Disease risk prediction method and system based on multi-modal data fusion

Abstract

The invention relates to the technical field of data analysis, provides a disease risk prediction method and system based on multi-modal data fusion, and solves the problems of poor interpretability and insufficient prediction adaptability caused by dependence on labeled data and neglect of rhythm collaboration. The method comprises the following steps: collecting physiological signals and inflammatory factor concentration of a user in a daily state; extracting a time domain feature and a frequency domain feature from the physiological signal, and extracting a concentration fluctuation feature from the concentration of the inflammatory factor; respectively calculating correlation intensity values between the time domain feature and the concentration fluctuation feature and between the frequency domain feature and the concentration fluctuation feature, and constructing a causal correlation map in combination with medical priori knowledge; analyzing physiological rhythm characteristics in the physiological signals and biochemical rhythm characteristics in the inflammatory factor concentration by using a biological rhythm analysis model, and determining rhythm coupling characteristics; rhythm coupling characteristics and node association information in a causal association map are analyzed through a containerized model, a disease risk prediction result of the user is generated, and the interpretability and individual adaptability of disease risk prediction are improved.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a method and system for predicting disease risk based on multimodal data fusion. Background Technology

[0002] In the context of early warning of chronic diseases and personalized health management, there is an urgent need for a technology that can integrate continuous physiological signals from wearable devices with biochemical indicators from laboratory tests to achieve dynamic assessment of an individual's health status and accurate prediction of future disease risks. Such applications require systems that can not only capture key features from multi-source heterogeneous data but also reveal the intrinsic relationships between different modalities, thereby supporting interpretable risk assessments.

[0003] Current research has proposed multimodal fusion models based on deep neural networks. These methods encode physiological time-series signals and inflammation-related biomarker levels separately, inputting them into a shared latent space for joint learning. An attention mechanism is then used to weight the contributions of different modalities, ultimately outputting the probability of disease occurrence. However, these methods rely excessively on a large number of labeled samples for training, making it difficult to effectively incorporate medical knowledge and resulting in poor model interpretability. Furthermore, their modeling of the dynamic interactions between physiological and biochemical indicators is relatively static, neglecting their rhythmic synergistic characteristics over time. This fails to accurately characterize the key coupling mechanisms in disease evolution, limiting the clinical reliability and individual adaptability of the predictions. Summary of the Invention

[0004] The purpose of this application is to provide a disease risk prediction method and system based on multimodal data fusion, so as to solve the problems of poor interpretability and insufficient predictive adaptability in the prior art due to reliance on labeled data and neglect of rhythmic coordination.

[0005] To address the aforementioned technical problems, in a first aspect, this application provides a disease risk prediction method based on multimodal data fusion, comprising:

[0006] Collect physiological signal data and inflammatory factor concentration data of users in their daily state;

[0007] Temporal and frequency domain features are extracted from the physiological signal data, and concentration fluctuation features are extracted from the inflammatory factor concentration data.

[0008] The correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features are calculated respectively. Combined with prior medical knowledge, the correlation between the physiological signal data and the inflammatory factor concentration data is analyzed to construct a causal correlation map.

[0009] Based on the causal relationship map, the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data are analyzed using a biological rhythm analysis model to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features.

[0010] By analyzing the rhythm coupling features and node association information in the causal relationship graph using a containerized model, the user's disease risk prediction results are generated.

[0011] Optionally, the association strength value includes a first association strength value and a second association strength value;

[0012] Calculate the correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features, respectively. Combined with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data, and construct a causal correlation map, including:

[0013] The direction of change of the temporal features and concentration fluctuation features at each time point is compared. If the direction of change is consistent, it is marked as a synchronous event. Combined with the total number of time points, the first correlation strength value between the temporal features and concentration fluctuation features is calculated.

[0014] The direction of change of the frequency domain features and concentration fluctuation features at each time point is compared to calculate the second correlation strength value between the frequency domain features and concentration fluctuation features;

[0015] Based on the characteristic change patterns in medical prior knowledge, the influence relationship between the first and second correlation strength values ​​is analyzed, and the analysis results are obtained.

[0016] Using time-domain features, frequency-domain features, and concentration fluctuation features as nodes, and combining the analysis results, a causal relationship map is constructed.

[0017] Optionally, based on the characteristic change patterns in prior medical knowledge, the influence relationship between the first and second association strength values ​​is analyzed to obtain the analysis results, including:

[0018] Based on prior medical knowledge and the logical relationships between different characteristics, the criteria for determining the type of impact, the defining conditions, and the basis for classifying the degree of impact are determined.

[0019] Extract the numerical features of the first correlation strength value and the numerical features of the second correlation strength value respectively;

[0020] The numerical features of the first association strength value and the second association strength value are compared item by item to obtain the comparison result. The comparison result is matched with the judgment criterion to determine the target influence type, which is a synergistic influence, an antagonistic influence, or no significant influence.

[0021] Based on the criteria for classifying the degree of influence, and combined with the numerical characteristics of the first and second correlation strength values, the degree of influence corresponding to the target influence type is determined, and the analysis results are formed by combining the target influence type.

[0022] Optionally, based on the causal association map, a biorhythm analysis model is used to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data to determine the rhythmic coupling features between the physiological rhythm features and the biochemical rhythm features, including:

[0023] Based on the node association information in the causal association map, the physiological signal data and inflammatory factor concentration data are split into separate data according to a preset time period using a biological rhythm analysis model, and the physiological rhythm features of the physiological signal data and the biochemical rhythm features of the inflammatory factor concentration data are extracted within each preset time period.

[0024] Based on the node association information and the target influence type between nodes, calculate the target parameter deviation values ​​of each core parameter in the physiological rhythm characteristics and biochemical rhythm characteristics;

[0025] Based on the target parameter deviation value and the node association information, the target coupling type and corresponding target coupling strength level of physiological rhythm characteristics and biochemical rhythm characteristics are determined using a biological rhythm analysis model.

[0026] The target coupling type, target coupling strength level, target parameter deviation value, and corresponding preset time period are integrated to form a rhythmic coupling feature.

[0027] Optionally, based on the target parameter deviation value and the node association information, a biological rhythm analysis model is used to determine the target coupling type and corresponding target coupling strength level of physiological rhythm features and biochemical rhythm features, including:

[0028] According to the preset rules of the biological rhythm analysis model, the association strength value in the node association information is divided into multiple association strength intervals, and a judgment weight is assigned to each association strength interval and a judgment coefficient is assigned to different target influence types.

[0029] Based on the correlation logic between different features, the target parameter deviation values ​​are classified to obtain multiple subsets of deviation values;

[0030] Establish rules for determining coupling type using a biological rhythm analysis model;

[0031] Each subset of deviation values ​​is compared with the determination rule of the coupling type to match the corresponding target coupling type;

[0032] Based on the judgment weight and the judgment coefficient, calculate the intensity judgment value of each subset of deviation values, and summarize all intensity judgment values ​​to obtain the total intensity judgment value.

[0033] The total number of intensity determination values ​​is matched with the preset level classification standard in the biological rhythm analysis model to determine the target coupling intensity level corresponding to the target coupling type.

[0034] Secondly, this application provides a disease risk prediction method and system based on multimodal data fusion, including:

[0035] The data acquisition module is used to collect physiological signal data and inflammatory factor concentration data of users in their daily state;

[0036] The extraction module is used to extract time-domain and frequency-domain features from the physiological signal data, and to extract concentration fluctuation features from the inflammatory factor concentration data;

[0037] The module is used to calculate the correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features, respectively, and, combined with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data to construct a causal correlation map.

[0038] The analysis module is used to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data based on the causal association map using a biorhythm analysis model, so as to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features;

[0039] The generation module is used to analyze the rhythm coupling features and node association information in the causal association graph through a containerized model to generate disease risk prediction results for the user.

[0040] Thirdly, this application provides an electronic device, comprising:

[0041] Memory, used to store computer programs;

[0042] A processor, configured to execute the computer program to implement the steps of a disease risk prediction method based on multimodal data fusion as described in the first aspect above.

[0043] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of a disease risk prediction method based on multimodal data fusion as described in the first aspect above.

[0044] The disease risk prediction method based on multimodal data fusion provided in this application has the following beneficial effects:

[0045] By collecting physiological signals and inflammatory factor concentration data from users in their daily lives, and extracting their time-frequency domain features and concentration fluctuation features, a causal relationship graph was constructed by combining prior medical knowledge, revealing the inherent logical relationship between multimodal data. On this basis, a biorhythm analysis model was introduced to analyze the rhythmic characteristics of the two types of data and identify their coupling patterns. Finally, a containerized model was used to fuse rhythmic coupling features and graph node information to generate risk prediction results. This significantly enhanced the model's ability to characterize the dynamic evolution of individual health status and its clinical interpretability without relying on large-scale labeled data.

[0046] Furthermore, this application, based on the node relationships in the causal relationship graph, splits the two types of data according to a preset time period and extracts their respective rhythmic features. It calculates the deviation of key parameters by combining the influence types between nodes, thereby determining the coupling type and intensity level, and integrating them to form a structured rhythmic coupling feature. This can accurately capture the synergistic change patterns of physiological and biochemical rhythms at different time scales, breaking through the technical bottleneck of existing solutions that are coarse in modeling cross-modal dynamic interactions and difficult to reflect the heterogeneity of individual rhythms, thus improving the adaptability and reliability of disease risk prediction in real health management scenarios. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating a disease risk prediction method based on multimodal data fusion, provided for an embodiment of this application;

[0049] Figure 2 A schematic diagram illustrating a specific implementation of a disease risk prediction method based on multimodal data fusion, provided in this application embodiment;

[0050] Figure 3 This is a schematic diagram of the structure of a disease risk prediction system based on multimodal data fusion, provided as an embodiment of this application. Detailed Implementation

[0051] To address the shortcomings of existing deep learning-driven multimodal fusion methods in integrating prior medical knowledge and lacking modeling of the dynamic synergistic relationship between physiological and biochemical indicators over time, this application provides a disease risk prediction method based on multimodal data fusion. The core idea of ​​this method is to start from causal reasoning and biological rhythm mechanisms, construct an association graph that integrates medical logic, introduce a rhythm analysis model to characterize the coupling characteristics of the two types of data in periodic changes, and then combine a containerized architecture to achieve dynamic prediction of individualized risks. This reduces the dependence on large-scale labeled data while improving the interpretability and clinical adaptability of the prediction results.

[0052] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] The core of this application is to provide a disease risk prediction method based on multimodal data fusion, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0054] Step 101: Collect physiological signal data and inflammatory factor concentration data of users in their daily state.

[0055] In this step, the "normal state" refers to the user's physical state of not receiving medical intervention, not being in a specific testing scenario, and maintaining a natural life, rest, and light activity. This is used to obtain raw data that reflects the user's actual physical condition and is adapted to the risk prediction needs of chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease.

[0056] Physiological signal data refers to continuous signal data of various physiological levels obtained from the user's body based on sensing acquisition methods, specifically including electrocardiogram signals, pulse signals, respiratory signals, and body movement signals. These signals are all related to the changes in the condition of chronic inflammatory diseases.

[0057] Inflammatory factor concentration data refers to the content data of various inflammatory factors obtained from peripheral blood or saliva samples of users' body fluids based on sensor detection methods. Specifically, these include interleukin-6, interleukin-1β, tumor necrosis factor-α, and C-reactive protein. Changes in the concentration of these inflammatory factors are the core characteristics of the occurrence and development of chronic inflammatory diseases.

[0058] In this embodiment, the user is first confirmed to be in a normal daily state, excluding interference factors such as recent medication, strenuous exercise, and acute infection. Then, the user's electrocardiogram, pulse, respiration, and body movement signals are continuously collected in real time for 72 hours using wearable sensing components. The raw signal values ​​are recorded at a frequency of once per minute, and the raw signal values ​​are structured and organized according to the time series to form continuous and resolvable physiological signal data. At the same time, the user's peripheral blood sample is collected once a day using an external sensing detection component to detect the content of interleukin-6, interleukin-1β, tumor necrosis factor-α, and C-reactive protein in the sample. The detection results are organized according to the collection time to form time-stamped inflammatory factor concentration data.

[0059] Step 102: Extract time-domain and frequency-domain features from the physiological signal data, and extract concentration fluctuation features from the inflammatory factor concentration data.

[0060] In this step, temporal features refer to the features obtained from the time dimension analysis of physiological signal data based on feature extraction algorithms. These features include the mean heart rate of electrocardiogram (ECG) signals, the difference between the maximum and minimum electrical signal conduction time from the atrium to the ventricle, the rate of change of the total time of ventricular electrical activity, the mean pulse wave conduction velocity of pulse signals, the range of pulse pressure fluctuations, the duration of pulse amplitude, the respiratory rate variation and tidal volume variation of respiratory signals, and the peak activity intensity and the proportion of exercise duration of body movement signals.

[0061] Frequency domain features refer to the features obtained by analyzing physiological signal data after converting it from the time dimension to the frequency dimension based on feature extraction algorithms. These features include the main frequency range, low frequency band, and high frequency band amplitude distribution of electrocardiogram signals; the fundamental frequency, harmonic amplitude ratio, and frequency band energy density of pulse signals; and the main frequency amplitude and frequency variation coefficient of respiratory signals.

[0062] Concentration fluctuation characteristics refer to the features that reflect the changes in concentration over time obtained from inflammatory factor concentration data based on feature extraction algorithms. These features include the rise and fall of interleukin-6, the trend of interleukin-1β, the fluctuation cycle of tumor necrosis factor-α, and the peak interval of C-reactive protein concentration.

[0063] In this embodiment, physiological signal data is first analyzed from a time dimension. The numerical changes of ECG, pulse, respiration, and body movement signals at different time points, the time and value of extreme values, the duration of the signal, and the rate of change are calculated sequentially to extract the corresponding time-domain features. Then, the physiological signal data is converted from the time dimension to the frequency dimension. The frequency domain features are extracted by analyzing the dominant frequency range, amplitude distribution of different frequency bands, frequency band coverage, and energy proportion of each physiological signal through spectral decomposition. Subsequently, the inflammatory factor concentration data is analyzed by feature parsing. The rise and fall amplitudes, overall rise and fall trends, time intervals of concentration fluctuations, and frequency of peak occurrences of interleukin-6, interleukin-1β, tumor necrosis factor-α, and C-reactive protein are analyzed within 72 hours to extract the corresponding concentration fluctuation features.

[0064] Step 103: Calculate the correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features, respectively. Combine with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data, and construct a causal correlation map.

[0065] In this step, the correlation strength value refers to the quantitative value of the degree of correlation between the time domain features and the frequency domain features and the concentration fluctuation features in the time dimension. The correlation strength value includes the first correlation strength value and the second correlation strength value.

[0066] Medical a priori knowledge refers to a set of medical conclusions related to the correlation between physiological signal characteristics and inflammatory factor concentration characteristics, the interaction logic of characteristic changes, and disease manifestations for chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease, which have been verified by clinical diagnosis and treatment and summarized by medical research.

[0067] The correlation can be understood as the interaction between the time-domain and frequency-domain features corresponding to physiological signal data and the concentration fluctuation features corresponding to inflammatory factor concentration data, based on the changing patterns of the time dimension. It reflects the intrinsic connection between physiological indicators and inflammatory indicators.

[0068] Causal association graphs refer to feature association networks constructed using physiological and inflammation-related features as nodes and the association strength values ​​and influence relationships between features as connecting edges. They are used to intuitively present the causal logic and the degree of association between various features.

[0069] In this embodiment of the application, step 103 specifically includes the following steps:

[0070] Step 301: Compare the direction of change of the temporal features and concentration fluctuation features at each time point. If the direction of change is consistent, mark it as a synchronous event. Combine the total number of time points to calculate the first correlation strength value between the temporal features and concentration fluctuation features.

[0071] In this step, a synchronous event refers to the phenomenon where the time-domain characteristics and concentration fluctuation characteristics exhibit the same direction of change at the same point in time.

[0072] In this embodiment, the numerical change direction of the time-domain features and concentration fluctuation features at all time points is first extracted. The change direction is divided into three categories: rising, falling, and stable. Then, the change direction of the two types of features at each time point is compared one by one. If the change direction of the two is the same, the time point is marked as a synchronous event. Then, the total number of all synchronous events is counted. The total number of synchronous events is divided by the total number of time points of the time-domain features and concentration fluctuation features. The resulting quantified value is the first correlation strength value between the time-domain features and concentration fluctuation features. This value is between zero and one. The higher the value, the stronger the correlation between the two types of features.

[0073] Step 302: Compare the direction of change of the frequency domain features and concentration fluctuation features at each time point to calculate the second correlation strength value between the frequency domain features and concentration fluctuation features.

[0074] In this embodiment, the numerical change direction of the frequency domain features and concentration fluctuation features at all time points is extracted in the same way. The change direction of the two types of features at each time point is compared one by one. If the change direction of the two is the same, the time point is marked as a synchronous event. Then, the total number of all synchronous events is counted. The total number of synchronous events is divided by the total number of time points of the frequency domain features and concentration fluctuation features. The resulting quantized value is the second correlation strength value between the frequency domain features and the concentration fluctuation features. This value is between zero and one. The higher the value, the stronger the correlation between the two types of features.

[0075] Step 303: Based on the characteristic change patterns in medical prior knowledge, analyze the influence relationship between the first correlation strength value and the second correlation strength value, and obtain the analysis results.

[0076] In this step, the characteristic change pattern refers to the interaction pattern between the time-domain and frequency-domain characteristics of physiological signals and the fluctuation characteristics of inflammatory factor concentrations as the disease progresses, which is summarized from medical prior knowledge for chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease. This pattern is the core basis for analyzing the influence relationship between the two types of correlation strength values.

[0077] The influence relationship refers to the interaction between the first and second correlation strength values ​​based on their respective numerical characteristics. In essence, it is the inherent logic of the correlation between time-domain features and frequency-domain features and concentration fluctuation features, reflecting the overall interaction law between the two core features of physiological signals and the concentration features of inflammatory factors.

[0078] The analysis results refer to the integrated conclusions obtained after systematically analyzing the influence relationship between the first and second correlation strength values ​​based on prior medical knowledge. They are the core results characterizing the type and strength of the correlation between physiological signals and inflammatory factor concentration data.

[0079] In this embodiment of the application, step 303 specifically includes the following steps:

[0080] Step 311: Based on prior medical knowledge and the logical connections between different characteristics, determine the criteria for judging the type of influence, the conditions for defining the influence, and the basis for classifying the degree of influence.

[0081] In this step, the correlation logic between different features refers to the inherent interaction and correlation between the time-domain and frequency-domain features of physiological signals and the fluctuation features of inflammatory factor concentrations, which are summarized by medical a priori knowledge and verified by clinical diagnosis and research for chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease.

[0082] The criteria for determining the type of influence refer to the quantitative reference criteria for distinguishing between three types of influence: synergistic influence, antagonistic influence, and no significant influence. It is the core basis for determining the interaction type of two types of association strength values.

[0083] The defining conditions refer to the range of numerical characteristics of the association strength value that must be met to determine various types of influence, and are the specific numerical requirements of the matching judgment criteria.

[0084] The criteria for classifying the degree of influence refer to the numerical characteristics of the correlation strength value, which quantifies the strength of various types of influence and is used to distinguish different degrees of influence under the same type of influence.

[0085] In this embodiment of the application, the criteria for determining the type of influence are first determined by combining the medical prior knowledge of chronic inflammatory diseases and the correlation logic between physiological characteristics and inflammatory characteristics. For example, one criterion is that the synergistic influence is that the two types of correlation strength values ​​change in the same trend and the values ​​are both in the high value range; the antagonistic influence is that the two types of correlation strength values ​​change in opposite trends and the difference in values ​​is in the high value range; and the no significant influence is that the two types of correlation strength values ​​are both in the low value range and the change trend has no obvious pattern.

[0086] Simultaneously, corresponding definition conditions are set. For example, one way to define synergistic effects is that the correlation strength values ​​of both types are greater than 0.6 and have the same trend; the definition condition for antagonistic effects is that the correlation strength values ​​of both types have opposite trends and the numerical difference is greater than 0.5; and the definition condition for no significant effect is that the correlation strength values ​​of both types are less than 0.4. Then, the criteria for classifying the degree of influence are established, and each type of influence is divided into three levels: mild, moderate, and severe according to the numerical range of the correlation strength values. For example, one method of classification is that for synergistic and antagonistic effects, the numerical value or difference is between the lower limit of the definition condition and 0.7, which is mild; 0.7 to 0.8 is moderate; and greater than 0.8 is severe. No significant effect is not classified.

[0087] Step 312: Extract the numerical features of the first association strength value and the numerical features of the second association strength value respectively.

[0088] In this step, numerical characteristics refer to indicators that can reflect the quantitative characteristics of the correlation strength value. These characteristics include core indicators such as the magnitude of the correlation strength value, the overall trend of change, the range of fluctuation, and the mean.

[0089] In this embodiment of the application, the specific numerical value, the overall trend of change in the time dimension, the range of numerical fluctuations in different time periods, and the average value of the whole time period are first extracted from the first correlation strength value to form the numerical characteristics of the first correlation strength value; then the specific numerical value, the overall trend of change, the range of fluctuations, and the average value are extracted from the second correlation strength value to form the numerical characteristics of the second correlation strength value.

[0090] Step 313: Compare the numerical features of the first association strength value and the second association strength value item by item to obtain the comparison result. Match the comparison result with the judgment criterion to determine the target influence type. The target influence type is synergistic influence, antagonistic influence, or no significant influence.

[0091] In this step, the comparison result refers to the conclusions on the differences and commonalities in features obtained by comparing the numerical features of the first and second correlation strength values ​​item by item.

[0092] The target influence type refers to the result that characterizes the interaction type between the first and second correlation strength values, reflecting the correlation type between time domain and frequency domain characteristics and concentration fluctuation characteristics.

[0093] In this embodiment, the numerical features of the first and second association strength values ​​are first compared item by item, specifically comparing their numerical ranges, overall trends, and mean ranges to clarify the commonalities and differences between the two types of numerical features and form a complete comparison result. Then, the comparison result is matched one by one with the criteria and conditions for determining the type of influence. If the comparison result meets the criteria and conditions for determining synergistic influence, the target influence type is determined to be synergistic influence. If it meets the criteria and conditions for determining antagonistic influence, the target influence type is determined to be antagonistic influence. If it meets the criteria and conditions for determining no significant influence, the target influence type is determined to be no significant influence.

[0094] Step 314: Based on the criteria for classifying the degree of influence, and combined with the numerical characteristics of the first correlation strength value and the second correlation strength value, determine the degree of influence level corresponding to the target influence type, and form an analysis result by combining the target influence type.

[0095] In this step, the impact level refers to the intensity level of the effect under the same target impact type, determined based on the impact level classification criteria. This level includes three levels: mild, moderate, and severe. There is no corresponding level for no significant impact.

[0096] The analysis results refer to the final conclusions formed after integrating the target impact types and corresponding impact degree levels, representing the impact relationship between the first and second correlation strength values.

[0097] In this embodiment, the influence level corresponding to the target influence type is first determined by combining the numerical characteristics of the first and second correlation strength values ​​and the criteria for classifying the degree of influence. If the target influence type is a synergistic or antagonistic influence, it is determined to be mild, moderate, or severe based on the specific range of the numerical value or difference. If the target influence type is no significant influence, there is no need to determine the degree of influence. Finally, the target influence type and the corresponding degree of influence are integrated to form a complete analysis result. This result can clearly characterize the interaction relationship and intensity between time domain, frequency domain characteristics and concentration fluctuation characteristics.

[0098] Step 304: Using time-domain features, frequency-domain features, and the concentration fluctuation features as nodes, and combining the analysis results, construct a causal relationship map.

[0099] In this embodiment, time-domain features, frequency-domain features, and concentration fluctuation features are first used as the three core nodes of the causal correlation graph. The first correlation strength value and the second correlation strength value are used as the quantitative attributes of the edges connecting the corresponding nodes, and the analysis results are used as the relational attributes of the edges. Each node and edge is clearly labeled in the graph. The nodes are labeled with feature names, and the edges are labeled with the corresponding correlation strength value, target influence type, and influence degree level. The correlation relationship between nodes and edges is built in the above way to form a complete causal correlation graph. This graph can intuitively present the causal logic, correlation tightness, and interaction type between the three types of features.

[0100] The embodiments of this application realize the quantification and visualization of the correlation between physiological characteristics and inflammatory characteristics, providing a clear basis for subsequent analysis of physiological and biochemical rhythm characteristics based on causal association maps, and improving the pertinence and logical rationality of disease risk prediction.

[0101] Step 104: Based on the causal association map, use the biological rhythm analysis model to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data, so as to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features.

[0102] In this step, the biological rhythm analysis model is a dedicated rhythm analysis model designed for chronic inflammatory diseases. The model has a modular structure and includes four core modules: data splitting module, parameter calculation module, coupling judgment module, and feature integration module. The application process of each module can be referred to steps 401 to 404 below, which will not be elaborated in this embodiment.

[0103] The model's preset parameters include thresholds for association strength interval division, thresholds for coupling type determination rules, strength determination weight coefficients, and coupling strength level classification standards, all based on prior medical knowledge and clinical data calibration of chronic inflammatory diseases. The modules work together to achieve the entire process of extracting physiological and biochemical rhythm features, calculating parameter deviations, determining coupling type and strength, and integrating rhythm coupling features, thus overcoming the limitations of single rhythm analysis and adapting to the rhythm coupling analysis needs of multimodal data.

[0104] It should be noted that this embodiment does not limit the specific structure of the biological rhythm analysis model, and can be set accordingly based on the actual situation.

[0105] Physiological rhythm features can be understood as features extracted from physiological signal data that reflect the periodic changes in human physiological indicators over time. These features include core attributes such as the fluctuation period of physiological signals, amplitude variation patterns, and peak occurrence periods, and are suitable for the physiological rhythm analysis needs of chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease.

[0106] Biochemical rhythm characteristics can be understood as features extracted from inflammatory factor concentration data that reflect the periodic changes in human inflammatory factor concentration over time. These characteristics include core attributes such as the fluctuation period of inflammatory factor concentration, peak interval, and concentration rise and fall rhythm, and are key indicators for characterizing the inflammatory activity rhythm of chronic inflammatory diseases.

[0107] Rhythm coupling characteristics can be understood as a comprehensive feature formed by integrating physiological rhythm characteristics and biochemical rhythm characteristics, including coupling type, coupling strength level, core parameter deviation value, and corresponding time period.

[0108] In the embodiments of this application, such as Figure 2 As shown, step 104, based on the causal association map, uses a biorhythm analysis model to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data, in order to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features, specifically including the following steps:

[0109] Step 401: Based on the node association information in the causal association map, the physiological signal data and inflammatory factor concentration data are split into separate data according to a preset time period using a biological rhythm analysis model, and the physiological rhythm features of the physiological signal data and the biochemical rhythm features of the inflammatory factor concentration data are extracted within each preset time period.

[0110] In this step, the preset time period refers to the time interval for splitting data that is pre-set based on the physiological and biochemical rhythm characteristics of chronic inflammatory diseases. This period includes a short period of 24 hours, a medium period of 7 days, and a long period of 30 days, which are adapted to the analysis needs of human diurnal rhythm, weekly rhythm, and monthly rhythm, respectively. It is the core preset parameter of the data splitting module of the biorhythm analysis model.

[0111] In this embodiment, the data splitting module of the biorhythm analysis model first reads the node association information in the causal association map and extracts the time identifier and feature association attributes. Then, according to the three time periods preset by the model (short-term, medium-term, and long-term), the physiological signal data and inflammatory factor concentration data are split into time dimensions to obtain the corresponding subsets for each period. Subsequently, based on the rhythm analysis rules of chronic inflammatory diseases, the module extracts physiological rhythm features such as fluctuation period, amplitude change pattern, and peak occurrence time from the physiological signal subset, and extracts biochemical rhythm features such as concentration fluctuation period, peak interval, and concentration rise and fall rhythm from the inflammatory factor concentration subset. Each feature is bound to the corresponding preset time period, providing a basis for subsequent parameter calculation.

[0112] Step 402: Based on the node association information and the target influence type between nodes, calculate the target parameter deviation value of each core parameter in the physiological rhythm feature and the biochemical rhythm feature.

[0113] In this step, the core parameters refer to key quantitative indicators that can characterize the essential laws of physiological rhythm characteristics and biochemical rhythm characteristics and reflect the coupling state of the two types of rhythm characteristics. The core parameters include the rhythm period, peak amplitude, and trend of physiological rhythm and biochemical rhythm.

[0114] The target parameter deviation value refers to the quantitative value of the deviation of the core parameters of physiological rhythm characteristics and biochemical rhythm characteristics from the reference value of their coordinated change.

[0115] In this embodiment, firstly, based on prior medical knowledge and node association information of the causal association map, the core parameter reference values ​​of the synergistic changes of physiological rhythms and biochemical rhythms are determined through the parameter calculation module of the biorhythm analysis model. These reference values ​​include rhythm cycle reference values, amplitude peak reference values, and change trend reference values. Then, the actual values ​​of each core parameter of the physiological rhythm features and biochemical rhythm features are extracted, and the coefficients of each reference value are corrected in combination with the target influence type between nodes to obtain the corrected reference values. For example, one correction method is to increase the reference value matching coefficient under synergistic influence and decrease the reference value matching coefficient under antagonistic influence. Finally, the actual value of each core parameter is subtracted from the corrected reference value to obtain the target parameter deviation value of each core parameter.

[0116] Step 403: Based on the target parameter deviation value and the node association information, use the biological rhythm analysis model to determine the target coupling type and the corresponding target coupling strength level of physiological rhythm characteristics and biochemical rhythm characteristics.

[0117] In this step, the target coupling type refers to the associated coupling type obtained by matching physiological rhythm features and biochemical rhythm features based on a subset of deviation values ​​and coupling type determination rules. Each type can be synchronous coupling, unidirectional temporal lag coupling, staggered coupling, or disordered coupling.

[0118] The target coupling strength level refers to the coupling strength level obtained by matching the total number of strength judgment values ​​with the preset level classification standard in the biological rhythm analysis model. The higher the level, the higher the degree of coupling abnormality between the two types of rhythm features.

[0119] In this embodiment of the application, step 403 specifically includes the following steps:

[0120] Step 411: According to the preset rules of the biological rhythm analysis model, the correlation strength value in the node correlation information is divided into multiple correlation strength intervals, and a judgment weight is assigned to each correlation strength interval and a judgment coefficient is assigned to different target influence types.

[0121] In this step, the preset rules refer to the unified operational guidelines pre-set for the biorhythm analysis model based on prior medical knowledge and clinical rhythm data related to chronic inflammatory diseases. These guidelines guide each module of the model in completing data processing and analysis. The preset rules include rules for dividing the correlation strength interval, rules for allocating judgment weights and coefficients, and rules for extracting rhythm features. The correlation strength interval refers to the interval divided according to the numerical range of the correlation strength value.

[0122] The judgment weight refers to the weight value assigned to each correlation strength interval for subsequent strength calculation.

[0123] The determination coefficient refers to the coefficient value assigned to different target impact types and used to correct the intensity calculation results.

[0124] In this embodiment, the association strength value is first extracted from the node association information. Combined with the preset rules of the biological rhythm analysis model, the association strength value is divided into multiple association strength intervals through the interval division unit of the coupling judgment module of the model. For example, one division method can be 0-0.4, 0.4-0.7, and 0.7-1.0, which correspond to low, medium, and high association strength intervals, respectively. Then, a judgment weight is assigned to each association strength interval. For example, one assignment method can be that the weight of the low association strength interval is 0.2, the weight of the medium association strength interval is 0.5, and the weight of the high association strength interval is 0.9. At the same time, a judgment coefficient is assigned to different target influence types. For example, one assignment method can be that the judgment coefficient for synergistic influence is 1.2, the judgment coefficient for antagonistic influence is 0.8, and the judgment coefficient for no significant influence is 1.0.

[0125] Step 412: Based on the correlation logic between different features, classify the target parameter deviation values ​​to obtain multiple subsets of deviation values.

[0126] In this step, the deviation value subset refers to the grouping of deviation values ​​obtained after classifying the target parameter deviation values ​​based on the association logic of different features. Each group corresponds to a core parameter association type of physiological rhythm and biochemical rhythm.

[0127] In this embodiment, based on the correlation logic between physiological rhythm characteristics and biochemical rhythm characteristics in chronic inflammatory diseases, the deviation classification unit of the coupling judgment module of the biological rhythm analysis model divides the target parameter deviation values ​​into a subset of rhythm period deviation values, a subset of amplitude peak deviation values, and a subset of change trend deviation values ​​according to the core parameter type. Each subset includes all target parameter deviation values ​​corresponding to the core parameter and is labeled according to a preset time period.

[0128] Step 413: Use the biological rhythm analysis model to set the rules for determining the coupling type.

[0129] In this step, the coupling type determination rule refers to the quantitative criteria that are pre-set based on medical prior knowledge and the rhythm coupling law of chronic inflammatory diseases, used to determine the coupling type of physiological and biochemical rhythms. The coupling types include synchronous coupling, unidirectional temporal lag coupling, staggered coupling, and disordered coupling.

[0130] In this embodiment, firstly, based on clinical rhythm data of chronic inflammatory diseases, the coupling type determination rules are set through the rule setting unit of the coupling determination module of the biorhythm analysis model. For example, one setting method is to set synchronous coupling as follows: the mean of the subset of rhythmic cycle deviation values, the subset of amplitude peak deviation values, and the subset of trend deviation values ​​are all in the low deviation range of 0-0.4, and the changing trends of physiological rhythm characteristics and biochemical rhythm characteristics are completely consistent with each other, with no time difference; unidirectional temporal lag coupling is set as follows: the mean of one type of deviation value subset is in the low deviation range of 0-0.4. In the low-bias range, the mean values ​​of the other two types of deviation value subsets are in the medium-bias range of 0.4-0.7, and there is a fixed time lag between physiological rhythm features and biochemical rhythm features. When interleaved coupling is defined as the mean values ​​of all three types of deviation value subsets being in the medium-bias range of 0.4-0.7, and the changing trends of physiological rhythm features and biochemical rhythm features alternating without a fixed synchronous or lagging pattern, the following definition is used. When disordered coupling is defined as the mean values ​​of all three types of deviation value subsets being in the high-bias range of 0.7-1.0, and there is no obvious periodic or trend correlation between the changes in physiological rhythm features and biochemical rhythm features.

[0131] Step 414: Compare each subset of deviation values ​​with the determination rules for the coupling type to match the corresponding target coupling type.

[0132] In this step, the target coupling type refers to the coupling type matched by comparing physiological rhythm features and biochemical rhythm features according to rules.

[0133] In this embodiment, the mean deviation value of each subset of deviation values ​​is first calculated and its deviation interval is determined. Then, the distribution of deviation intervals and the trend of rhythm changes are compared with the coupling type determination rules in the model rule base one by one. The corresponding coupling type is matched according to the principle of best matching the rule, and the coupling type is determined as the target coupling type. Each target coupling type is bound to the corresponding preset time period.

[0134] Step 415: Calculate the intensity judgment value of each subset of deviation values ​​according to the judgment weight and the judgment coefficient, and summarize all intensity judgment values ​​to obtain the total intensity judgment value.

[0135] In this step, the strength judgment value refers to the quantitative value calculated by combining the judgment weight and judgment coefficient of each subset of deviation values, which reflects the rhythmic coupling strength of a single core parameter.

[0136] The total strength assessment value refers to the comprehensive quantitative value obtained by summing up all strength assessment values, which is the core basis for determining the coupling strength level.

[0137] In this embodiment, the absolute value of the mean deviation value is first multiplied by the judgment weight and then by the judgment coefficient to obtain the intensity judgment value of each subset; then, all intensity judgment values ​​under the same preset time period are summed to obtain the total intensity judgment value under that period. The higher the value, the greater the deviation of the rhythm coupling.

[0138] Step 416: Match the total number of intensity determination values ​​with the preset level classification standard in the biological rhythm analysis model to determine the target coupling intensity level corresponding to the target coupling type.

[0139] In this step, the grading standard refers to the interval criteria for the total number of strength judgment values ​​that are pre-set by the biorhythm analysis model to determine the coupling strength level.

[0140] The target coupling strength level refers to the coupling strength level that matches the total value of the strength judgment. This level includes level one, level two, level three, and level four. The higher the level, the higher the degree of abnormality of the rhythm coupling.

[0141] In this embodiment, the model first reads the preset level classification standard. For example, a level classification standard may be that the total strength judgment value of 0-0.3 is level 1, 0.3-0.6 is level 2, 0.6-0.9 is level 3, and greater than 0.9 is level 4. Then, the total strength judgment value under each preset time period is compared with the standard, the corresponding level is matched and determined as the target coupling strength level, and the level is bound to the corresponding target coupling type and preset time period.

[0142] Step 404: Integrate the target coupling type, target coupling strength level, target parameter deviation value and corresponding preset time period to form rhythmic coupling features.

[0143] In this embodiment, firstly, according to the preset time period as the classification dimension, the feature integration module of the biological rhythm analysis model structurally integrates the target coupling type, target coupling strength level, and all target parameter deviation values ​​under each period to generate a rhythm coupling feature subset corresponding to each preset time period; then, the rhythm coupling feature subsets of each preset time period are summarized to form a complete rhythm coupling feature.

[0144] The embodiments of this application realize precise analysis of rhythm coupling of multimodal data, providing core analytical basis for subsequent disease risk prediction that fits the characteristics of chronic inflammatory diseases, improving the scientificity and pertinence of risk prediction, while avoiding the problems of insufficient feature analysis and logical discontinuity.

[0145] Step 105: Analyze the rhythm coupling features and node association information in the causal association graph using a containerized model to generate disease risk prediction results for the user.

[0146] In this step, the containerized model refers to a lightweight distributed analysis model designed for risk prediction of chronic inflammatory diseases such as rheumatoid arthritis, chronic bronchitis, and coronary heart disease. The model has a modular and hierarchical structure and includes four core functional modules: input adaptation module, task scheduling module, decision module, and generation module. Each module is configured with an independent analysis unit that corresponds to a time interval. The application process of each module in the containerized model can be referred to steps 401 to 404 below, which will not be elaborated in this embodiment.

[0147] The core parameters preset within the containerized model can include data interface specifications, risk assessment weight allocation thresholds, target parameter deviation value ranges, and disease risk level assessment thresholds, all based on clinical risk data and prior medical knowledge of chronic inflammatory diseases. Each module and analysis unit collaborate to achieve multi-source data structured processing, multi-dimensional risk analysis, risk score calculation, risk level assessment, and prediction result integration. The analysis unit pre-encapsulates risk analysis logic for the corresponding time dimension, supports parallel analysis across multiple time intervals, and ensures analysis efficiency and accuracy.

[0148] Node association information refers to the set of association attributes and interaction logics between feature nodes extracted from the causal association graph. This information includes the association strength value, target influence type, and influence level of each node, and also covers the intrinsic interaction relationships between three types of nodes: time domain features, frequency domain features, and concentration fluctuation features.

[0149] Disease risk prediction results refer to the standardized analysis results generated after multi-dimensional analysis by a containerized model. These results include information such as risk level, risk score, risk characteristic basis, and rhythm coupling anomalies, providing complete and traceable quantitative basis for clinical risk assessment of chronic inflammatory diseases.

[0150] In this embodiment of the application, step 105 specifically includes the following steps:

[0151] Step 501: The rhythm coupling features and node association information are structurally bound through the input adaptation module of the containerized model to generate an input data packet that conforms to the data interface specification of the containerized model.

[0152] In this step, the input adaptation module refers to the front-end data processing module of the containerized model, which is responsible for the format conversion, structure binding and interface adaptation of multi-source data. It is the bridge connecting the original feature information and the core analysis module of the model.

[0153] The data interface specification refers to the pre-defined multi-source data access format guidelines for containerized models. This specification includes unified requirements for field names, data types, time dimension identifiers, and feature attribute classifications.

[0154] The input data packet refers to the unified data carrier generated after structured binding and format verification, which integrates rhythm coupling characteristics, node association information and corresponding time dimension identifiers.

[0155] In this embodiment, core attribute fields are first extracted from rhythmic coupling features and node association information. These fields include a preset time period, target coupling type, target coupling strength level, target parameter deviation value, association strength value, and target influence type. Then, according to the preset data interface specifications of the containerized model, the input adaptation module of the containerized model performs time-dimensional attribute association and structured mapping on these core attribute fields, binding the field information of the same preset time period into a group. Subsequently, missing fields are standardized and completed, and abnormal data is verified for compliance. After the verification is passed, an input data package conforming to the model data interface specifications is generated.

[0156] Step 502: The task scheduling module of the containerized model splits the input data packet into different analysis units according to the duration dimension of the preset time period, assigns risk judgment weights to different target coupling types, and establishes numerical correlation rules between the target coupling strength level and the target parameter deviation value. Each analysis unit pre-encapsulates the risk analysis logic corresponding to the duration dimension.

[0157] In this step, the task scheduling module refers to the core scheduling module of the containerized model, which is responsible for dividing the time interval, matching analysis units, splitting input data packets, and distributing analysis tasks.

[0158] Risk assessment weights refer to the quantitative weight values ​​assigned to different target coupling types for calculating disease risk scores. They reflect the degree of contribution of different coupling types to the risk of chronic inflammatory diseases, and these weight values ​​are positively correlated with the degree of disease risk association.

[0159] Numerical association rules refer to rules that characterize the quantitative correspondence between the target coupling strength level and the target parameter deviation value, and are defined by the analysis unit based on pre-encapsulated risk analysis logic.

[0160] An analysis unit refers to an independent risk analysis unit in a containerized model, which corresponds to a time interval divided by the time dimension. Each analysis unit is pre-encapsulated with risk analysis logic adapted to the corresponding time interval.

[0161] In this embodiment of the application, step 502 specifically includes the following steps:

[0162] Step 511: The preset time period in the input data packet is divided into multiple time intervals according to different time length dimensions through the task scheduling module of the containerized model. Each time interval corresponds to an analysis unit, and the risk analysis logic of each analysis unit is adapted to the correlation analysis logic between the target coupling type and the disease risk within the corresponding time interval.

[0163] In this step, the duration interval refers to the independent time interval that subdivides the preset time period into short, medium, and long duration dimensions. It is the time basis for the multidimensional risk analysis of the containerized model. This embodiment does not limit the numerical value and number of dimensions of each duration dimension, and can be set accordingly according to the actual situation.

[0164] The logic of the correlation analysis between target coupling type and disease risk refers to the intrinsic correlation criteria between each target coupling type and the risk of chronic inflammatory diseases, which are summarized based on medical prior knowledge and adapted to a specific time interval. It reflects the influence of coupling type on disease risk under different time intervals.

[0165] In this embodiment, the preset time period identifier in the input data packet is first read by the task scheduling module of the containerized model. The preset time period in the input data packet is divided into three independent time intervals according to the length dimensions of short time (24 hours), medium time (7 days), and long time (30 days). Then, an independent analysis unit built into the containerized model is matched for each time interval to complete the exclusive binding between the analysis unit and the time interval. Subsequently, the compatibility of the pre-encapsulated risk analysis logic of each analysis unit with the correlation analysis logic of target coupling type and disease risk in the corresponding time interval is verified to ensure that the analysis logic such as the disordered coupling of disease risk correlation in the long time interval and the synchronous coupling of disease risk correlation in the short time interval is consistent with clinical patterns.

[0166] Step 512: Based on the correspondence between each analysis unit and the time interval, the input data packet is split into multiple sub-data packets.

[0167] In this step, the correspondence between the analysis unit and the time interval refers to the matching relationship between the three time intervals of short time / medium time / long time preset by the containerized model and the three independent analysis units. Each time interval corresponds to only one dedicated analysis unit, ensuring the uniqueness of data splitting and analysis.

[0168] Subdata packets refer to data packets generated after the input data packets are split according to time intervals, and are specifically matched to each analysis unit.

[0169] In this embodiment, the input data packet is first split based on the correspondence between the duration interval and the analysis unit, and all core attribute fields corresponding to the short, medium and long duration intervals are extracted respectively. Then, each split data packet is labeled with a unique analysis unit identifier and duration interval identifier to form an initial sub-data packet. Subsequently, the initial sub-data packet is checked for data integrity, invalid data is removed, and sub-data packets that can be directly parsed by each analysis unit are generated. Each sub-data packet is then distributed to the corresponding dedicated analysis unit.

[0170] Step 513: Based on the risk analysis logic pre-encapsulated in each analysis unit, assign risk judgment weights to the target coupling type in each sub-data packet, and divide the numerical range of target parameter deviation values ​​under different target coupling strength levels, so as to establish a numerical correlation rule between the target coupling strength level and the target parameter deviation value.

[0171] In this step, the numerical range refers to the range of target parameter deviation values ​​defined for different target coupling strength levels.

[0172] In this embodiment, after each analysis unit receives the corresponding sub-data packet, it calls its own pre-encapsulated risk analysis logic for the corresponding duration dimension. Based on the clinical risk patterns of chronic inflammatory diseases, it assigns risk judgment weights to the four target coupling types in the sub-data packet: synchronous coupling, unidirectional temporal lag coupling, interleaved coupling, and disordered coupling.

[0173] Next, according to different target coupling strength levels, the numerical range of the target parameter deviation value corresponding to each target coupling strength level is defined. For example, one way to divide it is that the numerical range of the first-level coupling strength level is 0-0.4, the numerical range of the second-level coupling strength level is 0.4-0.6, the numerical range of the third-level coupling strength level is 0.6-0.9, and the numerical range of the fourth-level coupling strength level is greater than 0.9. Then, based on the defined numerical range, the corresponding quantitative relationship between the target coupling strength level and the target parameter deviation value is established, forming a numerical association rule suitable for the corresponding duration dimension.

[0174] Step 503: Based on the numerical association rules and the risk judgment weights, calculate the individual risk scores for different preset time periods, and then weight and integrate the individual risk scores to obtain a comprehensive risk score.

[0175] In this step, the individual risk score refers to the quantitative value calculated by each analysis unit based on numerical association rules and risk judgment weights, which reflects the degree of risk of chronic inflammatory diseases in a single time interval. The higher the value, the higher the disease risk in the corresponding time interval.

[0176] The comprehensive risk score is a quantitative value that reflects the overall risk level of a user's chronic inflammatory diseases by weighting and summing the individual risk scores of each preset time period according to preset integration weights. It is the core basis for determining the disease risk level.

[0177] In this embodiment, each analysis unit first extracts the target coupling strength level corresponding to the target parameter deviation value in the sub-data packet based on numerical association rules, and calculates the mean of the target parameter deviation value; then, the mean is multiplied by the risk judgment weight assigned to the corresponding target coupling type to obtain the individual risk score for the corresponding preset time period. Each analysis unit feeds back the calculation results to the task scheduling module of the containerized model; subsequently, the task scheduling module collects the individual risk scores for three time intervals: short time, medium time, and long time. Based on the risk pattern of chronic inflammatory diseases, it assigns an integration weight to each time interval, and then multiplies each individual risk score by the corresponding integration weight and sums them to obtain the user's comprehensive risk score.

[0178] Step 504: Based on prior medical knowledge and the combined characteristics of different target impact types and target coupling types, the judgment module of the containerized model generates a judgment threshold range for the disease risk level, so as to determine the target disease risk level to which the comprehensive risk score belongs.

[0179] In this step, the judgment module refers to the core module for risk level judgment of the containerized model. It is responsible for generating the threshold range for disease risk level judgment based on prior medical knowledge and combined features, and completing the matching of comprehensive risk score and risk level.

[0180] The combined features of target impact type and target coupling type refer to the feature set formed by combining different target impact types and target coupling types, reflecting the comprehensive impact of the association type of physiological and inflammatory features on disease risk.

[0181] The threshold range for determining disease risk level refers to the range of comprehensive risk scores generated based on prior medical knowledge and combined characteristics, used to classify the risk level of chronic inflammatory diseases. It is a quantitative standard for determining risk level.

[0182] The target disease risk level refers to the disease risk level obtained by substituting the comprehensive risk score into the judgment threshold range. This risk level includes low risk, medium risk, high risk, and very high risk.

[0183] In this embodiment, firstly, based on prior medical knowledge and clinical risk grading standards for chronic inflammatory diseases, and combined with the combined characteristics of target impact type and target coupling type, the judgment module of the containerized model adjusts the comprehensive risk score range corresponding to high-risk combinations upwards and the range corresponding to low-risk combinations downwards. Based on the above adjustment method, a judgment threshold range for disease risk level is generated. Here, high-risk combinations can be combinations of antagonistic impact types and disordered coupling types, and low-risk combinations can be combinations of synergistic impact types and synchronous coupling types. For example, a judgment threshold range can be a low-risk score range of 0-0.2, a medium-risk score range of 0.2-0.5, a high-risk score range of 0.5-0.8, and an extremely high-risk score range greater than 0.8. Then, the comprehensive risk score is substituted into this threshold range, and the corresponding risk level is matched as the user's target disease risk level.

[0184] Step 505: Integrate the target disease risk level, comprehensive risk score, individual risk scores for different preset time periods, and corresponding rhythm coupling features and node association information through the containerized model generation module to generate disease risk prediction results.

[0185] In this step, the generation module refers to the result integration and output module of the containerized model, which is responsible for the structured integration of various risk analysis results and original feature information to generate standardized disease risk prediction results.

[0186] In this embodiment, rhythm coupling anomalies and high-risk feature combinations are first extracted from the rhythm coupling features and node association information corresponding to each individual risk score. Then, according to the standardized report format preset by the model, the target disease risk level, comprehensive risk score, individual risk scores for each preset time period, rhythm coupling anomalies, and high-risk feature combinations are structured and integrated, and a brief analysis of risk causes and clinical reference suggestions are added to finally generate a complete disease risk prediction result.

[0187] The embodiments of this application achieve accurate and efficient prediction of disease risk, providing comprehensive quantitative evidence for clinical diagnosis and treatment, while avoiding the problems of insufficient risk analysis and logical gaps, thus improving the scientificity and practicality of risk prediction.

[0188] Figure 3 This is a schematic diagram illustrating a specific implementation of a disease risk prediction system based on multimodal data fusion, as provided in this application embodiment. (Refer to...) Figure 3 The system may include:

[0189] The acquisition module 31 is used to collect physiological signal data and inflammatory factor concentration data of the user in daily life.

[0190] Extraction module 32 is used to extract time-domain features and frequency-domain features from the physiological signal data, and to extract concentration fluctuation features from the inflammatory factor concentration data;

[0191] The construction module 33 is used to calculate the correlation strength values ​​between the time domain features and frequency domain features and the concentration fluctuation features, respectively, and, combined with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data to construct a causal correlation map.

[0192] The parsing module 34 is used to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data based on the causal association map using a biorhythm analysis model, so as to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features.

[0193] The generation module 35 is used to analyze the rhythm coupling features and node association information in the causal association graph through a containerized model to generate disease risk prediction results for the user.

[0194] This application provides a disease risk prediction system based on multimodal data fusion to implement the aforementioned disease risk prediction method based on multimodal data fusion. Therefore, the specific implementation of the disease risk prediction system based on multimodal data fusion can be found in the embodiment section of the disease risk prediction method based on multimodal data fusion described above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0195] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described methods for predicting disease risk based on multimodal data fusion.

[0196] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described methods for predicting disease risk based on multimodal data fusion.

[0197] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0198] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the disease risk prediction method based on multimodal data fusion described above.

[0199] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0200] The above provides a detailed description of a disease risk prediction method and system based on multimodal data fusion provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A disease risk prediction method based on multimodal data fusion, characterized in that, include: Collect physiological signal data and inflammatory factor concentration data of users in their daily state; Temporal and frequency domain features are extracted from the physiological signal data, and concentration fluctuation features are extracted from the inflammatory factor concentration data. The correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features are calculated respectively. Combined with prior medical knowledge, the correlation between the physiological signal data and the inflammatory factor concentration data is analyzed to construct a causal correlation map. Based on the causal relationship map, the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data are analyzed using a biological rhythm analysis model to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features. By analyzing the rhythm coupling features and node association information in the causal relationship graph using a containerized model, the user's disease risk prediction results are generated.

2. The disease risk prediction method based on multimodal data fusion according to claim 1, characterized in that, The association strength value includes a first association strength value and a second association strength value; Calculate the correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features, respectively. Combined with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data, and construct a causal correlation map, including: The direction of change of the temporal features and concentration fluctuation features at each time point is compared. If the direction of change is consistent, it is marked as a synchronous event. Combined with the total number of time points, the first correlation strength value between the temporal features and concentration fluctuation features is calculated. The direction of change of the frequency domain features and concentration fluctuation features at each time point is compared to calculate the second correlation strength value between the frequency domain features and concentration fluctuation features; Based on the characteristic change patterns in medical prior knowledge, the influence relationship between the first and second correlation strength values ​​is analyzed, and the analysis results are obtained. Using time-domain features, frequency-domain features, and concentration fluctuation features as nodes, and combining the analysis results, a causal relationship map is constructed.

3. The disease risk prediction method based on multimodal data fusion according to claim 2, characterized in that, Based on the characteristic change patterns in prior medical knowledge, the influence relationship between the first and second correlation strength values ​​is analyzed, and the analysis results are obtained, including: Based on prior medical knowledge and the logical relationships between different characteristics, the criteria for determining the type of impact, the defining conditions, and the basis for classifying the degree of impact are determined. Extract the numerical features of the first correlation strength value and the numerical features of the second correlation strength value respectively; The numerical features of the first association strength value and the second association strength value are compared item by item to obtain the comparison result. The comparison result is matched with the judgment criterion to determine the target influence type, which is a synergistic influence, an antagonistic influence, or no significant influence. Based on the criteria for classifying the degree of influence, and combined with the numerical characteristics of the first and second correlation strength values, the degree of influence corresponding to the target influence type is determined, and the analysis results are formed by combining the target influence type.

4. The disease risk prediction method based on multimodal data fusion according to claim 1, characterized in that, Based on the causal relationship map, a biorhythm analysis model is used to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data to determine the rhythmic coupling features between the physiological rhythm features and the biochemical rhythm features, including: Based on the node association information in the causal association map, the physiological signal data and inflammatory factor concentration data are split into separate data according to a preset time period using a biological rhythm analysis model, and the physiological rhythm features of the physiological signal data and the biochemical rhythm features of the inflammatory factor concentration data are extracted within each preset time period. Based on the node association information and the target influence type between nodes, calculate the target parameter deviation values ​​of each core parameter in the physiological rhythm characteristics and biochemical rhythm characteristics; Based on the target parameter deviation value and the node association information, the target coupling type and corresponding target coupling strength level of physiological rhythm characteristics and biochemical rhythm characteristics are determined using a biological rhythm analysis model. The target coupling type, target coupling strength level, target parameter deviation value, and corresponding preset time period are integrated to form a rhythmic coupling feature.

5. The disease risk prediction method based on multimodal data fusion according to claim 4, characterized in that, Based on the target parameter deviation value and the node association information, the target coupling type and corresponding target coupling strength level of physiological rhythm characteristics and biochemical rhythm characteristics are determined using a biological rhythm analysis model, including: According to the preset rules of the biological rhythm analysis model, the association strength value in the node association information is divided into multiple association strength intervals, and a judgment weight is assigned to each association strength interval and a judgment coefficient is assigned to different target influence types. Based on the correlation logic between different features, the target parameter deviation values ​​are classified to obtain multiple subsets of deviation values; Establish rules for determining coupling type using a biological rhythm analysis model; Each subset of deviation values ​​is compared with the determination rule of the coupling type to match the corresponding target coupling type; Based on the judgment weight and the judgment coefficient, calculate the intensity judgment value of each subset of deviation values, and summarize all intensity judgment values ​​to obtain the total intensity judgment value. The total number of intensity determination values ​​is matched with the preset level classification standard in the biological rhythm analysis model to determine the target coupling intensity level corresponding to the target coupling type.

6. The disease risk prediction method based on multimodal data fusion according to claim 1, characterized in that, By analyzing the rhythm coupling features and node association information in the causal correlation graph using a containerized model, the user's disease risk prediction results are generated, including: The rhythm coupling features and node association information are structurally bound through the input adaptation module of the containerized model to generate an input data packet that conforms to the data interface specification of the containerized model. The task scheduling module of the containerized model splits the input data packet into different analysis units according to the duration dimension of the preset time period, assigns risk judgment weights to different target coupling types, and establishes numerical correlation rules between the target coupling strength level and the target parameter deviation value. Each analysis unit pre-encapsulates the risk analysis logic corresponding to the duration dimension. Based on the numerical association rules and the risk assessment weights, individual risk scores for different preset time periods are calculated, and the individual risk scores are weighted and integrated to obtain a comprehensive risk score. Based on prior medical knowledge and the combined characteristics of different target impact types and target coupling types, the judgment module of the containerized model generates a threshold range for judging the disease risk level, so as to determine the target disease risk level to which the comprehensive risk score belongs. The containerized model generation module integrates the target disease risk level, comprehensive risk score, individual risk scores for different preset time periods, and corresponding rhythm coupling features and node association information to generate disease risk prediction results.

7. The disease risk prediction method based on multimodal data fusion according to claim 6, characterized in that, The task scheduling module of the containerized model splits the input data packets into different analysis units according to the duration dimension of a preset time period, assigns risk judgment weights to different target coupling types, and establishes numerical correlation rules between target coupling strength levels and target parameter deviation values, including: The task scheduling module of the containerized model divides the preset time period in the input data packet into multiple time intervals according to different time length dimensions. Each time interval corresponds to an analysis unit, and the risk analysis logic of each analysis unit is adapted to the correlation analysis logic between the target coupling type and the disease risk within the corresponding time interval. Based on the correspondence between each analysis unit and the time interval, the input data packet is split into multiple sub-data packets; Based on the risk analysis logic pre-encapsulated in each analysis unit, risk judgment weights are assigned to the target coupling type in each sub-data packet, and the numerical ranges of target parameter deviation values ​​under different target coupling strength levels are divided to establish numerical correlation rules between target coupling strength levels and target parameter deviation values.

8. A disease risk prediction system based on multimodal data fusion, characterized in that, include: The data acquisition module is used to collect physiological signal data and inflammatory factor concentration data of users in their daily state; The extraction module is used to extract time-domain and frequency-domain features from the physiological signal data, and to extract concentration fluctuation features from the inflammatory factor concentration data; The module is used to calculate the correlation strength values ​​between the time-domain features and frequency-domain features and the concentration fluctuation features, respectively, and, combined with prior medical knowledge, analyze the correlation between the physiological signal data and the inflammatory factor concentration data to construct a causal correlation map. The analysis module is used to analyze the physiological rhythm features in the physiological signal data and the biochemical rhythm features in the inflammatory factor concentration data based on the causal association map using a biorhythm analysis model, so as to determine the rhythm coupling features between the physiological rhythm features and the biochemical rhythm features; The generation module is used to analyze the rhythm coupling features and node association information in the causal association graph through a containerized model to generate disease risk prediction results for the user.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a disease risk prediction method based on multimodal data fusion as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements a disease risk prediction method based on multimodal data fusion as described in any one of claims 1 to 7.

Images