Hypertension risk assessment method based on time-frequency component mixing guidance

Abstract

A hypertension risk assessment method based on time-frequency component mixing guidance, the method comprising: constructing an initial model, wherein the initial model comprises a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module; acquiring an initial data set, wherein the initial data set is a time series data set of a plurality of hypertensive patients; on the basis of the initial data set, training the initial model to obtain a target prediction model; and acquiring target data, and on the basis of the target prediction model, performing prediction on the target data to obtain hypertension risk prediction results of target patients, wherein the target data is hypertension-related data of the target patients, and there are a plurality of target patients. In the hypertension risk assessment method based on time-frequency component mixing guidance, a parameter-free component extraction module and feature enhancement module are designed, thereby effectively reducing network parameters, achieving the effect of decoupling data from a model, and also enabling more accurate prediction of the hypertension risk of patients.

Description

A Hypertension Risk Assessment Method Based on Time-Frequency Component Hybrid Guidance Technical Field

[0001] This invention relates to the field of computer-aided diagnostic technology, and in particular to a method for hypertension risk assessment based on time-frequency component hybrid guidance. Background Technology

[0002] With the development of society and economy, people's living standards are constantly improving, and the incidence of chronic diseases such as hypertension is also rising.

[0003] Hypertension is defined as a systolic blood pressure ≥140 mmHg and / or a diastolic blood pressure ≥90 mmHg at rest. It is one of the major risk factors for cardiovascular disease. Long-term hypertension can lead to cardiac hypertrophy, coronary heart disease, myocardial infarction, heart failure, and other diseases. It also increases the risk of cerebrovascular diseases, such as cerebral hemorrhage and cerebral infarction. Furthermore, hypertension may also lead to kidney disease, eye disease, or other complications. Therefore, studying and evaluating an individual's risk of developing hypertension can enable effective interventions to delay the onset of the disease and reduce its harmfulness.

[0004] Time series forecasting plays a crucial role in various application fields. With the rapid development of the Internet of Things (IoT), the scale and dimensionality of collected data are enormous, posing significant challenges to time series analysis and making time series forecasting a key research task in time series data mining and analysis. However, existing methods mainly focus on model design, relying too heavily on the model's adaptive learning capabilities and resulting in excessively high parameter counts. This leads to strong coupling between the data and the model, resulting in overfitting, poor model versatility, low prediction accuracy, and high training costs.

[0005] Therefore, existing technologies still need to be improved and enhanced.

[0006] Summary of the Invention

[0007] To address the aforementioned shortcomings of existing technologies, this paper proposes a hypertension risk assessment method guided by time-frequency component hybridization, aiming to solve the problems of high parameter count and poor model versatility in existing technologies.

[0008] A first aspect of the present invention provides a method for hypertension risk assessment based on time-frequency component hybrid guidance, comprising:

[0009] An initial model is constructed, which includes a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module;

[0010] Obtain an initial dataset, which is a time-series dataset of multiple hypertensive patients;

[0011] The initial model is trained based on the initial dataset to obtain the target prediction model;

[0012] Obtain target data, predict the target data based on the target prediction model, and obtain the hypertension risk prediction result of the target patient. The target data is hypertension-related data of the target patient, and there are multiple target patients.

[0013] In one implementation, the time component mixing module includes a period division module, a period pooling module, a flattening module, and a first splicing module;

[0014] The period division module is used to divide the time series data input to the model according to the period, extract the periodic components, and obtain periodic data;

[0015] The periodic pooling module is used to activate and pool the periodic data obtained by the periodic division module to obtain pooled features;

[0016] The flattening module is used to flatten the pooling features into vectors;

[0017] The first splicing module is used to splice the vectors flattened by the flattening module to obtain time component data.

[0018] In one implementation, the periodic pooling module includes a max pooling module and an average pooling module;

[0019] The max pooling module is used to activate the periodic data based on the Tanh function and then perform max pooling to obtain the main features within each period.

[0020] The average pooling module is used to activate the periodic data based on the Tanh function and then perform average pooling to obtain the average features within each period.

[0021] In one implementation, the frequency component mixing module includes a transformation module, an amplitude calculation module, a component extraction module, a local pooling module, a second splicing module, and an inverse transformation module;

[0022] The transformation module is used to perform a fast Fourier transform on the time series data to obtain frequency domain data;

[0023] The amplitude calculation module is used to obtain the amplitude of the frequency domain data;

[0024] The component extraction module is used to extract the Top-k components from the frequency domain data. The Top-k components are the k frequency components with the largest amplitude in the frequency domain data.

[0025] The local pooling module is used to perform local average pooling on the frequency domain data to obtain the local aggregated frequency components of the frequency domain data.

[0026] The second splicing module is used to splice the Top-k component and the local aggregation frequency component to obtain spliced ​​data;

[0027] The inverse transform module is used to perform an inverse fast Fourier transform on the spliced ​​data to obtain global frequency components.

[0028] In one implementation, the feature enhancement module includes a linear mapping module, a feature introduction module, an activation module, a regularization module, an anti-linear mapping module, a normalization module, and an enhancement module.

[0029] The linear mapping module is used to perform a linear transformation on the time component data and the global frequency component to obtain a target linear feature with uniform dimension.

[0030] The feature introduction module is used to introduce nonlinear features into the target linear features based on the ReLU function to obtain the features to be activated.

[0031] The activation module is used to perform point-to-point activation of the target enhancement features based on the memory matrix to obtain activated features;

[0032] The regularization module is used to regularize the activation features based on Dropout technology to obtain regularized features.

[0033] The inverse linear mapping module is used to inversely map the regularized features to restore them to the original dimension, thereby obtaining the features to be enhanced.

[0034] The normalization module is used to normalize the features to be enhanced in order to maintain the stability of the feature values;

[0035] The enhancement module is used to enhance the normalized features based on the Hadamard product to obtain enhanced features.

[0036] In one implementation, the step of predicting the target data based on the target prediction model to obtain the hypertension risk prediction result for the target patient includes:

[0037] Extract target information from the target data. The target information includes the height, gender, weight, body temperature, pulse, systolic blood pressure, diastolic blood pressure, respiration, alcohol consumption, and family history of the target patient at a first number of time points, wherein the time interval between each time point is the same.

[0038] Based on the target information, the predicted blood pressure values ​​of the target patient are obtained for a second number of future time points corresponding to the target patient, and the predicted blood pressure values ​​include diastolic blood pressure and systolic blood pressure.

[0039] Obtain a target threshold, and based on the target threshold and the predicted blood pressure value, obtain the hypertension risk prediction result for the target patient.

[0040] In one implementation, after obtaining the hypertension risk prediction result for the target patient, the method further includes:

[0041] A risk assessment report and health management strategy are generated based on the hypertension risk level of each target patient.

[0042] Real-time physiological data of each target patient is dynamically monitored and hypertension risk level is predicted to enable personalized health management for each target patient based on the latest risk level.

[0043] A second aspect of the present invention provides a hypertension risk assessment device based on time-frequency component hybrid guidance, comprising:

[0044] The model building module is used to build an initial model, which includes a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module.

[0045] The dataset acquisition module acquires an initial dataset, which is a time-series dataset of multiple hypertensive patients.

[0046] The training module is used to train the initial model based on the initial dataset to obtain the target prediction model;

[0047] The prediction module is used to acquire target data, predict the target data based on the target prediction model, and obtain the hypertension risk prediction result of the target patient. The target data is hypertension-related data of the target patient, and there are multiple target patients.

[0048] A third aspect of the present invention provides a terminal comprising: a processor and a storage medium communicatively connected to the processor, the storage medium being adapted to store a plurality of instructions, and the processor being adapted to invoke the instructions in the storage medium to execute the steps of implementing the hypertension risk assessment method based on time-frequency component hybrid guidance as described in any of the preceding claims.

[0049] In a fourth aspect, the present invention provides a storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the hypertension risk assessment method based on time-frequency component hybridization as described in any of the preceding claims.

[0050] Beneficial Effects: Compared with existing technologies, this invention provides a hypertension risk assessment method guided by time-frequency component mixing. In this method, when assessing the hypertension risk of a target patient, an initial model is constructed, including a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module. Then, an initial dataset is obtained, which is a time-series dataset of multiple hypertension patients. Based on this initial dataset, the initial model is trained to obtain a target prediction model. Finally, hypertension-related data of the target patients is obtained, and the hypertension-related data of the target patients is predicted based on the target prediction model to obtain the hypertension risk prediction result for the target patients. The target data refers to the hypertension-related data of multiple target patients. The hypertension risk assessment method guided by time-frequency component mixing provided by this invention, by constructing a time component mixing module, a frequency component mixing module, and a feature enhancement module, and designing parameter-free component extraction strategies in the time and frequency domains, effectively reduces network parameters and achieves data and model decoupling. Meanwhile, a feature enhancement module is designed to enhance features by increasing the module's versatility and utilizing component information, enabling the network to learn the temporal and frequency local and global patterns in the sequence, resulting in more comprehensive and accurate predictions. Attached Figure Description

[0051] Figure 1 is a flowchart of an embodiment of the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention;

[0052] Figure 2 is a model structure diagram in an embodiment of the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention;

[0053] Figure 3 is a diagram of the period segmentation strategy in an embodiment of the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention;

[0054] Figure 4 shows the maximum pooling effect within a cycle in an embodiment of the hypertension risk assessment method based on time-frequency component hybridization provided by the present invention.

[0055] Figure 5 shows the weekly average pooling effect in an embodiment of the hypertension risk assessment method based on time-frequency component hybridization provided by the present invention.

[0056] Figure 6 shows the Top-k frequency component extraction effect in an embodiment of the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention;

[0057] Figure 7 shows the sequence results of Top-k converted to the time domain in an embodiment of the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention;

[0058] Figure 8 is a schematic diagram of an embodiment of the hypertension risk assessment device based on time-frequency component hybrid guidance provided by the present invention;

[0059] Figure 9 is a structural schematic diagram of an embodiment of the terminal provided by the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0061] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0062] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0063] The present invention provides a hypertension risk assessment method based on time-frequency component hybrid guidance, which can be applied to a terminal with computing capabilities. The terminal can execute the hypertension risk assessment method based on time-frequency component hybrid guidance provided by the present invention to detect and locate the target position in the image to be processed.

[0064] Example 1

[0065] Currently, risk assessment methods for hypertension mainly focus on model design, relying too heavily on the model's adaptive learning capabilities. This leads to strong coupling between data and the model, resulting in overfitting, poor model versatility, low prediction accuracy, and high training costs. Therefore, this embodiment addresses these shortcomings by providing a hypertension risk assessment method guided by time-frequency component mixing. To reduce coupling and achieve more accurate predictions of hypertension and other chronic diseases with increased prediction length, deep learning is utilized to conduct in-depth research on time series prediction problems.

[0066] As shown in Figure 1, the hypertension risk assessment method based on time-frequency component hybrid guidance provided in this embodiment includes the following steps:

[0067] S100. Construct an initial model, which includes a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module.

[0068] Specifically, this embodiment aims to detect and intervene in the development of diseases in advance by jointly predicting the time series of multiple biomarkers.

[0069] First, the initial model is constructed. The structure of the initial model is shown in Figure 2, which mainly includes four main modules: the time component mixing module, the frequency component mixing module, the feature enhancement module, and the prediction module.

[0070] The time component mixing module proposes a novel parameter-free piecewise pooling strategy. This strategy allows the model to capture global smooth trend information while preserving local details of the sequence. The entire process is based on data characteristics and does not require the introduction of additional parameters, significantly reducing the number of parameters and thus better decoupling the data from the model, mitigating overfitting.

[0071] Specifically, referring to Figure 2, the time component mixing module includes a period division module, a period pooling module, a flattening module, and a first splicing module.

[0072] The period division module is used to divide the time series data input to the model according to the period, extract the periodic components, and obtain periodic data;

[0073] Referring to Figure 3, the period segmentation strategy is executed based on the period segmentation module.

[0074] Specifically, in the period division module, the historical observation data is first divided into S segments, where S is set to 8, and each segment contains... There are 10 time steps. Then, all the periodic segments are stacked along the row dimension, where each row corresponds to a continuous subset of the sequence. Each column corresponds to the combination of relatively consistent time steps in each periodic segment. Therefore, for an input sequence X... t After readjusting it, we obtain the sequence group X in the i-th row. i· :

[0075] Among them, X i· This corresponds to the i-th periodic segment. Additionally, the j-th sequence group X is also obtained. ·j :

[0076] Among them, X ·j This is equivalent to combining the j-th time step in all periodic segments, reflecting the value of the same relative moment in each period.

[0077] The periodic pooling module is used to activate and pool the periodic data obtained by the periodic division module to obtain pooled features.

[0078] The periodic pooling module includes a max pooling module and an average pooling module, and the pooling features include local salient component features and global fusion component features.

[0079] The max pooling module is used to activate the periodic data based on the Tanh function and then perform max pooling to obtain the main features within each period.

[0080] Based on the maximum pooling module, the maximum pooling strategy within the execution period is executed. Referring to Figure 4, Figure 4 is a diagram of the maximum pooling effect within the period.

[0081] Specifically, each period segment after segmentation contains local information within the period. To highlight significant features, this embodiment employs positive and negative max pooling to simulate the extraction of local features. Unlike traditional max pooling, considering that negative values ​​may also be significant, this embodiment designs a variant of max pooling.

[0082] First, the Tanh activation function is applied to the sequence, distributing the values ​​around the zero axis. By mapping the values ​​to the zero axis before pooling, this can be effectively combined with pooling operations that extract the largest positive or negative values, thus accurately extracting time steps with prominent features. Next, a pooling kernel of size 8 and stride 4 is set to highlight significant data by retaining the largest positive or negative values ​​within the range. Similarly, the same operation is performed on each row. As shown in Figure 4, the red sequence segments represent the results of combining pooling kernel data with the same relative position in each period segment, yielding the resulting features. The changes are obvious. The green line segments represent pooling operations that didn't account for negative values, leading to incorrect data extraction. Subsequently, the max pooling features are expanded to reveal key local information:

[0083] in, This refers to the local salient component features obtained by applying a max-pooling operation within a period. Maxpool1d ± This indicates that the maximum positive or maximum negative value is retained.

[0084] The average pooling module is used to activate the periodic data based on the Tanh function and then perform average pooling to obtain the average features within each period.

[0085] The average pooling module executes the weekly average pooling strategy. Referring to Figure 5, Figure 5 is a diagram of the weekly average pooling effect.

[0086] Specifically, X after segmentation ·j The combination of each time step within a given period segment corresponds to the trend result generated by combining time steps of the same time step across different periods. Averaging pooling (AvgPool1d) with a pooling kernel of size 8 and a step size of 4 is used to smooth the sequence and fuse information from the same relative time step across different periods. Similarly, the same averaging pooling operation is performed for each time step to obtain the trend of each time step within the entire time series. As shown in Figure 5, assuming a period has three time steps, the three different colored sequence segments represent the result of averaging pooling the data from the same time step across different period segments after segmentation, thus simulating the fusion of the global trend across various time steps. Subsequently, the features are expanded to obtain the features that fuse global information:

[0087] in, The global fused component features are obtained by applying average pooling operations over a period of time.

[0088] The flattening module is used to flatten the pooling features into vectors;

[0089] The first splicing module is used to splice the vectors flattened by the flattening module to obtain time component data.

[0090] Specifically, after the time component data is obtained by splicing, the time component data will be fused and enhanced by the feature enhancement module. The original features will be enhanced by the local salient features and the global fusion features to obtain time-enhanced features containing rich salient local features and global trend features.

[0091] In this embodiment, the frequency component mixing module includes a transformation module, an amplitude calculation module, a component extraction module, a local pooling module, a second splicing module, and an inverse transformation module.

[0092] The frequency component mixing module is used to extract and fuse global component features in the enhanced frequency domain. To reduce model parameters, decouple features, and effectively utilize frequency information, the frequency component mixing module employs a local frequency pooling strategy to fuse the amplitude of frequency components, simulating the extraction and processing of local information in the frequency domain. The use of local pooling helps to retain information distributed across multiple frequencies, thereby reducing the risk of information loss. The frequency components generated by pooling are similar to the block aggregation of the original frequency data, without directly performing inverse Fourier transform operations, thus enhancing the original features without introducing additional noise. In addition, the frequency component mixing module also employs a Top-k frequency component extraction method to simulate the extraction of global information in the frequency domain. This is used to suppress noise present in the entire sequence, reduce information redundancy, and obtain global components. In this way, by simulating the extraction of local and global features in a parameterless manner, the model complexity is significantly reduced, achieving the purpose of decoupling and improving the cost-effectiveness of network usage.

[0093] Specifically, the frequency component mixing module includes a transformation module, an amplitude calculation module, a component extraction module, a local pooling module, a second splicing module, and an inverse transformation module.

[0094] The transformation module is used to perform a fast Fourier transform on the time series data to obtain frequency domain data.

[0095] The amplitude calculation module is used to obtain the amplitude of the frequency domain data;

[0096] The component extraction module is used to extract the Top-k components from the frequency domain data. The Top-k components are the top k frequency components with the largest amplitude in the frequency domain data.

[0097] In this embodiment, to remove information deemed invalid or noisy from the sequence, filtering is performed based on the amplitude of the frequencies, filtering out and retaining the frequencies corresponding to the k largest amplitude values. Based on this, only the main components containing the sequence fitting trend can be extracted.

[0098] Specifically, the complex periodic function f(t) is represented by a sum of trigonometric functions. The Fourier series can be expressed as:

[0099] Where t and A represent time and amplitude, and ω and ψ represent angular frequency and initial phase, respectively. n represents different periods or frequencies (integer multiples of the original periodic function). It can be seen that the higher the frequency, the faster the sine wave changes, and the smaller the amplitude. Therefore, the faster the change, the greater the likelihood that the waveform represents noise. This indicates that removing the high-frequency components from the time series can effectively remove noise while retaining most of the information in the fitted curve. This suggests that low-frequency data typically has larger amplitudes and slower sine wave changes, representing the main information of the sequence. It is worth noting that this chapter does not directly extract the low-frequency components, but rather selects the k largest frequencies based on their amplitude. This is because real-world data is often complex, and a small amount of high-frequency data itself belongs to the main frequencies, which can help extract the fitted curve.

[0100] Referring to Figure 6, which shows the extraction effect of the Top-k frequency components (k in Figure 6 is set to 5), it can be seen that selecting the Top-k frequency can effectively remove noise and maintain the overall trend of the sequence.

[0101] The local pooling module is used to perform local average pooling on the frequency domain data to obtain the local aggregated frequency components of the frequency domain data.

[0102] Specifically, the local pooling module executes a frequency local pooling strategy. Specifically, assuming all information in the frequency range is available, the amplitudes are averaged and merged to aggregate adjacent frequency information, thereby enabling parameter-free correlation of the frequency components in the features to obtain a detailed local information fusion representation.

[0103] First, calculate the frequency domain data f. t Real numbers in complex numbers and imaginary numbers The sum of squares of parts is used to simulate the magnitude of the amplitude:

[0104] Where μ is set to 10 -6 To prevent the amplitude Amp calculated through the sum of squares from becoming too small and causing gradient vanishing in subsequent calculations, amplitude average pooling (AvgPool1d) with a kernel size of 8 and a stride of 4 is then used to fuse local features of the amplitude.

[0105] in, It is the local aggregation frequency component after average pooling.

[0106] The second splicing module is used to splice the Top-k component and the local aggregation frequency component to obtain spliced ​​data;

[0107] The inverse transform module is used to perform an inverse fast Fourier transform on the spliced ​​data to obtain global frequency components.

[0108] Referring to Figure 7, which shows the time-domain sequence result after performing an inverse Fourier transform (IFFT) on the selected frequencies (in Figure 7, k is set to 8), this allows for the ignoring of local noise details while preserving key frequency band information, selecting and extracting global frequency features that can fit the sequence trend. Specifically, the selection is based on the amplitude (Amp):

[0109] in, This represents the extracted global frequency components.

[0110] Specifically, after splicing the global frequency components, the original features are enhanced using the local aggregated frequency components and the global frequency component features to obtain frequency-enhanced features containing rich local and global frequency information.

[0111] After obtaining the time enhancement feature and the frequency enhancement feature, the method further includes: using the frequency enhancement feature to guide the time-based feature to obtain the time-frequency enhancement feature, and then enhancing the time-frequency enhancement feature based on the feature enhancement module.

[0112] The feature enhancement module includes a linear mapping module, a feature introduction module, an activation module, a regularization module, an anti-linear mapping module, a normalization module, and an enhancement module;

[0113] The linear mapping module is used to perform a linear transformation on the time component data and the global frequency component to obtain a target linear feature with uniform dimension.

[0114] The feature introduction module is used to introduce nonlinear features into the target linear features based on the ReLU function to obtain the features to be activated.

[0115] The activation module is used to perform point-to-point activation of the target enhancement features based on the memory matrix to obtain activated features;

[0116] The regularization module is used to regularize the activation features based on Dropout technology to obtain regularized features.

[0117] The inverse linear mapping module is used to inversely map the regularized features to restore them to the original dimension, thereby obtaining the features to be enhanced.

[0118] The normalization module is used to normalize the features to be enhanced in order to maintain the stability of the feature values;

[0119] The enhancement module is used to enhance the normalized features based on the Hadamard product to obtain enhanced features.

[0120] Specifically, referring to Figure 2, the components are first linearly combined, and then their dimensionality is reduced to M along the time axis. In this embodiment, M is set to 128 to capture their linear relationships and facilitate information fusion.

[0121] Where, X′ t and X″ t These are the sampled local and global components, H t It is the feature after information interaction and dimensionality reduction through linear projection. Indicates a splicing operation, Φ t ,b t σ represents the learnable parameters of the linear layer. t (·) represents the Tanh activation function, which maps input values ​​to the range between -1 and 1, making it more suitable for network design.

[0122] In other words, the generated information interaction features and the randomly initialized memory matrix Element-wise multiplication of the matrices is performed. Since the sequence was previously divided into periodic segments, the time steps within each period are relatively consistent. Therefore, this step allows the network to learn an adaptive mechanism to select features to be activated in a point-to-point manner, thereby increasing the module's generality. Next, a linear transformation is applied to the activated features for information interaction and projection back to their original dimensions, and a Softmax activation function is applied for normalization to obtain the enhancement score. This indicates the combined contribution of the components to the data.

[0123] W score =σ s (Φ s (W mem ⊙H t )+b s );

[0124] Here, ⊙ represents the Hadamard product operation, which is the element-wise multiplication of a matrix. Φ s and b s σ represents the learnable parameters of the linear unit, used for information exchange and projection back to the L dimension. s (·) represents the softmax activation function, used to normalize and obtain the enhancement score. Finally, the enhancement score is used for weighted features, and the enhancement features are generated through an aggregation process:

[0125] Among them, X t This is represented as a feature that needs to be enhanced. This represents the feature after component enhancement. In this embodiment, the simplified formula for the feature enhancement process is:

[0126] Its meaning is to utilize component X′ t and X″ t For feature X t Perform fusion enhancement and obtain the enhanced features.

[0127] Based on this, after obtaining the time-enhanced features and the frequency-enhanced features, the frequency-enhanced features are used to guide the time-based features, thereby obtaining time-frequency enhanced features. Then, the time-frequency enhanced features are enhanced based on the feature enhancement module, including:

[0128] The time-frequency enhancement features are enhanced based on the target feature enhancement formula;

[0129] The target feature enhancement formula is as follows:

[0130] Here, None indicates that only one component is used for the enhancement operation, i.e., FEM(·) is applied. In this implementation, only one feature is utilized. For X t Enhance it. Indicates the use of frequency enhancement features Temporal augmentation features Further information integration enhances the guiding results.

[0131] Finally, the enhanced features, which possess both local and global information in the time-frequency domain, are used to enhance and filter the feature information of the original sequence, resulting in the final prediction result:

[0132] Where FEM(·) is used For X t Enhancements are then performed. The Head is a predictor consisting of a single linear layer. This represents the final time series prediction result, which contains D future time steps of T.

[0133] After constructing the initial model based on the time component mixing module, the frequency component mixing module, the feature enhancement module, and the prediction module, the following steps are also included:

[0134] S200. Obtain the initial dataset, which is a time series dataset of multiple hypertensive patients;

[0135] S300. The initial model is trained based on the initial dataset to obtain the target prediction model.

[0136] Specifically, in this embodiment, the time series dataset of the hypertensive patients corresponds to the following ten indicators for each patient: height, gender, weight, body temperature, pulse, systolic blood pressure, diastolic blood pressure, respiration, alcohol consumption, and family history. In other embodiments, the time series data of the hypertensive patients may correspond to more indicators.

[0137] Then, the time step length for each hypertensive patient entering the network is 64, for example, including 64 blood pressure values ​​measured at fixed intervals. In more embodiments, the time step length can be set; the longer the length, the more accurate the prediction results.

[0138] Next, the number of patients in each training round is set to 32, meaning the number of people trained by the network each time is 32. In more embodiments, the number of patients can also be set; the more patients, the more accurate the prediction results.

[0139] In other words, in this embodiment, the input dimension of each network training is 32×64×10. The initial model is trained based on the known results in the initial dataset to obtain the target prediction model. The future data about blood pressure can be obtained by the network output based on the target prediction. Assuming the output time step length is 32, the output dimension of the network is 32×32×1.

[0140] S400. Obtain target data, predict the target data based on the target prediction model, and obtain the hypertension risk prediction result of the target patient. The target data is hypertension-related data of the target patient, and there are multiple target patients.

[0141] The step of predicting the target data based on the target prediction model to obtain the hypertension risk prediction result for the target patient includes:

[0142] Extract target information from the target data. The target information includes the height, gender, weight, body temperature, pulse, systolic blood pressure, diastolic blood pressure, respiration, alcohol consumption, and family history of the target patient at a first number of time points, wherein the time interval between each time point is the same.

[0143] Based on the target information, the predicted blood pressure values ​​of the target patient are obtained for a second number of future time points corresponding to the target patient, and the predicted blood pressure values ​​include diastolic blood pressure and systolic blood pressure.

[0144] Obtain a target threshold, and based on the target threshold and the predicted blood pressure value, obtain the hypertension risk prediction result for the target patient.

[0145] As can be seen, based on the target prediction network, it is possible to simultaneously predict the future hypertension time series of multiple target patients based on their past hypertension index and health data.

[0146] Based on the time series results output by the prediction module, the predicted blood pressure values ​​of the target patient at a second number of future time points can be obtained. In this embodiment, the second number is 32.

[0147] After determining the predicted blood pressure values ​​of the target patient at a second number of future time points, a target threshold is obtained. Based on the target threshold and the predicted blood pressure values, the hypertension risk prediction result of the target patient can be obtained.

[0148] In this embodiment, the hypertension risk prediction results can be divided into three types: high risk, medium risk, and low risk based on the target threshold.

[0149] Obtain a target threshold, and based on the target threshold and the predicted blood pressure value, obtain the hypertension risk prediction result for the target patient.

[0150] After obtaining the hypertension risk prediction results for the target patient, the method further includes:

[0151] A risk assessment report and health management strategy are generated based on the hypertension risk level of each target patient.

[0152] Real-time physiological data of each target patient is dynamically monitored and hypertension risk level is predicted to enable personalized health management for each target patient based on the latest risk level.

[0153] Specifically, for high-risk patients, health management strategies can be implemented, dynamically monitoring their real-time physiological data and predicting their hypertension risk level, so as to provide personalized health management for each target patient based on the latest level.

[0154] As can be seen, based on the method described in this embodiment, the number of parameters in the constructed initial model is significantly reduced, achieving decoupling between data and model, while reducing the overall time and space complexity to a linear relationship with the sequence length.

[0155] Specifically, parameter-free component extraction strategies were designed in the time and frequency domains to significantly reduce network parameters and achieve decoupling of data and model.

[0156] Furthermore, based on the aforementioned feature enhancement module, a representation containing comprehensive time-series information related to the period and time nodes can be obtained, further improving the module's versatility and providing more effective and accurate information for time series prediction. By increasing the module's versatility and utilizing component information to enhance features, the network learns the time-frequency local and global patterns in the sequence, resulting in more comprehensive and accurate predictions.

[0157] In summary, this embodiment provides a hypertension risk assessment method guided by time-frequency component mixing. When assessing the hypertension risk of a target patient, an initial model is constructed, including a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module. Then, an initial dataset is obtained, which is a time-series dataset of multiple hypertension patients. Based on this initial dataset, the initial model is trained to obtain a target prediction model. Finally, hypertension-related data of the target patients is obtained, and the hypertension-related data of the target patients is predicted based on the target prediction model to obtain the hypertension risk prediction result for the target patients. The target data refers to the hypertension-related data of multiple target patients. The hypertension risk assessment method based on time-frequency component mixing provided by this invention, by constructing a time component mixing module, a frequency component mixing module, and a feature enhancement module, designs parameterless component extraction strategies in the time and frequency domains, effectively reducing network parameters and achieving data-model decoupling. Simultaneously, a feature enhancement module is designed to increase the module's versatility and enhance features using component information, enabling the network to learn local and global time-frequency patterns in the sequence, resulting in more comprehensive and accurate predictions.

[0158] It should be understood that although the steps in the flowcharts shown in the accompanying drawings are displayed sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of the steps in this invention, and these steps can be executed in other orders. Moreover, at least a portion of the steps in this invention may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0159] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0160] Example 2

[0161] Based on the above embodiments, the present invention also provides a hypertension risk assessment device based on time-frequency component hybrid guidance, the functional module schematic diagram of which is shown in Figure 8. The hypertension risk assessment device based on time-frequency component hybrid guidance includes:

[0162] The model building module is used to build an initial model, which includes a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module, as described in Example 1.

[0163] The dataset acquisition module acquires an initial dataset, which is a time-series dataset of multiple hypertensive patients, as described in Example 1.

[0164] The training module is used to train the initial model based on the initial dataset to obtain the target prediction model, as described in Example 1.

[0165] The prediction module is used to acquire target data, predict the target data based on the target prediction model, and obtain the hypertension risk prediction result of the target patient. The target data is hypertension-related data of the target patient, and there are multiple target patients, as specifically described in Embodiment 1.

[0166] Example 3

[0167] Based on the hypertension risk assessment method guided by time-frequency component mixing described in Embodiment 1 above, the present invention also provides a terminal, the principle block diagram of which is shown in Figure 9. The terminal includes a memory 20 and a processor 10. The memory 20 stores a hypertension risk assessment program guided by time-frequency component mixing. When the processor 10 executes the computer program, it can at least implement the following steps:

[0168] An initial model is constructed, which includes a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module;

[0169] Obtain an initial dataset, which is a time-series dataset of multiple hypertensive patients;

[0170] The initial model is trained based on the initial dataset to obtain the target prediction model;

[0171] Obtain target data, predict the target data based on the target prediction model, and obtain the hypertension risk prediction result of the target patient. The target data is hypertension-related data of the target patient, and there are multiple target patients.

[0172] The time component mixing module includes a period division module, a period pooling module, a flattening module, and a first splicing module.

[0173] The period division module is used to divide the time series data input to the model according to the period, extract the periodic components, and obtain periodic data;

[0174] The periodic pooling module is used to activate and pool the periodic data obtained by the periodic division module to obtain pooled features;

[0175] The flattening module is used to flatten the pooling features into vectors;

[0176] The first splicing module is used to splice the vectors flattened by the flattening module to obtain time component data.

[0177] The periodic pooling module includes a maximum pooling module and an average pooling module;

[0178] The max pooling module is used to activate the periodic data based on the Tanh function and then perform max pooling to obtain the main features within each period.

[0179] The average pooling module is used to activate the periodic data based on the Tanh function and then perform average pooling to obtain the average features within each period.

[0180] The frequency component mixing module includes a transformation module, an amplitude calculation module, a component extraction module, a local pooling module, a second splicing module, and an inverse transformation module.

[0181] The transformation module is used to perform a fast Fourier transform on the time series data to obtain frequency domain data;

[0182] The amplitude calculation module is used to obtain the amplitude of the frequency domain data;

[0183] The component extraction module is used to extract the Top-k components from the frequency domain data. The Top-k components are the k frequency components with the largest amplitude in the frequency domain data.

[0184] The local pooling module is used to perform local average pooling on the frequency domain data to obtain the local aggregated frequency components of the frequency domain data.

[0185] The second splicing module is used to splice the Top-k component and the local aggregation frequency component to obtain spliced ​​data;

[0186] The inverse transform module is used to perform an inverse fast Fourier transform on the spliced ​​data to obtain global frequency components.

[0187] The feature enhancement module includes a linear mapping module, a feature introduction module, an activation module, a regularization module, an anti-linear mapping module, a normalization module, and an enhancement module.

[0188] The linear mapping module is used to perform a linear transformation on the time component data and the global frequency component to obtain a target linear feature with uniform dimension.

[0189] The feature introduction module is used to introduce nonlinear features into the target linear features based on the ReLU function to obtain the features to be activated.

[0190] The activation module is used to perform point-to-point activation of the target enhancement features based on the memory matrix to obtain activated features;

[0191] The regularization module is used to regularize the activation features based on Dropout technology to obtain regularized features.

[0192] The inverse linear mapping module is used to inversely map the regularized features to restore them to the original dimension, thereby obtaining the features to be enhanced.

[0193] The normalization module is used to normalize the features to be enhanced in order to maintain the stability of the feature values;

[0194] The enhancement module is used to enhance the normalized features based on the Hadamard product to obtain enhanced features.

[0195] The step of predicting the target data based on the target prediction model to obtain the hypertension risk prediction result for the target patient includes:

[0196] Extract target information from the target data. The target information includes the height, gender, weight, body temperature, pulse, systolic blood pressure, diastolic blood pressure, respiration, alcohol consumption, and family history of the target patient at a first number of time points, wherein the time interval between each time point is the same.

[0197] Based on the target information, the predicted blood pressure values ​​of the target patient are obtained for a second number of future time points corresponding to the target patient, and the predicted blood pressure values ​​include diastolic blood pressure and systolic blood pressure.

[0198] Obtain a target threshold, and based on the target threshold and the predicted blood pressure value, obtain the hypertension risk prediction result for the target patient.

[0199] The process of obtaining the hypertension risk prediction results for the target patient further includes:

[0200] A risk assessment report and health management strategy are generated based on the hypertension risk level of each target patient.

[0201] Real-time physiological data of each target patient is dynamically monitored and hypertension risk level is predicted to enable personalized health management for each target patient based on the latest risk level.

[0202] Example 4

[0203] The present invention also provides a storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the hypertension risk assessment method based on time-frequency component hybridization as described in the above embodiments.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the risk of hypertension based on hybrid guidance of time-frequency components, characterized in that, The method comprises the following steps: constructing an initial model, the initial model comprising a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module; obtaining an initial data set, the initial data set being a time series data set of a plurality of hypertension patients; training the initial model based on the initial data set to obtain a target prediction model; obtaining target data, predicting the target data based on the target prediction model to obtain a hypertension risk prediction result of a target patient, the target data being hypertension-related data of the target patient, and the target patient being one of the plurality of patients.

2. The time-frequency component mixture-based guidance based hypertension risk assessment method according to claim 1, characterized in that, The time component mixing module comprises a period division module, a period pooling module, a flattening module, and a first splicing module; The period division module is configured to divide the time series data of the model input by period, extract the periodic component, and obtain periodic data; The period pooling module is configured to activate and pool the periodic data obtained by the period division module to obtain pooled features; The flattening module is configured to flatten the pooled features into a vector; The first splicing module is configured to splice the vector flattened by the flattening module to obtain time component data.

3. The time-frequency component mixture-based guidance-based hypertension risk assessment method according to claim 2, characterized in that, The period pooling module comprises a maximum pooling module and an average pooling module; The maximum pooling module is configured to activate the periodic data based on a Tanh function and then perform maximum pooling to obtain the main features in each period; The average pooling module is configured to activate the periodic data based on a Tanh function and then perform average pooling to obtain the average features in each period.

4. The time-frequency component mixture-based guidance-based hypertension risk assessment method according to claim 3, characterized in that, The frequency component mixing module comprises a transformation module, an amplitude calculation module, a component extraction module, a local pooling module, a second splicing module, and an inverse transformation module; The transformation module is configured to perform fast Fourier transform on the time series data to obtain frequency domain data; The amplitude calculation module is configured to obtain the amplitude of the frequency domain data; The component extraction module is configured to extract Top-k components in the frequency domain data, the Top-k components being the first k frequency components with the largest amplitudes in the frequency domain data; The local pooling module is configured to perform local average pooling on the frequency domain data to obtain local aggregated frequency components of the frequency domain data; The second splicing module is configured to splice the Top-k components and the local aggregated frequency components to obtain spliced data; The inverse transformation module is configured to perform inverse fast Fourier transform on the spliced data to obtain global frequency components.

5. The time-frequency component mixture-based guidance based hypertension risk assessment method according to claim 4, characterized in that, The feature enhancement module comprises a linear mapping module, a feature introduction module, an activation module, a regularization module, an inverse linear mapping module, a normalization module, and an enhancement module; The linear mapping module is configured to perform linear transformation on the time component data and the global frequency components to obtain target linear features with unified dimensions; The feature introduction module is configured to introduce nonlinear features into the target linear features based on a ReLU function to obtain activated features; The activation module is configured to activate the target enhancement features point by point based on a memory matrix to obtain activated features; The regularization module is configured to regularize the activation features based on a Dropout technique to obtain regularized features; The inverse linear mapping module is configured to perform inverse mapping on the regularized features to restore the original dimension to obtain to-be-enhanced features; The normalization module is configured to perform normalization processing on the to-be-enhanced features to maintain the stability of feature values; The enhancement module is configured to enhance the to-be-enhanced features based on Hadamard product to obtain enhanced features.

6. The time-frequency component mixture-based guidance for hypertension risk assessment method according to claim 1, wherein, The target prediction model is used to predict the target data to obtain a hypertension risk prediction result of the target patient, including: Extracting target information in the target data, the target information including the target patient's height, gender, weight, temperature, pulse, high pressure, low pressure, respiration, alcohol consumption, family history corresponding to a first number of time points, wherein each time point is equally spaced in time; Based on the target information, the predicted blood pressure value of the target patient corresponding to a second number of future time points is obtained, including diastolic pressure and systolic pressure; A target threshold is obtained, and the hypertension risk prediction result of the target patient is obtained based on the target threshold and the predicted blood pressure value.

7. The time-frequency component mixture-based guidance for hypertension risk assessment method according to claim 1, characterized in that, After obtaining the hypertension risk prediction result of the target patient, it further includes: Based on the hypertension risk level of each target patient, a risk assessment report and a health management strategy are generated; Real-time physiological data of each target patient is dynamically monitored and a hypertension risk level is predicted to perform personalized health management on each target patient based on the latest level.

8. A hypertension risk assessment apparatus based on time-frequency component hybrid guidance, characterized by, The device includes: A model construction module is configured to construct an initial model, the initial model including a time component mixing module, a frequency component mixing module, a feature enhancement module, and a prediction module; A data set acquisition module acquires an initial data set, the initial data set being a time series data set of a plurality of hypertension patients; A training module is configured to train the initial model based on the initial data set to obtain a target prediction model; A prediction module is configured to acquire target data, predict the target data based on the target prediction model, and obtain a hypertension risk prediction result of a target patient, the target data being hypertension-related data of a target patient, and the target patient being a plurality of patients. The terminal includes a processor, a storage medium in communication with the processor, the storage medium being adapted to store a plurality of instructions, and the processor being adapted to invoke the instructions in the storage medium to perform the steps of the hypertension risk assessment method based on time-frequency component mixing guidance according to any one of claims 1-7.

9. A terminal, characterized by comprising: The storage medium stores one or more programs, which can be executed by one or more processors to implement the steps of the hypertension risk assessment method based on time-frequency component mixing guidance according to any one of claims 1-7.

10. A storage medium, characterized by ​

Images