Trend inference method, device and equipment of time sequence medical data and medium

Abstract

The present application relates to the technical field of intelligent decision-making, and discloses a trend inference method, device, equipment and medium for time-series medical data. The method comprises: performing data standardization processing on multi-dimensional medical data to obtain a standard data set; constructing a medical data constraint model according to medical theory data, embedding the medical data constraint model into a pre-constructed initial inference model to generate a trend inference model; performing parameter training on the trend inference model using the standard data set to obtain a target trend inference model; when receiving time-series medical data, extracting high-dimensional data features using the target trend inference model; generating a time state sequence according to the high-dimensional data features, and predicting a data trend according to the time state sequence. The present application can be applied to business system platforms such as financial technology and medical health, and can improve the accuracy of trend inference of time-series medical data.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to a method, apparatus, device, and medium for trend extrapolation of time-series medical data. Background Technology

[0002] With the rapid development of smart healthcare and precision medicine, trend prediction of time-series medical data has become one of the core research directions in the medical and health field. It aims to predict potential development trends such as disease staging, complication occurrence, disease deterioration and prognosis by relying on multi-dimensional medical data modeling and analysis of patients, thus breaking through the limitations of traditional medicine that can only achieve static disease screening.

[0003] In the fintech field, such as insurance, the trend projections of chronic diseases (e.g., diabetes, hypertension, chronic kidney disease) and critical illnesses (e.g., cancer, cardiovascular and cerebrovascular diseases) directly determine the probability of insurance claims, the setting of coverage amounts, and the effectiveness of health management plans. The progression of these diseases is not random but follows clear pathophysiological laws. Its core characteristic is the dynamic evolution of multiple physiological indicators in the body (e.g., blood sugar, blood pressure, tumor markers, liver and kidney function indicators, etc.) over time, and this evolution process is constrained by pathophysiological dynamic equations, belonging to a typical high-dimensional, strongly constrained temporal evolution problem.

[0004] However, existing technologies mainly rely on traditional statistical models or conventional machine learning models to roughly predict medical data trends by fitting historical medical data, thereby assisting in actuarial pricing, claims review, and health management decisions. However, due to the complex pathophysiological mechanisms of medical data such as chronic diseases and critical illnesses, the strong coupling of multiple indicators, and the significant individual differences and physical constraints in disease progression, existing technologies are insufficient to meet the needs of accurate prediction.

[0005] Therefore, improving the accuracy of trend extrapolation from time-series medical data has become an urgent problem to be solved. Summary of the Invention

[0006] This invention provides a method, apparatus, equipment, and medium for trend extrapolation of time-series medical data, the main purpose of which is to solve the problem of poor accuracy in trend extrapolation of time-series medical data.

[0007] In a first aspect, to achieve the above objectives, the present invention provides a method for trend extrapolation of time-series medical data, comprising: Acquire multi-dimensional medical data and medical theoretical data, and perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; A medical data constraint model is constructed based on the medical theoretical data, and the medical data constraint model is embedded into a pre-constructed initial inference model to generate a trend inference model; The trend projection model is trained with the standard dataset to obtain the target trend projection model; Upon receiving the medical time-series data of the user to be analyzed, the high-dimensional data features of the medical time-series data are extracted using the target trend inference model. A time-state sequence is generated based on the high-dimensional data features, and the data trend of the user to be analyzed is predicted based on the time-state sequence.

[0008] Secondly, the present invention also provides a trend extrapolation device for time-series medical data, comprising: The data processing module is used to acquire multi-dimensional medical data and medical theoretical data, and to perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; The trend extrapolation model construction module is used to construct a medical data constraint model based on the medical theoretical data, embed the medical data constraint model into the pre-constructed initial extrapolation model, and generate a trend extrapolation model. The model training module is used to train the parameters of the trend inference model using the standard dataset to obtain the target trend inference model. The high-dimensional data feature extraction module is used to extract high-dimensional data features of the medical time-series data when receiving the medical time-series data of the user to be analyzed, using the target trend inference model. The data trend extrapolation module is used to generate a time state sequence based on the high-dimensional data features, and to predict the data trend of the user to be analyzed based on the time state sequence.

[0009] Thirdly, the present invention also provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the trend extrapolation method for time-series medical data described above.

[0010] Fourthly, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the above-described trend extrapolation method for time-series medical data.

[0011] In this embodiment of the invention, by standardizing multi-dimensional medical data, the standard dataset has unified data specifications, complete data dimensions, and reliable data accuracy. A medical data constraint model is constructed based on medical theoretical data, and the medical data constraint model is embedded into a pre-constructed initial inference model to generate a trend inference model. This can achieve synergistic optimization of data distribution fitting and physical constraint satisfaction, ensuring the prediction accuracy and physical reliability of the model. By training the trend extrapolation model with standard datasets, a target trend extrapolation model can be obtained. This can further improve the generalization ability and training efficiency of the trend extrapolation model, ensuring that the model can still output accurate prediction results stably even in scenarios with small samples and missing data. When receiving medical time-series data of the user to be analyzed, the target trend extrapolation model is used to extract high-dimensional data features, generate time state sequences based on these features, and predict the data trend of the user to be analyzed based on these time state sequences. This approach can balance the fitting accuracy to historical data distribution with the satisfaction of physical constraints, effectively improving the accuracy of data trend prediction. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram illustrating an application environment of the trend extrapolation method for time-series medical data according to an embodiment of the present invention. Figure 2 A flowchart illustrating a method for trend extrapolation of time-series medical data according to an embodiment of the present invention; Figure 3 for Figure 2 A flowchart illustrating the data standardization process for multi-dimensional medical data in step S1. Figure 4 for Figure 2 A flowchart illustrating the process of constructing a medical data constraint model in step S2. Figure 5 for Figure 2 A flowchart illustrating the process of training the parameters of the trend extrapolation model using a standard dataset in step S3. Figure 6 for Figure 5 A flowchart illustrating the process of constraining the predicted trend data in step S52. Figure 7 for Figure 2A flowchart illustrating the process of extracting high-dimensional data features from medical time-series data in step S4. Figure 8 for Figure 2 A flowchart illustrating the process of predicting data trends based on time state sequences in step S5. Figure 9 A functional block diagram of a trend extrapolation device for time-series medical data provided in an embodiment of the present invention; Figure 10 A schematic diagram of an electronic device for implementing a trend extrapolation method for time-series medical data according to an embodiment of the present invention; Figure 11 Another schematic diagram of an electronic device for implementing a trend extrapolation method for time-series medical data, as provided in an embodiment of the present invention.

[0014] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] It should be noted that in the technical solutions disclosed in this invention, the acquisition of user information (personal image data (e.g., facial videos or pictures, facial feature videos or pictures, etc.) and personal privacy information (e.g., name, ID number, occupation, address, etc.)) is all completed with the user's knowledge and consent, and the acquisition of the relevant user information is legal and compliant.

[0016] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this disclosure.

[0017] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] This application provides a method for trend extrapolation of time-series medical data. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the device provided in this application: a server, a terminal, etc. In other words, the method for trend extrapolation of time-series medical data can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0019] The trend extrapolation method for time-series medical data of this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can obtain multi-dimensional medical data and theoretical medical data from the client, perform data standardization on the multi-dimensional medical data to obtain a standard dataset; construct a medical data constraint model based on the theoretical medical data, embed the medical data constraint model into a pre-constructed initial inference model to generate a trend inference model; train the parameters of the trend inference model using the standard dataset to obtain the target trend inference model; when receiving the medical time-series data of the user to be analyzed, extract high-dimensional data features of the medical time-series data using the target trend inference model; generate a time state sequence based on the high-dimensional data features, and predict the data trend of the user to be analyzed based on the time state sequence. This invention can be applied to business scenarios such as fintech and healthcare. By standardizing multi-dimensional medical data and training the parameters of a trend inference model, a target trend inference model is obtained. This improves the generalization ability and training efficiency of the trend inference model, ensuring that it can still output accurate prediction results stably even in scenarios with small samples and missing data. When receiving medical time-series data, the target trend inference model is used to predict the data trend of the user being analyzed, balancing the fitting accuracy to historical data distribution with the satisfaction of physical constraints, effectively improving the accuracy of data trend prediction. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0020] Reference Figure 2 The diagram shown is a flowchart illustrating a method for trend extrapolation of time-series medical data according to an embodiment of the present invention. In this embodiment, the method for trend extrapolation of time-series medical data includes: S1. Obtain multi-dimensional medical data and medical theoretical data, and perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset.

[0021] In this embodiment of the invention, the multi-dimensional medical data refers to the medical data of patients with chronic or serious illnesses, including time-series physiological indicator data, pathological diagnosis data, lifestyle data, and insurance business data. Specifically, the multi-dimensional medical data covers time-series monitoring data of the patient's basic physiological indicators (such as blood glucose, blood pressure, heart rate, tumor marker concentration, etc.), clinical pathological diagnosis reports, surgical and treatment records, medication use, dietary structure, exercise frequency and intensity, and other lifestyle data, as well as insurance business-related data (such as insured age, insurance duration, and past claims records).

[0022] In detail, medical theoretical data includes medical theoretical knowledge such as pathophysiological mechanisms, encompassing the physiological and pathological operating mechanisms of the human body (organ metabolism, substance diffusion, inflammatory evolution, hemodynamics, cell proliferation, toxin clearance, neurodegenerative mechanisms, etc.); the inherent changing patterns of physiological indicators: the causal relationship between indicator rises and falls, the upper limit of the rate of change, the reasonable range of indicator values, and the maximum gradient of change; biological boundary constraints: the upper and lower limits of physiological indicators (blood oxygen, blood glucose, and creatinine concentrations cannot be negative, and there are physiological extreme values); the law of conservation of matter, the law of diffusion, and the laws of kinetic reactions; and the pathways and mechanisms of disease progression: such as tumor proliferation and spread, the progression of fibrosis in chronic kidney disease, and the irreversible characteristics of neuronal damage.

[0023] Specifically, see Figure 3 As shown, the standardization process for the multi-dimensional medical data to obtain a standard dataset includes: S31. Identify outliers and missing values ​​in the multi-dimensional medical data; S32. Remove the outliers from the multi-dimensional medical data and fill the missing values ​​in the multi-dimensional medical data with preset values ​​to obtain filled medical data; S33. Perform data standardization processing on the filled medical data to obtain a standard dataset.

[0024] In this embodiment of the invention, multi-dimensional medical data includes structured medical data and semi-structured medical data. Corresponding defect identification rules are applied to different types of data: For structured data such as vital signs, laboratory indicators, and patient records, based on preset reasonable value ranges, data collection frequencies, and field integrity rules, missing content in each data field is detected, and the location, quantity, and type of missing values ​​are identified. Simultaneously, combined with medical industry data standard thresholds and normal clinical data ranges, abnormal data deviating from the normal medical data range and inconsistent with clinical diagnosis and treatment logic are identified, determining the specific location, type, and degree of abnormality. For semi-structured medical records and image annotation data, text field integrity checks and semantic compliance checks are used to identify defective data such as missing fields and disordered content.

[0025] Furthermore, based on the identified and marked missing and outlier values, the original multi-dimensional medical data is cleaned and filled, invalid and outlier data is removed, and missing data is repaired to obtain medical data that meets the standards of completeness and validity.

[0026] Specifically, all identified outliers are completely removed from the original multi-dimensional medical dataset. Outlier entries that do not conform to medical data logic, contain collection errors, or are invalid or distorted are deleted, eliminating the interference of abnormal noise data on the overall quality of the dataset. Next, for each type of missing value identified, targeted filling is performed using preset values ​​corresponding to preset filling rules. These preset values ​​can be set differently based on the dimensions, data types, and clinical application scenarios of the medical data. They may include compliant filling values ​​such as dimensional mean, industry standard default values, historical patient data from the same period, and routine clinical benchmark values. The preset values ​​are matched to different missing scenarios to complete the filling and repair, maximizing the restoration of the true characteristics of the medical data, ensuring the integrity of the data dimensions, and avoiding dataset failure caused by data gaps.

[0027] Furthermore, standardization processes include operations such as unifying data units, normalizing data formats, calibrating data precision, and normalizing data value ranges. For medical data across different dimensions, the differentiated collection standards and display formats of the original data are discarded. Instead, based on pre-defined medical big data processing standards, the units of measurement, data retention precision, field formats, and data value ranges for various vital sign data, test data, and diagnostic data are unified, eliminating data discrepancies caused by different data sources and collection devices. Simultaneously, correlation and consistency checks are performed on the standardized data across each dimension to ensure logical matching, format uniformity, and dimensional compatibility among multi-dimensional data.

[0028] In this embodiment of the invention, by standardizing multi-dimensional medical data, the standard dataset has unified data specifications, complete data dimensions, and reliable data accuracy, making it fully adaptable to various subsequent business scenarios such as medical big data statistical analysis, intelligent diagnosis and treatment model training, patient health status analysis, and medical quality control.

[0029] S2. Construct a medical data constraint model based on the medical theoretical data, embed the medical data constraint model into the pre-constructed initial inference model, and generate a trend inference model.

[0030] In this embodiment of the invention, the medical data constraint model involves structured analysis and mechanism decomposition of original medical theoretical data. The sources of the medical theoretical data can be prior data in the form of text, charts, and experimental conclusions, such as physiological literature, clinical pathology research conclusions, pharmacokinetics theories, and human biological regulation theories.

[0031] Further, see Figure 4 As shown, the step of constructing a medical data constraint model based on the medical theoretical data includes: S41. Extract theoretical data on different physiological mechanisms from the aforementioned medical theoretical data; S42. Fit the partial differential equations corresponding to each of the physiological mechanisms based on the theoretical data; S43. The partial differential equation is calibrated to obtain a medical data constraint model.

[0032] Specifically, medical theoretical data is deconstructed to distinguish independent physiological and pathological mechanisms of action. For example, it can be broken down into multiple independent mechanisms such as substance diffusion, cell proliferation, signal transduction, metabolic clearance, and tissue perfusion. Then, for each mechanism, core elements are extracted: participating variables (such as concentration, volume, temperature, lesion size, time, and spatial coordinates), causal relationships between variables, variable trends, boundary conditions, and conservation rules. Finally, unstructured medical theoretical knowledge is transformed into structured theoretical data. The theoretical data explicitly records: the state variables, independent variables (time and space), driving relationships of variable changes, and qualitative evolutionary rules under a given mechanism.

[0033] Furthermore, for each type of theoretical data, the independent variables and the state variables to be solved are determined. The independent variables generally include time and space dimensions; the function to be solved is the target medical indicator. Based on the changing patterns described by the theoretical mechanism, a partial differential operator form is constructed. By combining the various operators, a preliminary prototype of the partial differential equation corresponding to the mechanism is constructed. Then, combined with medical theory, initial and boundary conditions are given to form a complete boundary value problem.

[0034] In detail, the partial differential equation obtained in the previous step is only the equation structure. Internally, there is a set of unknown constant parameters (diffusion coefficient, reaction rate, attenuation coefficient, gain coefficient, etc.). The values ​​of these parameters directly determine whether the model's evolution trajectory conforms to real human physiological patterns, necessitating parameter calibration. Clinical medical data and medical research findings can be used to perform preliminary calibration of the parameters in the equation, ensuring its rationality and applicability.

[0035] In this embodiment of the invention, the partial differential equations (along with the corresponding initial conditions and boundary conditions) after the equation parameters are calibrated are encapsulated to form a medical data constraint model, which is a complete mathematical model that can be directly solved numerically and realize quantitative deduction.

[0036] For example, taking diabetes, a chronic disease, as an example, the medical data constraint model for the evolution of blood glucose concentration over time is expressed as follows:

[0037] in, express Blood glucose concentration at a specific time and location on the body is a core indicator of the physical state of the disease and can directly reflect the severity of the diabetic patient's condition. The blood glucose diffusion coefficient (unit: cm² / s) is determined by human physiological characteristics and varies among patients of different ages and physical conditions. Since the values ​​differ, specific values ​​can be obtained by fitting clinical data. The Laplace operator describes the spatial distribution of blood glucose concentration, which can characterize the diffusion pattern of blood glucose in different parts of the human body and avoid the bias caused by prediction in a single time dimension. The rate constant for blood glucose metabolism (unit: s) - ¹), related to insulin secretion and metabolic function, insufficient insulin secretion or abnormal metabolic function can lead to The value decreases, which in turn causes the blood glucose concentration to rise; The glucose production rate function (unit: mmol / (L·s)) is determined by lifestyle factors such as diet and exercise. ,in The time correlation coefficient reflects the differences in blood glucose production at different time periods (such as before meals, after meals, and at night). This is the spatial distribution coefficient, reflecting the differences in blood glucose production in different parts of the human body. This function can accurately characterize the impact of lifestyle habits on blood glucose concentration, thus improving the applicability of the equation.

[0038] For lung cancer, a serious disease, the medical data constraint model for tumor volume evolution over time is as follows:

[0039] in, The tumor volume at time t (unit: cm³) is a core indicator for measuring lung cancer progression. Changes in tumor volume directly reflect the deterioration or improvement of the condition. Tumor natural growth rate constant (unit: d) - ¹), determined by tumor type and differentiation degree; the higher the malignancy, The larger the value; The maximum volume of a tumor (unit: cm³) is determined by the body's immune function and the type of tumor. The stronger the body's immune function, the greater the tumor's growth potential. The smaller the value, the better it reflects the body's inhibitory effect on tumor growth; The effect coefficient of treatment intervention (unit: d) - ¹), different treatment methods (such as chemotherapy, radiotherapy, targeted therapy) Different values ​​indicate better treatment results. The larger the value; The intensity of treatment intervention (0≤T(t)≤1, where T(t)=0 indicates no treatment and T(t)=1 indicates maximum intensity treatment) can be dynamically adjusted according to the patient's treatment plan. This equation can accurately characterize the impact of treatment intervention on tumor growth and provide a reliable physical constraint for predicting lung cancer progression.

[0040] Furthermore, the pre-built initial inference model includes a feature extraction module, a time series generation module, and a loss function module; the feature extraction module uses a Transformer architecture to extract features from the input multi-dimensional medical data to obtain high-dimensional feature vectors. Where N is the number of samples and d is the feature dimension; this module adopts a multi-layer Transformer encoder and uses a self-attention mechanism to mine hidden relationships in multi-dimensional medical data, such as the coupling relationship between physiological indicators and the relationship between lifestyle habits and physiological indicators. It can effectively extract key features from medical data and avoid the shortcomings of traditional feature extraction methods (such as PCA and convolution operations) in capturing temporal features and high-dimensional correlation features. At the same time, a positional encoding mechanism is introduced to preserve the temporal information of medical data, ensuring the integrity and accuracy of feature extraction and providing high-quality feature support for subsequent time series generation.

[0041] Time series generation module: Employs a time series generation network based on feature vectors. The PDE (Partial Differential Equation) constraints of the medical data constraint model generate a data sequence corresponding to the constrained predicted data trend for the next T time points. ,in Here, m represents the number of physiological indicators. This module employs a temporal Transformer decoder, combining the high-dimensional feature vectors output by the feature extraction module with the constraints of the PDE constraint module. Through an autoregressive generation mechanism, it generates the physical state of the disease at each time node. Simultaneously, an attention mechanism is introduced to focus on the impact of historical physiological indicator changes on future states, improving the continuity and accuracy of time series generation. Furthermore, this module supports flexible adjustment of the number of time nodes T, allowing for different prediction durations to be set according to insurance business needs (such as short-term and long-term predictions), thereby enhancing the model's practicality.

[0042] Loss Function Module: Constructs a bi-objective loss function to balance the accuracy of data distribution fitting with the satisfaction of physical constraints. as follows:

[0043] in, The loss function value of the initial inference model. The loss is the data distribution fitting loss. For physical constraint loss, Preset constraint weights.

[0044] In this embodiment of the invention, Wasserstein distance is used to calculate the data distribution fitting loss, which measures the difference between the generated constrained prediction trend data and the actual medical data distribution. Compared with the traditional mean squared error (MSE) loss, Wasserstein distance can better characterize the difference between the two distributions and avoid the loss calculation bias caused by uneven data distribution. The data distribution fitting loss is calculated using the following formula. :

[0045] in For data fitting loss, A data sequence representing the true trend data. The data sequence corresponding to the predicted data trend of the initial inference model. and The joint distribution set between them It is an L1 norm. To preset the first weight, For the jointly distributed infinitesimal elements.

[0046] The physical constraint loss measures the degree to which the generated constraint prediction trend data satisfies the medical data constraint model, ensuring that the generated sequences conform to pathophysiological laws. The physical constraint loss is calculated using the following formula:

[0047] in, For physical constraint loss, Partial differential equations in medical data constraint models The right-hand side expression at time t, It is the L2 norm. For time The total number, for Time-constrained prediction data.

[0048] Specifically, Partial differential equations in medical data constraint models The expression on the right side at time (such as in the PDE of blood glucose concentration) The specific values ​​are obtained through numerical calculation; the L2 norm can be used to measure the squared error between the time derivative of the generated sequence and the expression on the right side of the PDE equation. The smaller the error, the more the generated sequence conforms to the PDE constraints. This is used to average the errors of T time points, avoiding excessive loss due to a large number of time points.

[0049] To constrain the weights, an adaptive adjustment mechanism is used, eliminating the need for manual parameter tuning and ensuring conflict-free optimization between data distribution fitting and physical constraints, achieving a Pareto optimal balance between the two. The adaptive adjustment mechanism dynamically adjusts the weights based on the loss changes during model training. The value of: when When the value is large (data distribution fitting accuracy is low), decrease The value of is primarily optimized to improve the accuracy of data distribution fitting; when When the value is large (low physical constraint satisfaction), increase The value of is selected with a focus on optimizing the satisfaction of physical constraints, ultimately achieving synergistic optimization between the two and avoiding the optimization imbalance problem caused by manual parameter tuning in existing technologies.

[0050] The initial parameter initialization of the initial model can adopt the Xavier initialization method to ensure that the initial values ​​of the model parameters are within a reasonable range, avoiding problems such as unstable model training and slow convergence speed caused by improper parameter initialization. At the same time, the parameters in the medical data-constrained model (such as diffusion coefficient and metabolic rate constant) are initially assigned values ​​based on clinical medical data and medical research results, laying the foundation for subsequent parameter optimization.

[0051] Furthermore, the medical data constraint model is deeply embedded into each stage of the initial inference model to achieve synergistic optimization of data distribution fitting and physical constraint satisfaction, avoiding the conflict between the two in existing models. Among them, each module is both independent and collaborative: the feature extraction module is responsible for mining key information in medical data; the medical data constraint model is responsible for forcing the model output to conform to pathophysiological laws; the time series generation module is responsible for predicting the future physical evolution trajectory of the disease; and the loss function module is responsible for balancing the dual optimization objectives of the model, jointly ensuring the model's prediction accuracy and physical reliability.

[0052] S3. Use the standard dataset to train the parameters of the trend inference model to obtain the target trend inference model.

[0053] In this embodiment of the invention, parameter training utilizes a standard dataset to iteratively adjust the unknown parameters within the trend inference model, minimizing the prediction error and enabling the trend inference model to have accurate prediction capabilities.

[0054] Specifically, see Figure 5 As shown, the step of training the trend projection model with parameters using the standard dataset to obtain the target trend projection model includes: S51. Use the trend inference model to extract the initial feature vector of the standard dataset, add feature perturbation to the initial feature vector to obtain the perturbation feature vector; S52. Generate predicted trend data using the perturbation feature vector, and constrain the predicted trend data using the medical data constraint model in the trend inference model to obtain constrained predicted trend data. S53. Calculate the loss function value of the trend inference model based on the constrained trend prediction data; S54. Update the parameters of the trend inference model in reverse according to the loss function value to obtain the target trend inference model.

[0055] In this embodiment of the invention, a standard dataset is input into a trend inference model to obtain a high-dimensional feature vector through a feature extraction module, and a disease physical state sequence is generated through a time series generation module. Specifically, time-series medical data in the standard dataset is input into the feature extraction module in chronological order, and after processing by a multi-layer Transformer encoder, a high-dimensional feature vector containing key features and time-series information is obtained.

[0056] Furthermore, a small random perturbation factor conforming to the distribution rules of medical data value range is selected. The perturbation factor is a random value with controllable amplitude that conforms to the fluctuation range of medical data, avoiding excessive perturbation that could lead to feature distortion or deviation from the reasonable range of clinical data. The perturbation factor of the preset dimension is superimposed on the feature components of each dimension of the initial feature vector to perform a small and compliant dynamic shift on the local values ​​of the initial features.

[0057] In detail, the initial feature vector is randomly perturbed to improve the model's generalization ability and adapt to small sample scenarios. Specifically, Gaussian noise perturbation is used to add small random noise (such as noise intensity of 0.02) to the high-dimensional feature vector output by the feature extraction module to simulate the physiological differences and data noise of different individuals, so that the trend inference model can learn more general features and avoid overfitting of the trend inference model to the training data.

[0058] In this embodiment of the invention, the time series generation module uses the perturbation feature vector to calculate the predicted trend data.

[0059] Specifically, see Figure 6 As shown, the step of using the medical data constraint model in the trend extrapolation model to constrain the predicted trend data to obtain constrained predicted trend data includes: S61. Divide the predicted trend data into windows to obtain windowed trend data; S62. Construct an optimization objective function for the window trend data using the medical data constraint model; S63. Solve the optimization objective function to obtain the corrected data subsequence; S64. Apply physiological value range constraints to the corrected data subsequence to obtain window constraint data for the window trend data; S65. Collect the constraint data of each window to obtain constraint prediction trend data.

[0060] In this embodiment of the invention, a fixed-length sliding time window is set, and the complete long-term time-series prediction trend data is segmented along the time dimension. Each segment of the extracted time-series data constitutes a set of window trend data. The medical data constraint model incorporates pathophysiological partial differential equations, dynamic evolution rules, and indicator change rate constraints. A joint optimization objective function is constructed for each set of window trend data to force the temporal evolution within the window to conform to the pathophysiological mechanism while preserving as much of the original prediction information as possible. This can be achieved by requiring the temporal evolution trajectory within the window to satisfy the pathophysiological partial differential dynamic equations.

[0061] Furthermore, the current window trend data is used as the initial value for optimization; a gradient-based optimization algorithm is used to minimize the above objective function; the physiological index values ​​at each time point within the window are continuously adjusted during the iteration process; optimization stops when the objective function converges or reaches the maximum number of iterations, and the optimized correction data subsequence is output; based on prior knowledge of pathophysiology, the legal value range of each physiological index and the upper and lower limits of the maximum rate of increase and decrease of the index are predefined; for example, the index concentration cannot be negative, and various test indicators have human physiological limits.

[0062] The index values ​​at each time step in the corrected data subsequence are verified element by element. If a value exceeds the valid physiological range, truncation or projection is performed to restrict the value to the valid interval. After range pruning, window-constrained data that satisfies both kinetic and physiological boundary constraints are obtained. Then, all window-constrained data corresponding to each window are reassembled according to the original time sequence. If there are overlapping areas between windows, methods such as mean fusion and weighted smoothing are used to eliminate potential gaps at window junctions. All segmented results are then spliced ​​and integrated to form a complete and continuous time series, which is the constrained prediction trend data.

[0063] Furthermore, the model parameters are updated using the backpropagation algorithm, and iterative optimization is performed until the loss function converges. Specifically, the AdamW optimizer is used for parameter updates. This optimizer can effectively alleviate the vanishing and exploding gradient problems, improving the stability and convergence speed of model training. In each iteration, the parameters are first calculated... and The specific value will then be determined based on the adaptive adjustment mechanism. The total loss is calculated by taking the value of . The loss value is propagated back to each module of the model through the backpropagation algorithm to update the module parameters and reduce the loss value. During the iterative training process, the change of the loss function is monitored in real time. When the change of the loss function is less than the preset convergence threshold (e.g., 1e-5) and the loss value does not decrease significantly for several consecutive iterations (e.g., 10 rounds), the trend inference model training is determined to be converged and training is stopped.

[0064] Preferably, a batch training strategy is adopted during the training process, in which a portion of the standard dataset is selected for training each time. This further improves the generalization ability and training efficiency of the trend inference model, ensuring that the trend inference model can still output accurate prediction results stably in scenarios with small samples and missing data.

[0065] S4. Upon receiving the medical time-series data of the user to be analyzed, the high-dimensional data features of the medical time-series data are extracted using the target trend inference model.

[0066] In this embodiment of the invention, receiving the medical time-series data of the user to be analyzed means that the model server receives the medical time-series observation sequence of a single target patient. The medical time-series data refers to longitudinal clinical information recorded sequentially in chronological order, including physiological test indicators, vital signs, imaging measurement results, medical history records, etc., collected at different consultation times; this sequence has a time dimension attribute and can reflect the evolution of the patient's physiological state over time.

[0067] Specifically, see Figure 7 As shown, the extraction of high-dimensional data features from the medical time-series data using the target trend extrapolation model includes: S71. Calculate the location encoding feature vector corresponding to the medical time series data; S72. The target trend inference model is used to perform multi-layer encoding on the location encoding feature vector to obtain the target encoding feature vector; S73. Perform self-attention calculation on the target encoded feature vector to obtain the high-dimensional data features of the medical time series data.

[0068] In this embodiment of the invention, the feature extraction module in the target trend inference model calculates a unique location-coded feature vector, assigning a unique temporal location feature to the medical data at each time-series node, thereby achieving the fusion of numerical features and temporal location features. Specifically, based on a preset location coding function, for each sampling node at each time step in the medical time-series data, the corresponding location coding value is calculated, constructing a location-coded feature vector that matches the dimension of the medical time-series data.

[0069] Positional encoding can transform the chronological order, time interval, and temporal arrangement patterns of time-series data into computable high-dimensional vector information, accurately representing the temporal attributes of medical data at different detection times. It solves the technical defects of traditional medical time-series features, such as lack of positional distinction and missing temporal logic, and provides complete temporal prior information for subsequent multi-layer feature encoding.

[0070] Furthermore, by using the trained target trend inference model to perform multi-level and multi-granular deep feature encoding mapping on the location encoding feature vector, the progressive transformation from low-dimensional original features to high-dimensional abstract features is achieved, thereby uncovering the hidden shallow change patterns and deep evolutionary features of medical data.

[0071] Furthermore, for the target encoded feature vector output by the multi-layer encoding, a query vector, a key vector, and a value vector are generated through a self-attention mechanism. By calculating the similarity between the query vector and the key vector, the association weight score between any two time-series node features is obtained. This weight score represents the degree of mutual influence of medical data at different times.

[0072] After normalizing the correlation weights, the corresponding value vectors are weighted and summed based on the weight scores to complete the adaptive weighted fusion of global time-series features. This process automatically focuses on key change nodes, abnormal fluctuation nodes, and trend turning points in medical time-series data, strengthening highly correlated effective time-series features and weakening ineffective and redundant features. Through global self-attention adaptive modeling of feature dependencies at all time-series positions, a high-dimensional data feature is finally output, fusing local detailed features with global time-series correlation features. This high-dimensional data feature fully preserves the time-series position information, multi-dimensional indicator correlation information, long and short time-series evolution patterns, and key node feature information of the medical time-series data, exhibiting strong robustness, strong interpretability, and high representation accuracy. This significantly improves the accuracy and stability of subsequent medical time-series trend extrapolation.

[0073] S5. Generate a time state sequence based on the high-dimensional data features, and predict the data trend of the user to be analyzed based on the time state sequence.

[0074] In this embodiment of the invention, the time series Transformer decoder of the time series generation module in the target trend inference model performs autoregressive iterative inference according to a preset time step. Specifically, at the current time step, the time series generation module receives the current hidden state of the high-dimensional data feature mapping, calculates it through the internal transformation unit of the network, and outputs the single-step state vector corresponding to the current time. At the same time, the time series generation module updates the internal hidden state, merges the current hidden state with the newly generated single-step state vector to obtain the updated hidden state, and passes it to the next time step. The above inference process is iterated until the preset prediction duration and preset total number of time steps are reached, at which point single-step generation stops, and the time state sequence is obtained.

[0075] Specifically, see Figure 8 As shown, predicting the data trend of the user to be analyzed based on the time state sequence includes: S81. Perform masked self-attention calculation on the time state sequence to obtain a self-attention feature sequence; S82. Perform cross-attention calculation on the self-attention feature sequence to obtain the cross-attention feature sequence; S83. Perform nonlinear residual mapping on the cross-attention feature sequence to obtain the deep hidden feature sequence; S84. Use a fully connected layer to map the deep hidden feature sequence to the original predicted value; S85. The original predicted value is constrained and corrected using the medical data constraint model in the target trend extrapolation model to obtain the data trend of the user to be analyzed.

[0076] In this embodiment of the invention, when the time series generation module generates the physical state sequence corresponding to the predicted trend of the next T time nodes, it introduces an attention mechanism to focus on the impact of changes in historical physiological indicators on the future state, thereby improving the continuity and accuracy of time series generation.

[0077] Specifically, masked self-attention computation is performed on the time-state sequence to prevent future information leakage by using temporal masking, and the dependencies between physiological indicators at different times within the sequence are mined to generate a self-attention feature sequence; then cross-attention computation is carried out to achieve information fusion between temporal features and pathophysiological prior features to construct a cross-attention feature sequence. Furthermore, by fitting the complex nonlinear laws of disease evolution through nonlinear residual mapping, deep hidden feature sequences are extracted. Fully connected layers are then used to perform dimensional mapping on these deep hidden features, generating original predicted values ​​for physiological indicators. Since the original predictions obtained solely through network fitting may contain unreasonable values ​​that violate medical principles, a medical data constraint model is used. This model, based on pathophysiological partial differential equations and physiological indicator boundary constraints, corrects the original predicted values, eliminating predictions that do not conform to physiological mechanisms. The final output is a biologically plausible data trend for the user being analyzed, i.e., a predicted data sequence at multiple prediction time points constrained by the medical data constraint model.

[0078] For example, in the healthcare field, data trends can predict the evolution of a disease's physical state and extract key time points (such as the time when the disease progresses to a severe stage or the time when complications occur). Meanwhile, in the fintech field, combined with insurance terms, the probability of claims at different time points can be calculated using the following probabilistic statistical model:

[0079] in for The probability of claims being paid at any given time. It is the Sigmoid activation function. To preset the weighting coefficients of physiological indicators, This is the offset. for Time-constrained prediction data.

[0080] In detail, through Map the output to the [0,1] interval to ensure that the probability of claim settlement conforms to the statistical laws of probability (0≤1). ≤1); The weighting coefficients for physiological indicators are determined based on the degree of influence of different physiological indicators on claims events. For example, the weighting coefficients for core indicators such as tumor volume and blood glucose concentration are relatively large, and specific values ​​can be obtained by fitting insurance actuarial data. The offset, obtained by fitting actuarial data, is used to adjust the baseline value of the claim probability, ensuring that the calculation of the claim probability conforms to the actual needs of insurance business. The extraction of key time nodes, combined with the provisions of the insurance terms regarding the triggering conditions for claims (such as critical illness claims requiring tumor volume to reach a specific threshold, blood glucose concentration to remain above the standard for a certain period of time, etc.), accurately extracts the time nodes that can trigger claims, calculates the claim probability of that node, and provides an accurate basis for insurance claim review.

[0081] In another embodiment of the present invention, based on the probability of claims settlement and the predicted data trend Set a threshold for adjusting the insured amount. When, it triggers an increase in the sum insured; when When this is triggered, the sum assured is reduced, among which , The preset threshold is determined by actuarial rules; specifically, The insured amount is increased by a threshold (e.g., 0.6). When the probability of claiming at a certain point in time reaches or exceeds this threshold, it indicates that the patient's condition has worsened and the risk of claiming has increased, triggering an increase in the insured amount (e.g., an increase of 20%) to protect the patient's insurance rights. The threshold for reducing the sum assured is set at 0.3. When the probability of a claim falls below this threshold at a certain point in time, it indicates that the patient's condition is under control and the risk of a claim is reduced, triggering a reduction in the sum assured (e.g., a 10% reduction), thereby reducing the operational risk of the insurance company. The extent of the sum assured adjustment is determined in conjunction with actuarial rules and the patient's disease progression to ensure the rationality and fairness of the adjustment.

[0082] Furthermore, based on the disease's physical state evolution trajectory corresponding to data trends, the time points and numerical ranges of abnormal fluctuations in physiological indicators are identified to determine health management intervention points. When physiological parameters exceed normal ranges, corresponding intervention suggestions are provided (such as dietary adjustments, medication, and regular checkups). Specifically, based on medical standards, normal ranges for various physiological indicators are set, and the constraint prediction data generated by the model is monitored in real time. When physiological indicators at a certain point in time exceed the normal range, that point is identified as a health management intervention point. Based on the type and degree of abnormality of the physiological indicators, targeted intervention suggestions are generated. For example, when blood glucose concentration is too high, it is recommended to reduce the intake of high-sugar and high-carbohydrate foods, increase aerobic exercise, and adjust the dosage of hypoglycemic drugs; when tumor volume grows abnormally, it is recommended to adjust the treatment plan and increase the frequency of follow-up examinations. At the same time, the intervention points and intervention suggestions are integrated into the insurance business support report to provide precise guidance for insurance companies to carry out health management services, help patients control their condition, and reduce claims risks.

[0083] In this embodiment of the invention, the data trend of the user to be analyzed is predicted by time state series, which can balance the model's fitting accuracy to historical data distribution and the degree of satisfaction with physical constraints, solve the problem of conflict between the two in the prior art, and effectively improve the accuracy of data trend prediction.

[0084] Meanwhile, data trends can serve as the trajectory of the physical state of the disease at different points in the future, thereby quantifying the probability of claims, determining the threshold for dynamic adjustment of the insured amount, and identifying health management intervention points. This eliminates the need for manual secondary processing, improving the efficiency and accuracy of insurance actuarial science, claims review, and health management, and reducing insurance business risks.

[0085] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0086] like Figure 9 The diagram shown is a functional block diagram of a trend extrapolation device for time-series medical data provided in an embodiment of the present invention.

[0087] In this embodiment of the disclosure, a trend extrapolation device for time-series medical data is provided, which corresponds one-to-one with the trend extrapolation method for time-series medical data described in the above embodiments. For example... Figure 9 As shown, the trend extrapolation device 100 for time-series medical data can be installed in an electronic device. According to its functions, the trend extrapolation device 100 for time-series medical data includes a data processing module 101, a trend extrapolation model construction module 102, a model training module 103, a high-dimensional data feature extraction module 104, and a data trend extrapolation module 105. Detailed descriptions of each functional module are as follows: Data processing module 101 is used to acquire multi-dimensional medical data and medical theoretical data, and to perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; The trend inference model construction module 102 is used to construct a medical data constraint model based on the medical theoretical data, embed the medical data constraint model into the pre-constructed initial inference model, and generate a trend inference model. Model training module 103 is used to train the parameters of the trend inference model using the standard dataset to obtain the target trend inference model; The high-dimensional data feature extraction module 104 is used to extract high-dimensional data features of the medical time-series data by using the target trend inference model when receiving the medical time-series data of the user to be analyzed. The data trend inference module 105 is used to generate a time state sequence based on the high-dimensional data features, and predict the data trend of the user to be analyzed based on the time state sequence.

[0088] In one embodiment, when the data processing module 101 performs data standardization processing on the multi-dimensional medical data to obtain a standard dataset, it is used to: Identify outliers and missing values ​​in the multi-dimensional medical data; The outliers are removed from the multi-dimensional medical data, and the missing values ​​in the multi-dimensional medical data are filled with preset values ​​to obtain filled medical data; The medical data is then standardized to obtain a standard dataset.

[0089] In one embodiment, when the trend extrapolation model construction module 102 executes the construction of a medical data constraint model based on the medical theoretical data, it is used to: Theoretical data on different physiological mechanisms were extracted from the aforementioned medical theoretical data; Fit the partial differential equations corresponding to each of the physiological mechanisms based on the theoretical data; The partial differential equations are calibrated to obtain a medical data constraint model.

[0090] In one embodiment, when the model training module 103 performs parameter training on the trend inference model using the standard dataset to obtain the target trend inference model, it is used to: The initial feature vector of the standard dataset is extracted using the trend inference model, and feature perturbation is added to the initial feature vector to obtain the perturbation feature vector; The perturbation feature vector is used to generate predicted trend data, and the medical data constraint model in the trend inference model is used to constrain the predicted trend data to obtain constrained predicted trend data. The loss function value of the trend inference model is calculated based on the constrained trend prediction data; The trend inference model is updated in reverse based on the loss function value to obtain the target trend inference model.

[0091] In one embodiment, when the model training module 103 executes the constraint model using the medical data constraint model in the trend extrapolation model to constrain the predicted trend data and obtain constrained predicted trend data, it is used to: The predicted trend data is divided into windows to obtain windowed trend data; The objective function for optimizing the window trend data is constructed using the aforementioned medical data constraint model. Solving the objective function yields the corrected data subsequence; Physiological range constraints are applied to the corrected data subsequence to obtain window constraint data for the window trend data; By aggregating the constraint data from each window, constraint prediction trend data is obtained.

[0092] In one embodiment, when the high-dimensional data feature extraction module 104 extracts high-dimensional data features from the medical time-series data using the target trend inference model, it is used to: Calculate the location-encoded feature vector corresponding to the medical time-series data; The target trend inference model is used to perform multi-layer encoding on the location encoding feature vector to obtain the target encoding feature vector; Self-attention calculation is performed on the target encoded feature vector to obtain the high-dimensional data features of the medical time series data.

[0093] In one embodiment, when the data trend inference module 105 performs the task of predicting the data trend of the user to be analyzed based on the time state sequence, it is used to: Perform masked self-attention calculation on the time state sequence to obtain a self-attention feature sequence; Perform cross-attention calculation on the self-attention feature sequence to obtain the cross-attention feature sequence; The deep hidden feature sequence is obtained by performing nonlinear residual mapping on the cross-attention feature sequence; The deep hidden feature sequence is mapped to the original predicted value using a fully connected layer; The original predicted values ​​are constrained and corrected using the medical data constraint model in the target trend extrapolation model to obtain the data trend of the user to be analyzed.

[0094] In this invention, the specific limitations of the trend extrapolation device for time-series medical data can be found in the limitations of the trend extrapolation method for time-series medical data described above, and will not be repeated here. Each module in the aforementioned trend extrapolation device for time-series medical data can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0095] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the server-side functions or steps of a trend extrapolation method for time-series medical data.

[0096] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the client-side functions or steps of a trend extrapolation method for time-series medical data.

[0097] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Acquire multi-dimensional medical data and medical theoretical data, and perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; A medical data constraint model is constructed based on the medical theoretical data, and the medical data constraint model is embedded into a pre-constructed initial inference model to generate a trend inference model; The trend projection model is trained with the standard dataset to obtain the target trend projection model; Upon receiving the medical time-series data of the user to be analyzed, the high-dimensional data features of the medical time-series data are extracted using the target trend inference model. A time-state sequence is generated based on the high-dimensional data features, and the data trend of the user to be analyzed is predicted based on the time-state sequence.

[0098] In the several embodiments provided by this invention, it should be understood that the disclosed devices and apparatuses can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0099] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0100] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0101] In some embodiments of this example, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in the above embodiments.

[0102] The readable storage medium of the present invention stores a computer program, which, when executed by a processor of an electronic device, can perform the following: Acquire multi-dimensional medical data and medical theoretical data, and perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; A medical data constraint model is constructed based on the medical theoretical data, and the medical data constraint model is embedded into a pre-constructed initial inference model to generate a trend inference model; The trend projection model is trained with the standard dataset to obtain the target trend projection model; Upon receiving the medical time-series data of the user to be analyzed, the high-dimensional data features of the medical time-series data are extracted using the target trend inference model. A time-state sequence is generated based on the high-dimensional data features, and the data trend of the user to be analyzed is predicted based on the time-state sequence.

[0103] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0104] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0105] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0106] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0107] Those skilled in the art will understand that all or part of the processes in 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, and when executed, it can include the processes of the embodiments of the above methods. Furthermore, any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0109] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0110] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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, and should all be included within the protection scope of the present invention.

[0111] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

Claims

1. A method for trend extrapolation of time-series medical data, characterized in that, The method includes: Acquire multi-dimensional medical data and medical theoretical data, and perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset; A medical data constraint model is constructed based on the medical theoretical data, and the medical data constraint model is embedded into a pre-constructed initial inference model to generate a trend inference model; The trend projection model is trained with the standard dataset to obtain the target trend projection model; Upon receiving the medical time-series data of the user to be analyzed, the high-dimensional data features of the medical time-series data are extracted using the target trend inference model. A time-state sequence is generated based on the high-dimensional data features, and the data trend of the user to be analyzed is predicted based on the time-state sequence.

2. The method for trend extrapolation of time-series medical data as described in claim 1, characterized in that, The process of standardizing the multi-dimensional medical data to obtain a standard dataset includes: Identify outliers and missing values ​​in the multi-dimensional medical data; The outliers are removed from the multi-dimensional medical data, and the missing values ​​in the multi-dimensional medical data are filled with preset values ​​to obtain filled medical data; The medical data is then standardized to obtain a standard dataset.

3. The method for trend extrapolation of time-series medical data as described in claim 1, characterized in that, The construction of the medical data constraint model based on the medical theoretical data includes: Theoretical data on different physiological mechanisms were extracted from the aforementioned medical theoretical data; Fit the partial differential equations corresponding to each of the physiological mechanisms based on the theoretical data; The partial differential equations are calibrated to obtain a medical data constraint model.

4. The method for trend extrapolation of time-series medical data as described in claim 1, characterized in that, The step of training the trend projection model with the parameters of the standard dataset to obtain the target trend projection model includes: The initial feature vector of the standard dataset is extracted using the trend inference model, and feature perturbation is added to the initial feature vector to obtain the perturbation feature vector; The perturbation feature vector is used to generate predicted trend data, and the medical data constraint model in the trend inference model is used to constrain the predicted trend data to obtain constrained predicted trend data. The loss function value of the trend inference model is calculated based on the constrained trend prediction data; The trend inference model is updated in reverse based on the loss function value to obtain the target trend inference model.

5. The trend extrapolation method for time-series medical data as described in claim 4, characterized in that, The step of using the medical data constraint model in the trend extrapolation model to constrain the predicted trend data to obtain constrained predicted trend data includes: The predicted trend data is divided into windows to obtain windowed trend data; The objective function for optimizing the window trend data is constructed using the aforementioned medical data constraint model. The optimization objective function is solved to obtain the corrected data subsequence; Physiological range constraints are applied to the corrected data subsequence to obtain window constraint data for the window trend data; By aggregating the constraint data from each window, constraint prediction trend data is obtained.

6. The method for trend extrapolation of time-series medical data as described in claim 1, characterized in that, The extraction of high-dimensional data features from the medical time-series data using the target trend extrapolation model includes: Calculate the location-encoded feature vector corresponding to the medical time-series data; The target trend inference model is used to perform multi-layer encoding on the location encoding feature vector to obtain the target encoding feature vector; Self-attention calculation is performed on the target encoded feature vector to obtain the high-dimensional data features of the medical time series data.

7. The method for trend extrapolation of time-series medical data as described in claim 1, characterized in that, The step of predicting the data trend of the user to be analyzed based on the time state sequence includes: Perform masked self-attention calculation on the time state sequence to obtain a self-attention feature sequence; Perform cross-attention calculation on the self-attention feature sequence to obtain the cross-attention feature sequence; The deep hidden feature sequence is obtained by performing nonlinear residual mapping on the cross-attention feature sequence; The deep hidden feature sequence is mapped to the original predicted value using a fully connected layer; The original predicted values ​​are constrained and corrected using the medical data constraint model in the target trend extrapolation model to obtain the data trend of the user to be analyzed.

8. A trend extrapolation device for time-series medical data, characterized in that, The trend extrapolation device for the time-series medical data includes: The data processing module is used to acquire multi-dimensional medical data and medical theoretical data, and to perform data standardization processing on the multi-dimensional medical data to obtain a standard dataset. The trend extrapolation model construction module is used to construct a medical data constraint model based on the medical theoretical data, embed the medical data constraint model into the pre-constructed initial extrapolation model, and generate a trend extrapolation model. The model training module is used to train the parameters of the trend inference model using the standard dataset to obtain the target trend inference model. The high-dimensional data feature extraction module is used to extract high-dimensional data features of the medical time-series data when receiving the medical time-series data of the user to be analyzed, using the target trend inference model. The data trend extrapolation module is used to generate a time state sequence based on the high-dimensional data features, and to predict the data trend of the user to be analyzed based on the time state sequence.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the trend extrapolation method for time-series medical data as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the trend extrapolation method for time-series medical data as described in any one of claims 1 to 7.

Images