A method, system, and terminal for analyzing electronic medical records of patients with myocardial infarction or heart failure.

Abstract

This invention discloses a method, system, and terminal for analyzing electronic medical records of patients with myocardial infarction or heart failure. The method includes: acquiring the patient's electronic medical record and inputting it into a spatiotemporal dynamic graph neural network model to obtain multiple time series graphs; acquiring sparse adjacency matrices and node attribute information from the multiple time series graphs, performing spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs; performing downsampling processing on the multiple multivariate time series graphs to obtain multiple multivariate time series features; performing ensemble learning processing on the spatiotemporal dynamic graph neural network model based on the multiple multivariate time series features, and outputting the electronic medical record analysis results corresponding to the patient's electronic medical record. This invention, by constructing a spatiotemporal dynamic graph neural network model, can dynamically extract spatiotemporal information corresponding to physiological parameters in the patient's electronic medical record, effectively improving the accuracy of the electronic medical record analysis results output.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for analyzing electronic medical records of patients with myocardial infarction or heart failure. Background Technology

[0002] Heart failure (HF) is a common complication following acute myocardial infarction (AMI). HF represents a severe and terminal stage of cardiovascular disease, especially after myocardial infarction, which has high rates of disability and mortality. However, methods for analyzing HF caused by myocardial infarction using electronic medical records of patients with myocardial infarction or HF have not yet been established.

[0003] To address the aforementioned issues, existing technologies typically extract static physiological variables from patients' electronic medical records using neural network models for analysis. However, these methods do not consider the interactions and dynamic changes between variables within the electronic medical record, resulting in inaccurate analysis results that fail to meet user needs.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide an electronic medical record analysis method, system, terminal, and computer-readable storage medium for patients with myocardial infarction or heart failure. This aims to solve the problem that existing technologies do not consider the interactions and dynamic changes between variables in the electronic medical records of patients with myocardial infarction or heart failure, resulting in inaccurate electronic medical record analysis results.

[0006] To achieve the above objectives, the present invention provides a method for analyzing electronic medical records of patients with myocardial infarction or heart failure, the method comprising the following steps:

[0007] The patient's electronic medical record is obtained and input into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs;

[0008] Obtain sparse adjacency matrices and node attribute information from multiple time series graphs, and perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs;

[0009] Multiple multivariate time series plots are downsampled to obtain multiple multivariate time series features;

[0010] The spatiotemporal dynamic graph neural network model is integrated and learned based on multiple multivariate time series features to output the electronic medical record analysis results corresponding to the patient's electronic medical record.

[0011] Optionally, the method for analyzing electronic medical records of patients with myocardial infarction or heart failure, wherein acquiring the patient's electronic medical record and inputting the patient's electronic medical record into a spatiotemporal dynamic graph neural network model to obtain multiple time series graphs specifically includes:

[0012] The patient's electronic medical record is obtained, the patient's electronic medical record is input into a spatiotemporal dynamic graph neural network model, and the multivariate time series in the patient's electronic medical record is extracted through the spatiotemporal dynamic graph neural network model.

[0013] The multivariate time series is segmented multiple times to obtain multiple segmented multivariate time series;

[0014] The multiple segmented multivariate time series are grouped according to their time slots to obtain multiple time series diagrams.

[0015] Optionally, in the electronic medical record analysis method for patients with myocardial infarction or heart failure, the spatiotemporal dynamic graph neural network model includes a time dynamic graph isomorphic network module.

[0016] The step of obtaining sparse adjacency matrices and node attribute information from multiple time series graphs, and performing spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs specifically includes:

[0017] Multiple time series graphs are input into the time dynamic graph isomorphic network module, and the sparse adjacency matrix and node attribute information of the multiple time series graphs are obtained through the time dynamic graph isomorphic network module.

[0018] The sparse adjacency matrix and node attribute information are aggregated, dynamically merged, self-loopedly added, and feature-connected to obtain multiple multivariate time series graphs.

[0019] Optionally, in the method for analyzing electronic medical records of patients with myocardial infarction or heart failure, the step of obtaining sparse adjacency matrices in multiple time-series graphs specifically includes:

[0020] Obtain connection weight information between different nodes in multiple time series graphs, and construct a graph adjacency matrix based on the connection weight information;

[0021] The graph adjacency matrix is ​​initialized to obtain a sparse adjacency matrix.

[0022] Optionally, in the method for analyzing electronic medical records of patients with myocardial infarction or heart failure, the initialization process of the graph adjacency matrix to obtain a sparse adjacency matrix specifically includes:

[0023] Obtain random sample values ​​from multiple time series graphs, and fill the graph adjacency matrix according to the random sample values ​​to obtain a filled graph adjacency matrix;

[0024] The sparse adjacency matrix of the filled graph is processed by sparse transformation to obtain a sparse adjacency matrix.

[0025] Optionally, in the electronic medical record analysis method for patients with myocardial infarction or heart failure, the spatiotemporal dynamic graph neural network model includes a graph pooling module;

[0026] The downsampling process performed on multiple multivariate time series plots to obtain multiple multivariate time series features specifically includes:

[0027] Multiple multivariate time series graphs are input into the graph pooling module, and the graph pooling module obtains the number of nodes in the multiple multivariate time series graphs.

[0028] Multiple node numbers are downsampled using convolution operations to obtain multiple multivariate time series features.

[0029] Optionally, in the electronic medical record analysis method for patients with myocardial infarction or heart failure, the spatiotemporal dynamic graph neural network model includes an average pooling layer and a fully connected layer.

[0030] The step of integrating and learning the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features to output the electronic medical record analysis results corresponding to the patient's electronic medical record specifically includes:

[0031] Multiple multivariate time series features are sequentially input into the average pooling layer and the fully connected layer to obtain multiple multivariate time series results;

[0032] An ensemble learning strategy is employed to obtain data features corresponding to multiple time scales in the multiple multivariate time series results, and the spatiotemporal dynamic graph neural network model is subjected to ensemble learning processing based on the data features corresponding to the multiple time scales.

[0033] The spatiotemporal dynamic graph neural network model, after ensemble learning processing, outputs the electronic medical record analysis results corresponding to the patient's electronic medical record.

[0034] Furthermore, to achieve the above objectives, the present invention also provides an electronic medical record analysis system for patients with myocardial infarction or heart failure, wherein the electronic medical record analysis system for patients with myocardial infarction or heart failure includes:

[0035] The electronic medical record input module is used to acquire the patient's electronic medical record and input the patient's electronic medical record into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs;

[0036] The spatial information learning module is used to acquire sparse adjacency matrices and node attribute information from multiple time series graphs, and to perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs.

[0037] The downsampling processing module is used to perform downsampling processing on multiple multivariate time series plots to obtain multiple multivariate time series features;

[0038] The medical record analysis result output module is used to perform integrated learning processing on the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features, and output the electronic medical record analysis result corresponding to the patient's electronic medical record.

[0039] In addition, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an electronic medical record analysis program for patients with myocardial infarction or heart failure stored in the memory and executable on the processor, wherein when the electronic medical record analysis program for patients with myocardial infarction or heart failure is executed by the processor, the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described above are implemented.

[0040] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an electronic medical record analysis program for patients with myocardial infarction or heart failure, and the electronic medical record analysis program for patients with myocardial infarction or heart failure, when executed by a processor, implements the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described above.

[0041] In this invention, a patient's electronic medical record (EMR) is acquired and input into a spatiotemporal dynamic graph neural network model to obtain multiple time series graphs. Sparse adjacency matrices and node attribute information are obtained from these time series graphs. Spatial information learning processing is performed on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs. Downsampling processing is applied to these multivariate time series graphs to obtain multiple multivariate time series features. The spatiotemporal dynamic graph neural network model is then integrated with these multivariate time series features to output the EMR analysis results corresponding to the patient's EMR. This invention, by constructing a spatiotemporal dynamic graph neural network model, can dynamically extract spatiotemporal information corresponding to physiological parameters in the patient's EMR and accurately capture the dynamic dependencies and interactions between these physiological parameters. This significantly improves the accuracy of the EMR analysis results output and provides more reliable support for hospitals' clinical decision-making for patients with myocardial infarction or heart failure. Attached Figure Description

[0042] Figure 1 This is a flowchart of a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0043] Figure 2 This is a schematic diagram of the overall framework of STDGNN, a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0044] Figure 3 This is a schematic diagram of the TDGIN framework of a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0045] Figure 4 This is a schematic diagram of the TGPool framework of a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0046] Figure 5 This is a schematic diagram of a complete list of physiological variables used for experimental analysis in a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0047] Figure 6 This is a schematic diagram of the data processing process of a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention;

[0048] Figure 7 This is a schematic diagram of the baseline model and STDGNN cross-dataset AUROC scoring benchmark of a preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure of the present invention.

[0049] Figure 8 This is a structural diagram of a preferred embodiment of the electronic medical record analysis system for patients with myocardial infarction or heart failure of the present invention;

[0050] Figure 9 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages 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.

[0052] This invention relates to the field of electronic medical record analysis for in-hospital patients with myocardial infarction or heart failure, specifically an application that uses a deep learning model to analyze patient medical records in a real clinical environment and generate analysis results for physician reference. Currently, predicting the severity of in-hospital myocardial infarction or heart failure faces two major challenges. First, the low sampling frequency of many physiological parameters results in a limited number of data available for early detection and analysis of the severity of myocardial infarction / heart failure. Therefore, the complexity of human dynamic systems and the lack of sufficient information or data make this multi-step early prediction a difficult and challenging task. Second, using multivariate time series (MTS) in electronic medical records for prediction and analysis remains a complex and long-standing area of ​​focus. A core characteristic of MTS is that each individual time series not only contains its own dynamic change pattern but also interacts with other series, exhibiting rich temporal correlations. For example, an increased heart rate may be associated with elevated blood pressure, and this relationship may change over time and with the patient's condition. This inherent dynamism and interconnectivity makes MTS data extremely rich in information, but also increases the difficulty of analysis. However, current work on predicting the severity of myocardial infarction or heart failure rarely considers the dynamism of time series and the correlations between different variables. Therefore, this invention addresses the above-mentioned technical problems by proposing a novel graph neural network—the Spatio-Temporal Dynamic Graph Neural Network (STDGNN). This network dynamically extracts spatiotemporal information of physiological parameters from patients' electronic medical records within hospitals to determine the severity of myocardial infarction or heart failure, thus aiding hospitals and physicians in further analysis. Specifically, STDGNN constructs graphical representations of multiple physiological parameters and employs temporal convolution and a graph neural network framework to learn the features of each variable. Furthermore, this invention develops a dynamic graph mechanism based on MTS data. This mechanism integrates multiple time periods of varying lengths and adaptively learns the dynamic evolution of the graph structure through gradient descent. This mechanism aims to enhance the model's ability to capture dynamic dependencies between adjacent variables, thereby improving the effectiveness of graph learning in complex medical data contexts.

[0053] One of the most widely used scoring systems for predicting the severity of in-hospital patient illness is the Acute Physiology and Chronic Health Evaluation II (APACHE II) algorithm. This system uses 15 variables, including physiological measurements, to provide a snapshot of a patient's recent condition. With the proliferation of Electronic Health Records (EHRs) and the deployment of patient workflow management software, the calculation process for APACHE II has been automated. However, a major drawback of this algorithm is its failure to utilize available historical patient information. While subsequent versions have optimized and improved upon this limitation, adding more detail, they have also increased the required number of variables. Therefore, due to its high complexity, the APACHE II algorithm remains largely unused in clinical practice.

[0054] Another widely used scoring system is the Simplified Acute Physiology Score II (SAPS-II). SAPS-II assesses the severity of a patient's illness using demographic information, vital signs, and laboratory test results, and simplifies the calculation process of APACHE II. Because the SAPS-II score exhibits an S-shaped relationship with severity, it can be used to predict the severity of illness in hospitalized patients. SAPS-III, a subsequent version of SAPS-II, uses 17 variables, including physiological and disease manifestation variables, further improving predictive accuracy.

[0055] The Oxford Acute Severity of Illness Score (OASIS) is also a system used to assess the severity of in-hospital illness, especially in the first 24 hours after admission. OASIS predicts in-hospital severity based on various clinical variables, such as demographics, comorbidities, vital signs, and laboratory results. Another scoring system is the Sequential Organ Failure Assessment (SOFA), which calculates the risk of organ failure. Although not specifically designed to predict severity, it is positively correlated with in-hospital severity.

[0056] It is important to note that these critical illness scoring systems are designed for a broad patient population and are not specifically designed for patients with heart failure. Current scoring systems for assessing heart failure primarily focus on detecting the risk and severity of heart failure or cardiovascular disease; however, these systems are not suitable for directly predicting the severity of myocardial infarction or heart failure.

[0057] Furthermore, the essence of traditional scoring systems is based on capturing the linear relationship between predictor variables and outcomes, while the relationship between physiological parameters and severity is usually very complex and may not be strictly linear. Capturing non-linear relationships can significantly enhance the predictive ability of severity.

[0058] With the increasing prevalence of end-stage renal disease (EHR) and the continuous advancement of computing technology, machine learning (ML) and deep learning (DL) methods have been gradually applied to predict the severity of heart failure patients. For example, various machine learning models, including Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Light Gradient Boosting Machine (LGBM), are used to predict the severity of heart failure patients. Alternatively, Extreme Gradient Boosting (XGBoost) models can be used to predict the severity of heart failure patients, and the prediction results can be interpreted using SHapley Additive exPlanations (SHAP). On the same dataset, the same model can be used to further predict the stratified severity of heart failure patients, with external validation added. Long Short-Term Memory (LSTM) network models can be used to analyze the impact of various combinations of physiological parameters on the severity of heart failure, or interpretable deep learning models based on Transformer can be used to model and predict the occurrence of heart failure.

[0059] Significant progress has been made in multivariate time series (MTS) models in recent years. While some landmark MTS models have not yet been specifically used to predict the severity of heart failure, they have demonstrated great potential in this task. For example, recurrent neural networks (RNNs), by introducing recurrent connections, can model and remember sequential data, effectively capturing time dependencies and performing well when dealing with missing data in the sequence. Residual networks (ResNets), through the stacking of residual blocks, address the vanishing gradient and model degradation problems in deep neural network training. The improved ResNetPlus introduces attention mechanisms and cross-layer connections, further enhancing the model's attention to and integration of important features in the sequence.

[0060] Other representative time series models include InceptionTime and InceptionTimePlus, which enhance the model's ability to understand long-term dependencies and local features by capturing multi-scale features. Furthermore, the Multilayer Perceptron (MLP) and its improved version, the Gated Multilayer Perceptron (GMLP), better capture global dependencies in sequences through self-attention mechanisms. TimesNet combines Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) and further improves predictive power by capturing local features and global dependencies in time series through temporal attention mechanisms. Utilizing the Transformer architecture and introducing spatial pyramid pooling significantly enhances the model's ability to capture complex patterns, becoming a leading technique for time series prediction.

[0061] In recent years, Graph Neural Networks (GNNs) have seen rapid development in the application of multivariate time series (MTS) data. GNNs process graph information through local aggregation mechanisms, where the embedding vector of each node is computed by recursively aggregating and transforming information from its neighbors. Over the years, several variants of GNNs have been proposed, such as Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and Graph Isomorphism Networks (GINs). These variants perform well in handling dependencies in time series data. For example, they utilize graph learning layers to capture hidden spatial relationships between time series data, and further, by capturing correlations between sequences and temporal dependencies in the spectral domain, they effectively improve the prediction accuracy of MTS.

[0062] Despite their impressive performance, most graphical models still only construct a single static graph of the entire time series data, neglecting the dynamic processes that change over time. Especially in hospital environments, where human physiological states are often unstable, capturing the dynamic dependencies between physiological variables is crucial for accurate prediction.

[0063] While some traditional scoring systems and machine learning / deep learning-based models offer some predictive ability in current in-hospital patient severity prediction methods, they suffer from several limitations. First, traditional severity scoring systems, such as APACHE II, SAPS-II, and SOFA, primarily rely on static physiological variables, failing to effectively capture the complex and dynamic temporal correlations between variables. These systems assume a linear relationship between predictor variables and severity, neglecting the potential nonlinear and dynamic patterns of change between physiological parameters and severity. Furthermore, existing MTS-based methods often only address changes in single time series, paying less attention to interactions and time dependencies between different variables. A core characteristic of MTS is that each time series not only contains its own dynamic changes but also exhibits complex interconnections with other series, relationships that have not been fully explored and utilized in current techniques. Therefore, existing technologies remain significantly insufficient in capturing the dynamic dependencies between physiological variables and the interactions between multiple variables in in-hospital patients.

[0064] To address the shortcomings of existing technologies, this invention proposes a novel Spatio-Temporal Dynamic Graph Neural Network (STDGNN) aimed at improving the performance of patient severity prediction models for myocardial infarction or heart failure in capturing time-series dynamics and multivariate interactions. By dynamically extracting spatiotemporal information of physiological parameters from patients' electronic medical records in hospitals, this invention constructs graphical representations of multiple physiological variables and effectively captures the feature changes of each variable using a framework of temporal convolution and graph neural networks. Its core lies in designing a dynamic graph mechanism based on MTS (Mean Transformation Theory), which adaptively learns the graph structure that evolves over time and combines multiple time scales to capture the dynamic dependencies and interactions of physiological parameters. Through this method, this invention can more accurately capture the complex relationships between patients' physiological parameters, significantly improving the predictive ability for the severity of myocardial infarction or heart failure and providing more reliable support for in-hospital clinical decision-making.

[0065] The preferred embodiment of the electronic medical record analysis method for patients with myocardial infarction or heart failure described in this invention, such as... Figure 1 As shown, the electronic medical record analysis method for patients with myocardial infarction or heart failure includes the following steps:

[0066] Step S10: Obtain the patient's electronic medical record and input the patient's electronic medical record into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs.

[0067] Specifically, the patient's electronic medical record is acquired, input into a spatiotemporal dynamic graph neural network model, and multivariate time series are extracted from the patient's electronic medical record through the spatiotemporal dynamic graph neural network model; the multivariate time series are segmented multiple times to obtain multiple segmented multivariate time series; the multiple segmented multivariate time series are grouped according to their time slots to obtain multiple time series graphs.

[0068] like Figure 2 As shown, this invention designs a Spatiotemporal Dynamic Graph Neural Network (STDGNN). To capture dynamic changes caused by local time periods, the input MTS (i.e., the multivariate time series in the patient's electronic medical record of this invention) is segmented multiple times (multiple segmentations generate multiple dynamic graphs, using fixed-length time segments; for example, a multivariate time series with a total length of 24 can be divided into two segments of 12, three segments of 8, and four segments of 6), and these segments are grouped into different time slots. For each group, the time series graph is constructed based on its respective time slot, with each variable representing a node in the time series graph.

[0069] Step S20: Obtain sparse adjacency matrices and node attribute information from multiple time series graphs, and perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs.

[0070] This invention utilizes the feature information of graph nodes and learnable parameter modules to initialize a set of graph adjacency matrices (graph adjacency matrices enable time-series graphs to be computed in graph neural networks; all parameters in the graph adjacency matrix are continuously updated as the model is trained; subsequently, a sparse adjacency matrix is ​​obtained through the initialization process of the graph adjacency matrix) to express the initial relationships between different nodes. At the same time, the elements of the graph adjacency matrix are iteratively updated through the graph learning module.

[0071] In this invention, TDGIN (the temporal dynamic graph isomorphic network module in this invention) and TGPool (the graph pooling module in this invention) are designed together to form a graph learning module. Each generated time series graph is sent to TDGIN for learning, and TGPool is used for graph pooling. TDGIN can learn the dynamic relationships between different nodes through the dynamic changes of the time series graph.

[0072] Specifically, multiple time series graphs are input into the time dynamic graph isomorphic network module. The connection weight information between different nodes in the multiple time series graphs is obtained through the time dynamic graph isomorphic network module. A graph adjacency matrix is ​​constructed based on the connection weight information. Random sampled values ​​from the multiple time series graphs are obtained and the graph adjacency matrix is ​​filled based on the random sampled values ​​to obtain a filled graph adjacency matrix. The filled graph adjacency matrix is ​​then subjected to sparse transformation to obtain a sparse adjacency matrix.

[0073] Given the complexity of the human body system, the impact of various physiological parameters on disease is deeply interconnected and constantly changing. Therefore, constructing appropriate time series graphs to address this issue is particularly important. This invention treats each physiological parameter as a node in a time series graph. When switching time slots, different time series graphs provide the model with information about the changes in connection weights between different nodes. This invention uses a graph adjacency matrix to represent the hidden dependencies between variables (i.e., nodes in the time series graph). For the initialization of the graph adjacency matrix A, a learnable weight matrix W is first set, and then random sampled values ​​from a uniform distribution [-limit, limit] are used to fill the weight matrix W. Where limit is a random sample value, N s It is the source node, N t This is the target node. Subsequently, the diagonal elements of the graph adjacency matrix A are set to -∞ to eliminate self-loops (this invention captures the relationships between variables (i.e., nodes in the sequence graph), and a self-loop node is connected to itself, therefore self-loops are unnecessary). Finally, by implementing Top-k sparsity, the most important k node connections for each node are preserved, and the remaining values ​​are set to zero. This transforms the graph adjacency matrix A into a sparse form, reducing computational cost. The formula is as follows: A sparse =Top-k(A, k), where A sparse Let A be a sparse adjacency matrix, Top-k(·) is the sparsification process, A is the graph adjacency matrix, and k is the number of nodes to be retained.

[0074] Furthermore, node attribute information from multiple time series graphs is obtained, and the sparse adjacency matrix and node attribute information are aggregated, dynamically merged, self-loopedly added, and feature-connected to obtain multiple multivariate time series graphs.

[0075] For the TDGIN (Time Dynamic Graph Isomorphic Network module in this invention): the spatial relationships of different physiological parameters changing over time can be explored through the interactions generated by node features in the time series graph. GNNs are powerful in processing graph-structured data, focusing on information transmission and feature aggregation between nodes. Existing GNN methods have shown excellent learning capabilities in processing static graph data. However, for dynamic graphs composed of multiple time series graphs transformed by MTS, existing methods are not suitable for learning the dynamic changes in the time dimension of the time series graph. Therefore, in this invention, an improved GIN method is proposed, called Time Dynamic Graph Isomorphic Network (TDGIN), as follows: Figure 3 As shown, its purpose is to extract spatiotemporal information from the dynamic graph.

[0076] Each time series graph contains the same nodes, so the feature vectors corresponding to the nodes form a feature vector set. (where v is a graph node, This is the feature vector corresponding to the first graph node in the feature vector set. It is the number of nodes. (A positive integer). The feature vector set consists of slices of the MTS sequence (i.e., segmented by time periods, used in this invention). (This represents a slice, where d is the slice length). The spatial information in the graph is learned by combining the structural information of the time sequence graph with the attribute information of the nodes. The initialized sparse adjacency matrix A... sparse The formula for representing the structural information of a graph is as follows: Where N is the total number of nodes; in, The final graph obtained is the multivariate time series graph in this invention. Let be the transpose of the sparse adjacency matrix; where, It is A sparse The standardization of , where i is the row vector of the sparse adjacency matrix, and j is the vertical dimension of the sparse adjacency matrix. A sparse These are learnable parameters that can be continuously updated during training. In the dynamic graph processing step, for each node, the feature vector set... It will be added to its previous timing diagram The node feature values ​​are used to integrate feature information from the previous time slot. Then, self-loops are added to each graph to mitigate information loss within the nodes themselves. Indicates dynamic time merging; This indicates a self-cyclic addition, where ∈ is a correction factor used to reflect the impact of time or parameter variations on the sum of its parts. The impact.

[0077] Next, an MLP layer is designed to learn the feature changes of each node in the dynamic graph (obtained from the time series graph). The MLP contains two temporal convolutional layers to capture the dynamic changes in the time dimension, and then the features of all time slots in the dynamic graph are concatenated to obtain the output of TDGIN (i.e., the MTS graph, multivariate time series graph). The core formula of TDGIN is as follows:

[0078] in, Let represent the GIN output of node v in layer k, where s∈N* is the number of time slots, and s>1. This represents the weight of the edge from node i to node j in the graph at different times.

[0079] Step S30: Downsample the multiple multivariate time series plots to obtain multiple multivariate time series features.

[0080] This invention includes a TGPool (i.e., the graph pooling module in this invention). TGPool is a differentiable method with learnable dynamic change information in the time dimension. Integrating it into the hierarchical pooling method can avoid the problem of flat pooling. The result of each group is obtained after average pooling and fully connected layers.

[0081] Regarding the TGPool set in this invention: Pooling is a common operation in deep learning and is necessary after the classifier generates new features. The essence of pooling is feature downsampling, but graphs are limited to non-Euclidean data structures that are difficult to manipulate. General pooling methods, such as max pooling, mean pooling, and summation pooling, aggregate a vector or even a feature matrix into multiple features, which can lead to the potential problem of flat pooling, resulting in the loss of a large amount of information. Furthermore, this invention constructs an MTS graph, which preserves the dynamic information of the time series. However, there is currently no mainstream pooling method that can handle this problem. Therefore, this invention proposes a novel time series graph pooling method called TGPool, as follows... Figure 4 As shown, it incorporates temporal features into graph pooling, alleviates the problems of flat pooling through hierarchical pooling, and preserves the features of MTS as much as possible.

[0082] Specifically, multiple multivariate time series graphs are input to the graph pooling module, and the graph pooling module obtains multiple node numbers in the multiple multivariate time series graphs; convolution operation is used to downsample the multiple node numbers to obtain multiple multivariate time series features.

[0083] This invention uses convolution to downsample the number of nodes, where the number of nodes is the number of channels in the convolutional layer. To avoid disrupting the receptive field window, the kernel size for this convolution is set to the same as the temporal convolution kernel size in TDGIN. For example, the shape of the graph in the input graph pooling module is (b, c, n, f), where b is the batch size, c is the number of graphs, n is the number of nodes, and f is the node features. This is achieved by transforming the multivariate time series graph into its first and second dimensions to ensure that the convolution operation downsamples the nodes. n... in and n out Let these be the number of input and output channels for the convolution, respectively. The specific formulas are as follows: in, The final feature obtained is K, which represents the convolution kernel. Therefore, K c It is the convolution kernel corresponding to the graph, K f It is the convolution kernel corresponding to the node features. This indicates that the input tensor is at position (b, n) in The element at position c+i, f+j). The element position (n) of the convolution kernel out n inThe tensor at (i, j).

[0084] Step S40: Perform integrated learning processing on the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features, and output the electronic medical record analysis results corresponding to the patient's electronic medical record.

[0085] To synthesize the dynamic changes of different time series graphs, this invention uses ensemble learning to integrate all network groups together, and the final output of the model is jointly determined based on the convergence degree of different local dynamic information in the time series. This improves the model's prediction accuracy and robustness while reducing overfitting.

[0086] Specifically, multiple multivariate time series features are sequentially input into the average pooling layer and the fully connected layer to obtain multiple multivariate time series results; an ensemble learning strategy is used to obtain data features corresponding to multiple time scales in the multiple multivariate time series results, and the spatiotemporal dynamic graph neural network model is subjected to ensemble learning processing based on the data features corresponding to the multiple time scales; the spatiotemporal dynamic graph neural network model after ensemble learning processing outputs the electronic medical record analysis results corresponding to the patient's electronic medical record.

[0087] In this invention, multiple multivariate time series features are processed through an average pooling layer and a fully connected layer to obtain multiple multivariate time series results, which are then integrated for learning.

[0088] Regarding the ensemble learning approach used in this invention: Ensemble learning is a powerful tool. By employing an ensemble learning strategy, the STDGNN model can be integrated for different time series segment lengths, significantly improving the model's predictive performance and generalization ability. Different time series segment lengths allow the model to capture data features across multiple time scales. Longer time series can reveal trends and long-term dependencies, while shorter time series can capture more subtle patterns and short-term fluctuations. By integrating these models at different time scales, information from each scale can be synthesized, enhancing the model's understanding of the complex dynamics of time series. Furthermore, ensemble learning typically reduces model variance, making the model more robust to noise and overfitting under single parameter settings. Therefore, the ensemble learning method used in this invention can improve the accuracy and reliability of the model without sacrificing interpretability. The prediction and training errors of ensemble learning are expressed as follows:

[0089] Let be the prediction expression for ensemble learning, where, This represents the predicted output of the ensemble learning model. This value is typically the average of the predictions from multiple models, used to enhance the accuracy and stability of the model's predictions; T represents the number of base models in the ensemble model, f t(X) represents the prediction result of the t-th base model for the input X. Each base model in the ensemble model will give a prediction value for the input X. This means averaging the predictions of all T base models to obtain the final ensemble prediction output. This is the training error expression for ensemble learning.

[0090] The novel deep learning model STDGNN set in this invention integrates a dynamic graph mechanism based on MTS data, which dynamically extracts spatiotemporal information from the electronic medical records of in-hospital patients to enhance the prediction of the severity of HF patients.

[0091] Regarding the design of the basic blocks in this invention: TDGIN and TGPool are designed to establish a dynamic dependency model. TDGIN can effectively extract spatiotemporal features from dynamic graphs. The TGPool method integrates temporal features into the graph structure, significantly improving the representational power of MTS.

[0092] Regarding the interpretability of deep learning in this invention: This invention reveals the complex and diverse dynamic relationships between physiological parameters through the dynamic graph of the STDGNN model. This analysis is crucial for understanding the patient's health status and predicting outcomes, providing valuable insights for clinical decision-making. In summary, the spatiotemporal dynamic graph neural network model constructed in this invention can extract spatiotemporal information from the electronic medical records of patients with myocardial infarction or heart failure. It can accurately obtain the interactions between parameters and variables in the medical records, as well as the dynamic changes of parameters and variables, thereby obtaining the output electronic medical record analysis results. These results can serve as a reference for hospitals or medical staff regarding the current status of patients with myocardial infarction or heart failure, facilitating further analysis of the patient's condition.

[0093] Technical effects of the present invention:

[0094] 1. Dynamically capturing multivariate time dependencies: STDGNN can adaptively learn the dynamic dependencies between different physiological parameters through a dynamic graph mechanism, especially the changes in time series. Compared with traditional static graph or time series methods, this can more effectively capture the complex interactions of physiological variables over time and improve the accuracy of prediction.

[0095] 2. Multi-timescale modeling capability: STDGNN combines multiple timescales, allowing the model to handle short-term and long-term dependencies simultaneously within different time windows. This makes STDGNN more advantageous than traditional models that rely on only a single timescale when capturing the complex and variable physiological states of hospitalized patients.

[0096] 3. No feature extraction required, reducing the impact of individual differences: STDGNN learns features directly from raw physiological data without the need for manual feature extraction beforehand. This not only reduces the complexity of data processing but also effectively reduces the impact of individual patient differences on prediction performance, ensuring the model's broad applicability across different patients.

[0097] Experimental verification process:

[0098] In this invention, two widely recognized and extensive public hospital databases, MIMIC-III and MIMIC-IV, are used to validate the model. These datasets provide a rich repository of information for analyzing the clinical characteristics and severity prediction of heart failure patients, including comprehensive analysis of dynamic physiological signals and multiple physiological variables. The MIMIC-III dataset includes demographic data, bedside vital sign measurements, laboratory test results, and patient severity-related information, laying the foundation for early identification and risk assessment of heart failure patients. As a subsequent product, the MIMIC-IV dataset expands the data scope by incorporating additional patient information, detailed medical process records, and updated clinical information, making it more suitable for medical research related to heart failure. To enhance the model's generalizability and applicability, commonly used physiological variables are extracted from both datasets for model input. Figure 5 As shown, data types can be categorized into static variables (SV), hourly chart variables (HCV), and waveform signals (WAVE). This experiment consists of two parts: a comparative experiment and an AHE prediction experiment. To ensure the accuracy of the results, this invention employs five-fold cross-validation for all experiments.

[0099] For the entire process of sample selection and data processing, such as Figure 6As shown, the first step of this invention involved identifying patients with heart failure when selecting samples from two large datasets. Patients under 18 years of age were subsequently excluded. Furthermore, patients transferred to the hospital or admitted more than twice during a single hospitalization were also excluded. This exclusion criterion was based on the understanding that patients with in-hospital stays of less than 24 hours typically undergo postoperative monitoring, and their monitoring data could lead to misleading predictions of in-hospital severity. Therefore, considering the research objectives and the above reasons, it was decided to exclude patients with in-hospital stays of less than 48 hours. The main objective of this invention is to investigate the influence of various physiological parameters on the severity of heart failure patients. In terms of data processing, to ensure the generalizability of the model, simplified methods commonly used for processing in-hospital datasets were employed. Specifically, SV and HCV data were extracted from clinical databases. Notably, the HCV variable was initially recorded at irregular time points. To standardize these data, this invention utilized interpolation techniques, selecting the closest recorded values ​​around specific time points to estimate the hourly value of each HCV variable. Additionally, for the WAVE data, the official hourly data version was used. Subsequently, these three types of data were merged, and data from the first 48 hours of each patient's hospitalization were extracted for analysis. Furthermore, it is worth noting that multiple hospitalizations of the same patient are considered separate samples. For example... Figure 7 As shown, the subset extracted from the MIMIC-III dataset consists of 549 positive samples and 7091 negative samples. Meanwhile, the subset derived from the MIMIC-IV dataset includes 723 positive samples and 8078 negative samples.

[0100] This invention divides the samples into training, validation, and test sets in an 8:1:1 ratio, ensuring that data from the same patient appears in only one set. Regarding model training parameters, the weight decay is set to 5, momentum to 0.9, and the initial learning rate to 0.01. All experiments were conducted using six NVIDIA RTX A6000 GPUs. For CPU, an Intel(R) Xeon(R) Gold 6240 CPU at 2.60GHz was used. For optimization, the AdamW optimizer was employed with an initial learning rate of 1e-3. The training epochs were set to 100, and the batch size to 64. To ensure fairness and model reliability, all baseline models involved in the experiments had their internal parameters optimized using grid search to achieve the best results. Specifically, the baseline model adjustments were strictly performed according to the official code. Furthermore, the STDGNN model of this invention is uniquely designed with three different time slot lengths: 6, 4, and 3, to more effectively capture dynamic changes in patient status. Each model undergoes 5x cross-validation to mitigate overfitting and optimize hyperparameters, thereby enhancing the robustness of the research results.

[0101] This invention compares the performance of various severity prediction models on the MIMIC-III and MIMIC-IV datasets, including rule-based SAPS II, traditional machine learning models, and deep learning models. The aim of this comparison is to evaluate their effectiveness in handling complex medical data, with a particular focus on the performance of the STDGNN model proposed in this invention.

[0102] Experimental results show that on the MIMIC-III dataset, the linear regression model exhibited the highest recall rate of 97.01%, surpassing other models in its ability to identify positive cases. Conversely, the XGBoost model led the field with an accuracy of 53.69%, highlighting its efficiency in positive prediction accuracy. Most notably, the STDGNN model developed in this invention performed exceptionally well on key performance metrics, achieving an AUC of 83.93% and an Fbeta score of 64.14%, demonstrating a significant advantage in overall performance. A similar pattern was observed in the analysis of the MIMIC-IV dataset. The linear regression model again achieved the highest recall rate, this time at 97.91%, though lower than its consistent ability to capture true positive results. The XGBoost model continued to lead in accuracy, reaching 63.76%, further demonstrating its precision strength. Notably, the STDGNN model of this invention maintained its exemplary performance advantage, achieving a highest AUC of 81.74% and a highest Fbeta score of 62.24%. These results demonstrate the robustness and superior predictive accuracy of the STDGNN model across various medical datasets.

[0103] In addition to the technical content of the embodiments of the present invention, the following can be added to the present invention: 1. The parameters and hyperparameters of each part of the model of the present invention can be changed; 2. The construction of the dynamic graph of the present invention can be designed according to the specific task; 3. The design of the present invention can be further extended to other medical fields, such as predicting the progression of chronic diseases or the long-term rehabilitation of patients through the STDGNN framework; 4. The present invention can analyze the weight of multiple variables on the prediction results, and therefore is also suitable for such tasks.

[0104] Furthermore, such as Figure 8 As shown, based on the above-mentioned electronic medical record analysis method for patients with myocardial infarction or heart failure, the present invention also provides an electronic medical record analysis system for patients with myocardial infarction or heart failure, wherein the electronic medical record analysis system for patients with myocardial infarction or heart failure includes:

[0105] The electronic medical record input module 51 is used to acquire the patient's electronic medical record and input the patient's electronic medical record into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs.

[0106] The spatial information learning module 52 is used to acquire sparse adjacency matrices and node attribute information in multiple time series graphs, and to perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs.

[0107] The downsampling processing module 53 is used to perform downsampling processing on multiple multivariate time series plots to obtain multiple multivariate time series features;

[0108] The medical record analysis result output module 54 is used to perform integrated learning processing on the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features, and output the electronic medical record analysis result corresponding to the patient's electronic medical record.

[0109] Furthermore, such as Figure 9 As shown, based on the above-mentioned electronic medical record analysis method and system for patients with myocardial infarction or heart failure, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 9 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0110] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores an electronic medical record analysis program 40 for patients with myocardial infarction or heart failure, which can be executed by the processor 10 to implement the electronic medical record analysis method for patients with myocardial infarction or heart failure of this application.

[0111] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the electronic medical record analysis method for patients with myocardial infarction or heart failure.

[0112] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0113] In one embodiment, when the processor 10 executes the electronic medical record analysis program 40 for patients with myocardial infarction or heart failure in the memory 20, the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described above are implemented.

[0114] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an electronic medical record analysis program for patients with myocardial infarction or heart failure, and the electronic medical record analysis program for patients with myocardial infarction or heart failure, when executed by a processor, implements the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described above.

[0115] In summary, this invention provides a method, system, terminal, and computer-readable storage medium for analyzing electronic medical records of patients with myocardial infarction or heart failure. The method includes: acquiring the patient's electronic medical record and inputting the patient's electronic medical record into a spatiotemporal dynamic graph neural network model to obtain multiple time series graphs; acquiring sparse adjacency matrices and node attribute information from the multiple time series graphs, performing spatial information learning processing on the sparse adjacency matrices and the node attribute information to obtain multiple multivariate time series graphs; performing downsampling processing on the multiple multivariate time series graphs to obtain multiple multivariate time series features; performing ensemble learning processing on the spatiotemporal dynamic graph neural network model based on the multiple multivariate time series features, and outputting the electronic medical record analysis results corresponding to the patient's electronic medical record. This invention, by constructing a spatiotemporal dynamic graph neural network model, can dynamically extract the spatiotemporal information corresponding to physiological parameters in the patient's electronic medical record, and can accurately capture the dynamic dependencies and interactions between physiological parameters in the patient's electronic medical record, significantly improving the accuracy of the electronic medical record analysis results output, and providing more reliable support for hospitals' clinical decision-making for patients with myocardial infarction or heart failure.

[0116] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0117] Of course, 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 (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0118] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for analyzing electronic medical records of patients with myocardial infarction or heart failure, characterized in that, The electronic medical record analysis method for patients with myocardial infarction or heart failure includes: The patient's electronic medical record is obtained and input into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs; Obtain sparse adjacency matrices and node attribute information from multiple time series graphs, and perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs; The spatiotemporal dynamic graph neural network model includes a time dynamic graph isomorphic network module; The step of obtaining sparse adjacency matrices and node attribute information from multiple time series graphs, and performing spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs specifically includes: Multiple time series graphs are input into the time dynamic graph isomorphic network module, and the sparse adjacency matrix and node attribute information of the multiple time series graphs are obtained through the time dynamic graph isomorphic network module. The sparse adjacency matrix and node attribute information are aggregated, dynamically merged, self-loopedly added, and feature-connected to obtain multiple multivariate time series graphs. The step of obtaining the sparse adjacency matrices in the multiple time series graphs specifically includes: Obtain connection weight information between different nodes in multiple time series graphs, and construct a graph adjacency matrix based on the connection weight information; The graph adjacency matrix is ​​initialized to obtain a sparse adjacency matrix; The initialization process of the graph adjacency matrix to obtain a sparse adjacency matrix specifically includes: Obtain random sample values ​​from multiple time series graphs, and fill the graph adjacency matrix according to the random sample values ​​to obtain a filled graph adjacency matrix; The sparse adjacency matrix of the filled graph is processed by sparse transformation to obtain a sparse adjacency matrix; Multiple multivariate time series plots are downsampled to obtain multiple multivariate time series features; The spatiotemporal dynamic graph neural network model is integrated and learned based on multiple multivariate time series features to output the electronic medical record analysis results corresponding to the patient's electronic medical record. The spatiotemporal dynamic graph neural network model includes an average pooling layer and a fully connected layer; The step of integrating and learning the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features to output the electronic medical record analysis results corresponding to the patient's electronic medical record specifically includes: Multiple multivariate time series features are sequentially input into the average pooling layer and the fully connected layer to obtain multiple multivariate time series results; An ensemble learning strategy is employed to obtain data features corresponding to multiple time scales in the multiple multivariate time series results, and the spatiotemporal dynamic graph neural network model is subjected to ensemble learning processing based on the data features corresponding to the multiple time scales. The spatiotemporal dynamic graph neural network model, after ensemble learning processing, outputs the electronic medical record analysis results corresponding to the patient's electronic medical record.

2. The electronic medical record analysis method for patients with myocardial infarction or heart failure according to claim 1, characterized in that, The process of acquiring the patient's electronic medical record and inputting it into a spatiotemporal dynamic graph neural network model to obtain multiple time series graphs specifically includes: The patient's electronic medical record is obtained, the patient's electronic medical record is input into a spatiotemporal dynamic graph neural network model, and the multivariate time series in the patient's electronic medical record is extracted through the spatiotemporal dynamic graph neural network model. The multivariate time series is segmented multiple times to obtain multiple segmented multivariate time series; The multiple segmented multivariate time series are grouped according to their time slots to obtain multiple time series diagrams.

3. The electronic medical record analysis method for patients with myocardial infarction or heart failure according to claim 1, characterized in that, The spatiotemporal dynamic graph neural network model includes a graph pooling module; The downsampling process performed on multiple multivariate time series plots to obtain multiple multivariate time series features specifically includes: Multiple multivariate time series graphs are input into the graph pooling module, and the graph pooling module obtains the number of nodes in the multiple multivariate time series graphs. Multiple node numbers are downsampled using convolution operations to obtain multiple multivariate time series features.

4. An electronic medical record analysis system for patients with myocardial infarction or heart failure, characterized in that, The electronic medical record analysis system for patients with myocardial infarction or heart failure is applied to the electronic medical record analysis method for patients with myocardial infarction or heart failure according to any one of claims 1-3, wherein the electronic medical record analysis system for patients with myocardial infarction or heart failure comprises: The electronic medical record input module is used to acquire the patient's electronic medical record and input the patient's electronic medical record into the spatiotemporal dynamic graph neural network model to obtain multiple time series graphs; The spatial information learning module is used to acquire sparse adjacency matrices and node attribute information from multiple time series graphs, and to perform spatial information learning processing on the sparse adjacency matrices and node attribute information to obtain multiple multivariate time series graphs. The downsampling processing module is used to perform downsampling processing on multiple multivariate time series plots to obtain multiple multivariate time series features; The medical record analysis result output module is used to perform integrated learning processing on the spatiotemporal dynamic graph neural network model based on multiple multivariate time series features, and output the electronic medical record analysis result corresponding to the patient's electronic medical record.

5. A terminal, characterized in that, The terminal includes: a memory, a processor, and an electronic medical record analysis program for patients with myocardial infarction or heart failure stored in the memory and executable on the processor. When the electronic medical record analysis program for patients with myocardial infarction or heart failure is executed by the processor, it implements the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described in any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an electronic medical record analysis program for patients with myocardial infarction or heart failure, which, when executed by a processor, implements the steps of the electronic medical record analysis method for patients with myocardial infarction or heart failure as described in any one of claims 1-3.

Images