A method and system for intelligent assessment of carbon monoxide poisoning encephalopathy data

Abstract

The present application relates to the technical field of medical equipment and artificial intelligence, and particularly relates to a brain disease data intelligent evaluation method and system for carbon monoxide poisoning. The brain disease data intelligent evaluation method for carbon monoxide poisoning comprises the following steps: S1: a dynamic data acquisition framework covering the whole cycle of the disease is designed, and core data at key time points after carbon monoxide poisoning is collected, wherein the core data comprises multi-modal image data, clinical evaluation data and laboratory data; S2: a time series data analysis engine is constructed, multi-modal image data collected is subjected to space-time registration through the time series data analysis engine, batch extraction of key brain region microstructure parameters is performed, and dynamic trajectory data of each parameter is calculated, wherein the dynamic trajectory data comprises a current value, a change rate and an evolution mode of the parameter. Through construction of a dynamic data acquisition and intelligent analysis system covering the whole cycle of the disease, the present application realizes evaluation of the evolution process of brain injury after poisoning.

Description

Technical Field

[0001] This invention relates to the field of medical equipment and artificial intelligence, and in particular to a smart assessment method and system for brain disease data in carbon monoxide poisoning. Background Technology

[0002] Carbon monoxide poisoning is one of the most common accidental poisoning causes in my country, and the resulting neurological damage, especially delayed encephalopathy (DEA), is a significant factor contributing to poor patient prognosis. DEA often develops suddenly after a "false recovery period" following poisoning, resulting in a high rate of disability and a complex pathogenesis. Currently, clinical assessment and early warning of DEA mainly rely on observation of clinical symptoms after poisoning, neuropsychological scale scoring, and routine brain imaging examinations. However, these traditional methods have significant limitations: First, the assessment time point is often singular, mostly limited to the acute phase or the period of obvious symptoms, making it difficult to fully capture the dynamic evolution of brain injury from the acute to the delayed phase; second, scale assessments are greatly influenced by physician experience and patient cooperation, lacking objective and quantitative biological markers; third, routine imaging examinations are not sensitive to early microstructural changes, failing to identify high-risk individuals before the onset of clinical symptoms; furthermore, existing methods are also unable to provide individualized prediction of the patient's recovery trajectory, thus failing to provide precise timing and target guidance for clinical intervention.

[0003] Therefore, there is an urgent need to develop a smart assessment method and system for encephalopathy data in carbon monoxide poisoning to overcome the shortcomings of the aforementioned assessment methods, such as single assessment time points, reliance on subjective experience, lack of early objective warnings, and inability to provide individualized recovery predictions and intervention guidance. While some existing technologies have attempted assessments based on single MRI or CT images, none have achieved the fusion analysis of multi-timepoint and multi-modal data, nor have they established a complete smart assessment closed loop from data acquisition, time-series analysis, dynamic early warning to rehabilitation guidance. These shortcomings fail to meet the urgent clinical need for comprehensive, dynamic, and precise management of encephalopathy (DNS). Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, such as single assessment time points, difficulty in capturing the dynamic evolution of brain injury, over-reliance on doctors' subjective experience, lack of early objective warning indicators, and inability to provide individualized recovery prediction and precise intervention guidance, this invention provides a smart assessment method and system for brain disease data in carbon monoxide poisoning.

[0005] The technical solution is: a smart assessment method for encephalopathy data in carbon monoxide poisoning, comprising the following steps: S1: Design a dynamic data acquisition framework covering the entire disease cycle, and collect core data at key time points after carbon monoxide poisoning. The core data includes multimodal imaging data, clinical assessment data, and laboratory data. S2: Construct a time-series data analysis engine to perform spatiotemporal registration on the collected multimodal image data, batch extraction of key brain region microstructure parameters, and calculation of dynamic trajectory data of each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. S3: Based on the microstructural parameters of the key brain regions and their dynamic trajectory data, a visual, individualized atlas is constructed using an algorithmic model; S4: By analyzing dynamic trajectory data, clinical assessment data, and laboratory data in real time, an individualized risk score is calculated using a built-in algorithm, and intervention recommendations are provided based on the clinical assessment data and evolution patterns; S5: Generate a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. Generate an intervention early warning strategy based on the comprehensive report, and formulate a rehabilitation guidance plan based on the intervention early warning strategy.

[0006] Preferably, the design covers a dynamic data acquisition framework spanning the entire disease cycle, collecting core data at key time points after carbon monoxide poisoning. This core data includes multimodal imaging data, clinical assessment data, and laboratory data, including: key time points encompassing the acute phase, early observation phase, false recovery phase, pre-symptom phase, and recovery tracking phase; a standardized data acquisition scheme is designed for each key time point, where individualized baseline data is established for the acute phase, microstructural changes are detected during the early observation phase, potential deterioration risks are detected at key points during the false recovery phase, the risk of delayed encephalopathy is assessed during the pre-symptom phase, and the recovery progress is quantified during the recovery tracking phase.

[0007] Preferably, the construction of the time-series data analysis engine involves performing spatiotemporal registration on the acquired multimodal image data, batch extraction of key brain region microstructural parameters, and calculation of dynamic trajectory data for each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. This includes: rigid registration of multiple scans of the same patient to eliminate differences in head position and orientation; using a B-spline-based nonlinear registration algorithm to correct morphological changes in brain tissue at different time points caused by disease or scan variations; uniformly registering all time-point multimodal image data and standard brain atlases onto the standard brain atlas space to achieve spatiotemporal registration; performing region mapping and identification on the registered multimodal image data and standard brain atlas to obtain key brain regions; obtaining raw signals based on the key brain regions using diffusion magnetic resonance imaging; obtaining microstructural parameters based on the raw signals through a microstructural model; and calculating the dynamic trajectory data for each parameter based on the microstructural parameters.

[0008] Preferably, the step of obtaining microstructure parameters from the original signal through a microstructure model and calculating dynamic trajectory data of each parameter based on the microstructure parameters includes: obtaining anisotropy fraction and average diffusion rate from the original signal by fitting a diffusion tensor model; separating and quantifying intraneuronal volume fraction and directional dispersion using a multi-compartment model of neurite directional dispersion and density imaging; obtaining the magnetic susceptibility value of iron precipitation through quantitative magnetic susceptibility mapping; calculating the rate of change of each parameter at adjacent time points using the central difference method, and using forward or backward difference for the first and last time points; fitting the time series of each brain region parameter into a trend line through piecewise linear fitting, extracting its slope and curvature as key features, and applying a Bayesian online change point detection algorithm to identify turning points in the time series where the mean and variance change abruptly; combining the extracted key features and identified turning points into a high-dimensional feature vector, and using unsupervised clustering to summarize typical evolution patterns; the evolution patterns include continuous deterioration, early decline followed by plateau, and V-shaped recovery.

[0009] Preferably, the visualization of an individual-specific atlas based on the microstructural parameters and dynamic trajectory data of the key brain regions is constructed using an algorithmic model, including: the visualization of an individual-specific atlas includes an individual evolution trajectory map showing the temporal changes of parameters, a heatmap indicating spatial risk distribution, a recovery potential map predicting long-term outcomes, and an efficacy map comparing intervention effects; wherein the individual evolution trajectory map showing the temporal changes of parameters is obtained through Gaussian process regression and change point detection algorithms, and the data used comes from microstructural parameters at multiple time points; the heatmap indicating spatial risk distribution is obtained through an isolated forest and Gaussian mixture model, based on the dynamic trajectory data; the recovery potential map predicting long-term outcomes is obtained through an LSTM prediction model, based on a sequence of microstructural parameters of the key brain regions arranged in chronological order; the efficacy map of intervention effects is obtained through a difference-in-differences model and a graph neural network, wherein the difference-in-differences model is based on microstructural parameters, and the key brain regions in the graph neural network are nodes, with edges being anisotropy scores and average diffusion rates obtained by fitting a diffusion tensor model.

[0010] Preferably, the step of calculating an individualized risk score using a built-in algorithm by analyzing dynamic trajectory data, clinical assessment data, and laboratory data in real time, and providing intervention recommendations based on the clinical assessment data and evolution patterns, includes: obtaining the individualized risk score using an individualized risk score formula; the intervention recommendations including evidence-based recommendations for optimal intervention timing and prognostic prediction; constructing a causal knowledge base mapping individual evolution patterns, intervention measures, and prognostic outcomes by collecting data from patients diagnosed with carbon monoxide poisoning and their long-term follow-up results; identifying the optimal intervention timing associated with the current disease stage by matching the evolution pattern with the mapping relationship in the causal knowledge base; and generating individualized evidence-based recommendations for prognostic prediction based on clinical assessment data and by comparing the long-term outcomes of similar cases according to the prognostic prediction model.

[0011] Preferably, obtaining the individualized risk score using the individualized risk scoring formula includes: defining a baseline value by collecting microstructural parameters of various brain regions from healthy individuals, and calculating the standard deviation of the corresponding parameters using the baseline value; obtaining the individualized risk score using the individualized risk scoring formula, wherein the individualized risk scoring formula is: ; in For individualized risk scoring; For the Sigmoid function; For the first The current value of the microstructure parameter; For the first The baseline values ​​for the microstructure parameters; For the first The standard deviation of the microstructure parameters; These are normalized clinical assessment data; These are normalized laboratory data; This represents the total number of microstructure parameters obtained through the microstructure model. For the first The weights of the microstructure parameters; This is the adjustment factor for clinical assessment data; This is the adjustment factor for laboratory data.

[0012] Preferably, the step of generating a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations, generating an intervention early warning strategy based on the comprehensive report, and formulating a rehabilitation guidance plan based on the intervention early warning strategy includes: the intervention early warning strategy is a three-tiered intervention early warning strategy system based on risk grading, pattern recognition, and dynamic feedback; triggering three levels of early warning based on the individualized risk score: when the individualized risk score is higher than the high-risk threshold, a strengthened comprehensive intervention early warning is triggered; when the individualized risk score is higher than the primary risk threshold but lower than the high-risk threshold, a standard detection and prevention early warning is triggered; when the individualized risk score is lower than the primary risk threshold, an educational follow-up early warning is triggered; accurately recommending intervention timing and indicating brain region targets based on the dynamic trajectory data, visualized individual-specific maps, and intervention recommendations; defining the early warning validity period; dynamically adjusting the three levels of early warning and rehabilitation intervention plan based on the review results, and feeding the outcome data back to the causal knowledge base; and implementing the rehabilitation intervention plan based on the generated intervention early warning strategy.

[0013] Preferably, the implementation of the rehabilitation intervention plan based on the generated intervention early warning strategy includes: initiating a high-intensity integrated rehabilitation plan, an active prevention and function maintenance plan, and a personalized health education plan respectively according to the three-level early warning provided by the intervention early warning strategy; setting corresponding start time limits and core objectives for different plans; initiating a re-evaluation after the start time limit is reached; updating and generating relevant data and generating re-evaluation results; defining the suggested brain region target points as rehabilitation target points; quantifying and recommending training intensity and frequency; and dynamically monitoring the rehabilitation response through the efficacy map of the intervention effect.

[0014] Preferably, the intelligent assessment system for encephalopathy data in carbon monoxide poisoning further includes: The multi-source data acquisition module is designed with a dynamic data acquisition framework covering the entire disease cycle. It collects core data at key time points after carbon monoxide poisoning, including multimodal imaging data, clinical assessment data, and laboratory data. The time-series data processing module constructs a time-series data analysis engine, which performs spatiotemporal registration on the acquired multimodal image data, extracts key brain region microstructure parameters in batches, and calculates the dynamic trajectory data of each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. The visualization map construction module constructs a visualized, individual-specific map based on the microstructural parameters and dynamic trajectory data of the key brain regions using an algorithmic model. The risk assessment and decision-making module analyzes dynamic trajectory data, clinical assessment data, and laboratory data in real time, uses built-in algorithms to calculate individualized risk scores, and provides intervention recommendations based on the clinical assessment data and evolution patterns. The early warning management module generates a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. It then generates an intervention early warning strategy based on the comprehensive report and formulates a rehabilitation guidance plan according to the intervention early warning strategy.

[0015] This invention offers the following advantages: By constructing a dynamic data acquisition and intelligent analysis system covering the entire disease cycle and integrating multi-temporal, multi-modal data, it achieves precise and dynamic assessment of the evolution of brain injury after carbon monoxide poisoning. Compared to existing technologies, it has significant advantages: it can significantly advance the warning time, identifying high-risk patients 2-3 weeks before the onset of clinical symptoms, providing a critical time window for early intervention; through multi-temporal data analysis, it effectively reduces the random error of a single assessment, significantly improving the objectivity and accuracy of the assessment; it establishes a personalized evolution model and exclusive atlas for each patient, avoiding the generalization bias of group data and achieving truly individualized assessment; the quantitative risk score, visual atlas, and specific intervention suggestions output by the system directly support clinical decision-making, improving clinical practicality and operability; it optimizes the allocation of medical resources based on precise risk grading, avoiding over-intervention for low-risk patients; and, relying on early evolution models to predict long-term recovery trends, it helps to formulate scientific individualized rehabilitation plans, thereby comprehensively improving the early warning, dynamic monitoring, precise intervention, and prognostic assessment capabilities for encephalopathy after carbon monoxide poisoning. Attached Figure Description

[0016] Figure 1 A flowchart of a smart assessment method for brain disease data in carbon monoxide poisoning; Figure 2 This is a schematic diagram of a smart assessment system for brain disease data related to carbon monoxide poisoning. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Example 1: A smart assessment method for encephalopathy data in carbon monoxide poisoning, such as... Figure 1 As shown, it includes the following steps: S1: Design a dynamic data acquisition framework covering the entire disease cycle, and collect core data at key time points after carbon monoxide poisoning. The core data includes multimodal imaging data, clinical assessment data, and laboratory data. The key time points include the acute phase, early observation phase, false recovery phase, pre-symptom phase, and recovery tracking phase. A standardized data collection scheme is designed for the key time points, including establishing individualized baseline data for the acute phase, detecting microstructural changes for the early observation phase, detecting potential deterioration risk for the false recovery phase, assessing the risk of delayed encephalopathy for the pre-symptom phase, and quantifying the recovery process for the recovery tracking phase.

[0019] It should also be noted that, for the acute phase, a standardized data collection plan was designed to establish individualized baseline data; for the early observation phase, multimodal imaging analysis was used to detect changes in brain region microstructure; for key points in the false recovery phase, a time-series data analysis engine was used to detect potential deterioration risks; for the pre-symptom phase, the risk of delayed encephalopathy was assessed by combining clinical and imaging characteristics; and for the recovery tracking phase, the progress of neurological function recovery was quantified based on dynamic trajectory data.

[0020] The multimodal imaging data includes ultra-high resolution diffusion tensor imaging, neurite orientation dispersion and density imaging, and quantitative magnetic susceptibility mapping data; the clinical assessment data includes standardized neuropsychological scales and gait analysis data; and the laboratory data includes blood biochemical indicators and inflammatory markers.

[0021] S2: Construct a time-series data analysis engine to perform spatiotemporal registration on the collected multimodal image data, batch extraction of key brain region microstructure parameters, and calculation of dynamic trajectory data of each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. Rigid registration was performed on multiple scans of the same patient to eliminate differences in head position and orientation. A nonlinear registration algorithm based on B-splines was used to correct morphological changes in brain tissue at different time points caused by disease or scan variations. All multimodal image data and standard brain atlases at all time points were uniformly registered onto the standard brain atlas space to achieve spatiotemporal registration. Region mapping and identification were performed on the registered multimodal image data and standard brain atlas to obtain key brain regions. Based on the key brain regions, diffusion magnetic resonance imaging was used to obtain raw signals. Microstructural parameters were obtained through microstructural modeling based on the raw signals. Dynamic trajectory data of each parameter were calculated based on the microstructural parameters.

[0022] Based on the original signal, anisotropy fraction and average diffusion rate are obtained by fitting a diffusion tensor model. The intraneuronal volume fraction and directional dispersion are obtained by using a multi-compartment model of neurite directional dispersion and density imaging. The magnetic susceptibility of iron deposits is obtained through quantitative magnetic susceptibility mapping. The rate of change of each parameter at adjacent time points is calculated using the central difference method, with forward or backward difference used for the first and last time points. The time series of parameters from each brain region are fitted into trend lines through piecewise linear fitting, and their slope and curvature are extracted as key features. A Bayesian online change point detection algorithm is applied to identify turning points in the time series where the mean and variance abruptly change. The extracted key features and identified turning points are combined into a high-dimensional feature vector, and unsupervised clustering is used to summarize typical evolution patterns. These evolution patterns include continuous deterioration, early decline followed by a plateau, and V-shaped recovery.

[0023] It should also be noted that, to analyze brain scan data of the same patient acquired at different time points, rigorous image alignment processing is required first. The initial rigid registration step aims to correct for overall displacement and rotation caused by differences in head position and angle, laying the spatial foundation for subsequent observations. Subsequently, a nonlinear fine registration is performed using a B-spline-based free deformation model to compensate for local brain tissue deformation caused by disease progression or subtle changes in scanning conditions. Through this series of registration operations, multimodal image data from all time points are uniformly mapped to a standard brain atlas space, thereby achieving accurate spatial normalization across time series.

[0024] The anisotropy fraction and average diffusivity were calculated by processing the original diffusion magnetic resonance signal using a fitted diffusion tensor model to extract the diffusion characteristics of water molecules in brain tissue, and the average diffusivity was calculated based on the tensor eigenvalues. , ,in Tensor eigenvalues ​​representing the three principal axes; anisotropy fractions are also calculated. , ,right The calculation is based on tensor eigenvalues ​​and the calculated values. The intraneuronal volume fraction and directional dispersion are obtained by using a multi-compartment model of neuronal directional dispersion and density imaging to separate the diffuse signal into multiple components, quantify the intraneuronal volume fraction to assess axonal density and integrity, and describe the spatial arrangement consistency of neurites through the directional dispersion parameter; the magnetic susceptibility value is obtained by processing phase magnetic resonance data using quantitative magnetic susceptibility mapping technology to reconstruct the tissue magnetic susceptibility distribution map, correct background field interference through post-processing algorithms and extract brain region magnetic susceptibility values. The magnetic susceptibility value is positively correlated with iron deposition concentration and is used to assess abnormal iron loading in the brain.

[0025] Based on the acquired temporal microstructural parameters, the central difference method was used to calculate the rate of parameter change between adjacent time points, and forward or backward differencing was performed at the beginning and end of the sequence to maintain data integrity. By performing piecewise linear fitting on the time series parameters of each brain region, key geometric features characterizing their dynamic evolution trends, such as the slope and curvature of each segment, were extracted. Furthermore, a Bayesian online change point detection algorithm was used to automatically identify critical time points in the time series where significant changes in mean or variance occurred. Finally, dynamic trajectory features of each brain region's parameters were extracted based on multi-time-point microstructural parameters, and combined to form a high-dimensional feature vector representing the individual evolution process. Hierarchical clustering was used to divide the feature vectors of all patients into groups, and the optimal number of clusters was determined by the silhouette coefficient or elbow rule. The original trajectory curves of patients within each cluster were visualized and overlaid to extract common trends. Semantic annotation of the clusters was performed in conjunction with clinical outcomes, ultimately summarizing them into typical evolution patterns: continuously deteriorating, early decline followed by a plateau, and V-shaped recovery.

[0026] S3: Based on the microstructural parameters of the key brain regions and their dynamic trajectory data, a visual, individualized atlas is constructed using an algorithmic model; The visualized individual-specific atlas includes an individual evolution trajectory map showing the temporal changes of parameters, a heatmap indicating spatial risk distribution, a recovery potential map predicting long-term outcomes, and an efficacy map comparing intervention effects. The individual evolution trajectory map showing the temporal changes of parameters is obtained through Gaussian process regression and change point detection algorithms, using data derived from microstructural parameters at multiple time points. The heatmap indicating spatial risk distribution is obtained through an isolated forest and Gaussian mixture model, based on the dynamic trajectory data. The recovery potential map predicting long-term outcomes is obtained through an LSTM prediction model, based on a time-ordered sequence of microstructural parameters from key brain regions. The efficacy map of intervention effects is obtained through a difference-in-differences model and a graph neural network. The difference-in-differences model is based on microstructural parameters, and the graph neural network uses key brain regions as nodes, with edges representing anisotropy scores and average diffusion rates obtained by fitting a diffusion tensor model.

[0027] It should also be noted that the individual evolution trajectory map is input as a sequence of key brain region microstructural parameters at multiple time points. Gaussian process regression is used to model the continuous trend and uncertainty of these parameters over time, and a change point detection algorithm is superimposed to automatically identify significant turning points in the trajectory. The final output is a curve with time as the x-axis and parameter values ​​as the y-axis, with change points and confidence intervals marked. The heatmap indicating spatial risk distribution is based on dynamic trajectory data of each brain region. The isolated forest algorithm is used to detect high-risk brain regions with abnormal evolution characteristics. A Gaussian mixture model is then used to perform probabilistic clustering and grading of the whole-brain risk level, ultimately mapping the risk levels to a standard brain atlas space to generate a risk heatmap colored by brain region. The recovery potential map is... The brain region microstructural parameter sequences of patients arranged chronologically are input into an LSTM prediction model. The model is trained to learn the mapping relationship between parameter evolution and long-term outcomes. Based on existing time-series data, the model extrapolates and predicts parameter values ​​and recovery trajectories at future time points, outputting a hierarchical map of recovery potential in probabilistic form, intuitively presenting the probability of achieving a good outcome. The efficacy map uses a difference-in-differences model to quantify the real effect of the intervention by comparing the differences in microstructural parameters between the intervention group and the control group before and after the intervention. At the same time, a graph neural network is constructed with key brain regions as nodes and inter-brain structural connections as edges to analyze changes in brain network properties after the intervention. Finally, the results of both are integrated to generate a multidimensional efficacy atlas that shows the global efficacy and brain region-specific responses.

[0028] S4: By analyzing dynamic trajectory data, clinical assessment data, and laboratory data in real time, an individualized risk score is calculated using a built-in algorithm, and intervention recommendations are provided based on the clinical assessment data and evolution patterns; The individualized risk score is obtained through an individualized risk scoring formula; the intervention recommendations include evidence-based recommendations for optimal intervention timing and prognostic prediction; a causal knowledge base is constructed by collecting data from patients diagnosed with carbon monoxide poisoning and their long-term follow-up results, mapping the individual evolution pattern, intervention measures, and prognostic outcomes; by matching the evolution pattern with the mapping relationship in the causal knowledge base, the optimal intervention timing associated with the current disease stage is identified; and based on the prognostic prediction model and clinical assessment data, individualized evidence-based recommendations for prognostic prediction are generated by comparing the long-term outcomes of similar cases.

[0029] A baseline value is defined by collecting microstructural parameters of various brain regions from healthy individuals, and the standard deviation of the corresponding parameter is calculated using the baseline value. The individualized risk score is obtained using an individualized risk scoring formula, wherein the individualized risk scoring formula is: ; in For individualized risk scoring; For the Sigmoid function; For the first The current value of the microstructure parameter; For the first The baseline values ​​for the microstructure parameters; For the first The standard deviation of the microstructure parameters; These are normalized clinical assessment data; These are normalized laboratory data; This represents the total number of microstructure parameters obtained through the microstructure model. For the first The weights of the microstructure parameters; This is the adjustment factor for clinical assessment data; This is the adjustment factor for laboratory data.

[0030] It should also be noted that by recruiting a large-scale cohort of age- and sex-matched healthy individuals, and using the same equipment and protocols as patients to collect multimodal imaging data, the distribution of microstructural parameters in each key brain region was analyzed. The mean or median of the parameter in that brain region for the healthy population was defined as the biological baseline value for that parameter. Based on the microstructural parameter data of the healthy population cohort in each brain region, the sample standard deviation of each parameter in a specific brain region was calculated. The calculated standard deviation was used as the normalization scaling factor in the subsequent individualized risk scoring formula. And store it in the standard parameter library.

[0031] In the individualized scoring formula middle For the Sigmoid function; via The calculation reflects the relative degree of deviation of the patient's brain microstructure from the norm; the clinical assessment data and laboratory data in the formula have been Z-score standardized and Min-Max normalized before being used in the calculation; the weights of the microstructure parameters are based on a large amount of follow-up data of carbon monoxide poisoning patients in a historical database, with the dynamic trajectory characteristics of microstructure parameters in each brain region as independent variables and the occurrence of delayed encephalopathy or neurofunctional outcome as dependent variables. A multivariate Cox regression algorithm is used to calculate the contribution of each parameter to the outcome, and the contribution is normalized to serve as the initial weight of the corresponding parameter; the adjustment coefficient of the clinical assessment data is obtained by collecting neuropsychological data synchronously with the microstructure parameters. Principal component analysis was used to normalize multidimensional clinical indicators into a comprehensive index using scale scores and gait analysis data. Using training set data, the model's discriminative power was maximized through Bayesian optimization, with the goodness of fit between individualized risk scores and actual outcomes as the optimization objective. This determined the relative contribution of clinical assessment data to the risk score. The laboratory data adjustment coefficient was obtained by extracting and normalizing patients' concurrent blood biochemical and inflammatory marker data to form a comprehensive laboratory index. Based on a historical cohort, elastic network regression was used to screen laboratory indicators significantly associated with brain injury evolution, and the synergistic effect between these indicators and microstructural deviation was fitted. The regression coefficients were then scaled and transformed to determine the adjustment coefficients.

[0032] S5: Generate a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. Generate an intervention early warning strategy based on the comprehensive report, and formulate a rehabilitation guidance plan based on the intervention early warning strategy.

[0033] The intervention and early warning strategy is a three-tiered system based on risk grading, pattern recognition, and dynamic feedback. It triggers three levels of early warning based on the individualized risk score: when the individualized risk score is higher than the advanced risk threshold, a comprehensive intervention early warning is triggered; when the individualized risk score is higher than the primary risk threshold but lower than the advanced risk threshold, a standard detection and prevention early warning is triggered; and when the individualized risk score is lower than the primary risk threshold, an educational follow-up early warning is triggered. Based on the dynamic trajectory data, visualized individualized atlas recommendations, and intervention suggestions, the timing of intervention is precisely recommended, and brain region targets are indicated. The validity period of the early warning is defined, and the three levels of early warning and rehabilitation intervention plan are dynamically adjusted based on the review results. Outcome data is fed back to the causal knowledge base. The rehabilitation intervention plan is implemented based on the generated intervention and early warning strategy.

[0034] Based on the three-level early warning provided by the intervention and early warning strategy, high-intensity integrated rehabilitation program, proactive prevention and function maintenance program, and personalized health education program are initiated respectively. For each program, corresponding start time limits and core objectives are set. After the start time limit is reached, a re-evaluation is initiated, relevant data is updated and generated, and re-evaluation results are generated. The brain regions that are indicated are defined as rehabilitation targets, and the recommended training intensity and frequency are quantified and recommended. The rehabilitation response is dynamically monitored through the efficacy map of the intervention effect.

[0035] It should also be noted that the aforementioned three-tiered intervention and early warning strategy system based on risk grading, pattern recognition, and dynamic feedback uses individualized risk scoring. As the core decision variable, when When the advanced risk threshold is exceeded, a standard detection and prevention warning is triggered. The primary goal at this stage is to protect neurological function in the acute phase. A high-intensity integrated program, including multimodal rehabilitation training, hyperbaric oxygen therapy, and neuromodulation, is initiated. The intervention window is strictly limited to the critical plasticity period after injury. The advanced risk threshold is determined by collecting acute risk scores from patients diagnosed with delayed encephalopathy or with severe residual neurological dysfunction. The risk function inflection point identification algorithm in survival analysis is used to automatically divide the natural boundary between medium-risk and high-risk groups. The boundary value is determined by maximizing the Youden index to ensure that the sensitivity and specificity of identifying high-risk groups are not less than 80%.

[0036] When the individualized risk score is higher than the primary risk threshold but lower than the advanced risk threshold, a standard detection and prevention warning is triggered. This stage is guided by active monitoring and functional maintenance, and a standardized intervention package including cognitive training, exercise therapy, and metabolic regulation is implemented, with the focus on blocking the continued deterioration of the subclinical state. The primary risk threshold is based on a patient cohort in the historical database who has not developed delayed encephalopathy or has a good long-term prognosis. The individualized risk score at the time of the first assessment is extracted, and the initial threshold is set using the percentile method. By comparing the actual outcome rates of the low-risk group and the intermediate-risk group, the threshold is calibrated to a clinically acceptable balance between the false positive rate and the missed diagnosis rate using ROC curves.

[0037] When the individualized risk score falls below the primary risk threshold, an educational follow-up warning is triggered. At this stage, the intervention shifts from clinical treatment to behavioral guidance, relying on personalized health education modules to enhance patients' self-symptom monitoring and life management abilities, thereby reducing the risk of long-term relapse.

[0038] The complete data chain generated by each early warning and intervention event—from risk scores, evolution patterns, intervention parameters to short-term functional outcomes and long-term prognoses—is structured, stored, and incorporated into a causal knowledge base. This continuously enriches and improves the mapping relationship between individual evolution patterns, intervention measures, and prognostic outcomes, providing more precise early warning and intervention services for subsequent patients.

[0039] Example 2: Based on Example 1, a smart assessment system for encephalopathy data in carbon monoxide poisoning, such as... Figure 2 As shown, it also includes: The multi-source data acquisition module is designed with a dynamic data acquisition framework covering the entire disease cycle. It collects core data at key time points after carbon monoxide poisoning, including multimodal imaging data, clinical assessment data, and laboratory data. The time-series data processing module constructs a time-series data analysis engine, which performs spatiotemporal registration on the acquired multimodal image data, extracts key brain region microstructure parameters in batches, and calculates the dynamic trajectory data of each parameter, which includes the rate of change and evolution pattern. The visualization map construction module constructs a visualized, individual-specific map based on the microstructural parameters and dynamic trajectory data of the key brain regions using an algorithmic model. The risk assessment and decision-making module analyzes dynamic trajectory data, clinical assessment data, and laboratory data in real time, uses built-in algorithms to calculate individualized risk scores, and provides intervention recommendations based on the clinical assessment data and evolution patterns. The early warning management module generates a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. It then generates an intervention early warning strategy based on the comprehensive report and formulates a rehabilitation guidance plan according to the intervention early warning strategy.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart assessment method for encephalopathy data in carbon monoxide poisoning, characterized in that, Includes the following steps: S1: Design a dynamic data acquisition framework covering the entire disease cycle, and collect core data at key time points after carbon monoxide poisoning. The core data includes multimodal imaging data, clinical assessment data, and laboratory data. S2: Construct a time-series data analysis engine to perform spatiotemporal registration on the collected multimodal image data, batch extraction of key brain region microstructure parameters, and calculation of dynamic trajectory data of each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. S3: Based on the microstructural parameters of the key brain regions and their dynamic trajectory data, a visual, individualized atlas is constructed using an algorithmic model; S4: By analyzing dynamic trajectory data, clinical assessment data, and laboratory data in real time, an individualized risk score is calculated using a built-in algorithm, and intervention recommendations are provided based on the clinical assessment data and evolution patterns; S5: Generate a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. Generate an intervention early warning strategy based on the comprehensive report, and formulate a rehabilitation guidance plan based on the intervention early warning strategy.

2. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 1, characterized in that, The design encompasses a dynamic data acquisition framework covering the entire disease cycle, collecting core data at key time points after carbon monoxide poisoning. This core data includes multimodal imaging data, clinical assessment data, and laboratory data. The key time points include the acute phase, early observation phase, false recovery phase, pre-symptom phase, and recovery tracking phase. A standardized data acquisition scheme is designed for each key time point, including establishing individualized baseline data for the acute phase, detecting microstructural changes during the early observation phase, detecting potential deterioration risks during the false recovery phase, assessing the risk of delayed encephalopathy during the pre-symptom phase, and quantifying the recovery progress during the recovery tracking phase.

3. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 1, characterized in that, The construction of the time-series data analysis engine involves performing spatiotemporal registration on the acquired multimodal image data, batch extraction of key brain region microstructural parameters, and calculation of dynamic trajectory data for each parameter. This dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameters. The process includes: rigid registration of multiple scans of the same patient to eliminate differences in head position and orientation; using a B-spline-based nonlinear registration algorithm to correct morphological changes in brain tissue at different time points caused by disease or scan variations; uniformly registering all time-point multimodal image data and standard brain atlases onto the standard brain atlas space to achieve spatiotemporal registration; performing region mapping and identification on the registered multimodal image data and standard brain atlas to obtain key brain regions; obtaining raw signals based on the key brain regions using diffusion magnetic resonance imaging; obtaining microstructural parameters based on the raw signals through a microstructural model; and calculating the dynamic trajectory data for each parameter based on the microstructural parameters.

4. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 3, characterized in that, The process of obtaining microstructural parameters from the original signal using a microstructural model and calculating dynamic trajectory data for each parameter based on these parameters includes: obtaining anisotropy fraction and average diffusion rate by fitting a diffusion tensor model to the original signal; separating and quantifying intraneuronal volume fraction and directional dispersion using a multi-compartment model of neurite directional dispersion and density imaging; obtaining the magnetic susceptibility value of iron precipitate through quantitative magnetic susceptibility mapping; calculating the rate of change of each parameter at adjacent time points using the central difference method, employing forward or backward difference for the first and last time points; fitting the time series of each brain region parameter into a trend line through piecewise linear fitting, extracting its slope and curvature as key features; identifying turning points in the time series where the mean and variance abruptly change; combining the extracted key features and identified turning points into a high-dimensional feature vector; and using unsupervised clustering to summarize typical evolution patterns; the evolution patterns include continuous deterioration, early decline followed by a plateau, and V-shaped recovery.

5. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 1, characterized in that, The visualization of an individual-specific atlas based on the microstructural parameters and dynamic trajectory data of the key brain regions is constructed using an algorithmic model. This includes: an individual evolution trajectory map showing temporal changes in parameters, a heatmap indicating spatial risk distribution, a recovery potential map predicting long-term outcomes, and an efficacy map comparing intervention effects. The individual evolution trajectory map showing temporal changes in parameters is obtained using Gaussian process regression and change point detection algorithms, with data derived from microstructural parameters at multiple time points. The heatmap indicating spatial risk distribution is obtained using an isolated forest and Gaussian mixture model, based on the dynamic trajectory data. The recovery potential map predicting long-term outcomes is obtained using an LSTM prediction model, based on a chronological sequence of microstructural parameters from the key brain regions. The efficacy map of intervention effects is obtained using a difference-in-differences model and a graph neural network. The difference-in-differences model is based on microstructural parameters, and the graph neural network uses key brain regions as nodes, with edges representing anisotropy scores and average diffusion rates obtained by fitting a diffusion tensor model.

6. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 1, characterized in that, The process involves real-time analysis of dynamic trajectory data, clinical assessment data, and laboratory data, using a built-in algorithm to calculate an individualized risk score, and providing intervention recommendations based on the clinical assessment data and evolution patterns. This includes: obtaining the individualized risk score using an individualized risk score formula; the intervention recommendations containing evidence-based recommendations for optimal intervention timing and prognostic prediction; constructing a causal knowledge base mapping individual evolution patterns, intervention measures, and prognostic outcomes by collecting data from diagnosed carbon monoxide poisoning patients and their long-term follow-up results; identifying the optimal intervention timing associated with the current disease stage by matching the evolution patterns with the mapping relationships in the causal knowledge base; and generating individualized evidence-based prognostic prediction recommendations based on clinical assessment data and by comparing the long-term outcomes of similar cases using a prognostic prediction model.

7. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 6, characterized in that, The process of obtaining the individualized risk score using a personalized risk scoring formula includes: defining baseline values ​​by collecting microstructural parameters of various brain regions from healthy individuals, and calculating the standard deviation of the corresponding parameters using the baseline values; and obtaining the individualized risk score using a personalized risk scoring formula, wherein the personalized risk scoring formula is: ； in For individualized risk scoring; For the Sigmoid function; For the first The current value of the microstructure parameter; For the first The baseline values ​​for the microstructure parameters; For the first The standard deviation of the microstructure parameters; These are normalized clinical assessment data; These are normalized laboratory data; This represents the total number of microstructure parameters obtained through the microstructure model. For the first Weights of microstructure parameters; This is the adjustment factor for clinical assessment data; This is the adjustment factor for laboratory data.

8. The intelligent assessment method for encephalopathy data in carbon monoxide poisoning according to claim 1, characterized in that, The process involves generating a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. An intervention early warning strategy is then generated based on this comprehensive report, and a rehabilitation guidance plan is formulated according to the intervention early warning strategy. This includes: the intervention early warning strategy is a three-tiered system based on risk grading, pattern recognition, and dynamic feedback; triggering three levels of early warning based on the individualized risk score: when the individualized risk score is higher than the high-risk threshold, a reinforced comprehensive intervention early warning is triggered; when the individualized risk score is higher than the primary risk threshold but lower than the high-risk threshold, a standard detection and prevention early warning is triggered; and when the individualized risk score is lower than the primary risk threshold, an educational follow-up early warning is triggered; precisely recommending intervention timing and suggesting brain region targets based on the dynamic trajectory data, visualized individual-specific maps, and intervention recommendations; defining the early warning validity period; dynamically adjusting the three levels of early warning and rehabilitation intervention plan based on the review results; and feeding the outcome data back to the causal knowledge base; and implementing the rehabilitation intervention plan according to the generated intervention early warning strategy.

9. A method for intelligent assessment of encephalopathy data in carbon monoxide poisoning according to claim 8, characterized in that, The implementation of the rehabilitation intervention plan based on the generated intervention early warning strategy includes: initiating a high-intensity integrated rehabilitation plan, an active prevention and function maintenance plan, and a personalized health education plan respectively based on the three-level early warning provided by the intervention early warning strategy; setting corresponding start time limits and core objectives for different plans; initiating a re-evaluation after the start time limit is reached; updating and generating relevant data and generating re-evaluation results; defining the suggested brain region target points as rehabilitation target points; quantifying and recommending training intensity and frequency; and dynamically monitoring the rehabilitation response through the efficacy map of the intervention effect.

10. A smart assessment system for encephalopathy data in carbon monoxide poisoning, used to implement the smart assessment method for encephalopathy data in carbon monoxide poisoning as described in any one of 1-9, characterized in that, Also includes: The multi-source data acquisition module is designed with a dynamic data acquisition framework covering the entire disease cycle. It collects core data at key time points after carbon monoxide poisoning, including multimodal imaging data, clinical assessment data, and laboratory data. The time-series data processing module constructs a time-series data analysis engine, which performs spatiotemporal registration on the acquired multimodal image data, extracts key brain region microstructure parameters in batches, and calculates the dynamic trajectory data of each parameter. The dynamic trajectory data includes the current value, rate of change, and evolution pattern of the parameter. The visualization map construction module constructs a visualized, individual-specific map based on the microstructural parameters and dynamic trajectory data of the key brain regions using an algorithmic model. The risk assessment and decision-making module analyzes dynamic trajectory data, clinical assessment data, and laboratory data in real time, uses built-in algorithms to calculate individualized risk scores, and provides intervention recommendations based on the clinical assessment data and evolution patterns. The early warning management module generates a comprehensive report based on dynamic trajectory data, visualized individual-specific maps, individualized risk scores, and intervention recommendations. It then generates an intervention early warning strategy based on the comprehensive report and formulates a rehabilitation guidance plan according to the intervention early warning strategy.

Images