Individualized cognitive decline risk prediction method and system based on multi-mode data fusion

Abstract

The invention relates to the technical field of individualized cognition, in particular to an individualized cognitive decline risk prediction method and system based on multi-mode data fusion. According to the method, multi-modal longitudinal evidence-based data is integrated on the basis of a group-level pathology law, a multi-scale causal link map capable of reflecting the complete period of cognitive decline is formed, pathology time axis alignment and individualized feature normalization are achieved, the dynamic pathology process of individual specificity can be captured, and the accuracy of the pathology process is improved. By fitting individual exclusive multi-scale causal parameters and constructing a time sequence cascade digital twinborn body, forward causal simulation can be executed on multiple time scales, a cognitive decline core driving path is directly revealed, and individual risk quantification and intervention targets are generated. Detection of early hidden pathological signals, dynamic modeling of individual heterogeneity and comprehensive analysis of a multi-channel pathological mechanism are achieved, the ultra-early prediction precision of cognitive decline and individualized intervention suitability are remarkably improved, and a full-period prediction system capable of being iteratively optimized continuously is formed.

Description

Technical Field

[0001] This invention relates to the field of personalized cognitive technology, and in particular to a method and system for predicting the risk of personalized cognitive decline based on multi-modal data fusion. Background Technology

[0002] Individualized cognitive decline risk refers to the quantitative assessment of the probability, rate of cognitive decline, and risk of adverse outcomes of a single individual within a specific timeframe by combining multi-dimensional individual-specific information such as genetic background, physiological and pathological characteristics, lifestyle, environmental exposure, and neuropsychological performance. This enables precise stratification and dynamic early warning of an individual's cognitive decline trajectory.

[0003] Currently, existing technologies primarily quantify an individual's baseline multi-cognitive domain function using standardized neuropsychological scales, then incorporate validated clinical risk factors, and employ traditional statistical algorithms to fit the weighting coefficients of each variable on the cognitive decline outcome to construct a standardized risk scoring equation. Finally, these equations are substituted into the individual's measured indicators to calculate their individualized risk probability of developing cognitive decline. However, these methods rely solely on single-modal phenotypic data—neuropsychological scales and clinical risk factors—for prediction, failing to capture the latent pathophysiological changes in the very early stages of cognitive decline. Furthermore, the single-dimensional information is insufficient to cover the individual heterogeneity and multi-pathway pathological mechanisms of cognitive decline, resulting in a lack of predictive efficacy in the very early stages and limited accuracy in individualized predictions. Summary of the Invention

[0004] The main objective of this invention is to provide a personalized cognitive decline risk prediction method based on multi-modal data fusion, aiming to solve the technical problems in the prior art.

[0005] This invention proposes a personalized cognitive decline risk prediction method based on multi-modal data fusion, comprising: We will acquire multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade patterns of cognitive decline throughout the entire cycle, and construct a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph based on the multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data. Multimodal longitudinal monitoring data and clinical risk factor time series data of the target individual are acquired, and pathological time axis alignment and normalization processing are performed on the multimodal longitudinal monitoring data and clinical risk factor time series data according to the time series cascade causal graph framework to obtain the individual standardized multimodal time series feature set; Based on the aforementioned time-series cascaded causal graph framework and individual standardized multimodal time-series feature set, individual-specific multi-scale causal link dynamic parameters are obtained through fitting, and an individual-specific multi-scale time-series cascaded causal digital twin is constructed based on the individual-specific multi-scale causal link dynamic parameters. Multi-timescale forward causal simulation is performed on the time-series cascaded causal digital twin to obtain the forward causal simulation results. Based on the forward causal simulation results, the individualized cognitive decline risk quantification results and the source tracing results of the core driving pathogenic pathways of the target individual are obtained. Based on the individualized cognitive decline risk quantification results, a cognitive decline grading early warning strategy is generated. Based on the source tracing results of the core driving pathogenic pathways, the core pathogenic targets that can be intervened in individuals are located. Based on the core pathogenic targets, an individualized intervention guidance plan matching the individual's pathological characteristics is generated. The system acquires new multimodal time-series monitoring data for target individuals and adaptively iterates and updates the time-series cascaded causal digital twin based on the multimodal time-series monitoring data. It also simultaneously optimizes the individualized cognitive decline risk quantification results and the core driving pathogenic pathway tracing results to update the hierarchical early warning strategy and individualized intervention guidance plan in real time, thereby continuously improving the accuracy of full-cycle risk prediction and the adaptability of intervention.

[0006] This application also provides a personalized cognitive decline risk prediction system based on multi-modal data fusion, including: The first building module is used to obtain multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade pattern of cognitive decline throughout the entire cycle, so as to build a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph. The alignment module is used to acquire multimodal temporal longitudinal monitoring data and clinical risk factor temporal data of the target individual, and perform pathological time axis alignment and normalization processing on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data according to the temporal cascade causal graph basic framework to obtain a standardized multimodal temporal feature set of the individual. The second construction module is used to fit the individual-specific multi-scale causal link dynamic parameters based on the time-series cascaded causal graph basic framework and the individual standardized multimodal time-series feature set to construct an individual-specific multi-scale time-series cascaded causal digital twin. The deduction module is used to perform multi-timescale forward causal simulation deduction on the time-series cascaded causal digital twin to obtain the individualized cognitive decline risk quantification results and core driving pathogenic pathway tracing results for the target individual; The generation module is used to generate a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locate the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generate an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets. The iterative update module is used to acquire newly added multimodal time-series monitoring data of target individuals to adaptively iteratively update the time-series cascaded causal digital twin, and simultaneously optimize the individualized cognitive decline risk quantification results and core driver pathogenesis tracing results to update the graded early warning strategy and individualized intervention guidance plan, continuously improving the accuracy of full-cycle risk prediction and intervention adaptability.

[0007] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for predicting the risk of individualized cognitive decline based on multi-modal data fusion.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for predicting the risk of individualized cognitive decline based on multi-modal data fusion.

[0009] The beneficial effects of this invention are as follows: By constructing a multimodal, multi-scale, time-series cascaded causal digital twin, this invention achieves refined quantification and dynamic prediction of individual cognitive decline risk, overcoming the limitations of existing technologies that rely on single-modal phenotypic data and cannot capture ultra-early hidden pathological changes. By integrating multimodal longitudinal evidence-based data on the basis of population-level pathological patterns, a multi-scale causal link map reflecting the entire cycle of cognitive decline is formed, achieving pathological time axis alignment and individualized feature normalization. It can capture individual-specific dynamic pathological processes. By fitting individual-specific multi-scale causal parameters and constructing a time-series cascaded digital twin, positive causal simulation can be performed at multiple time scales, directly revealing the core driving pathways of cognitive decline, and generating individualized risk quantification and intervention targets. It realizes the detection of early hidden pathological signals, dynamic modeling of individual heterogeneity, and comprehensive analysis of multi-pathway pathological mechanisms, significantly improving the accuracy of ultra-early prediction of cognitive decline and the adaptability of individualized intervention, forming a sustainable iterative optimization full-cycle prediction system, and solving the problem that existing single-dimensional risk scoring methods cannot accurately capture early pathological changes and individual differences. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0015] like Figure 1 As shown, this application provides a personalized cognitive decline risk prediction method based on multi-modal data fusion, including: S1. Obtain multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade pattern of cognitive decline throughout the entire cycle, and construct a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph based on the multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data. S2. Obtain multimodal temporal longitudinal monitoring data and clinical risk factor temporal data of the target individual, and perform pathological time axis alignment and normalization processing on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data according to the temporal cascade causal graph basic framework to obtain the individual standardized multimodal temporal feature set; S3. Based on the time-series cascaded causal graph framework and the individual standardized multimodal time-series feature set, fit the individual-specific multi-scale causal link dynamic parameters, and construct an individual-specific multi-scale time-series cascaded causal digital twin based on the individual-specific multi-scale causal link dynamic parameters. S4. Perform multi-timescale forward causal simulation on the time-series cascaded causal digital twin to obtain the forward causal simulation results, and obtain the individualized cognitive decline risk quantification results and core driving pathogenic pathway tracing results of the target individual based on the forward causal simulation results. S5. Generate a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locate the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generate an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets. S6. Acquire new multimodal time-series monitoring data for the target individual, and adaptively iterate and update the time-series cascaded causal digital twin based on the multimodal time-series monitoring data. Simultaneously optimize the individualized cognitive decline risk quantification results and the core driving pathogenic pathway tracing results to update the hierarchical early warning strategy and individualized intervention guidance plan in real time, and continuously improve the accuracy of full-cycle risk prediction and intervention adaptability.

[0016] As described in steps S1-S6 above, this invention, by introducing multimodal longitudinal follow-up evidence-based data and constructing a causal atlas based on temporal cascade patterns, not only solves the problem of single-dimensional information in single-modal data, but more importantly, establishes a temporal causal logical chain from molecular / cellular pathological changes to the appearance of clinical symptoms at the population level. The constructed basic framework includes multi-scale hierarchical rules for pathological occurrence, development, and outcome, ensuring that the model can identify latent pathological features that have occurred before the appearance of symptoms. By aligning the pathological timeline through the atlas framework, the physiological clocks of different individuals are calibrated to a unified pathological clock, enabling precise capture of the ultra-early stage (i.e., before the presence of pathogens). For individuals whose scores have decreased but pathology has already begun, a specific normalization process maps data with completely different dimensions, such as scale scores, imaging data, and biochemical indicators, to a unified feature space, solving the heterogeneity problem in multimodal data fusion. The digital twin is a dynamic parameterized model based on causal graph fitting, simulating the complete physiological / pathological process from biological pathology to clinical manifestations, realizing the digital mapping of the individual's life system. By fitting individual-specific causal link parameters, the model can capture the unique pathological weight distribution of a specific individual, overcoming the one-size-fits-all defect of traditional statistical models and improving the model's ability to represent individual-specific pathological mechanisms.

[0017] This application utilizes the dynamic characteristics of digital twins through positive causal simulation to extrapolate pathological states at future points in time. This allows for the prediction of future decline risk even before significant abnormalities are observed on neuropsychological scales. Through causal mapping-based extrapolation, the core driving pathways leading to increased risk can be clearly identified. By combining the dual outputs of risk quantification and pathway tracing, pathology-mechanism-oriented intervention strategies can be generated. For example, if the tracing results indicate insulin resistance, a metabolic intervention plan is generated; if it indicates neuroinflammation, an anti-inflammatory plan is generated. Through precise matching based on etiology, this addresses the problems of generalized intervention plans and lack of individualized targeting in existing technologies, improving the effectiveness of interventions. This is achieved by establishing a monitoring, prediction, and... The adaptive closed loop of intervention, feedback, and update continuously refines the parameters of the digital twin using new data. The model can evolve in real time to follow changes in the individual's pathophysiological state. Through an adaptive iteration mechanism, it overcomes the shortcomings of existing technology models that have declining predictive effectiveness over time, ensuring that risk prediction and intervention recommendations are always based on the latest individual status throughout the entire disease cycle. This achieves truly long-term, high-precision individualized management. This invention not only introduces multimodal data to solve the problem of single information dimensions, but also captures the ultra-early hidden pathological changes of cognitive decline through pathological timeline alignment, causal inference, and adaptive update mechanisms, effectively covering the individual's pathological heterogeneity and multi-pathway mechanisms.

[0018] In one embodiment, step S1, which involves acquiring multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade patterns throughout the cognitive decline cycle to construct a basic framework for a population-level multi-modal, multi-scale temporal cascade causal graph, includes: S11. Based on the pathological cascade pattern of the entire cognitive decline cycle, five degradation scales with temporal causal relationships are divided: molecular level, cellular circuit level, brain network structure level, neuropsychological phenotype level, and clinical outcome level. Each degradation scale is matched with corresponding exclusive modal data and temporal resolution to form a corresponding scale modal feature set. S12. Based on the multimodal longitudinal follow-up evidence-based data and the feature set of each scale modality, the continuous time marginal structure model and the temporal Granger causality test are used to obtain the cross-scale temporal causal correlation degree and the pathological transmission time delay coefficient. S13. Construct acyclic causal edges with time delay effects across scales based on the cross-scale temporal causal correlation degree and pathological transmission time delay coefficient, and remove false related causal edges in the acyclic causal edges according to the preset causal confidence threshold to obtain an effective cross-scale causal link set. S14. Based on the multi-scale temporal hierarchical rules and the effective cross-scale causal link set, perform topological binding of multi-scale levels and causal links to construct the basic framework of the population-level multimodal multi-scale temporal cascaded causal graph.

[0019] As described in steps S11-S14 above, this invention explicitly introduces molecular and cellular circuit levels as independent degradation scales and matches them with highly sensitive biomarkers or imaging modalities. Architecturally, this shifts the observation perspective forward to the starting point of pathological development. By matching the corresponding temporal resolution for each degradation scale, a priori constraints for multimodal data fusion are established. Through hierarchical matching based on pathological biological mechanisms, the fusion barrier of multi-source heterogeneous data in the temporal dimension is effectively overcome, ensuring that subsequent models can accurately capture the dynamic characteristics of pathological changes at each scale. This is achieved through continuous temporal boundary connections. The model and temporal Granger causality test not only quantified the strength of the association between variables, but also identified the time delay coefficient of pathological transmission at different scales. It simulated the biological law that molecular pathologies lag by several years before manifesting as cognitive decline in the real world, providing core parameters for building models with time prediction capabilities. By obtaining the pathological transmission time delay coefficient, the time window from the appearance of hidden pathophysiological changes to the appearance of symptoms was quantified at the algorithm level. This enabled the model to infer and predict the occurrence time of future cognitive decline based on current molecular / imaging changes, thereby achieving accurate capture of ultra-early risks.

[0020] It should be noted that by constructing acyclic causal edges with time delay effects, the model is forced to follow the temporal logic of pathological occurrence, i.e., the cause must occur before the effect. Combined with causal confidence thresholds for pruning, the interference of irrelevant variables is effectively eliminated, while the core pathogenic pathways that truly drive cognitive decline are preserved. This significantly improves the specificity and robustness of the population graph framework. By explicitly constructing causal edges, the risk prediction results output by the model have a clear biological path explanation, solving the problem that existing technologies cannot reveal the specific pathogenic mechanisms of individuals. By topologically binding rules, causal correlation, and time delay coefficients, discrete medical knowledge is transformed into a computer-computable mathematical model, providing a benchmark coordinate system for the mapping and alignment of individual data in subsequent steps. Only by establishing a complete population framework that includes multiple pathways and mechanisms can the specific nodes in which a particular individual deviates from the normal path be identified within this framework, thereby accurately locating the heterogeneous pathological characteristics of individuals.

[0021] In one embodiment, step S2, which involves performing pathological time axis alignment and normalization processing on multimodal temporal longitudinal monitoring data and clinical risk factor temporal data based on the temporal cascade causal graph framework to obtain an individual standardized multimodal temporal feature set, includes: S21. Obtain the pathological timeline benchmarks and physiological reference intervals of healthy individuals of the same age based on the timeline cascade causal graph framework, and obtain the corresponding individual modal pathological zero-point determination threshold based on each pathological timeline benchmark and physiological reference interval. S22. Based on the multimodal time-series longitudinal monitoring data and clinical risk factor time-series data, obtain the measured time-series sequences of each modality and the time-series change sequences of risk factors, and locate the pathological zero-point time of the corresponding modality and risk factor based on each measured time-series sequence, time-series change sequence and pathological zero-point determination threshold. S23. Based on each pathological zero point time, perform temporal rearrangement and phase calibration on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data to achieve pathological time axis alignment processing, and obtain the corresponding aligned multimodal temporal dataset; S24. Obtain the normalized baseline coefficients of each scale data according to the multi-scale weighting rules of the time-series cascaded causal graph basic framework, and perform amplitude normalization and outlier removal processing on the corresponding aligned multimodal time series dataset according to each normalized baseline coefficient to obtain the corresponding standard multimodal time series data. S25. Integrate the features of multiple standard multimodal time series data to obtain an individual standardized multimodal time series feature set.

[0022] As described in steps S21-S25 above, this invention utilizes the pathological timeline benchmark and physiological reference interval of the population map to calculate an individualized pathological zero-point determination threshold for each modality of data. This quantifies the critical point between healthy and pathological states, enabling the model to identify individuals in the ultra-early stage where scale scores have not yet declined but pathology has already begun. By introducing the physiological reference interval of healthy individuals of the same age, population knowledge is integrated into individual threshold calculation, ensuring that the threshold setting conforms to biological laws and reflects the individual's specificity in dimensions such as age and genetic background. This avoids false positives or false negatives caused by using a uniform threshold, thus improving pathological detection. By dynamically comparing time-series measured sequences with the zero-point threshold, the model identifies the pathological zero-point moment when each modality's data first deviates from the healthy state, thus establishing a dedicated timeline for each individual with the pathological initiation as the origin. This solves the problem that traditional methods cannot effectively compare data from different individuals due to the asynchronous pathological process. By locating the pathological zero point, the model can clarify the relative order and absolute time point of pathological changes in each modality, providing a key time dimension input for prediction. This enables the model to infer future decline risks based on the time from the current state to the pathological zero point, significantly improving the accuracy of ultra-early prediction.

[0023] This invention achieves strict alignment of different individuals and modalities across different stages of pathological progression by uniformly mapping all modal data to a pathological timeline starting from the pathological zero point. For example, if the pathological zero points of the imaging data of two individuals of different ages both occur in the second year of the aligned pathological timeline, the model can consider them to be at the same stage of pathological development, thus enabling effective comparison and modeling. The alignment of the pathological timeline eliminates temporal misalignment caused by different onset times among individuals, allowing the model to integrate multimodal information at the same pathological stage (such as simultaneously analyzing imaging data and scale scores in the second year of pathology). Through temporally consistent multimodal fusion, the model's ability to analyze complex pathological mechanisms is significantly enhanced. By using multi-scale weighting rules of the population atlas for normalization, not only are numerical range differences eliminated, but more importantly, the values ​​of each modality are mapped to a unified pathological contribution coordinate system. For example, the variation amplitude of imaging data is scaled according to its weight in the causal atlas, aligning it with the scale scores. The comparable impact of each component in the model ensures that the model can focus on key pathological signals. By combining the weighting rules of the population map for outlier removal, this invention can identify and filter out outliers that are significantly inconsistent with the pathological patterns of the population. This is more biologically reasonable than methods that rely solely on statistical thresholds, significantly improving the robustness and generalization ability of the model. By integrating time-aligned and normalized multimodal features to form a comprehensive feature set containing multi-scale information such as molecular, imaging, and behavioral data, the model can simultaneously capture pathological changes at different levels, thereby more comprehensively characterizing the multi-pathway mechanisms of cognitive decline and significantly improving the accuracy of individualized predictions. The standardized time-series feature set not only solves the problems of data heterogeneity and time-series misalignment, but also provides a high-quality data foundation for subsequent construction of digital twins and causal simulations. Through the temporal consistency and pathological relevance of its features, it ensures that subsequent dynamic modeling can accurately simulate the evolution of individual pathology, thereby supporting accurate risk prediction and intervention decisions.

[0024] In one embodiment, step S3, which involves fitting individual-specific multi-scale causal link dynamic parameters based on the temporal cascade causal graph framework and the individual standardized multimodal temporal feature set to construct an individual-specific multi-scale temporal cascade causal digital twin, includes: S31. Obtain the multi-scale causal topology and the initial weight set of the population-level link based on the time-series cascaded causal graph basic framework, and obtain the individual pathological boundary conditions and initial state parameters based on the individual standardized multimodal time-series feature set. S32. A continuous-time causal state-space model with Bayesian recursive updates is used to fit and dynamically update the multi-scale causal topology, the initial weight set of the group-level link, the individual pathological boundary conditions and the initial state parameters link by link to obtain the individual-specific multi-scale causal link dynamic parameters. S33. Based on the individual-specific multi-scale causal link dynamic parameters, the multi-scale causal topology structure is personalized and modified to obtain an individual-specific causal topology framework. Based on the individual-specific causal topology framework and the individual standardized multimodal temporal feature set, an individual-specific multi-scale temporal cascade causal digital twin that can replicate the individual's full-cycle pathological cascade evolution law is constructed.

[0025] As described in steps S31-S33 above, after the step of constructing an individual-specific multi-scale time-series cascaded causal digital twin, the method further includes obtaining historical longitudinal follow-up outcome data of the target individual and using counterfactual causal inference to verify the causal validity of the individual-specific multi-scale time-series cascaded causal digital twin; blocking a single upstream causal link in the digital twin through counterfactual simulation and obtaining the simulated change value of the corresponding downstream phenotype based on the simulation results; comparing the simulated change value of the downstream phenotype with the actual change value of the historical longitudinal follow-up outcome data of the target individual, and if the deviation between the two exceeds a preset threshold, then the false causal link is eliminated, and the causal topology of the digital twin is optimized.

[0026] This invention utilizes a multi-scale causal topology and a population-level initial weight set to treat population-level causal association patterns as prior knowledge for modeling. Simultaneously, it combines standardized temporal characteristics of individuals to determine their pathological boundary conditions and initial state parameters. This approach leverages the statistical regularities of population data while preserving individual differences, significantly enhancing the model's ability to characterize complex pathological mechanisms. It addresses the problems of poor generalization and insensitivity to rare variations in existing technologies due to a lack of population knowledge. The initial state parameters determined by individual temporal characteristics provide an accurate starting point for subsequent iterative optimization of the dynamic causal model. Pathological boundary conditions limit the model's solution space, preventing prediction failures caused by parameter overfitting or divergence. This ensures the model can capture the current pathological state of individuals and reasonably infer future evolutionary trends based on population topology, achieving continuous time... The causal state-space model models pathological evolution as a dynamic system that changes continuously over time. Both state variables and causal parameters can be updated over time. Furthermore, by combining Bayesian recursive updates, the model can correct the posterior distribution of parameters in real time based on new observation data, achieving a closed-loop iteration of prediction, observation, and correction. This dynamic modeling capability enables the model to accurately capture instantaneous changes and turning points in the pathological process, thus solving the problems of prediction lag and insensitivity to rapidly progressing individuals caused by static modeling in existing technologies. By using the initial weights of the population as the prior for Bayesian updates and combining them with individual temporal characteristics for recursive optimization, the dynamic parameters of the multi-scale causal links specific to each individual are finally obtained. This quantifies the unique pathological interaction patterns of each individual across multiple scales and modalities, allowing the model to break away from standardized equations and achieve truly individualized predictions.

[0027] By adapting and modifying the topology based on individual-specific dynamic parameters, the model can flexibly adjust the weights and directions of causal edges or add / remove secondary edges within the population framework. For example, for patients carrying specific genetic mutations, the model may enhance the direct causal edges of the corresponding genes, i.e., biomarkers. For individuals with significant compensatory mechanisms, the model may identify additional protective causal pathways, enabling the model to accurately reflect the unique pathological network of an individual. By combining personalized topological frameworks with standardized temporal features, a digital twin capable of replicating the cascading evolution of an individual's entire pathological cycle is constructed. This not only extrapolates and predicts future pathological states but also supports reverse causal tracing and intervention simulation. For instance, by adjusting the virtual value of a certain risk factor in the digital twin, the cascading impact of the intervention on downstream pathological events can be simulated, providing a basis for precise intervention decisions.

[0028] In one embodiment, step S4, which involves performing multi-timescale forward causal simulation on the time-series cascaded causal digital twin to obtain individualized cognitive decline risk quantification results and core driver pathogenesis tracing results for the target individual, includes: S41. Obtain individual pathological initial state parameters and multi-scale causal transmission constraints based on the time-series cascaded causal digital twin, and set multi-time-scale deduction nodes and deduction step size according to the pathological evolution law of cognitive decline. S42. Perform forward causal iterative deduction on the temporal cascade causal digital twin based on the individual pathological initial state parameters, multi-scale causal transmission constraints, multi-timescale deduction nodes and deduction step size, and obtain the full-scale pathological feature evolution sequence and cross-scale causal link activation weight sequence for each deduction node. S43. Obtain the cognitive impairment progression risk value and cognitive function decline rate of the target individual at the corresponding inference node based on each full-scale pathological feature evolution sequence and cross-scale causal link activation weight sequence, and integrate multiple cognitive impairment progression risk values ​​and cognitive function decline rates to obtain an individualized cognitive decline risk quantification result. S44. Based on the cross-scale causal link activation weight sequence and preset weight threshold, core contributing causal links are selected, and based on the pathological scale and modal characteristics corresponding to the core contributing causal links, the source tracing results of the core driving pathogenic pathway of the target individual are obtained.

[0029] As described in steps S41-S44 above, the step of screening core contributing causal links involves extracting the degradation contribution of each causal link in the individual-specific multi-scale time-series cascaded causal digital twin based on the positive causal simulation results, sorting all causal links from high to low according to the degradation contribution, and selecting causal links whose degradation contribution exceeds a preset contribution threshold as core contributing causal links.

[0030] This invention, by setting multi-timescale extrapolation nodes and step sizes, enables the model to perform refined extrapolations across the natural timescales of different pathological processes. This ensures that the model can capture ultra-early microscopic molecular changes and predict mid- to long-term macroscopic clinical outcomes. By utilizing the individual pathological initial state parameters of the digital twin and multi-scale causal transmission constraints, the model starts from the individual's actual state during extrapolation while following the biological laws of pathological evolution, avoiding randomness or physiological infeasibility in the extrapolation process. This significantly improves the reliability and clinical relevance of predictions. Through positive causal iterative extrapolation, the model starts from the current state and progresses along the digital twin... The personalized causal topology framework simulates pathological changes at multiple scales, including molecular, imaging, and cognitive levels, and outputs the full-scale pathological feature evolution sequence of each extrapolation node. This enables the model to extrapolate the complete pathological trajectory of an individual over the next few years or even decades, breaking through the limitation of existing technologies that can only extrapolate a limited time range based on current data. This significantly improves the temporal coverage and foresight of the prediction. By recording the activation weight sequence of cross-scale causal links during the extrapolation process, the model tracks the dynamic intensity changes of causal relationships between different scales in real time. This not only reflects the temporal dependence between different pathological events but also quantifies the relative importance of each causal path in disease progression.

[0031] By combining the risk values ​​for cognitive impairment progression and the rate of cognitive function decline at different time points, the model's output risk results not only include the traditional probability of disease but also incorporate the dynamic characteristics of disease progression. This allows for a more accurate assessment of individual risk and provides richer decision-making information for stratified intervention. The risk value calculation is based on a full-scale pathological evolution sequence, integrating information on the changes over time from multiple dimensions of data, including molecular, imaging, and cognitive data. A multi-source data fusion strategy overcomes the noise and bias of single-modal data, making the prediction results more robust to measurement errors or short-term fluctuations. Through the integration of multiple time points, the model can smooth out the impact of individual outliers, further improving the accuracy of predictions. The model enhances stability and accuracy by setting preset weight thresholds to screen core contributing causal links. It can identify the key pathological pathways that contribute most to an individual's cognitive decline, enabling diagnosis to move from a general high-risk level to a specific biological mechanism level. This provides clinicians with actionable intervention targets. By clarifying the pathological scale and modal characteristics corresponding to the core pathways, the model provides precise biological markers and intervention windows for drug development, lifestyle interventions, or neuromodulation. If the source tracing shows that a patient's core pathway is driven by synaptic toxicity, drugs targeting synaptic protection can be tested first, rather than using cognitive-enhancing supplements indiscriminately, thereby significantly improving the accuracy and efficiency of intervention.

[0032] In one embodiment, step S5, which involves generating a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locating the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generating an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets, includes: S51. Based on the individualized cognitive decline risk quantification results, obtain the individual cognitive decline risk probability, cognitive function decline rate level and pathological progression stage, and generate a corresponding level of cognitive decline graded early warning strategy based on the cognitive decline risk probability, cognitive function decline rate level, pathological progression stage and preset graded early warning threshold. S52. Based on the source tracing results of the core driving pathogenic pathway, obtain the pathological scale type, modal abnormality characteristics and key nodes of pathological transmission corresponding to the pathogenic pathway, and locate the core pathogenic target that can be intervened in an individual based on the pathological scale type, modal abnormality characteristics and key nodes of pathological transmission. S53. Based on the core pathogenic target, the individual standardized multimodal temporal feature set, and the preset multidimensional intervention scheme library, obtain the target matching intervention items, individual pathological adaptation parameters, and intervention timing rules, and generate an individualized intervention guidance scheme that matches the individual pathological characteristics based on the target matching intervention items, individual pathological adaptation parameters, and intervention timing rules.

[0033] As described in steps S51-S53 above, this invention integrates the probability of individual cognitive decline risk, the rate of cognitive function decline, and the stage of pathological progression, and combines this with preset graded early warning thresholds to classify individuals into different early warning levels, such as low-risk static monitoring, medium-risk intensive follow-up, and high-risk emergency intervention. This multidimensional dynamic early warning mechanism overcomes the shortcomings of traditional one-size-fits-all methods, enabling clinicians to accurately formulate differentiated management strategies based on the severity of individual risk and the speed of progression. It solves the problem of premature or late intervention caused by coarse early warning. By considering both probability and rate, the model can distinguish between high-risk individuals with slow progression and low-risk individuals with rapid deterioration, providing a priority basis for resource allocation. For example, for high-risk patients with rapid progression, intensive treatment can be initiated immediately, while for low-risk patients with slow progression, lifestyle intervention can be prioritized. By analyzing the pathological scale type, modal abnormality characteristics, and key nodes of pathological transmission in the core driving pathogenic pathways, the abstract risk is transformed into specific biological intervention targets. By screening the core nodes that can be intervened, the model avoids locating uncontrollable pathological endpoints, ensuring that the recommended targets have practical therapeutic value.

[0034] By matching core pathogenic targets with target-matching intervention items in a pre-defined multi-dimensional intervention protocol library, and combining this with individual standardized multimodal temporal feature sets to determine individual pathological adaptation parameters, the intervention protocol can be precisely adapted to the individual's current state and pathological dynamics. For example, the brain stimulation frequency can be adjusted based on the rate of synaptic plasticity decline, or the antibody dosage can be titrated based on the Aβ deposition rate. This solves the problems of large efficacy differences and high side effect risks caused by the one-size-fits-all approach of traditional protocols. The pre-defined intervention library integrates validated population-level effective interventions, ensuring the clinical feasibility and safety of the protocols, while individual temporal features enable the protocols to be adapted to the population-level dynamics. Based on the experience, the model enables personalized fine-tuning. For example, for different individuals with the same core pathway, the model can dynamically adjust the timing of intervention or combination strategies according to the differences in their biomarker trajectories, maximizing the treatment effect and reducing the probability of ineffective intervention. By clarifying the timing rules of intervention (such as the order of clearing Aβ first and then improving cerebral perfusion, or implementing different interventions in stages according to the disease stage), the model overcomes the shortcomings of existing technologies that only provide a static list of interventions and cannot guide the implementation order and dynamic adjustment. This enables the intervention to accurately block key pathological links while avoiding the potential risks that may be brought about by unreasonable intervention timing.

[0035] In one embodiment, step S6, which involves acquiring newly added multimodal time-series monitoring data of the target individual to adaptively and iteratively update the time-series cascaded causal digital twin, and simultaneously optimizing the individualized cognitive decline risk quantification results and the core driver pathogenesis pathway tracing results to synchronously update the graded early warning strategy and individualized intervention guidance plan, includes: S61. Based on the aforementioned temporal cascade causal graph framework, the newly added multimodal temporal monitoring data of the target individual is subjected to pathological time axis calibration and amplitude normalization processing to obtain a newly added standardized multimodal temporal feature set; S62. Based on the newly added standardized multimodal temporal feature set and the current causal link parameters of the temporal cascaded causal digital twin, a Bayesian recursive update algorithm is used to obtain the parameter correction increment, and the multi-scale causal link dynamic parameters of the twin are adaptively adjusted according to the parameter correction increment to obtain the adjusted causal link parameters. S63. Based on the adjusted causal link parameters, complete the iterative reconstruction of the time-series cascaded causal digital twin to obtain the reconstructed twin, and re-execute the multi-timescale forward causal simulation based on the reconstructed twin to obtain the optimized simulation results. S64. Update the individualized cognitive decline risk quantification results and the core driving pathogenic pathway tracing results based on the optimized deduction results. Based on the updated risk quantification results and pathogenic pathway tracing results, adjust the warning level and intervention sequence of the graded early warning strategy in real time, and adjust the target adaptability and execution intensity of the individualized intervention guidance plan to continuously improve the accuracy of full-cycle risk prediction and intervention adaptability.

[0036] As described in steps S61-S64 above, this invention unifies data from different sources and with different sampling frequencies into the pathological evolution timeframe of the digital twin through pathological timeline calibration. This solves the problem of information misalignment caused by inconsistent monitoring times. Amplitude normalization further eliminates the differences in the dimensions of data from different modalities, ensuring balanced weights of multimodal information during fusion. This allows the model to seamlessly integrate newly added monitoring data from ultra-early stages and multiple scales, solving the problem that existing technologies cannot effectively utilize dynamic information due to data temporal heterogeneity. This significantly improves the model's sensitivity to capturing latent pathological changes. By generating a new standardized multimodal temporal feature set, the model ensures that subsequent parameter correction and structural optimization are based on a unified and comparable data foundation, avoiding update bias caused by improper data preprocessing. This provides high-quality input for the adaptive evolution of the digital twin, thereby mitigating the risk of model degradation caused by data noise or heterogeneity. By combining newly added standardized temporal feature sets with current causal link parameters using a Bayesian recursive update algorithm, the model incorporates new evidence into the posterior correction increment of parameters calculated using a Bayesian inference method. This allows the model to dynamically adjust the weights and thresholds of causal links at various scales based on real-time data, thereby tracking the evolution of individual pathological mechanisms in real time. This solves the problem of prediction results drifting over time due to model rigidity. The introduction of the Bayesian framework means that the update process not only depends on new data but also incorporates the population causal topology learned during the construction of the digital twin. This avoids overfitting or deviation from biological laws due to individual abnormal data. For example, when new data indicates that a certain causal path is abnormally active, the model will make reasonable corrections by referring to the typical weight range of that path in the population, ensuring a balance between the biological rationality of parameter adjustment and individual specificity. This significantly enhances the model's adaptability to individual heterogeneity and dynamic changes in multiple pathways.

[0037] By iteratively reconstructing a time-series cascaded causal digital twin, the latest pathological patterns learned by the model from new data are embedded into the entire simulation system. Subsequently, based on the reconstructed twin, multi-timescale positive causal simulations are re-executed. The model can generate evolutionary trajectories and causal weight sequences that better reflect the current real pathological state. Through closed-loop iteration, the prediction results are continuously optimized with the accumulation of monitoring data, solving the long-term predictive efficacy decay problem caused by the inability of existing technologies to evolve. By continuously optimizing the simulation results, the model provides real-time and reliable decision-making basis for subsequent strategy adjustments. For example, when the reconstructed twin shows a significant change in the weight of a certain pathway, the intervention plan can be immediately adjusted to ensure that the treatment always targets the most critical pathogenic mechanism, thus overcoming the shortcomings of traditional static models that cannot respond to the dynamic evolution of the disease, leading to intervention failure. By updating the individualized cognitive decline risk quantification results and the source tracing results of the core driving pathogenic pathways in real time, the model provides a basis for prediction. The dynamic adjustment of warning strategies and intervention plans provides a quantitative basis. For example, when the risk probability increases significantly, the warning level can be automatically raised and a more aggressive intervention can be initiated. When pathway tracing reveals the emergence of new pathogenic mechanisms, intervention measures targeting that pathway can be supplemented in a timely manner. This enables the entire system to have the autonomous evolutionary ability to monitor, evaluate, and adjust, solving the problem of intervention lag or ineffectiveness caused by rigid strategies. By real-time correction of the warning level and intervention sequence of the graded warning strategy, the model can intelligently adjust the management intensity and intervention timing according to the current risk status. At the same time, it can adjust the target adaptability and execution intensity of the individualized intervention guidance plan so that the treatment can accurately match the latest pathological characteristics. For example, it can adjust the drug dosage or the order of combined treatment according to the changes in causal weight. Through fine-grained adjustments based on real-time data, overtreatment or undertreatment is avoided, significantly improving the timeliness, effectiveness, and safety of intervention, which fully meets the needs of precision medicine for dynamic and personalized medicine.

[0038] like Figure 2 As shown, this application also provides a personalized cognitive decline risk prediction system based on multi-modal data fusion, including: The first building module is used to obtain multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade pattern of cognitive decline throughout the entire cycle, so as to build a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph. The alignment module is used to acquire multimodal temporal longitudinal monitoring data and clinical risk factor temporal data of the target individual, and perform pathological time axis alignment and normalization processing on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data according to the temporal cascade causal graph basic framework to obtain a standardized multimodal temporal feature set of the individual. The second construction module is used to fit the individual-specific multi-scale causal link dynamic parameters based on the time-series cascaded causal graph basic framework and the individual standardized multimodal time-series feature set to construct an individual-specific multi-scale time-series cascaded causal digital twin. The deduction module is used to perform multi-timescale forward causal simulation deduction on the time-series cascaded causal digital twin to obtain the individualized cognitive decline risk quantification results and core driving pathogenic pathway tracing results for the target individual; The generation module is used to generate a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locate the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generate an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets. The iterative update module is used to acquire newly added multimodal time-series monitoring data of target individuals to adaptively iteratively update the time-series cascaded causal digital twin, and simultaneously optimize the individualized cognitive decline risk quantification results and core driver pathogenesis tracing results to update the graded early warning strategy and individualized intervention guidance plan, continuously improving the accuracy of full-cycle risk prediction and intervention adaptability.

[0039] It should be noted that each module and unit in the personalized cognitive decline risk prediction system based on multi-mode data fusion corresponds one-to-one with the steps in the personalized cognitive decline risk prediction method based on multi-mode data fusion.

[0040] like Figure 3 As shown, this application also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all the data required for the process of a personalized cognitive decline risk prediction method based on multi-modal data fusion. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the personalized cognitive decline risk prediction method based on multi-modal data fusion.

[0041] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0042] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for predicting the risk of individualized cognitive decline based on multi-modal data fusion.

[0043] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0044] It should be noted that, in this document, the terms include, encompass, or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "including a…" does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0045] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A personalized cognitive decline risk prediction method based on multi-modal data fusion, characterized in that, include: To acquire multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade patterns of cognitive decline throughout the entire cycle, so as to construct a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph; Multimodal longitudinal monitoring data and clinical risk factor time series data of the target individual are acquired, and pathological time axis alignment and normalization processing are performed on the multimodal longitudinal monitoring data and clinical risk factor time series data according to the time series cascade causal graph framework to obtain the individual standardized multimodal time series feature set; Based on the aforementioned temporal cascaded causal graph framework and individual standardized multimodal temporal feature set, individual-specific multi-scale causal link dynamic parameters are fitted to construct an individual-specific multi-scale temporal cascaded causal digital twin; Perform multi-timescale forward causal simulation on the aforementioned time-series cascaded causal digital twin to obtain individualized cognitive decline risk quantification results and core driver pathogenesis tracing results for the target individual; Based on the individualized cognitive decline risk quantification results, a cognitive decline grading early warning strategy is generated. Based on the source tracing results of the core driving pathogenic pathways, the core pathogenic targets that can be intervened in individuals are located. Based on the core pathogenic targets, an individualized intervention guidance plan matching the individual's pathological characteristics is generated. We acquire new multimodal time-series monitoring data for target individuals to adaptively and iteratively update the time-series cascaded causal digital twin, and simultaneously optimize the individualized cognitive decline risk quantification results and core driver pathogenesis tracing results to update the hierarchical early warning strategy and individualized intervention guidance plan, continuously improving the accuracy of full-cycle risk prediction and intervention adaptability.

2. The method for predicting the risk of cognitive decline based on multi-modal data fusion according to claim 1, characterized in that, The steps for obtaining multi-scale temporal stratification rules and multimodal longitudinal follow-up evidence-based data corresponding to the pathological cascade patterns of cognitive decline throughout the entire cycle to construct a basic framework for a population-level multimodal multi-scale temporal cascade causal graph include: Based on the pathological cascade pattern of the entire cognitive decline cycle, multiple degradation scales with temporal causal relationships are divided, and corresponding exclusive modal data and temporal resolution are matched for each degradation scale to form a corresponding scale modal feature set; Based on the multimodal longitudinal follow-up evidence-based data and the feature set of each scale modality, the continuous time marginal structure model and the temporal Granger causality test were used to obtain the cross-scale temporal causal correlation degree and the pathological transmission time delay coefficient. Based on the cross-scale temporal causal correlation degree and pathological transmission time delay coefficient, acyclic causal edges are constructed, and spurious causal edges in the acyclic causal edges are removed according to a preset causal confidence threshold to obtain an effective cross-scale causal link set. Based on the multi-scale temporal hierarchical rules and the effective cross-scale causal link set, the topological binding of multi-scale levels and causal links is carried out to construct the basic framework of the population-level multimodal multi-scale temporal cascaded causal graph.

3. The personalized cognitive decline risk prediction method based on multi-modal data fusion according to claim 1, characterized in that, The step of performing pathological time axis alignment and normalization processing on multimodal temporal longitudinal monitoring data and clinical risk factor time series data based on the aforementioned temporal cascade causal graph framework to obtain an individual standardized multimodal temporal feature set includes: Based on the aforementioned temporal cascade causal graph framework, pathological temporal benchmarks and physiological reference intervals are obtained to determine the corresponding individual modal pathological zero-point determination threshold. Based on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data, the measured temporal sequences of each modality and the temporal change sequences of risk factors are obtained, and the pathological zero point time of the corresponding modality and risk factor is located by combining the pathological zero point determination threshold. Based on each pathological zero point, the multimodal time-series longitudinal monitoring data and clinical risk factor time-series data are time-series rearranged and phase-calibrated to achieve pathological time axis alignment, resulting in a corresponding aligned multimodal time-series dataset. Based on the multi-scale weighting rules of the aforementioned time-series cascaded causal graph framework, normalized baseline coefficients for each scale are obtained to perform amplitude normalization and outlier removal on the corresponding aligned multimodal time-series dataset, thereby obtaining the corresponding standard multimodal time-series data. By integrating the features of multiple standard multimodal time series data, an individual standardized multimodal time series feature set is obtained.

4. The method for predicting the risk of cognitive decline based on multi-modal data fusion according to claim 1, characterized in that, The step of fitting individual-specific multi-scale causal link dynamic parameters based on the time-series cascaded causal graph framework and the individual standardized multimodal time-series feature set to construct an individual-specific multi-scale time-series cascaded causal digital twin includes: Based on the aforementioned temporal cascaded causal graph framework, a multi-scale causal topology and a population-level initial weight set of links are obtained, and individual pathological boundary conditions and initial state parameters are obtained based on the aforementioned individual standardized multimodal temporal feature set. A continuous-time causal state-space model with Bayesian recursive updates is used to fit and dynamically update the multi-scale causal topology, the initial weight set of the group-level link, the individual pathological boundary conditions, and the initial state parameters link by link, so as to obtain the individual-specific multi-scale causal link dynamic parameters. Based on the individual-specific multi-scale causal link dynamic parameters, the multi-scale causal topology is personalized and adapted to obtain an individual-specific causal topology framework. Based on the individual-specific causal topology framework and the individual standardized multimodal temporal feature set, an individual-specific multi-scale temporal cascade causal digital twin that can replicate the individual's full-cycle pathological cascade evolution law is constructed.

5. The personalized cognitive decline risk prediction method based on multi-modal data fusion according to claim 1, characterized in that, The steps of performing multi-timescale forward causal simulation on the time-series cascaded causal digital twin to obtain individualized cognitive decline risk quantification results and core driver pathogenesis tracing results for the target individual include: Based on the aforementioned time-series cascaded causal digital twin, the initial state parameters of individual pathology and multi-scale causal transmission constraints are obtained, and multi-time-scale inference nodes and inference step sizes are set according to the pathological evolution law of cognitive decline. Based on the individual pathological initial state parameters, multi-scale causal transmission constraints, multi-timescale inference nodes, and inference step size, a forward causal iterative inference is performed on the temporal cascade causal digital twin to obtain the full-scale pathological feature evolution sequence and cross-scale causal link activation weight sequence for each inference node. Based on each full-scale pathological feature evolution sequence and cross-scale causal link activation weight sequence, the risk value of cognitive impairment progression and the rate of cognitive function decline of the target individual at the corresponding inference node are obtained and integrated to obtain the individualized cognitive decline risk quantification result. The core contributing causal links are selected based on the cross-scale causal link activation weight sequence and preset weight threshold, and the source tracing results of the core driving pathogenic pathways of the target individual are obtained based on the pathological scale and modal characteristics corresponding to the core contributing causal links.

6. The personalized cognitive decline risk prediction method based on multi-modal data fusion according to claim 1, characterized in that, The steps of generating a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locating the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generating an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets include: Based on the individualized cognitive decline risk quantification results, the individual cognitive decline risk probability, cognitive function decline rate level, and pathological progression stage are obtained, and combined with the preset graded early warning threshold, a corresponding graded early warning strategy for cognitive decline is generated. Based on the source tracing results of the core driving pathogenic pathway, the pathological scale type, modal abnormality characteristics and key nodes of pathological transmission corresponding to the pathogenic pathway are obtained to locate the core pathogenic targets that can be intervened in individuals. Based on the core pathogenic targets, the individual standardized multimodal temporal feature set, and the preset multidimensional intervention program library, target matching intervention items, individual pathological adaptation parameters, and intervention timing rules are obtained to generate individualized intervention guidance programs that match individual pathological characteristics.

7. The method for predicting the risk of cognitive decline based on multi-modal data fusion according to claim 1, characterized in that, The steps of acquiring newly added multimodal time-series monitoring data for target individuals to adaptively and iteratively update the time-series cascaded causal digital twin, and simultaneously optimizing the individualized cognitive decline risk quantification results and core driver pathogenesis pathway tracing results to synchronously update the graded early warning strategy and individualized intervention guidance plan, include: Based on the aforementioned temporal cascade causal graph framework, the newly added multimodal temporal monitoring data of the target individual is subjected to pathological time axis calibration and amplitude normalization processing to obtain a newly added standardized multimodal temporal feature set; Based on the newly added standardized multimodal temporal feature set and the current causal link parameters of the temporal cascaded causal digital twin, a Bayesian recursive update algorithm is used to obtain parameter correction increments to adaptively adjust the dynamic parameters of the multi-scale causal links of the twin, thus obtaining the adjusted causal link parameters. Based on the adjusted causal link parameters, the iterative reconstruction of the time-series cascaded causal digital twin is completed to obtain the reconstructed twin. Then, the multi-timescale forward causal simulation is re-executed to obtain the optimized simulation results. Based on the optimized simulation results, update the individualized cognitive decline risk quantification results and the core driving pathogenic pathway tracing results, and in real time correct the warning level and intervention sequence of the graded early warning strategy, adjust the target adaptability and execution intensity of the individualized intervention guidance plan, and continuously improve the accuracy of full-cycle risk prediction and intervention adaptability.

8. A personalized cognitive decline risk prediction system based on multi-modal data fusion, characterized in that, include: The first building module is used to obtain multi-scale temporal stratification rules and multi-modal longitudinal follow-up evidence-based data corresponding to the pathological cascade pattern of cognitive decline throughout the entire cycle, so as to build a basic framework for a population-level multi-modal multi-scale temporal cascade causal graph. The alignment module is used to acquire multimodal temporal longitudinal monitoring data and clinical risk factor temporal data of the target individual, and perform pathological time axis alignment and normalization processing on the multimodal temporal longitudinal monitoring data and clinical risk factor temporal data according to the temporal cascade causal graph basic framework to obtain a standardized multimodal temporal feature set of the individual. The second construction module is used to fit the individual-specific multi-scale causal link dynamic parameters based on the time-series cascaded causal graph basic framework and the individual standardized multimodal time-series feature set to construct an individual-specific multi-scale time-series cascaded causal digital twin. The deduction module is used to perform multi-timescale forward causal simulation deduction on the time-series cascaded causal digital twin to obtain the individualized cognitive decline risk quantification results and core driving pathogenic pathway tracing results for the target individual; The generation module is used to generate a cognitive decline grading and early warning strategy based on the individualized cognitive decline risk quantification results, locate the individual's interventionable core pathogenic targets based on the core driving pathogenic pathway tracing results, and generate an individualized intervention guidance plan matching the individual's pathological characteristics based on the core pathogenic targets. The iterative update module is used to acquire newly added multimodal time-series monitoring data of target individuals to adaptively iteratively update the time-series cascaded causal digital twin, and simultaneously optimize the individualized cognitive decline risk quantification results and core driver pathogenesis tracing results to update the graded early warning strategy and individualized intervention guidance plan, continuously improving the accuracy of full-cycle risk prediction and intervention adaptability.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images