Post-stroke cognitive impairment intelligent assessment method based on brain network dynamic reconstruction

Abstract

The invention relates to the technical field of medical care informatics, and discloses an intelligent post-stroke cognitive impairment assessment method based on brain network dynamic reconstruction, which comprises the following steps: acquiring multi-modal brain image data of a patient, dynamically compensating the difference between haemodynamic delay and white matter conduction velocity through a nerve-blood vessel coupling model, and calculating the post-stroke cognitive impairment; precise space-time alignment of function and structure signals is realized; a pulse neural network dynamic entropy change model is constructed based on the alignment data, functional connection strength change is coded into a pulse distribution sequence, and Gamma wave band synchronization enhancement and other compensatory features are extracted; constraining rationality of brain network reconstruction by adopting an energy optimal transmission model, and generating a compensatory thermodynamic diagram; and real-time calculation of the edge end is realized through pulse time sequence compression coding. The limitation of traditional static analysis is avoided, the algorithm achieves microsecond response on embedded equipment through the pulse coding technology, and a reliable stroke rehabilitation evaluation tool is provided for primary hospitals.

Description

Technical Field

[0001] The present invention relates to an intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks, belonging to the technical field of healthcare informatics. Background Art

[0002] In the rehabilitation assessment of stroke patients, dynamically capturing the spatio-temporal evolution law of the compensatory path of the brain network based on multi-modal medical data is a key link to improve the accuracy of the intelligent assessment system. At present, clinical assessment methods mainly rely on static functional connectivity analysis techniques, such as fMRI data processing based on a fixed time window, and independent assessment combined with structural imaging (such as DTI white matter tractography). Although this assessment method can locate the damaged area, there are the following technical bottlenecks: 1. Insufficient capture of dynamic compensatory characteristics: Static analysis cannot resolve the millisecond-level oscillation characteristics (such as enhanced Gamma-band synchronization) caused by synaptic plasticity, resulting in missed detection of the initiation signal of the compensatory path. The industry generally compensates for sensitivity by extending the sampling time window, but it will blur the temporal resolution and misjudge the compensatory window period.

[0003] 2. Multi-modal spatio-temporal mismatch: The hemodynamic delay of fMRI and the difference in white matter conduction velocity of DTI are not corrected. The independent processing of functional and structural data leads to cross-modal signal misalignment and generates false connections. Existing solutions usually use simple time interpolation alignment, but it is difficult to adapt to the asynchronous characteristics under non-uniform neural activities.

[0004] 3. Computational resource dependence restricts grass-roots applications: Dynamic network modeling relies on the high computing power support of GPU clusters, and traditional algorithms cannot achieve real-time analysis on embedded devices in grass-roots hospitals, such as mobile MRI workstations. The industry tries to relieve the computing power pressure by reducing the model complexity, but sacrifices the physiological rationality of the compensatory heat map.

[0005] Based on this, existing assessment methods have problems such as high misjudgment rate of compensatory paths, poor real-time performance, and difficulties in grass-roots deployment. In recent years, although some studies have tried to introduce dynamic graph networks or lightweight models, it is still difficult to balance temporal accuracy and physiological rationality, and lack of modeling of metabolic cost constraints, which is easy to generate results that violate the law of neurovascular coupling. Therefore, how to achieve dynamic and accurate alignment of multi-modal data, millisecond-level analysis of compensatory oscillation characteristics, and real-time and reliable assessment in a low-computing power environment has become a technical problem to be solved in this field. Summary of the Invention

[0006] The present invention provides an intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks, and its main purpose is to solve the problems of insufficient capture of dynamic compensatory characteristics, multi-modal spatio-temporal mismatch, and computational resource dependence restricting grass-roots applications.

[0007] To achieve the above object, an intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks provided by the present invention includes the following steps: Step 1, obtaining multi-modal brain imaging data of post-stroke patients, where the multi-modal brain imaging data at least includes blood oxygenation level-dependent (BOLD) signal data of functional magnetic resonance imaging (fMRI) and white matter fiber tracking data of diffusion tensor imaging (DTI); Step 2, for the multi-modal brain imaging data, using a delayed adaptive alignment algorithm to dynamically match the BOLD signal data and the white matter fiber tracking data on a non-uniform time axis, specifically including: constructing a neurovascular coupling model, and dynamically estimating and correcting the cross-modal time axis misalignment caused by hemodynamic delay and the difference in white matter fiber conduction velocity to generate a spatio-temporally aligned multi-modal brain imaging data stream; where is estimated based on the following formula: , where is the neurovascular coupling coefficient, is the white matter fiber length, is the white matter fiber conduction velocity; Step 3, based on the spatio-temporally aligned multi-modal brain imaging data stream, constructing a spiking neural network (SNN) dynamic entropy change model to encode the temporal oscillation characteristics of post-stroke brain network reconstruction, specifically including: encoding the change in dynamic functional connection strength between brain network nodes into a spike train of the SNN, and extracting temporal entropy change features characterizing the stability of the compensatory path and the resource consumption efficiency by analyzing the membrane potential oscillation characteristics of the spike train, where the temporal entropy change features include a synchronous enhancement mode in the Gamma band (30 - 80 Hz); Step 4, based on the spatio-temporally aligned multi-modal brain imaging data stream, constructing an energy optimal transport model to constrain the rationality of post-stroke brain network dynamic reconstruction, specifically including: modeling the post-stroke brain network dynamic reconstruction process as an energy optimal transport problem, introducing a neurovascular coupling equation as a constraint condition, calculating the reconstruction efficiency and resource allocation rationality of key hub nodes, and generating a compensatory heat map reflecting the resource allocation of the compensatory path; Step 5, using the spike train compression encoding (STCE) technology to sparsify the spike train of the SNN to generate a sparse event stream, and real-time reconstructing the temporal entropy change features and the compensatory heat map based on the sparse event stream on an edge embedded chip to achieve rapid intelligent assessment of post-stroke cognitive impairment.

[0008] In a preferred embodiment, in Step 2, the delayed adaptive alignment algorithm further includes: iteratively optimizing the parameters of the neurovascular coupling model to minimize the residual between the BOLD signal data and the white matter fiber tracking data on the time axis.

[0009] In a preferred embodiment, in step 3, the temporal entropy change model quantifies the transient response of synaptic weight adjustment during the compensation process by analyzing the frequency change and phase synchronization of membrane potential oscillations, and identifies specific compensatory oscillation patterns related to cognitive function recovery.

[0010] In a preferred embodiment, in step 4, the energy-optimal transmission model further introduces a metabolic cost function, which is constructed based on the energy consumption relationship between neuronal activity and hemodynamics, and is used to suppress the generation of compensatory paths that do not conform to physiological laws and consume excessive energy.

[0011] In a preferred embodiment, in step 5, the spike timing compression coding technology adopts an adaptive sparsification strategy based on dynamically reconstructed spike density, specifically including: real-time monitoring the spike calculation density of each brain region in the brain network, and when the spike density of the key compensatory brain region exceeds a preset threshold, triggering a local spike cluster merging mechanism to integrate high-frequency micro-spikes into low-frequency macro-spikes and retain their oscillation phase information; implementing event-driven truncation coding on the spike sequences of non-key brain regions, filtering low-information entropy spikes through a synaptic weight threshold, and only retaining key conduction events of cross-hemisphere interaction; deploying a dynamic scheduling algorithm for the spike buffer pool at the edge, and dynamically adjusting the parallel processing queue depth of the spike clusters according to the remaining real-time computing resources to ensure the real-time nature of the evaluation.

[0012] In a preferred embodiment, when generating macro-spikes, the local spike cluster merging mechanism embeds a time stamp vector carrying the original oscillation phase to ensure the integrity of oscillation synchronization analysis in subsequent SNN modeling.

[0013] In a preferred embodiment, in the event-driven truncation coding, the synaptic weight threshold is dynamically adjusted based on the brain network reconstruction data and cognitive function recovery of historical stroke patients.

[0014] In a preferred embodiment, it further includes the steps of: predicting the cognitive function recovery trajectory of stroke patients based on temporal entropy change features and compensatory heat maps, and generating personalized rehabilitation suggestions.

[0015] In a preferred embodiment, the machine learning model includes but is not limited to a support vector machine, a neural network, or a regression model.

[0016] In a preferred embodiment, after step 5, it further includes the step of: visually displaying the temporal entropy change features and compensatory heat maps.

[0017] Compared with the problems of the background art, the beneficial effects of the present invention are: 1. Through the neurovascular coupling model, the time axis offset of multimodal data is dynamically compensated, the cross-interference of hemodynamic delay and white matter conduction velocity difference is eliminated, and the precise spatiotemporal matching of functional and structural signals is ensured. On this basis, the pulse sequence encoding mechanism unique to the spiking neural network is used to convert the transient fluctuations of the dynamic functional connection strength into a resolvable membrane potential oscillation pattern, effectively capturing the compensatory path initiation signals that are difficult to identify by traditional static analysis methods, such as the initial activation characteristics of synaptic plasticity. This deep coupling of spatiotemporal calibration and dynamic encoding enables the system to extract key compensatory patterns such as Gamma band synchronization enhancement from millisecond-level neural oscillations, providing a basis for clinical judgment of the timing of intervention at an earlier stage.

[0018] 2. The brain network reconstruction process is modeled as an optimal energy transmission problem. The physiological rationality of the compensatory path is constrained by the neurovascular coupling equation to suppress the risk of false functional connections caused by ignoring metabolic costs in traditional algorithms. On this basis, the dynamic entropy change characteristics are sparsely processed by pulse timing compression coding technology. Through the pulse density adaptive resource scheduling strategy, the computing priority of key compensatory brain areas is achieved on the embedded chip. This collaborative design of energy constraint and hardware adaptation ensures that the compensatory thermogram conforms to the laws of cerebral hemodynamics, and enables the evaluation system to maintain microsecond response in a low computing power environment, breaking the real-time bottleneck of primary medical scenarios, realizing the deep integration of brain imaging physiological parameters and computational models, and improving the intelligence level of health data processing in edge computing.

[0019] 3. Imitate the regulation principle of glial cells on synaptic activity, design local pulse cluster merging and phase retention mechanism, compress the computing load while retaining the oscillation synchronization information completely, further combine with the metabolic cost dynamic decision-making model, prioritize the allocation of computing resources to cross-frequency coupling and other clinical high-value compensatory feature analysis, realize the accurate capture of key biomarkers and intelligent filtering of redundant data, this pulse coding strategy is linked with metabolic value assessment, so that the system can maintain low power operation while avoiding the feature confusion and phase distortion problems caused by full-band calculations in traditional methods, and realize the intelligent pulse processing mechanism to enhance the automatic decoding of medical data and clinical feature extraction capabilities.

[0020] 4. Through the organic integration of the pulsed neural network dynamic entropy change model and machine learning algorithms, a multi-dimensional correlation analysis is carried out on the temporal entropy change feature, the enhanced slope of the Gamma band synchronization and the resource allocation mode of the compensatory heat map, automatically identifying the compensatory oscillation mode strongly correlated with the recovery of cognitive function, avoiding the over-reliance on experience in traditional evaluation methods, and providing interpretable decision support for the formulation of personalized rehabilitation programs by quantifying the stability of the compensatory path and the resource utilization efficiency, significantly improving the accuracy and timeliness of rehabilitation intervention. Moreover, under the constraint of the neurovascular coupling equation, the compensatory heat map generated by the energy optimal transmission model can accurately reflect the rationality of resource reallocation between brain regions, avoiding the generation of connection patterns that violate physiological laws by the algorithm. At the same time, the dynamic scheduling algorithm of the pulse buffer pool automatically adjusts the parallel processing depth according to the real-time computing resources, ensuring the stable output of clinical-level evaluation results on the commonly used equipment in primary hospitals. Description of the Drawings

[0021] Figure 1 This is a schematic diagram of the process for adaptive alignment of brain imaging data with delay in the present invention.

[0022] Figure 2 This is a flow chart of spatio-temporal alignment of multi-modal brain imaging data and pulsed neural network modeling in the present invention.

[0023] Figure 3 This is a calibration flow chart of multi-modal brain imaging alignment data based on dynamic compensation in the present invention.

[0024] Figure 4 This is a flow chart of pulse density sparse compression coding and sparse event stream generation in the present invention.

[0025] The realization of the object, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0026] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] The embodiments of the present application provide an intelligent assessment method for post-stroke cognitive impairment based on dynamic brain network reconstruction, including the following steps: Step 1, obtaining multi-modal brain imaging data of post-stroke patients, where the multi-modal brain imaging data at least includes blood oxygenation level-dependent (BOLD) signal data of functional magnetic resonance imaging (fMRI) and white matter fiber tracking data of diffusion tensor imaging (DTI); Step 2, for the multi-modal brain imaging data, using a delay adaptive alignment algorithm to dynamically match the BOLD signal data and the white matter fiber tracking data on a non-uniform time axis, specifically including: constructing a neurovascular coupling model, and dynamically estimating and correcting the hemodynamic delay based on the neurovascular coupling model and the cross-modal time axis misalignment caused by the difference in white matter fiber conduction velocity to generate a spatio-temporally aligned multi-modal brain imaging data stream; wherein, is estimated based on the following formula: , wherein, is the neurovascular coupling coefficient, is the white matter fiber length, is the white matter fiber conduction velocity; Step 3, based on the spatio-temporally aligned multi-modal brain imaging data stream, construct a spiking neural network (SNN) dynamic entropy change model to encode the temporal oscillation characteristics of post-stroke brain network reconstruction, specifically including: encoding the change in dynamic functional connection strength between brain network nodes as a spike train of the SNN, and extracting the temporal entropy change features characterizing the stability of the compensatory path and the resource consumption efficiency by analyzing the membrane potential oscillation characteristics of the spike train, wherein the temporal entropy change features include the synchronous enhancement mode in the Gamma band (30 - 80 Hz); Step 4, based on the spatio-temporally aligned multi-modal brain imaging data stream, construct an energy optimal transport model to constrain the rationality of post-stroke brain network dynamic reconstruction, specifically including: modeling the process of post-stroke brain network dynamic reconstruction as an energy optimal transport problem, introducing the neurovascular coupling equation as a constraint condition, calculating the reconstruction efficiency and resource allocation rationality of key hub nodes, and generating a compensatory heat map reflecting the resource allocation of the compensatory path; Step 5, adopt the spiking timing compression encoding (STCE) technology to sparsify the spike train of the SNN to generate a sparse event stream, and based on the sparse event stream, real-time reconstruct the temporal entropy change features and the compensatory heat map on the edge embedded chip to realize the rapid intelligent assessment of post-stroke cognitive impairment.

[0028] In a preferred embodiment, in Step 2, the delay adaptive alignment algorithm further includes: iteratively optimizing the parameters of the neurovascular coupling model to minimize the residual between the BOLD signal data and the white matter fiber tracking data on the time axis.

[0029] In a preferred embodiment, in Step 3, the temporal entropy change model quantifies the transient response of synaptic weight adjustment during the compensatory process by analyzing the frequency change and phase synchrony of the membrane potential oscillation, and identifies specific compensatory oscillation patterns related to cognitive function recovery.

[0030] In a preferred embodiment, in Step 4, the energy optimal transport model further introduces a metabolic cost function, which is constructed based on the energy consumption relationship between neuronal activity and hemodynamics, and is used to suppress the generation of compensatory paths that do not conform to physiological laws and consume excessive energy.

[0031] In a preferred embodiment, in step 5, the pulse timing compression coding technology adopts an adaptive sparsification strategy based on dynamically reconstructed pulse density, which specifically includes: real-time monitoring of the pulse calculation density of each brain region in the brain network. When the pulse density of the key compensatory brain region exceeds the preset threshold, a local pulse cluster merging mechanism is triggered to integrate high-frequency micro-pulses into low-frequency macro-pulses and retain their oscillation phase information; event-driven truncation coding is implemented for the pulse sequences in non-key brain regions, and low-information-entropy pulses are filtered through the synaptic weight threshold, and only the key conduction events of cross-hemisphere interaction are retained; a pulse buffer pool dynamic scheduling algorithm is deployed at the edge end to dynamically adjust the parallel processing queue depth of the pulse clusters according to the real-time remaining computing resources to ensure the real-time nature of the evaluation.

[0032] In a preferred embodiment, when generating macro-pulses, the local pulse cluster merging mechanism embeds a time stamp vector carrying the original oscillation phase to ensure the integrity of the oscillation synchronization analysis in subsequent SNN modeling.

[0033] In a preferred embodiment, in the event-driven truncation coding, the synaptic weight threshold is dynamically adjusted based on the brain network reconstruction data and cognitive function recovery of historical stroke patients.

[0034] In a preferred embodiment, it further includes the steps of: predicting the cognitive function recovery trajectory of stroke patients based on the time series entropy change characteristics and compensatory heat map, and generating personalized rehabilitation suggestions.

[0035] In a preferred embodiment, the machine learning model includes but is not limited to a support vector machine, a neural network, or a regression model.

[0036] In a preferred embodiment, after step 5, it further includes the steps of: visually displaying the time series entropy change characteristics and compensatory heat map.

[0037] Example 1: In this example, represents the time offset between the BOLD signal and DTI data caused by hemodynamic lag and white matter conduction velocity difference. To ensure the time alignment of multi-modal data in the context of dynamic neural activity, this offset needs to be modeled and calculated based on the individual brain region neurovascular coupling characteristics. Among them, is the white matter tract path length between specific brain regions obtained by DTI white matter fiber tracking, with the unit of millimeter (mm), which can be extracted from the three-dimensional structure reconstruction of diffusion tensor imaging after standard atlas registration. is the average conduction velocity of white matter fibers, with the unit of millimeter per millisecond (mm / ms), which is usually set to 0.2 to 0.6 mm / ms according to the range reported in clinical literature, or the initial estimated value can be derived from the P300 latency in individual functional excitation tests. is the coupling coefficient, which is used to correct the neuro-vascular structural differences not explicitly modeled in the model. It is usually set as a dimensionless empirical constant and can be obtained through small-sample iterative optimization in practical applications. Its initial value can be set between 0.8 and 1.2, and the specific value is automatically updated according to the minimum criterion of the time-axis residual.

[0038] During the processing, the BOLD signal and DTI data are respectively mapped to a unified spatial coordinate system, and the time axis is reconstructed by non-uniform sampling, so that each data point carries its spatial position and estimated timestamp. For each BOLD sampling point, according to the calculated based on its corresponding spatial position and the DTI data association path, its timestamp is corrected. The resulting spatio-temporal aligned data stream has the synchronous expression characteristics of cross-modal neural signals under the same reference time axis.

[0039] For the neural network modeling part, to improve the interpretability of the spiking neural network (SNN) in identifying the compensatory pathways after stroke, during the implementation process, the original spatio-temporal aligned signal is first subjected to brain region registration and network node mapping. Each node represents a specific functional area group, and the connection weights between nodes are constructed according to the change of functional connection strength and encoded as a discretized spike train. This sequence is input into a pre-trained SNN model, and the change rule of its membrane potential can be mapped to the frequency component fluctuation in the time series, with a focus on the enhanced synchronization phenomenon in the Gamma band (30 - 80 Hz).

[0040] To avoid relying solely on high-frequency oscillations as the evaluation basis, the system further introduces the membrane potential temporal entropy index as a joint feature of compensatory stability and resource utilization efficiency. During the implementation process, the entropy value is calculated based on the multi-window sliding aggregation method, sequentially counting the distribution changes of the inter-spike intervals of key nodes in the spike train, so as to quantify its rhythmic and regular fluctuations. The enhanced synchronization in the Gamma band is manifested as a decrease in the entropy value and an enhanced frequency concentration trend within the corresponding window. Combining the metabolic cost constraint to construct a compensatory heat map, the resource distribution of the compensatory pathway is evaluated from the spatial dimension to determine whether it is consistent with the neuro-vascular distribution logic, avoiding misjudging pseudo-compensatory pathways with extremely high energy consumption as effective connections. In the STCE compression stage, the system counts the change of spike density per unit time in units of neural nodes. If the spike firing density in a certain area exceeds the device's carrying capacity within a short time window, the spike cluster merging mechanism is activated. This mechanism integrates adjacent spikes into a single macro-spike in a phase-preserving manner, and at the same time embeds a timestamp vector representing the original oscillation rhythm. This merging strategy ensures that the SNN can still accurately analyze the original synchronization characteristics in the subsequent processing.

[0041] For non-critical brain regions, when the pulse density is low and the information entropy is insufficient, the system implements event-driven truncation coding. Based on the synaptic weight distribution threshold in historical samples, it dynamically filters pulses with lower information content and retains the event stream related to cross-hemisphere interaction or critical path coupling, further reducing the computational burden on the embedded system. During this process, the synaptic weight threshold is obtained through training based on the rehabilitation trajectory data of historical stroke patients and is dynamically adjusted according to the current prediction error of the model to avoid information loss or resource waste caused by a fixed threshold. At the edge execution layer, a dynamic scheduling module for the pulse buffer pool is implemented and deployed. According to the remaining computing resources and processing delay of the chip fed back in real time, it dynamically adjusts the parallel processing depth and data extraction rate of the pulse sequence to ensure the stable output of the heat map and entropy change index for clinical reference without affecting real-time performance. These are all extended implementation methods known to those of ordinary skill in the art.

[0042] Embodiment 2: This embodiment further illustrates the technical solution of the present invention in combination with Figures 1 to 4 ,. Figure 1 It is a schematic diagram of the processing flow for delay adaptive alignment of brain imaging data, mainly used to describe how to dynamically estimate and correct the input blood oxygenation level-dependent (BOLD) signal data and white matter fiber tracking data through a neurovascular coupling model to generate a spatio-temporally aligned multi-modal brain imaging data stream. This process first imports the BOLD signal data and white matter fiber tracking data through the input data module. Then, in the delay adaptive alignment processing module, a physical constraint relationship between the two types of data is established by constructing a neurovascular coupling model. Based on this model, the hemodynamic delay and conduction velocity difference are further estimated to obtain preliminary estimation parameters, and iterative optimization of the model parameters is combined for repeated iterative updates to improve the estimation accuracy. Subsequently, the system enters the dynamic estimation and correction step to perform alignment processing on the data under a non-uniform time axis, and the final result is output through the module for generating a spatio-temporally aligned data stream, and then provided as output data with a spatio-temporally aligned multi-modal brain imaging data stream.

[0043] Figure 2It shows the main steps from the acquisition of original brain images to neural network modeling and their logical associations. First, in a multimodal acquisition system, BOLD signals and white matter fiber tracking data are synchronously acquired to form the original input for subsequent processing. Through the spatio-temporal calibration step, this original data enters the dynamic alignment engine module, where operations of blood flow delay compensation and conduction velocity correction are successively performed to correct the time-axis mismatch between BOLD signals and DTI data caused by hemodynamic lag and white matter fiber conduction differences. The above two processing measures jointly complete the dynamic time registration to ensure the synchrony of multimodal data. After alignment, through the membrane potential encoding process, functional connectivity changes are converted into pulse sequences and input into the spiking neural network module, which is used to extract the temporal oscillation characteristics and compensatory patterns in the post-stroke brain network reconstruction.

[0044] Figure 3 It is a calibration flowchart of multimodal brain image alignment data based on dynamic compensation, showing the whole process of data calibration through the dynamic compensation mechanism based on the fusion of fMRI BOLD signals and DTI white matter fibers. First, the system respectively receives fMRI BOLD signals and DTI white matter fiber data as inputs and introduces neurovascular coupling coefficients and conduction velocity correction parameters according to their respective characteristics. On this basis, the two types of data are unified to the same time-axis reference framework through the spatio-temporal alignment engine and sent to the dynamic compensation module for in-depth correction processing. In this module, independent identification and modeling are respectively carried out for hemodynamic delay and white matter conduction differences. The former mainly reflects the response lag of BOLD signals relative to neural activities, and the latter depicts the non-uniformity of conduction velocity along DTI paths. The two cooperate to form a complete dynamic compensation path. After correction, the system outputs the calibrated data stream, that is, the multimodal fusion data stream with double deviations of hemodynamics and conduction velocity corrected, for subsequent analysis steps such as neural network modeling or compensatory pattern recognition.

[0045] Figure 4It is a flowchart of pulse density sparsification compression encoding and sparse event stream generation, which describes the density monitoring and encoding compression process experienced by the original pulse sequence before entering the edge computing terminal. First, the system receives the input of the original pulse sequence and dynamically monitors the pulse activity frequency through the pulse density monitoring module. If the monitoring result exceeds the threshold, a local pulse cluster merging operation is triggered to aggregate adjacent high-frequency pulses, and then macro-pulse generation is performed to construct a macroscopic coding unit that retains the original oscillation phase information, and finally it is output to the sparse event stream; if it does not exceed the set threshold, it enters the event-driven truncation encoding stage, implementing a resource-saving compression strategy for the low-density pulse sequence, and only screening out events that are meaningful for cross-brain region interaction through the key conduction event retention mechanism, also forming a sparse event stream. This process realizes the branched processing and compression optimization of the original pulse signal, ensuring that high-value sparse event data can still be output for subsequent timing modeling under the limited resources of the embedded platform.

[0046] Example 3: In this example, the specific details in the dynamic alignment of multimodal data and the modeling process of the spiking neural network are specifically described. First, in the dynamic alignment process, the calculation of the hemodynamic delay term has been defined by the formula . Among them, the variable represents the white matter fiber path length obtained by diffusion tensor imaging (DTI), with the unit of millimeters (mm). In practice, the connection distance between corresponding brain regions can be extracted through standard brain atlas registration; is the white matter fiber conduction speed, with the unit of millimeters per millisecond (mm / ms). Usually, the reference setting range is 0.2–0.6 mm / ms, and the individual initial value can be estimated through experimental means such as P300 evoked potential; is a dimensionless coupling correction coefficient, which mainly reflects the individual differences between neural activity and blood flow response. It is recommended that the initial value be taken in the range of 0.8–1.2, and it is iteratively updated by minimizing the BOLD signal and DTI timing residuals.

[0047] In the actual implementation process, for each BOLD data point, according to its spatial mapping relationship to the DTI path, the above formula is used to estimate its time offset, and its timestamp is corrected through non-uniform time axis interpolation to ensure the fusion of functional signals and structural signals under the same reference time axis, thereby improving the consistency and analysis accuracy of cross-modal data. In the construction stage of the spiking neural network (SNN), the aligned brain network node signals are mapped into a pulse sequence, and the membrane potential change trend is calculated in a discrete time window manner. This sequence is input into the pre-trained spiking neural network, which not only retains the spatial connection structure but also encodes the high-frequency change characteristics of functional activities on the time axis.

[0048] In this embodiment, it is further illustrated that the calculation of the entropy value is based on a multi-window sliding aggregation strategy. In each sliding window, the pulse interval sequence of the key nodes is statistically analyzed, its time distribution density is estimated through a probability distribution function, and its uncertainty index is calculated in the form of Shannon entropy, so as to evaluate the rhythm stability of neural activities. This method not only provides a quantitative judgment basis for synchronous enhancement, but also provides time-domain feature support for the physiological rationality analysis of subsequent compensatory paths. The decrease of the entropy value usually indicates the concentration of frequencies and the enhancement of the rhythm of the neural network, and this phenomenon is particularly significant in the Gamma band, which can be used to identify effective compensatory pathways between brain regions. In this embodiment, it is further defined that the width of the sliding window is set between 100–200 ms, and it is dynamically adjusted according to the sampling frequency and the neural discharge cycle to ensure statistical stability without sacrificing time resolution. After the pulse density exceeds the set threshold, the system will start the macro-pulse integration mechanism, and generate a macro time-stamp vector under the principles of time continuity and phase retention to ensure that the analysis of frequency synchronization in subsequent analysis is not interfered by sparsification. Moreover, the generation of the sparse event stream during the compression process does not simply compress the data scale, but performs weight screening according to the neural network structure characteristics and the information entropy weight of the physiological signal. For example, in the deployment of the edge computing platform, if the non-critical brain regions meet the conditions of low pulse density and low synaptic weight at the same time, their event streams will be actively truncated by the system to save computing resources; the event streams of the critical brain regions will enter the real-time processing channel through the priority queue mechanism to ensure the integrity of the decision-making features.

[0049] And the synaptic weight threshold is not fixed, but is dynamically adjusted according to the median of the weight distribution curve constructed based on the rehabilitation data of historical stroke patients, and is updated reversely in combination with the current network prediction error to form a neural path screening strategy with adaptive learning ability, which helps to avoid missing important compensatory features due to signal clipping, taking into account both computational efficiency and the credibility of physiological mechanisms. In terms of system deployment, for example, the entire modeling and optimization process can be completed based on mainstream open-source platforms such as PyTorch (for SNN construction) and TensorFlow Lite for edge deployment. It is recommended to use a combination of a standardized fMRI acquisition protocol TR = 2000 ms and DTI imaging (30 vectors, b = 1000 s / mm²) for data acquisition to ensure the consistency and reproducibility of the model input, and these are all extended implementation methods known to those of ordinary skill in the art.

[0050] Embodiment 4: In the process of realizing spatio-temporal alignment in this embodiment, first, a neurovascular coupling model is used to perform compensation processing on multi-modal brain image data such as fMRI BOLD signals and DTI white matter fiber tracking data. Specifically, first, the hemodynamic delay is calculated: and the conduction velocity difference of white matter fibers. In the formula, The calculation of is as follows, where represents the white matter fiber path length (unit: millimeter), and represents the conduction velocity of white matter fibers (unit: millimeter / millisecond). is the neurovascular coupling coefficient, dimensionless, mainly used to compensate for the neuro-vascular differences not explicitly modeled in the model. In practical applications, The initial value of can be set to 0.8 to 1.2 and dynamically optimized by minimizing the temporal axis residuals between the BOLD signal and DTI data. During this process, this formula is used to correct the time offset of each BOLD signal sampling point. By associating with the corresponding white matter fiber path in the DTI data, the timestamp of each BOLD signal is corrected, so as to ensure the consistency of functional and structural data on the time axis, generate a spatio-temporally aligned data stream, and in the formula, : Hemodynamic delay, used to compensate for the time offset caused by the difference between blood flow delay and white matter fiber conduction velocity; : White matter fiber path length, extracted from DTI data, unit is millimeter; : Conduction velocity of white matter fibers, usually in the range (0.2–0.6 mm / ms) set in the reference literature, or obtained through P300 evoked potential experiments; : Neurovascular coupling coefficient, used to correct the relationship between blood flow and neural activity, with a value range of 0.8 to 1.2.

[0051] The construction goal of the spiking neural network SNN is to extract compensatory oscillation features based on the aligned multi-modal data stream. In this step, the nodes of the brain network are encoded by the change of functional connection strength and transformed into spike train sequences. The change of membrane potential of the spike train sequence reflects the temporal characteristics of neural activity. In particular, the synchronous enhancement phenomenon in the Gamma band (30 - 80Hz) is concerned, and this phenomenon can characterize the stability and resource consumption efficiency of the compensatory path. By analyzing the membrane potential oscillation characteristics of these spike train sequences, the temporal entropy change index representing the compensatory characteristics can be extracted.

[0052] Based on the spatio-temporal aligned data stream, an energy-optimal transmission model is further constructed to constrain the rationality of brain network reconstruction. This model regards the dynamic reconstruction process of the brain network as an energy-optimal transmission problem and introduces the neurovascular coupling equation as a constraint condition. This model can effectively suppress the generation of false functional connections and ensure that the compensatory path conforms to the neurovascular coupling law. On this basis, data sparsification is performed through the Spatio-Temporal Compressive Encoding (STCE) technique. Specifically, an adaptive sparsification strategy based on the pulse density of dynamic reconstruction is adopted. When the pulse density in some key regions of the brain network exceeds the set threshold, a local pulse cluster merging mechanism is triggered. Through this mechanism, high-frequency micro-pulses are integrated into low-frequency macro-pulses while retaining their oscillation phase information. This operation not only effectively reduces the computational burden but also retains the temporal synchrony of the compensatory path. For non-critical brain regions, if the pulse density is low and the information entropy is small, resource-saving processing is performed through event-driven truncation coding, and only the event stream with high information entropy and cross-brain region interaction is retained. This method can effectively reduce the consumption of computing resources while ensuring that the information on the key compensatory path is not lost. Through the above steps, the generated temporal entropy change features and compensatory heat maps will be output as real-time computed data streams and processed by the edge embedded chip. This chip adjusts the parallel processing queue depth of the pulse clusters according to the real-time computing resources to ensure that the system can still stably output clinical-level evaluation results in a low-computing power environment. All of these belong to the extended implementation manners known to those of ordinary skill in the art.

[0053] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0054] 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 preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks, characterized in that, Including the following steps: Step 1: Obtain multimodal brain imaging data of patients after stroke. The multimodal brain imaging data includes at least blood oxygenation level-dependent (BOLD) signal data of functional magnetic resonance imaging (fMRI) and white matter fiber tracking data of diffusion tensor imaging (DTI). Step 2: For the multi-modal brain imaging data, use the delayed adaptive alignment algorithm to dynamically match the BOLD signal data and the white matter fiber tracking data on the non-uniform time axis, specifically including: constructing a neurovascular coupling model, and dynamically estimating and correcting the cross-modal time axis misalignment caused by the hemodynamic delay and the difference in white matter fiber conduction velocity, and generating a spatio-temporally aligned multi-modal brain imaging data stream; Step 3: Based on the spatio-temporally aligned multimodal brain imaging data stream, construct a spiking neural network (SNN) dynamic entropy change model to encode the temporal oscillation characteristics of post-stroke brain network reconstruction. Specifically, it includes encoding the change in dynamic functional connection strength between brain network nodes as a spike train of the SNN, and extracting temporal entropy change features characterizing the stability of the compensatory path and the resource consumption efficiency by analyzing the membrane potential oscillation characteristics of the spike train. Step 4: Based on the spatio-temporally aligned multimodal brain imaging data stream, construct an energy optimal transport model to constrain the rationality of post-stroke brain network dynamic reconstruction. Specifically, it includes modeling the process of post-stroke brain network dynamic reconstruction as an energy optimal transport problem, introducing a neurovascular coupling equation as a constraint condition, calculating the reconstruction efficiency and resource allocation rationality of key hub nodes, and generating a compensatory heat map reflecting the resource allocation of the compensatory path. Step 5: Adopt the spike timing compression encoding (STCE) technique to sparsify the spike train of the SNN, generate a sparse event stream, and reconstruct the temporal entropy change features and the compensatory heat map in real time on the edge embedded chip based on the sparse event stream to achieve rapid and intelligent assessment of post-stroke cognitive impairment.

2. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 1, wherein In the said step 2, the delay adaptive alignment algorithm further includes: iteratively optimizing the parameters of the neurovascular coupling model to minimize the residuals of the BOLD signal data and the white matter fiber tracking data on the time axis, and, estimating based on the following formula: ， Among them, is the neurovascular coupling coefficient, is the white matter fiber length, is the white matter fiber conduction velocity.

3. The intelligent assessment method for post-stroke cognitive impairment based on dynamic brain network reconstruction according to claim 1, characterized in that, In step 4, the energy optimal transport model further introduces a metabolic cost function, which is constructed based on the energy consumption relationship between neuronal activity and hemodynamics. And in step 3, the temporal entropy change features include a synchronous enhancement pattern in the Gamma band of 30 - 80 Hz.

4. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 1, characterized in that, In step 5, the spike timing compression encoding technique adopts an adaptive sparsification strategy based on dynamically reconstructed pulse density. Specifically, it includes: real-time monitoring the pulse calculation density of each brain region in the brain network, triggering a local pulse cluster merging mechanism when the pulse density of the key compensatory brain region exceeds a preset threshold, integrating high-frequency micro-pulses into low-frequency macro-pulses and retaining their oscillation phase information; implementing event-driven truncation encoding for the pulse sequence of non-key brain regions, filtering low-information entropy pulses through a synaptic weight threshold, and only retaining key conduction events of cross-hemisphere interaction; deploying a dynamic scheduling algorithm for the pulse buffer pool at the edge, and dynamically adjusting the parallel processing queue depth of the pulse cluster according to the real-time remaining computing resources.

5. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 4, characterized in that, When generating macro-pulses, the local pulse cluster merging mechanism embeds a timestamp vector carrying the original oscillation phase.

6. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 5, characterized in that, In the event-driven truncation encoding, the synaptic weight threshold is dynamically adjusted based on the brain network reconstruction data and cognitive function recovery of historical stroke patients.

7. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to any one of claims 1 to 6, characterized in that It further includes the step of: based on the temporal entropy change features and the compensatory heat map, using a machine learning model to predict the cognitive function recovery trajectory of post-stroke patients and generate personalized rehabilitation suggestions.

8. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 7, characterized in that, The machine learning model includes a support vector machine, a neural network, or a regression model.

9. The intelligent assessment method for post-stroke cognitive impairment based on dynamic reconstruction of brain networks according to claim 1, characterized in that, After step 5, it further includes the step of: visually displaying the temporal entropy change features and the compensatory heat map.

Images