted as... ,in This represents the number of dimensions of the state vector; the time-varying environmental parameter data is denoted as... ,in This indicates the number of dimensions of the environmental parameters.

[0117] In the embedding space, it is necessary to quantify the information transfer strength between the components of each dimension of the hypoxia susceptibility state vector and the components of each dimension of the time-varying environmental parameter data. Mutual information is used as a metric, as it can capture the nonlinear dependencies between variables and is not limited by the data distribution. For the state vector... Each component and environmental parameters Each component their mutual information The calculation is based on the joint probability distribution and the marginal probability distribution. In actual calculations, a statistical analysis of historical data is performed to construct... and The joint frequency distribution histogram divides the numerical interval into several equally wide bins, counts the frequency of data points falling into each bin, and then estimates the probability density. A larger mutual information value indicates stronger information transfer between the two variables and a higher degree of interdependence.

[0118] A cross-modal correlation matrix is ​​constructed by calculating the mutual information between all state components and environmental parameter components. The rows of this matrix correspond to the dimensions of the hypoxia susceptibility state vector, and the columns correspond to the dimensions of the time-varying environmental parameters. The matrix elements... That is The dimension of the cross-modal correlation matrix is... This intuitively reflects the coupling structure between the state space and the environment space. The elements with larger values ​​in the matrix indicate environmental parameters that significantly affect hypoxia susceptibility; these parameters are the environmental stress factors.

[0119] To extract a set of environmental stress factors from the cross-modal correlation matrix, a preset weight threshold is set. The selection of this threshold needs to comprehensively consider the signal-to-noise ratio of the data and clinical practice experience. It can usually be... Set it to 1.5 to 2 times the mean of all elements in the matrix, or select a threshold that ensures the number of retained elements is 10% to 20% of the total number of elements. Iterate through each column of the cross-modal correlation matrix, for the Column, calculate the maximum value of all elements in the column. If the maximum value is greater than Then it is considered that the first One environmental parameter has a significant impact on hypoxia susceptibility state and is included in the set of environmental stress factors. Furthermore, it is necessary to examine the combined effect of each environmental parameter on multiple state components and calculate the... Weighted sum of column elements The weight Reflecting the If the weighted sum of each state component in the overall hypoxia risk assessment exceeds another preset threshold, that environmental parameter is also included in the set of stress factors. Through the above screening process, the set of environmental stress factors is obtained. ,in The number of stress factors, usually much smaller This enabled the dimensionality reduction of environmental parameters and the identification of key factors.

[0120] Based on the set of environmental stress factors and the hypoxia susceptibility state vector, a nonlinear mapping function is constructed. This function describes the current time. state vector and stress factors Under the influence of time step The post-state vector evolves into The process involves modeling nonlinear mapping functions using multinomial regression, kernel methods, or neural networks. Radial basis function networks (RBF) are used as the implementation of the mapping function. The input layer receives a concatenated vector of the state vector and the stress factor, while the hidden layer contains several RBF units, each corresponding to a center point and a width parameter. The output layer generates the predicted state vector through linear combination. Network parameters are obtained through training on historical data, using mean squared error as the loss function and gradient descent to optimize the parameters.

[0121] To achieve forward-looking trajectory extrapolation, it is necessary to obtain the predicted sequence of environmental parameters within a future time window. The length of the time window is determined based on the specific application requirements; in high-altitude mountaineering scenarios, it is typically set to 30 to 60 minutes to allow sufficient time for decision-making and intervention. The predicted sequence of environmental parameters can be generated by comprehensively considering meteorological forecast data, individual travel planning information, and historical environmental change trends. The predicted environmental parameter sequence will then be... The nonlinear mapping functions are input sequentially, and the evolution trajectory of the state vector is calculated recursively. First, the calculation... , and then with As a new initial state, calculate This process continues until the entire time window has been calculated, yielding the state vector evolution trajectory. .

[0122] In the dynamic model of hypoxia susceptibility, a compensatory limit dynamic boundary has been determined. This boundary forms a hypersurface in the state space. When the state vector touches or crosses this boundary, it indicates that the individual's compensatory mechanism is about to fail, and the risk of hypoxia increases sharply. The compensatory limit dynamic boundary can be represented as an implicit function. ,in This indicates that the state vector is located in the safe region. This indicates that the area has entered a danger zone. For each predicted state on the evolutionary trajectory... Calculate the distance between it and the compensatory limit dynamic boundary. The distance is calculated using Euclidean or Mahalanobis distance in the state space; specifically, it involves finding the distance on the boundary. nearest point , making and Minimum, this minimum distance is denoted as .

[0123] In one optional implementation, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction based on the aforementioned forward-looking risk assessment benchmark. This process traces back and quantifies the triggering factors and propagation paths of hypoxia risk, outputting structured prediction results including:

[0124] Based on the arrival time in the forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is extrapolated in multiple forward steps. By taking the tensor at the current time as the initial state and combining it with the evolution trajectory predicted by the nonlinear mapping function, the tensor state at each future time step is iteratively calculated to construct the spatiotemporal evolution tensor sequence.

[0125] Risk pattern recognition is performed on the spatiotemporal evolution tensor sequence. By detecting abrupt changes in the spatial distribution of brain oxygen saturation and instability points in the brain inter-regional coordination patterns in the spatiotemporal evolution tensor sequence, time nodes and brain functional area nodes are marked.

[0126] Using the time nodes and brain functional area nodes as anchor points, reverse causal chain reconstruction is performed. By constructing a time-series dependency graph of the brain inter-brain oxygen metabolism coordination mode and performing backpropagation on the time-series dependency graph, the precursor brain functional area node sequence and precursor environmental stress factor sequence that cause the oxygen saturation mutation at the brain functional area node are identified.

[0127] The contribution of the precursor brain functional area node sequence and the precursor environmental stress factor sequence is quantified. By calculating the causal contribution intensity of each node in the precursor brain functional area node sequence to the oxygen saturation mutation and the driving contribution intensity of each factor in the precursor environmental stress factor sequence to the evolution of the hypoxia susceptibility state vector, a structured prediction result is constructed.

[0128] like Figure 2 As shown, the method includes:

[0129] After obtaining the prospective risk assessment baseline, a forward multi-step extrapolation of the spatiotemporal evolution tensor of brain oxygen saturation is required. The prospective risk assessment baseline contains crucial arrival time information, which represents the expected time point at which environmental stressors act on an individual's physiological system and induce significant changes in hypoxia susceptibility. The current time... Spatiotemporal evolution tensor of brain oxygen saturation As the initial state, the dimension of this tensor is ,in Indicates the number of brain oxygen monitoring channels. Indicates the number of brain functional areas. This represents the feature dimension. It incorporates the nonlinear mapping function constructed from the forward-looking risk assessment benchmark. This function describes the mechanism by which environmental stress factors affect hypoxia susceptibility and achieves tensor state prediction for future time steps through iterative calculation.

[0130] Specifically, for the future... time step (in (with time step), tensor state prediction is achieved through... Implementation, in which For tensor evolution operators, Indicates time Environmental stress factor vector This represents the hypoxia susceptibility state vector from the previous time step. Tensor evolution operator. This study comprehensively considers the temporal dependence of brain inter-regional oxygen metabolism coordination patterns, the multi-scale characteristics of physiological feedback loops, and the nonlinear driving effect of environmental factors. It starts from the current moment... Begin iterating step by step to the target prediction time domain (Typically set to the next 30 to 120 minutes), construct a complete spatiotemporal evolution tensor sequence. .

[0131] After obtaining the spatiotemporal evolution tensor sequence, risk pattern recognition needs to be performed on the sequence to capture potential hypoxia risk events. Abrupt points in the spatial distribution of brain oxygen saturation manifest as drastic changes in the tensor's spatial dimension. This can be identified by calculating the spatial gradient norm of the tensor between adjacent time steps. (in Represents the spatial gradient operator, The Frobenius norm is used for detection. Exceeding the dynamic threshold At that time, mark that moment as a spatial mutation point. Dynamic threshold. It adaptively adjusts based on historical gradient statistical characteristics to avoid false detections or missed detections caused by fixed thresholds.

[0132] The instability points of inter-brain regional coordination patterns reflect the breakdown of the coordinated relationship of oxygen metabolism between different brain functional areas. This can be addressed by calculating the coordination matrix between brain functional areas. The elements of the matrix Indicates brain functional areas With brain functional areas At any moment The degree of synergy in oxygen metabolism. Synergy can be quantified using methods such as mutual information, correlation coefficient, or phase synchronization index. When the eigenvalue distribution of the synergy matrix deviates significantly from the normal pattern, i.e., the maximum eigenvalue... with the second largest eigenvalue ratio Exceeding the threshold When the condition number of the cooperation degree matrix increases sharply, that moment is marked as a cooperative instability point. For the detected mutation points and instability points, the corresponding set of time points is recorded. And the set of brain functional areas that exhibit abnormalities at these time points. .

[0133] Using marked time nodes and brain functional area nodes as anchors, a reverse causal chain reconstruction is performed to trace the origin of hypoxia risk. A time-dependent graph of the coordinated oxygen metabolism pattern between brain regions is constructed. vertex set The edge set contains nodes of all brain functional areas at all time steps. This represents the causal dependencies between nodes. Edge weights. Indicates brain functional areas At any moment brain functional areas At any moment The causal influence strength is estimated by methods such as Granger causality test, transitive entropy, or conditional mutual information.

[0134] Perform backpropagation on the time dependency graph, starting from the anchor node. Begin by tracing back along the edge in the opposite direction to the previous node that influenced the current node's state. During the backpropagation process, for the current node... Search all that meet the criteria predecessor node (in , (This is the causality strength threshold). The search proceeds recursively until the initial time is reached. Or, if the causal strength is below a threshold, construct a complete sequence of precursor brain functional area nodes. .

[0135] Simultaneously, it is necessary to identify the sequence of precursor environmental stress factors, including altitude, air pressure, temperature, humidity, and exercise intensity. This can be achieved by analyzing the environmental stress factor vector. With hypoxia susceptibility state vector The time-lag correlation between these factors was analyzed to identify environmental factors that significantly drive hypoxia susceptibility before the anchor point time. Specifically, the environmental factors were calculated. The components of the hypoxia susceptibility state vector The time-delay cross-correlation function between ,when At a certain time delay Exceeding the threshold At that time, environmental factors At any moment These are labeled as precursor environmental stress factors. A sequence of precursor environmental stress factors is constructed by traversing all environmental factors and all time lags. .

[0136] The contribution of precursor brain functional region node sequences and precursor environmental stress factor sequences was quantified to clarify the relative importance of each factor to hypoxia risk. For precursor brain functional region nodes... The causal contribution strength of its contribution to the abrupt change in oxygen saturation at the anchor point Calculated using the path integral method. From node To anchor node There are multiple causal paths, and the contribution of each path is the product of the weights of all edges along that path. (Node) The total contribution strength is the sum of the contributions of all paths passing through that node. To avoid path explosion, path length is usually limited or only the top few paths with the highest weights are considered.

[0137] For precursor environmental stress factors Its driving contribution strength to the evolution of the hypoxia susceptibility state vector Calculated using sensitivity analysis. Specifically, at time... Environmental factors Apply small perturbation The impact of this disturbance on the oxygen susceptibility state vector at the anchor point was calculated through forward extrapolation. The driving contribution intensity is defined as This quantitative indicator reflects the amplifying effect of changes in environmental factors on the evolution of the system state.

[0138] A second aspect of this invention provides an intelligent fusion analysis system for brain oxygen signals in high-altitude environments, comprising:

[0139] The brain oxygen data unit is used to acquire multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. The multi-channel brain oxygen saturation data is then subjected to spatiotemporal multi-scale separation and reconstruction to mine the features of brain inter-brain oxygen metabolism synergistic patterns and to construct a brain oxygen saturation spatiotemporal evolution tensor.

[0140] The hypoxia susceptibility unit is used to construct a multi-scale time-dependent hypoxia susceptibility dynamic model based on the nonlinear correlation between the characteristics of the brain region oxygen metabolism coordination mode and the multi-dimensional physiological parameter data. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector.

[0141] The risk assessment unit is used to perform cross-modal semantic alignment between the hypoxia susceptibility state vector and the time-varying environmental parameter data, and to generate a forward-looking risk assessment benchmark by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state.

[0142] The risk extrapolation unit is used to perform multi-step extrapolation and risk causal chain reconstruction on the spatiotemporal evolution tensor of brain oxygen saturation based on the aforementioned forward-looking risk assessment benchmark, reverse trace and quantify the triggering factors and propagation paths of hypoxia risk, and output structured prediction results.

[0143] The model dynamic unit is used to continuously track subsequent multi-channel brain oxygen saturation data based on the structured prediction results, and to dynamically recalibrate the multi-scale temporal dependencies in the hypoxia susceptibility dynamic model by using the deviation between the actual observed data and the predicted trajectory.

[0144] A third aspect of the present invention provides an electronic device, comprising:

[0145] processor;

[0146] Memory used to store processor-executable instructions;

[0147] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0148] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0149] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent fusion analysis of brain oxygenation signals in high-altitude environments, characterized in that, include: The system acquires multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. It performs spatiotemporal multi-scale separation and reconstruction on the multi-channel brain oxygen saturation data, mines the features of brain inter-brain oxygen metabolism synergistic patterns, and constructs a brain oxygen saturation spatiotemporal evolution tensor. Based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensatory limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector. The hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A forward-looking risk assessment benchmark is generated by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state. Based on the aforementioned forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction. The triggering factors and propagation paths of hypoxia risk are traced and quantified in reverse, and structured prediction results are output. Based on the structured prediction results, subsequent multi-channel brain oxygen saturation data are continuously tracked, and the multi-scale temporal dependence in the hypoxia susceptibility dynamic model is dynamically recalibrated in parameter space using the deviation between the actual observed data and the predicted trajectory.

2. The method according to claim 1, characterized in that, The multi-channel brain oxygen saturation data were subjected to spatiotemporal multi-scale separation and reconstruction to mine the features of coordinated oxygen metabolism patterns between brain regions, and a spatiotemporal evolution tensor of brain oxygen saturation was constructed, including: Variational mode decomposition was performed on multi-channel brain oxygen saturation data, and the signals of each channel were adaptively decomposed into a set of intrinsic mode components representing different physiological time scales. The intrinsic mode component sets corresponded to the rapid brain oxygen fluctuation layer, the respiratory heart rate coupling layer and the baseline metabolic drift layer in descending order of characteristic frequency. The instantaneous frequency envelope and instantaneous phase envelope of each intrinsic mode component were extracted as multi-scale time domain feature vectors. Based on the cross-scale phase coupling relationship between the intrinsic modal component set and the channels of different brain functional areas, a brain inter-modal coupling tensor is constructed. The multi-scale temporal domain feature vector is spatiotemporally embedded with the brain inter-modal coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modal scale, and time evolution as each dimension, the spatiotemporal evolution tensor of brain oxygen saturation is obtained.

3. The method according to claim 2, characterized in that, The multi-scale temporal feature vectors are spatiotemporally embedded with the brain inter-region modality coupling tensor. By constructing a tensor structure with brain functional area channels, intrinsic modality scales, and temporal evolution as dimensions, the spatiotemporal evolution tensor of brain oxygen saturation is obtained, which includes: Based on the phase-locking index and phase amplitude modulation index in the brain inter-modal coupling tensor, a cross-scale coupling propagation operator is constructed. The cross-scale coupling propagation operator maps the local representation of multi-scale temporal feature vectors in each intrinsic modal scale dimension to a global representation that integrates cross-scale synergistic effects by encoding the nonlinear interaction mechanism between different intrinsic modal scales. The global representation is tensor-organized according to the spatial topological relationship of the brain functional area channels. Based on the anatomical location and functional connectivity strength of the brain functional area corresponding to each channel, a brain region spatial adjacency structure is constructed. The brain region spatial adjacency structure embeds the global representation of discrete channels into tensor slices that preserve the spatial dependence between brain regions by quantifying the topological distance and functional synergy between brain functional area nodes. The tensor slices are stacked sequentially along the time evolution direction to construct a third-order tensor structure with brain functional area channels as the spatial dimension, intrinsic modality scale as the scale dimension, and time series as the temporal dimension. The third-order tensor structure is rearranged dimensionally according to the topological centrality index of brain functional area nodes in the spatial dimension and the characteristic frequency of intrinsic modal components in the scale dimension. This eliminates redundant dependencies between dimensions and highlights the dominant cooperative mode, resulting in the spatiotemporal evolution tensor of brain oxygen saturation.

4. The method according to claim 1, characterized in that, Based on the nonlinear correlation between the brain region oxygen metabolism coordination pattern characteristics and the multidimensional physiological parameter data, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed. This hypoxia susceptibility dynamic model characterizes the dynamic boundary between an individual's brain oxygen metabolism reserve capacity and compensatory limit through multi-time-scale physiological feedback loops, resulting in a hypoxia susceptibility state vector including: We quantify the cross-scale coupling strength of brain region oxygen metabolism coordination pattern characteristics and multidimensional physiological parameter data by constructing a conditional transfer entropy matrix under multiple time scale windows. Based on the conditional transfer entropy matrix, closed-loop physiological feedback loops are identified. By detecting brain oxygen metabolism coordination patterns and physiological parameter pairs that exhibit bidirectional transfer entropy and fast-slow transfer delay characteristics, a set of feedback loops is extracted. The set of feedback loops is constructed as a coupled set of dynamic differential equations. By defining the brain oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established. Phase space trajectory analysis was performed on the hypoxia susceptibility dynamic model. By solving the phase space evolution trajectory under different initial reserve states and physiological disturbance intensities, the set of critical bifurcation points that lead to trajectory escape from the stable attraction domain was identified. The set of critical bifurcation points constitutes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit. The phase space coordinates of the dynamic boundary are represented as a hypoxia susceptibility state vector.

5. The method according to claim 4, characterized in that, The feedback loop set is constructed as a coupled set of dynamic differential equations. By defining the cerebral oxygen metabolism reserve state as the system phase variable and multidimensional physiological parameters as external driving variables, a multi-scale time-dependent hypoxia susceptibility dynamic model is established, including: The feedback loops in the feedback loop set are decomposed into a structure. By identifying the brain oxygen metabolism coordination mode nodes and physiological parameter nodes in each feedback loop, the causal transmission paths and transmission delays between nodes are extracted, and a feedback loop topology diagram is constructed. Based on the feedback loop topology diagram, the system phase variables and external driving variables are defined. The brain oxygen metabolism reserve state and its consumption rate and recovery rate at different time scales are defined as the system phase variable set, and the multidimensional physiological parameters and their fluctuation amplitude and change trend at different time scales are defined as the external driving variable set. Based on the causal propagation path and propagation delay in the feedback loop topology diagram, each feedback loop is expressed as a differential equation, and by introducing delay terms and coupling terms, the differential equations of multiple feedback loops are coupled into a set of dynamic differential equations. The dynamic differential equations are subjected to time-scale separation transformation. By solving the rapidly changing system phase variables and the slowly changing system phase variables at different time scales, a multi-scale time-dependent hypoxia susceptibility dynamic model is constructed.

6. The method according to claim 1, characterized in that, The hypoxia susceptibility state vector is cross-modal semantically aligned with the time-varying environmental parameter data. A forward-looking risk assessment benchmark is generated by constructing a nonlinear mapping function between environmental stress factors and hypoxia susceptibility states, including: Cross-modal semantic projection is performed on the hypoxia susceptibility state vector and time-varying environmental parameter data. By calculating the mutual information transmission strength between each dimension component of the hypoxia susceptibility state vector and each dimension component of the time-varying environmental parameter data in the embedding space, a cross-modal correlation matrix is ​​constructed. Based on the cross-modal correlation matrix, an environmental stress factor set is extracted. By identifying environmental parameter dimensions in the matrix whose influence weight exceeds a preset weight threshold, an environmental stress factor set is constructed. The set of environmental stress factors and the hypoxia susceptibility state vector are constructed into a nonlinear mapping function. The nonlinear mapping function is then prospectively extrapolated. By inputting the predicted sequence of environmental parameters within a future time window into the nonlinear mapping function, the evolution trajectory of the hypoxia susceptibility state vector is solved. The minimum distance between the evolution trajectory and the dynamic boundary of the compensation limit and its corresponding arrival time are calculated. The minimum distance and the arrival time constitute a prospective risk assessment benchmark.

7. The method according to claim 1, characterized in that, Based on the aforementioned forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is subjected to multi-step extrapolation and risk causal chain reconstruction. The triggering factors and propagation paths of hypoxia risk are traced and quantified in reverse, and the structured prediction results are output, including: Based on the arrival time in the forward-looking risk assessment benchmark, the spatiotemporal evolution tensor of brain oxygen saturation is extrapolated in multiple forward steps. By taking the tensor at the current time as the initial state and combining it with the evolution trajectory predicted by the nonlinear mapping function, the tensor state at each future time step is iteratively calculated to construct the spatiotemporal evolution tensor sequence. Risk pattern recognition is performed on the spatiotemporal evolution tensor sequence. By detecting abrupt changes in the spatial distribution of brain oxygen saturation and instability points in the brain inter-regional coordination patterns in the spatiotemporal evolution tensor sequence, time nodes and brain functional area nodes are marked. Using the time nodes and brain functional area nodes as anchor points, reverse causal chain reconstruction is performed. By constructing a time-series dependency graph of the brain inter-brain oxygen metabolism coordination mode and performing backpropagation on the time-series dependency graph, the precursor brain functional area node sequence and precursor environmental stress factor sequence that cause the oxygen saturation mutation at the brain functional area node are identified. The contribution of the precursor brain functional area node sequence and the precursor environmental stress factor sequence is quantified. By calculating the causal contribution intensity of each node in the precursor brain functional area node sequence to the oxygen saturation mutation and the driving contribution intensity of each factor in the precursor environmental stress factor sequence to the evolution of the hypoxia susceptibility state vector, a structured prediction result is constructed.

8. A brain oxygen signal intelligent fusion analysis system for high-altitude environments, used to implement the method as described in any one of claims 1-7, characterized in that, include: The brain oxygen data unit is used to acquire multi-channel brain oxygen saturation data, multi-dimensional physiological parameter data, and time-varying environmental parameter data of target individuals in high-altitude environments. The multi-channel brain oxygen saturation data is then subjected to spatiotemporal multi-scale separation and reconstruction to mine the features of brain inter-brain oxygen metabolism synergistic patterns and to construct a brain oxygen saturation spatiotemporal evolution tensor. The hypoxia susceptibility unit is used to construct a multi-scale time-dependent hypoxia susceptibility dynamic model based on the nonlinear correlation between the characteristics of the brain region oxygen metabolism coordination mode and the multi-dimensional physiological parameter data. The hypoxia susceptibility dynamic model describes the dynamic boundary of an individual's brain oxygen metabolism reserve capacity and compensation limit through a multi-time-scale physiological feedback loop, and obtains a hypoxia susceptibility state vector. The risk assessment unit is used to perform cross-modal semantic alignment between the hypoxia susceptibility state vector and the time-varying environmental parameter data, and to generate a forward-looking risk assessment benchmark by constructing a nonlinear mapping function of environmental stress factors to hypoxia susceptibility state. The risk extrapolation unit is used to perform multi-step extrapolation and risk causal chain reconstruction on the spatiotemporal evolution tensor of brain oxygen saturation based on the aforementioned forward-looking risk assessment benchmark, reverse trace and quantify the triggering factors and propagation paths of hypoxia risk, and output structured prediction results. The model dynamic unit is used to continuously track subsequent multi-channel brain oxygen saturation data based on the structured prediction results, and to dynamically recalibrate the multi-scale temporal dependencies in the hypoxia susceptibility dynamic model by using the deviation between the actual observed data and the predicted trajectory.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images