A Fetal Heart Rate Signal Classification Method Based on Spatiotemporal Graph Convolutional Networks

Abstract

This invention provides a fetal heart rate (FHR) signal classification method based on a spatiotemporal graph convolutional network, comprising the following steps: acquiring and preprocessing fetal heart rate signals; extracting node feature attributes from the fetal heart rate signals to generate an initial static graph; constructing a dynamic graph based on the initial static graph and an adaptive adjacency matrix, updating the information propagation of nodes in the dynamic graph at each time step, and generating an updated dynamic graph; constructing an optimized spatiotemporal graph convolutional network model, inputting the updated dynamic graph into the optimized spatiotemporal graph convolutional network model, and outputting the classification result of the fetal heart rate signal. This invention introduces a dynamic graph mechanism and optimizes traditional spatiotemporal graph convolutional networks, effectively capturing the spatial and temporal dependencies between nodes, realizing cross-time dimension information transmission, inferring the pathological causality of FHR data, and improving the accuracy of FHR signal classification.

Description

Technical Field

[0001] This invention relates to the field of modern medical signal processing technology, and in particular to a method for classifying fetal heart rate signals based on spatiotemporal graph convolutional networks. Background Technology

[0002] Acute and chronic fetal distress, including fetal hypoxia, acidosis, and abnormal fetal movement, are among the most significant factors affecting fetal health and safety. Timely detection of chronic fetal distress and the implementation of effective diagnostic and treatment strategies are crucial. Clinically, cardiotocography (CTG) is widely used in prenatal and intrapartum diagnosis. Fetal heart rate (FHR) signals provide valuable information about fetal health and are the only directly available source of information for clinicians. However, current mainstream CTG interpretation relies primarily on subjective diagnosis by obstetricians and gynecologists, leading to inconsistencies between doctors and even within the same doctor at different times, thus affecting the accuracy of CTG interpretation. Therefore, research on intelligent CTG classification models based on FHR signals is essential for achieving accurate and timely diagnosis of chronic fetal distress.

[0003] While advances in machine learning have provided various tools for intelligent assisted diagnosis of chronic fetal distress, limitations such as the inability to fully capture spatiotemporal dependencies within and between time series data hinder the further clinical application of data-driven methods. Furthermore, compared to most univariate time series classification tasks, FHR signals, as typical medical data, exhibit significant temporal instability and complex, variable pathological characteristics. Unlike time-invariant classification tasks, the information hidden in FHRs in chronic fetal distress diagnosis is richer and more difficult to uncover. Therefore, a method capable of accurately handling the spatiotemporal dependencies of FHRs is needed.

[0004] In view of this, there is an urgent need to provide a classification algorithm for FHR signals to effectively process the complex spatiotemporal dependencies of FHR signals, thereby achieving a more intelligent and accurate pathological diagnosis of chronic fetal distress. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art and provide a fetal heart rate signal classification method based on spatiotemporal graph convolutional networks. The invention designs a dynamic graph mechanism, combines a static graph structure and an adaptive adjacency matrix, and uses an improved spatiotemporal graph neural network (STGNNs) model architecture to process message propagation of dynamic graph data, thereby achieving effective classification based on FHR signals.

[0006] To solve the above problems, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for classifying fetal heart rate signals based on spatiotemporal graph convolutional networks, comprising the following steps:

[0008] Step S1: Acquire fetal heart rate signal and perform preprocessing;

[0009] Step S2: Extract node feature attributes from the fetal heart rate signal to generate an initial static graph;

[0010] Step S3: Construct a dynamic graph based on the initial static graph and the adaptive adjacency matrix, update the information propagation of nodes in the dynamic graph at each time step, and generate the updated dynamic graph;

[0011] Step S4: Construct an optimized spatiotemporal graph convolutional network model, input the updated dynamic graph into the optimized spatiotemporal graph convolutional network model, and output the classification results of the fetal heart rate signal.

[0012] The optimized spatio-temporal graph convolutional network model includes three cascaded ST-GCN (Spatio-Temporal Graph Convolutional Networks) modules, an average pooling layer, a fully connected layer, and a softmax activation function. Each ST-GCN module includes a graph convolutional network block OGCN, a temporal convolutional network block TCN, and a residual block. The residual block is used to implement residual connections between the graph convolutional network block OGCN and the temporal convolutional network block TCN.

[0013] Furthermore, the graph convolutional network block OGCN comprises a 3×3 channel-wise convolution, a 1×1 pointwise convolution, and an activation function, which are sequentially concatenated.

[0014] Furthermore, the Temporal Convolutional Network (TCN) block includes a convolutional layer and an activation function.

[0015] Furthermore, step S3 includes the following steps:

[0016] S3.1 Construct the node embedding matrix of the source node and the target node, and perform sparsification and normalization on the node embedding matrix to obtain the adaptive adjacency matrix.

[0017] S3.2. Combine the initial static graph and the adaptive adjacency matrix to construct the dynamic graph;

[0018] S3.3 Introduce diffusion graph convolution, and update the node features of the current dynamic time series through a bidirectional random walk model to obtain the updated dynamic graph.

[0019] Secondly, the present invention provides a fetal heart rate signal classification system for implementing the above method, comprising the following modules:

[0020] The data processing module is used to acquire fetal heart rate signals and perform preprocessing.

[0021] The graph construction module is used to generate a static graph based on the preprocessed fetal heart rate signal, construct a dynamic graph based on the static graph, and update the information propagation of nodes.

[0022] The fetal heart rate signal classification module inputs the updated dynamic graph into the optimized spatiotemporal graph convolutional network model to obtain the fetal heart rate signal classification results.

[0023] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing machine-executable instructions executable by the processor, the processor executing the machine-executable instructions to implement the above-described method.

[0024] Fourthly, the present invention provides a machine-readable storage medium storing machine-executable instructions that, when invoked and executed by a processor, cause the processor to implement the above-described method.

[0025] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0026] (1) The present invention constructs a graph structure learning module, which structurates multiple clinical feature indicators into a static graph structure, automatically learns the hidden spatiotemporal dependencies, is suitable for time series without a fixed graph structure, and overcomes the defect that most GNNs methods in the prior art often rely on predefined graph structures.

[0027] (2) This invention adds a dynamic graph mechanism, which constructs a dynamic graph structure based on the constructed static graph. By reducing the dependence on the static graph structure, it enhances the learning of the dynamic changes of nodes, thus solving the problem of randomness that may occur in daily CTG monitoring and thus causing error accumulation.

[0028] (3) The present invention designs an optimized spatiotemporal graph convolutional network model to process this specific dynamic graph structure, effectively captures the spatial and temporal dependencies between nodes, realizes information transmission across time dimensions, infers the pathological causality of FHR data, and improves the accuracy of FHR signal classification. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention.

[0030] Figure 2 This is a schematic diagram of the ST-GCN module structure. Detailed Implementation

[0031] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0032] like Figure 1 As shown, this embodiment provides a fetal heart rate signal classification method based on a spatiotemporal graph neural network, including the following steps:

[0033] S1: Building a Graph Structure Learning Module

[0034] First, fetal heart rate (FHR) signals are acquired and preprocessed. For FHR signal data, the FHR data of the i-th pregnant woman is labeled as X. i Set the time step to s, the sliding window interval to w, and slice the data to obtain n = X. i / s of node information. For each node information, a total of 25 physiological and pathological features across 4 categories are structured to form node feature attributes α. i =[a i1 ,a i2 ,...,a im (m=25). The first 23 features are clinical expert consensus summarized from authoritative guidelines and practices of the International Federation of Gynecology and Obstetrics (see Table 1 below), while the last two are important medical record text data of the fetus and pregnant woman. These micro-data are the gold standard for clinical heart rate variability analysis, and some features are obtained by the following formula:

[0035] Low-frequency normalization (LFnorm):

[0036]

[0037] Intermediate frequency normalization (MFnorm):

[0038]

[0039] High Frequency Normalization (HFnorm):

[0040]

[0041] Among them, VLFpower, LFpower, MFpower, and HFpower represent very low frequency power, low frequency power, medium frequency power, and high frequency power, respectively, and Totalpower represents the total power.

[0042] Table 1

[0043]

[0044]

[0045] Therefore, we can obtain X.i Node attribute matrix A i =[α1,α2,...,α n This matrix will be considered as the initial adjacency matrix for the i-th pregnant woman, with all elements being learnable parameters. Then, the values ​​of the adjacency matrix are optimized through graph structure learning training to obtain static graph data.

[0046] S2: Provides a dynamic graph mechanism to solve the randomness problem in CTG monitoring.

[0047] Step S2 includes the following steps:

[0048] 1) Constructing the node embedding matrix. This invention randomly initializes two nodes E1, E2 ∈ R with learnable parameters. N ×d E1 and E2 represent the node embedding matrices of the source node and the target node, respectively, and N and d represent the number of nodes and features, respectively.

[0049] 2) Obtain the adaptive adjacency matrix. The spatial weights obtained by the dot product of E1 and E2 are sparsified using the ReLU activation function to eliminate weak connections between spatial dependencies; then, Softmax is used for normalization to obtain the adaptive adjacency matrix A. adp .

[0050] 3) Obtain the dynamic graph structure. Combine this with the static graph data A. s Together they form the dynamic graph structure A of each node g The entire process can be represented by the following formula:

[0051]

[0052] 4) Update the graph structure of the dynamic time series. Introduce diffusion graph convolution, using a bidirectional random walk model to simulate information flow between nodes, and update the node features of the current dynamic time series X. The modeling process of diffusion graph convolution is shown in the following equation:

[0053]

[0054] Where θ1, θ2, and θ3 are trainable parameters that vary with different input samples; K represents the kernel size; D in and D out Let A be the degree matrix, where the first matrix represents the in-degree matrix and the second matrix represents the out-degree matrix. These are calculated from the topological structure of the dynamic graph and are used for the normalization operation of the adaptive adjacency matrix; A is the original adjacency matrix; A d T is the adjacency matrix obtained by performing a Laplace transformation on the original adjacency matrix, where T is the transpose.

[0055] This invention captures the information propagation of nodes at each time step through multi-step forward and backward diffusion to obtain the spatial features of the nodes. This reduces the dependence on the static structure of the graph, enhances the learning of dynamic changes in nodes, and thus improves the model's performance in graph data processing. Finally, it outputs a dynamic graph X. G .

[0056] S3: Establish a feature extraction model based on an optimized spatiotemporal graph convolutional network, update and process the message propagation of the dynamic graph, and realize the classification of fetal heart rate signals.

[0057] The optimized spatiotemporal graph convolutional network model proposed in this invention is a feature extraction network model with multiple optimized ST-GCN (Spatial Temporal Graph Convolutional Networks) structural units. It originates from the basic units of classic STGNNs but has been structurally improved to utilize dynamic graphs X. G As input to the model, the final output is the classification result of the FHR signal, including normal signals and disease signals, used to classify the fetal health status.

[0058] The optimized spatiotemporal graph convolutional network model includes three cascaded ST-GCN modules, an average pooling layer, a fully connected layer, and a Softmax activation function.

[0059] like Figure 2 As shown, each ST-GCN module includes an Optimized Graph Convolutional Network (OGCN), a Temporal Convolutional Network (TCN), and a residual block. The OGCN block and the TCN block are residually connected through the residual block. The OGCN block includes a 3×3 channel-wise convolutional layer, a 1×1 pointwise convolutional layer, and an activation function layer, which are sequentially connected in series. The TCN block includes a convolutional layer and an activation function layer.

[0060] The OGCN block aims to capture the spatial dependencies between different nodes, while the TCN block is used to simulate the temporal dynamics of FHR signals, enabling the model to consider the evolution of these signals over time. The improved OGCN employs a lightweight network structure, first applying a 3×3 channel-wise convolution to extract node information within each neighborhood, followed by a cascaded 1×1 pointwise convolution for channel-wise fusion. The features are then further input into the TCN layer to refine and enhance the temporal dependency features of the time series.

[0061] Specifically, the node output of each convolution operation in the spatial domain can be represented as follows:

[0062]

[0063] Where N(v) ti )={v ti |d(v ti ,v tj )≤D} represents node v ti The set of neighbors, d(v) ti ,v tj ) is node v tj to v ti Minimum distance metric for any path; normalization term Z ti (v tj )={v tu |y ti (v tu )=y ti (v tj )} represents the cardinality of the corresponding subset, y ti (v tj p(v) is the current label mapping; ti ,v tj ) and w(v ti ,v tj ) are the sampling function and the weighting function, respectively.

[0064] Furthermore, the optimized STGCNs model of this invention expands the spatial graph convolutional network to the spatiotemporal domain, so that the neighborhood information also includes the temporal information of the nodes, thereby obtaining a new set of spatiotemporal neighbors:

[0065]

[0066]

[0067] Where k represents the time dimension frame, kt defines the time neighbor range; D′ represents node v tj to v ti The minimum distance; H is the time range of the control neighborhood; y ST This is a label mapping in the spatiotemporal domain.

[0068] To further verify the effectiveness of the method of the present invention, 552 sets of CTG data from the CTU-UHB intrapartum CTG database jointly collected by the Czech Technical University and Brno University Hospital were selected for comparative testing. The accuracy of pathological diagnosis was compared with four benchmark models: LS-SVM+GA (based on genetic algorithm and least squares support vector machine), DT-CTNet (based on two-layer AI model architecture), Hybrid-FHR (based on cross-modal feature fusion), LSTM (based on long short-term memory network), and ViT (based on VisionTransformer). The accuracy (ACC), F1 score, and precision (P) of the test set are shown in Table 2 below.

[0069] Table 2

[0070]

[0071]

[0072] As shown in Table 2, the method of the present invention has significant advantages.

[0073] The above embodiments are not intended to limit the present invention, and the present invention is not limited to the above embodiments. Any embodiment that meets the requirements of the present invention is within the protection scope of the present invention.

Claims

1. A method for classifying fetal heart rate signals based on spatiotemporal graph convolutional networks, characterized in that, Includes the following steps: Step S1: Acquire fetal heart rate signal and perform preprocessing; Step S2: Extract node feature attributes from the fetal heart rate signal to generate an initial static graph; Step S3: Construct a dynamic graph based on the initial static graph and the adaptive adjacency matrix, update the information propagation of nodes in the dynamic graph at each time step, and generate the updated dynamic graph; Specifically, the following steps are included: S3.1 Construct the node embedding matrix of the source node and the node embedding matrix of the target node ,right , The spatial weights obtained by dot product are then sparsified and normalized to obtain an adaptive adjacency matrix. ; S3.2, Initial static diagram and adaptive adjacency matrix By combining these elements, a dynamic graph can be constructed. ; indicates as: (4)； S3.3 Introduce diffusion graph convolution to update the current dynamic time series through a bidirectional random walk model. The node features are used to obtain the updated dynamic graph; the modeling process of the diffusion graph convolution is shown in the following equation: (5)； in All of these are trainable parameters that vary with different input samples; K represents the kernel size. and The first matrix represents the in-degree matrix, and the second matrix represents the out-degree matrix. These are calculated from the topological structure of the dynamic graph and are used for the normalization operation of the adaptive adjacency matrix. This is the original adjacency matrix; This is the adjacency matrix obtained by performing a Laplace transformation on the original adjacency matrix. For transpose; Step S4: Construct an optimized spatiotemporal graph convolutional network model. Input the updated dynamic graph into the optimized spatiotemporal graph convolutional network model and output the classification result of the fetal heart rate signal. The optimized spatiotemporal graph convolutional network model includes three cascaded ST-GCN modules, an average pooling layer, a fully connected layer, and a Softmax activation function. Each ST-GCN module includes a graph convolutional network block OGCN, a temporal convolutional network block TCN, and a residual block. The residual block is used to implement the residual connection between the graph convolutional network block OGCN and the temporal convolutional network block TCN. The graph convolutional network block OGCN includes a 3×3 channel-wise convolution, a 1×1 pointwise convolution, and an activation function, which are sequentially connected in series.

2. The fetal heart rate signal classification method based on spatiotemporal graph convolutional networks according to claim 1, characterized in that, The Temporal Convolutional Network (TCN) consists of a convolutional layer and an activation function.

3. The fetal heart rate signal classification method based on spatiotemporal graph convolutional networks according to claim 1 or 2, characterized in that, The activation functions in the temporal convolutional network block TCN and the graph convolutional network block OGCN are ReLU activation functions.

4. A fetal heart rate signal classification system implementing the method as described in any one of claims 1-3, characterized in that, Includes the following modules: The data processing module is used to acquire fetal heart rate signals and perform preprocessing. The graph construction module is used to generate a static graph based on the preprocessed fetal heart rate signal, construct a dynamic graph based on the static graph, and update the information propagation of nodes. The fetal heart rate signal classification module inputs the updated dynamic graph into the optimized spatiotemporal graph convolutional network model to obtain the fetal heart rate signal classification results.

5. An electronic device comprising a processor and a memory, characterized in that, The memory stores machine-executable instructions that can be executed by the processor to implement the method as described in any one of claims 1-3.

6. A machine-readable storage medium storing machine-executable instructions, characterized in that, When the machine-executable instructions are invoked and executed by the processor, the machine-executable instructions cause the processor to implement the method as described in any one of claims 1-3.

Images