A multimodal brain-computer interface system for ultra-early stroke warning that integrates eye diagnosis and traditional Chinese medicine syndromes

By constructing a brain-computer multimodal AI early warning system that combines eye diagnosis with traditional Chinese medicine syndrome differentiation, the problems of non-invasiveness, specificity, and individual adaptability in stroke screening have been solved, achieving highly efficient early warning of stroke in its very early stages and possessing commercial potential.

CN123058751APending Publication Date: 2026-11-10ZHENGZHOU ZHENGHUA ENTROPY MACHINE HUMAN RESOURCES MANAGEMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611265586.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-20
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing stroke screening methods cannot achieve continuous and non-invasive risk monitoring. Brain-computer interface signals lack specificity in identifying stroke precursors. Traditional Chinese medicine stroke precursor syndromes lack calculable and verifiable digital models. There is a lack of quantitative correlation between abnormal ocular microcirculation and cerebral perfusion decline based on anatomical homology. Existing early warning systems lack the ability to dynamically adapt to individual baseline states and are unable to distinguish between pathological abnormalities and individual differences in constitution.

Method used

We will construct a brain-computer multimodal AI early warning system for stroke that integrates eye diagnosis and TCM syndrome differentiation. By combining ocular microcirculation imaging, brain perfusion monitoring and TCM syndrome differentiation with Bayesian network architecture, we will achieve causal consistency reasoning based on multi-source evidence, construct an individual brain perfusion baseline model, and provide dynamic early warning.

Benefits of technology

It achieves efficient and reliable early warning of stroke in the very early stage, reduces the risk of false alarms in a single modality, has a commercialization path, adapts to individual physical differences, and is clinically operable.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This invention discloses a multimodal brain-computer interface (BCI) AI early warning system for stroke that integrates eye diagnosis and traditional Chinese medicine (TCM) syndrome differentiation, belonging to the interdisciplinary field of medical AI and brain-computer interface. The system includes: a one-second eye diagnosis rapid screening module, which extracts scleral choroidal vessel morphological features based on an eye-brain blood supply homology model and outputs a retinal-cerebral small vessel disease risk index; a multimodal brain perfusion acquisition module, which simultaneously acquires EEG, fNIRS, and cerebral blood flow fluctuation data to identify pre-hypoperfusion features; a TCM stroke prodromal syndrome quantification module, which maps liver wind, phlegm stasis, qi deficiency, and yin deficiency syndromes into risk weight vectors associated with carotid artery IMT, blood pressure variability, and clinical stroke risk scores; and an AI multi-source coupling inference early warning module, which constructs a three-way causal consistency map based on a Bayesian network. When abnormal eye diagnosis and decreased brain perfusion have upstream and downstream correlations in the anatomical blood supply chain, and the syndrome matching and brain perfusion decline trend are consistent and the correlation continuously exceeds a preset threshold, it is determined to be an ultra-early high-risk state. The system introduces an individual dynamic baseline model of brain perfusion and a hierarchical deployment architecture, reducing the risk of false alarms from a single modality through multi-source verification. This invention fills the gap in the lack of continuous non-invasive monitoring tools in the ultra-early window before stroke onset.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence and brain-computer interface technology, and in particular to an AI early warning system for stroke that integrates ocular microcirculation imaging, brain perfusion monitoring and TCM syndrome quantification. Background Technology

[0002] Stroke is characterized by high incidence, high disability rate, and high mortality rate. According to the "China Stroke Report 2023," approximately 3.94 million new stroke cases occur in my country each year, with over 28 million patients currently in the disease. Clinical evidence shows that reversible hypoperfusion and microcirculatory disturbances occur in brain tissue several hours to days before a stroke. Effective early warning and intervention at this stage can significantly reduce disability and mortality rates. Existing stroke screening methods mainly include carotid ultrasound, cranial MRI, CT perfusion imaging, and ABCD² scoring, but these methods mostly rely on large medical equipment, making it difficult to achieve continuous, non-invasive, and low-cost risk monitoring. Existing brain-computer interface (BCI) systems primarily focus on post-stroke rehabilitation training, with few studies addressing ultra-early warning systems before stroke onset. Meanwhile, traditional Chinese medicine (TCM) has accumulated rich experience in identifying stroke precursor symptoms (such as "internal liver wind" and "phlegm and blood stasis obstructing the collaterals"), but lacks structured, quantifiable digital representations, making it difficult to integrate with modern physiological signals for inference. Furthermore, single-modal approaches (such as EEG or fNIRS) exhibit poor specificity and high false alarm rates in stroke precursor identification, lacking mechanisms for verifying pathological consistency across multi-source data. The ocular microcirculation (retina and scleral choroid) shares a common blood supply system with cerebral small vessels (the ophthalmic artery is the first branch of the internal carotid artery), and changes in retinal microvessels can reflect the risk of cerebral small vessel disease. Existing studies (such as the ARIA-stroke system from the Chinese University of Hong Kong) have shown that retinal image analysis can assess the risk of ischemic stroke, with reportedly high sensitivity and specificity. However, the specificity of a single ocular diagnostic modality decreases in asymptomatic individuals in the community and remains insufficient to independently support clinical early warning decisions. Summary of the Invention

[0003] (a) Technical problems to be solved This invention aims to solve the following technical problems: Current stroke screening methods cannot achieve continuous, non-invasive risk monitoring; Brain-computer interface signals (EEG / fNIRS) lack specificity in stroke precursor identification and are easily affected by physiological fluctuations. Traditional Chinese medicine lacks calculable and verifiable digital models for stroke prodromal symptoms, making it difficult to integrate with modern medical indicators. There is a lack of quantitative association based on anatomical homology between ocular microcirculation abnormalities and cerebral perfusion decline; Existing early warning systems lack the ability to dynamically adapt to an individual's baseline state, making it difficult to distinguish between pathological abnormalities and individual differences in physical condition. (II) Technical Solution To address the aforementioned technical problems, this invention provides a brain-computer interface multimodal AI early warning system for stroke that integrates eye diagnosis and traditional Chinese medicine syndrome differentiation, comprising: The one-second eye diagnosis rapid screening module is configured to acquire high-definition images of the eye within a preset exposure time. It automatically segments the scleral choroid region using a semantic segmentation model based on deep learning, extracts vascular density, arteriovenous diameter ratio, and microvascular morphological abnormalities, and generates a retinal-cerebral small vessel disease risk index based on a homologous blood supply model between the ophthalmic artery and the internal carotid artery. The brain-computer interface multimodal brain perfusion acquisition module simultaneously acquires EEG microrhythm signals, fNIRS near-infrared brain blood oxygen perfusion parameters, and cerebral blood flow fluctuation data. Through the signal synchronization unit, it realizes timestamp alignment and artifact removal of multi-source signals, and extracts α / β band asymmetry index, θ band burst rate, δ band slow wave diffusion coefficient, oxyhemoglobin slope attenuation, and deoxyhemoglobin retention time to identify the precursor features of brain tissue hypoperfusion. The TCM stroke prodromal syndrome knowledge base quantitative module stores structured data models of liver wind, phlegm stasis, qi deficiency, and yin deficiency syndromes. It maps syndrome elements into risk weight vectors associated with carotid artery IMT, homocysteine, blood pressure variability, and clinical scores (CHA2DS2-VASc, ABCD²), and outputs syndrome matching scores through a dynamic reasoning engine. The AI ​​multi-source coupled inference and early warning module constructs a causal consistency map among the ocular microcirculation abnormality index, the cerebral perfusion decline index, and the stroke syndrome matching index. This map is built on a Bayesian network architecture, using the three indices as observation nodes and "early-stage high-risk state of stroke" as the latent variable. A directed acyclic graph structure between nodes is learned from the training data using a structural learning algorithm (such as the PC algorithm or the snow-mountain climbing algorithm), and the conditional probability table for each node is calculated using maximum likelihood estimation. During the inference phase, after evidence is input from the three observation nodes, the posterior probability of the latent variable is calculated using the message passing algorithm of the Bayesian network. An early warning is triggered when the posterior probability exceeds a preset threshold. An early-stage high-risk state of stroke is determined when all three indices meet the following conditions: (1) There is an upstream and downstream correlation between the abnormal area of ​​eye diagnosis and the area of ​​decreased brain perfusion in the anatomical blood supply chain. The specific determination method is as follows: First, locate the abnormal eye diagnosis in the four quadrants of the scleral choroid (superior temporal, inferior temporal, superior nasal, inferior nasal). Based on the anatomical correspondence between the branches of the ophthalmic artery and the blood supply areas of the middle cerebral artery and the anterior cerebral artery (the temporal branch of the ophthalmic artery mainly corresponds to the blood supply area of ​​the middle cerebral artery, and the nasal branch mainly corresponds to the blood supply area of ​​the anterior cerebral artery), map the fundus quadrant to the corresponding blood supply area of ​​the ipsilateral cerebral hemisphere. Then, determine the brain region where the brain perfusion is decreased through the spatial localization information of fNIRS. When the brain region mapped by the abnormal eye diagnosis quadrant is anatomically consistent with the low perfusion brain region detected by fNIRS, it is determined that the upstream and downstream correlation is established. (2) The pathological tendency indicated by the syndrome matching index is consistent with the time-varying trend of cerebral perfusion decline; (3) The above correlation persists for more than a preset time threshold on the time axis—based on clinical pre-trial data, this threshold is set at 30 minutes. The physiological basis for this setting is that physiological fluctuations (such as changes in body position, emotional fluctuations, and transient hypotension) are usually short-lived (from a few seconds to a few minutes), while the persistence of pathological hypoperfusion is a key characteristic of tissue ischemia progressing to irreversible damage. In the clinical pre-trial, when the threshold was 30 minutes, the correlation between the warning event and the clinically diagnosed stroke within the following 7 days reached the optimal balance. This threshold can be further optimized and adjusted in subsequent large-scale clinical trials. Meanwhile, this module constructs an individual brain perfusion baseline model based on historical continuous monitoring data. When the current parameter deviates from the baseline by more than the standard and triggers multi-source mutual verification, it outputs an individualized early warning result. The system also includes an optional expandable physiological sensing submodule, including carotid artery MEMS blood flow vibration sensing, PPG peripheral perfusion and HRV acquisition, and IMU posture and gait imbalance sensing, to supplement vascular stiffness, autonomic nerve imbalance and micro-impairment characteristics of motor function. The system adopts a hierarchical deployment architecture: Community / Home Screening Level: Integrates an eye diagnosis module and a simplified PPG+IMU module for periodic risk trend monitoring; Clinical / Emergency Review Level: After the screening-level warning is triggered, the brain-computer interface multimodal brain perfusion acquisition module and the carotid artery MEMS module are activated to perform high-precision diagnostic review. (III) Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects: This invention is the first to construct a causal consistency coupling mechanism among ocular microcirculation imaging, brain-computer multimodal brain perfusion monitoring, and TCM stroke precursor syndrome quantification in the ultra-early warning scenario of stroke (based on Bayesian network architecture to realize causal consistency reasoning of multi-source evidence), which has higher reasoning interpretability and decision reliability than traditional threshold alarm methods. The system utilizes multi-source mutual verification logic, employing pathological consistency verification among three information sources: eye diagnosis (microcirculation abnormalities), cerebral perfusion (brain function monitoring), and traditional Chinese medicine syndromes (pathological tendencies), to effectively reduce the false alarm risk of a single modality. The technical basis of each single modality is as follows: In terms of ocular diagnosis modalities, the ARIA-stroke system from the Chinese University of Hong Kong has demonstrated the technical feasibility of stroke risk assessment based on retinal image analysis. This system, building upon scleral choroid imaging, further strengthens the pathological association between microcirculatory abnormalities and cerebral small vessel disease through eye-brain blood supply homology mapping. In the quantification of TCM syndromes, existing research has constructed an intelligent diagnostic model for Qi deficiency and blood stasis syndrome in stroke based on the characteristics of the scleral veins, providing a methodological foundation for the objective quantification of syndrome elements. Building upon this, our system expands the syndrome types to four categories: liver wind, phlegm stasis, Qi deficiency, and Yin deficiency, and establishes mapping relationships with modern medical indicators. Regarding cerebral perfusion monitoring, previous studies have shown a temporal correlation between electroencephalogram (EEG) changes in patients with severe acute cerebrovascular disease and TCM syndrome elements, suggesting that neurophysiological signals can serve as effective indicators for assessing brain function. This system further integrates fNIRS cerebral oxygenation monitoring and cerebral vascular blood flow fluctuation data to construct a multimodal cerebral perfusion assessment system. The system's overall early warning performance needs to be validated through a prospective, multi-center clinical trial (sample size ≥1200 cases, with the gold standard being cranial MRI, DSA, and 90-day clinical follow-up). Furthermore, this invention, by introducing an individual dynamic baseline model of cerebral perfusion, can distinguish between pathological abnormalities and individual differences in physical condition, adapt to seasonal blood pressure fluctuations and age-related vascular aging, and fill the gap in effective monitoring tools for the ultra-early window before stroke onset. The tiered deployment architecture balances the universality of screening with the accuracy of verification, enabling low-cost continuous monitoring in community / home settings and providing diagnostic-level decision support in clinical settings, with a clear path to commercialization. The early warning results are directly linked to clinical actions such as carotid ultrasound, MRI examinations, or medication adjustments, making the system clinically operable. Detailed Implementation Example 1: System Overall Architecture The system consists of a wearable front-end data acquisition device and a cloud / local AI processing unit. The front end includes an eye examination camera module (RGB + near-infrared), a dry electrode EEG cap (8-16 channels), a dual-wavelength fNIRS probe (4×4 arrangement), a PPG photoelectric sensor, a triaxial IMU, and a carotid artery MEMS vibration sensor (clinical-grade configuration). All sensors achieve millisecond-level timestamp alignment through a unified clock source. Example 2: Implementation of the One-Second Eye Diagnosis Module The system acquires high-resolution images of both eyes within a preset exposure time (≤1 second), with the user facing the device. The semantic segmentation model is based on the U-Net++ architecture (using ResNet50 as the backbone network in the encoder and nested dense skip connections in the decoder; training uses the Adam optimizer with an initial learning rate of 1e-4, a batch size of 16, and a total of 200 epochs; the loss function is a weighted combination of Dice loss and cross-entropy loss). It is trained on a proprietary dataset containing 5000 annotated scleral choroid images (annotations were independently completed by at least two ophthalmologists, including scleral vessel boundaries, bifurcation points, and microvascular abnormalities; annotation consistency was evaluated using the Dice coefficient, with samples ≥0.85 included in the training set). The model's segmentation accuracy has a Dice coefficient ≥0.92. Extracted vascular morphology features are converted into a retinal-cerebral small vessel disease risk index (range 0~1) using an eye-brain homology mapping function. Example 3: Implementation of a Brain-Computer Multimodal Brain Perfusion Module After bandpass filtering of EEG signals from 0.5 to 45 Hz, the relative power of each frequency band was calculated (frequency domain analysis was performed using Fast Fourier Transform, with a window length of 2 s and an overlap rate of 50%). HbO2 and Hb concentration changes in fNIRS signals were calculated using a modified Beer-Lambert law (the correction factor was calibrated based on the differences in optical properties of different tissue types). When slow wave diffusion (>2 s duration) and HbO2 slope decay (>0.1 μM / s) simultaneously occur in the same brain region, it is marked as a candidate event for hypoperfusion precursors. The spatiotemporal correlation of multimodal signals was determined using the following method: the spatial coordinates of each EEG electrode channel were spatially registered with the sensitive regions of each fNIRS channel. When the spatial distance between the brain region containing the abnormal EEG channel and the brain region containing the abnormal fNIRS channel is less than a preset threshold (≤2 cm), spatiotemporal correlation was considered established. Example 4: Implementation of the TCM Syndrome Quantification Module The knowledge base contains a structured data model trained on 2000 clinical cases (the model uses a random forest algorithm with 500 trees, a maximum depth of 10, and Gini coefficient as the feature splitting criterion; training employs five-fold cross-validation for model selection and hyperparameter tuning). Each case is independently labeled by at least two associate chief physicians or above, with Kappa consistency ≥ 0.75. The syndrome element digitization unit, based on the trained model, maps the input objective physical signs data (four diagnostic methods, laboratory test indicators, etc.) to matching scores (0-1 range) for four syndrome types. The risk weight vector for each syndrome type is determined by combining the feature importance score of the random forest with weighting by clinical experts. A weighted mapping is established between the risk weight vector and carotid IMT (threshold ≥ 0.9 mm), 24-hour systolic blood pressure variability (threshold ≥ 15 mmHg), and CHA2DS2-VASc score (threshold ≥ 2 points) (weight coefficients are determined through multivariate logistic regression analysis, with clinically diagnosed stroke events as the dependent variable and each indicator as an independent variable). Example 5: Implementation of the Three-Party Mutual Verification Noise Reduction Algorithm The system is configured with a three-level early warning logic: Level 1 (Note): If a single dimension exceeds the threshold, it is recorded as a trend deviation and no warning is triggered; Level 2 (Warning): If any two dimensions exceed the threshold simultaneously and a spatiotemporal correlation is established, a "review suggestion" will be sent. Level 3 (High Risk): If all three dimensions simultaneously meet the causal consistency condition and continue to exceed the preset time threshold, a message will be sent saying "High risk in the very early stage, it is recommended to seek medical attention immediately". The construction and reasoning process of the causal consistency map is as follows: (1) Atlas Structure: Using the ocular microcirculation abnormality index (X1), cerebral perfusion decline index (X2), and stroke syndrome matching index (X3) as observation nodes, and "early-stage high-risk state of stroke" (H) as a binary latent variable node (H=1 represents a high-risk state, H=0 represents a non-high-risk state), a Bayesian network with 4 nodes was constructed. The network structure is: H→X1, H→X2, H→X3, that is, given a high-risk state, the three observation indicators are conditionally independent. (2) Parameter learning: Based on historical clinical case data (including gold standard outcome annotation), the conditional probability table for each node was calculated using maximum likelihood estimation. Specifically, this includes: the prior probability of P(H), and the conditional probability distributions of P(X1|H), P(X2|H), and P(X3|H). Among them, the probability distribution parameters of each observation node under the conditions of H=1 and H=0 were obtained through frequency statistics of the training data. (3) Reasoning and Decision-Making: In real-time monitoring, after obtaining the specific values ​​of the three observation indicators, they are discretized into binary states of "abnormal / normal" (the abnormal threshold is set according to the independent verification results of each module) and input into a Bayesian network. The posterior probability P(H=1|X1,X2,X3) is calculated through a variable elimination algorithm. When P(H=1|X1,X2,X3) exceeds the preset decision threshold (initially set at 0.85, which can be adjusted according to clinical verification results), a level 3 high-risk warning is triggered. (4) Atlas update: As clinical data accumulates, the conditional probability table is re-estimated every quarter to keep the model adaptable to changes in population characteristics. Example 6: Implementation of Individual Baseline Model The system establishes a 30-day rolling baseline window for each user. (The baseline model is calculated as follows: for each brain perfusion parameter, the mean μ and standard deviation σ of the data within the window are calculated; during real-time monitoring, when the current value x of a parameter satisfies |x-μ|>2σ, it is determined to deviate from the baseline. To reduce the impact of outliers on baseline estimation, the exponentially weighted moving average (EWMA) algorithm is used to smooth the data within the window, with a smoothing coefficient λ=0.2). The baseline model includes: the mean and standard deviation of brain perfusion parameters, the individual physical baseline of the eye diagnosis risk index, and the seasonal offset of the syndrome matching degree. When a real-time parameter deviates from the baseline by more than 2σ and triggers three-way mutual verification, it is determined to be an individualized early warning event. Example 7: System Performance Verification Scheme The early warning performance of this system needs to be validated through a prospective, multi-center whole-system clinical trial. The clinical trial design is as follows: (1) Sample size: no less than 1,200 subjects, covering people of different age groups (40-80 years old) and different stroke risk levels. (2) Gold standard: The gold standard for diagnosis is cranial MRI (to assess brain perfusion status and new infarcts), DSA (to assess cerebral vascular stenosis and collateral circulation status) and 90-day clinical follow-up (to assess the occurrence of stroke events). (3) Validation indicators: Calculate the sensitivity, specificity, positive predictive value, negative predictive value and area under the receiver operating characteristic curve (AUC) of the system's early warning. (4) Validation process: After enrollment, subjects were monitored by the system and independently evaluated using the gold standard. The monitoring data and the gold standard results were compared in a blinded manner to evaluate the system's early warning performance. (5) Statistical analysis: The bootstrap method (resampled 1000 times) was used to estimate the 95% confidence interval of each validation index. Attached Figure Description

[0004] Figure 1 This is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the causal consistency graph (Bayesian network) of the present invention; Figure 3 This is a flowchart of the three-level early warning logic of the present invention; Figure 4 This is a schematic diagram of the hierarchical deployment architecture of the present invention; Figure 5 This is a schematic diagram illustrating the deviation determination of the individual brain perfusion baseline model of the present invention.

Claims

1. A multimodal brain-computer interface system for ultra-early stroke warning that integrates eye diagnosis and traditional Chinese medicine syndrome differentiation, characterized in that, include: The one-second eye diagnosis rapid screening module is configured to acquire high-definition images of the eye, extract morphological features of scleral choroidal vessels based on the homologous blood supply model of the ophthalmic artery and the internal carotid artery, and output the risk index of retinal-cerebral small vessel disease. The brain-computer interface multimodal brain perfusion acquisition module is configured to simultaneously acquire EEG microrhythm signals, fNIRS near-infrared brain blood oxygen perfusion parameters, and cerebral blood flow fluctuation data, and identify precursor features of brain tissue hypoperfusion. The TCM stroke prodromal syndrome knowledge base quantitative module is configured to store structured data models of liver wind, phlegm stasis, qi deficiency, and yin deficiency syndromes, and maps syndrome elements into risk weight vectors associated with carotid intima-media thickness, blood pressure variability, and clinical stroke risk scores. The AI ​​multi-source coupled reasoning and early warning module is configured to construct a causal consistency map among the ocular microcirculation abnormality index, the cerebral perfusion decline index, and the stroke syndrome matching index. The causal consistency map is constructed based on a Bayesian network architecture, using the ocular microcirculation abnormality index, the cerebral perfusion decline index, and the stroke syndrome matching index as observation nodes, and "early-stage high-risk state of stroke" as a latent variable node. A directed acyclic graph structure between nodes is established through a structural learning algorithm, and the conditional probability table of each node is calculated using maximum likelihood estimation to achieve causal consistency reasoning based on multi-source evidence. When the three satisfy the preset spatiotemporal correlation and pathological consistency conditions, an early-stage stroke warning result is output. The early warning results are used to indicate the risk of stroke during the ultra-early intervention window before the onset of the disease.

2. The system according to claim 1, characterized in that, The one-second eye diagnosis rapid initial screening module includes: An eye image acquisition unit is configured to acquire high-resolution images of the eye within a preset exposure time. The feature extraction unit uses a deep learning-based semantic segmentation model to automatically segment the scleral choroid region and extract features such as blood vessel density, arteriovenous diameter ratio, and abnormal microvascular morphology. The homology mapping unit, based on the anatomical blood supply relationship between the ophthalmic artery and the internal carotid artery, maps the above features into a retinal-cerebral small vessel disease risk index, which serves as the prior weight of the cerebral perfusion decline index.

3. The system according to claim 1, characterized in that, The brain-computer interface multimodal brain perfusion acquisition module includes: The EEG acquisition submodule is configured to extract the α / β band asymmetry index, the θ band burst rate, and the δ band slow wave diffusion coefficient. The fNIRS acquisition submodule is configured to calculate the slope decay of oxyhemoglobin and the residence time of deoxyhemoglobin. The blood flow fluctuation acquisition submodule is configured to acquire cerebral vascular pulsation waveform data. The signal synchronization unit is configured to achieve timestamp alignment and artifact removal of EEG signals, fNIRS signals, and blood flow fluctuation signals, and to mark candidate events for low perfusion precursors within the same brain region and the same time window.

4. The system according to claim 1, characterized in that, The quantitative module of the TCM stroke prodrome knowledge base includes: The digital unit of syndrome elements is configured to map the syndrome elements of liver wind, phlegm stasis, qi deficiency, and yin deficiency into a multidimensional risk weight vector associated with carotid intima-media thickness, homocysteine ​​level, 24-hour systolic blood pressure variability, and CHA2DS2-VASc score or ABCD² score. The dynamic reasoning engine is configured to calculate and output the matching score of each syndrome based on the input objective physical signs data through weighted calculation, which is used to correct the warning threshold of the brain perfusion decline index.

5. The system according to claim 1, characterized in that, The AI ​​multi-source coupled inference early warning module is configured as follows: A causal consistency map was constructed among the ocular microcirculation abnormality index, cerebral perfusion decline index, and stroke syndrome matching index. The map was based on a Bayesian network architecture, with each index as an observation node and "early-stage high-risk state of stroke" as a latent variable. A directed acyclic graph structure between nodes was established through a structural learning algorithm, and a conditional probability table was calculated. When there is an upstream and downstream correlation between abnormal areas of eye diagnosis and areas of decreased cerebral perfusion in the anatomical blood supply chain, and the pathological tendency pointed to by the syndrome matching index is consistent with the time-varying trend of cerebral perfusion decline, and the above correlation continues to exceed a preset time threshold, it is judged as a high-risk state of ultra-early stroke; wherein, the preset time threshold is set to 30 minutes based on clinical pre-experiment data. When data in only one dimension exceeds the threshold, it is determined to be a physiological fluctuation or physical characteristic, and a high-risk warning is not triggered.

6. The system according to claim 1, characterized in that, Also includes: The expandable physiological sensing submodule includes at least one of the following: The carotid artery MEMS blood flow vibration sensing submodule is configured to collect the micro-tremor spectrum of the carotid artery wall and vascular stiffness parameters. The PPG peripheral perfusion and HRV heart rate variability acquisition submodule is configured to acquire peripheral blood flow volume pulse wave and autonomic nerve parameters. The IMU posture and gait imbalance sensing submodule is configured to collect data on minute limb gait deviations and posture stability. The AI ​​multi-source coupled reasoning early warning module is also configured to input the data collected by the sub-module as supplementary features into the coupled operation.

7. The system according to claim 6, characterized in that, The system is configured with a hierarchical deployment architecture: The first configuration version includes the one-second eye diagnosis rapid initial screening module, the brain-computer multimodal brain perfusion acquisition module, the traditional Chinese medicine stroke precursor syndrome knowledge base quantification module, and the AI ​​multi-source coupled reasoning early warning module, which are used for community or home screening scenarios. The second configuration version, based on the first configuration version, integrates all the modules in the expandable physiological sensing submodule, and is used for clinical or emergency review scenarios.

8. The system according to claim 1, characterized in that, The AI ​​multi-source coupled inference early warning module is also configured as follows: Based on historical continuous monitoring data, an individual brain perfusion baseline model was constructed. The baseline model used a 30-day rolling window to calculate the mean and standard deviation of each brain perfusion parameter, and the data within the window was smoothed using an exponentially weighted moving average algorithm. When the deviation of the current brain perfusion parameters from the individual baseline exceeds a preset threshold, and at the same time, abnormal eye diagnosis and syndrome matching are triggered, it is determined to be an individualized stroke ultra-early warning event. The individual baseline model is dynamically updated every 30 days. The update method is as follows: take 30 days of continuous monitoring data backward from the current date, recalculate the mean and standard deviation of each parameter, and replace the old baseline parameters.

9. The system according to claim 1, characterized in that, The warning results output by the system are associated with at least one of the following clinical actions: Initiate a carotid ultrasound or cranial MRI examination; Adjust the anticoagulant or antiplatelet medication regimen; Triggering pre-activation of the emergency green channel.

10. The system according to claim 1, characterized in that, The structured data model in the quantitative module of the TCM stroke prodromal syndrome knowledge base is generated based on historical clinical case data. The historical clinical case data includes syndrome types marked by TCM experts and corresponding objective sign data. The objective sign data is statistically significantly correlated with at least one of carotid intima-media thickness, homocysteine ​​level and blood pressure variability (P<0.05).