Method for fusing morphological and timing features, electronic device and computer readable storage medium

Abstract

The present application relates to a kind of morphological and timing feature fusion method, electronic equipment and computer readable storage medium;First data after pre-processing is input to first processing, and multiple sub-images are generated based on multi-scale decomposition;Multiple sub-images contain information features representing different direction features;Feature extraction is carried out to multiple sub-images, and feature map is reconstructed by inverse transform, and morphological feature vector is obtained after pooling processing;First data after pre-processing is input to second processing, and is segmented into ordered subgraph sequence based on sliding window;Feature extraction is carried out to each subgraph in subgraph sequence, forms feature sequence, and timing feature vector is extracted by timing model;Morphological feature vector and timing feature vector are spliced, and fusion feature vector is generated;Fusion feature vector is input to classifier, and the detection result obtained;The beneficial effects of the present application increase feature fusion effect.

Description

Technical Field

[0001] This invention relates to the fields of computer science and biomedical engineering, specifically to a method for fusing morphological and temporal features, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Cardiovascular disease is a major chronic disease threatening human health, and early and accurate diagnosis is crucial for its prevention and treatment. Current methods for analyzing high-resolution electrocardiograms (ECGs) largely rely on physician experience, resulting in high subjectivity and inefficiency. Although machine learning techniques have been introduced into ECG analysis, current methods often focus only on single-modal features, emphasize waveform morphology while ignoring temporal variations, or analyze only rhythm patterns while neglecting morphological features, leading to insufficient utilization of key diagnostic information. Particularly when attempting to fuse multiple features, the failure to effectively address feature heterogeneity issues results in poor feature fusion performance, hindering further improvements in detection accuracy and efficiency. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method, electronic device and computer-readable storage medium for fusing morphological and temporal features, so as to solve the problem of fusing multiple features mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for fusing morphological and temporal features, comprising the following steps; The preprocessed first data is input into the first processing, and multiple sub-images are generated based on multi-scale decomposition. Multiple sub-images contain information features representing characteristics in different directions; Feature extraction is performed on multiple sub-images, and feature maps are reconstructed through inverse transformation. After pooling, morphological feature vectors are obtained. The preprocessed first data is input into the second processing, and is divided into an ordered sequence of subgraphs based on a sliding window; Feature extraction is performed on each subgraph in the subgraph sequence to form a feature sequence, and a temporal feature vector is extracted using a temporal model. The morphological feature vector and the temporal feature vector are concatenated to generate a fused feature vector; The fused feature vector is input into the classifier to obtain the detection result.

[0005] A further optimization of the present invention includes multi-scale decomposition comprising two-dimensional wavelet transform.

[0006] Used to reduce the original input size and improve computational efficiency.

[0007] A further optimization of the present invention is that multiple sub-images include an approximate component (LL), a horizontal detail component (HL), a vertical detail component (LH), and a diagonal detail component (HH).

[0008] A further optimization of this invention involves using a convolutional network to extract the main morphological features cA, cH, cV, and cD of LL, HL, LH, and HH respectively, and reconstructing the feature maps through inverse wavelet transform.

[0009] The main features are extracted using cA, and the remaining parts are used for inverse transformation to recover the size. The reconstructed feature maps are easy to input into the next layer of the network.

[0010] A further optimization of the present invention involves using a convolutional network to extract features from each subgraph in the subgraph sequence, forming a feature sequence.

[0011] Fully extract electrocardiogram information at various scales.

[0012] In a further optimization of the present invention, the classifier includes a linear classifier or a support vector machine.

[0013] In order to obtain the result from the vector.

[0014] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method of fusing morphological and temporal features as described above.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of fusing morphological and temporal features as described above.

[0016] The beneficial effects of this invention are: 1. Based on multi-scale decomposition, multiple sub-images are generated to achieve effective separation of the global morphology and local details of electrocardiogram signals.

[0017] 2. By extracting features from sub-images and reconstructing them through inverse transformation, noise interference is effectively suppressed while preserving key waveform details.

[0018] 3. By using a sliding window to segment electrocardiogram images into an ordered sub-image sequence, static images were successfully transformed into dynamic sequence data.

[0019] 4. By concatenating and fusing morphological and temporal feature vectors, the advantages of static structural information and dynamic temporal information can be complemented, thereby improving the ability to identify complex pathological manifestations. Attached Figure Description

[0020] Figure 1 This is a diagram showing the sequential arrangement of the 12 leads in this invention.

[0021] Figure 2This is a schematic diagram of the first processing operation of the present invention.

[0022] Figure 3 This is a schematic diagram of the second processing operation of the present invention.

[0023] Figure 4 This is a schematic diagram of the morphological and temporal feature fusion operation of the present invention.

[0024] Figure 5 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the embodiments shown in the accompanying drawings.

[0026] Reference Figure 1-5 As shown, the steps are as follows: Step 1.1: Collect 12-lead ECG signals. Divide each dataset into training and test sets according to the ratio of 70%, 15%, and 15%. Perform preprocessing operations such as downsampling, Gaussian filtering, and baseline drift removal on the data to unify the sampling rate and signal length of all ECG signals from all sources.

[0027] Step 1.2: Arrange the processed ECG signal waveforms according to the lead order (I, II, III, aVL, aVF, aVR, V1, V2, V3, V4, V5, V6) and plot them as high-resolution images, such as... Figure 2 As shown.

[0028] Step 2: Morphological feature extraction: Step 2.1: Perform a two-dimensional wavelet transform on the electrocardiogram (ECG) image, decomposing it into four sub-images (cA, cH, cV, cD), each one-quarter the size of the original image, representing the main features, vertical waveform features, horizontal waveform features, and oblique waveform features, respectively. The two-dimensional wavelet transform formula is as follows: ; ; ; ; Where I 2i,2j I 2i,2j+1 I 2i+1,2j、I2i+1,2j+1就是图像块I分解后的四个部分； Step 2.2: Apply a convolutional network to cA to extract the main morphological features of the ECG waveform. After convolution, cA becomes cA*. The convolutional network formula is as follows: ; in These are the learnable parameters of the convolutional network. For the extracted feature tensor; Step 2.3: Add cA * The cH, cV, and cD features are reconstructed using inverse wavelet transform to form a feature map enhanced by a convolutional module. The inverse wavelet transform formula is as follows: ; ; ; ; Step 2.4: Repeat the above steps to finally obtain the morphological feature vector V through global pooling. shape The entire process of step 2 is as follows: Figure 2 As shown; Step 3: Temporal Feature Extraction: Step 3.1: Use the sliding window method to segment the ECG image into an ordered sequence of sub-images, each sub-image containing a local waveform of a single lead; the sliding window formula is as follows: ; Where w and h are the window sizes, and L... k This is a subgraph of the original image; Step 3.2: Use a convolutional network to extract features for each sub-image, and add a blank image to the end of the feature sequence formed by all sub-images to store the information of each lead; the convolutional network formula is as follows: ; in These are the learnable parameters of the convolutional network. For the extracted feature tensor Step 3.3: Input the feature sequence into a time series model to finally obtain the time series feature vector V of the electrocardiogram. time The entire process of step 3 is as follows: Figure 3 As shown; the time series model formula is as follows: ; in These are the parameters of the time series model; Step 4: Integration and Classification: Step 4.1: Convert the morphological feature vector V shape and time series feature vector V time Concatenate into a fused feature vector V fusion The fusion formula is as follows: ; Step 4.2: Convert the feature vector V fusion The input is a linear classifier, which yields the detection results for cardiovascular diseases. The entire processing flow in step 4 is as follows: Figure 4As shown; Step 5: Obtain and test the model Step 5.1: Update network parameters using cross-entropy loss to train and optimize the model. The trained model is then validated for generalization performance on multiple datasets. The cross-entropy loss formula is as follows: ; Where N is the total number of samples, y i For real labels, V i fusion The fused feature vector Step 5.2: The new data is preprocessed in the same way as in Step 1, and then used to detect cardiovascular diseases using the model obtained in Step 3.

[0029] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for fusing morphological and temporal features, characterized in that, Includes the following steps; The preprocessed first data is input into the first processing, and multiple sub-images are generated based on multi-scale decomposition. Multiple sub-images contain information features representing characteristics in different directions; Feature extraction is performed on multiple sub-images, and feature maps are reconstructed through inverse transformation. After pooling, morphological feature vectors are obtained. The preprocessed first data is input into the second processing, and is divided into an ordered sequence of subgraphs based on a sliding window; Feature extraction is performed on each subgraph in the subgraph sequence to form a feature sequence, and a temporal feature vector is extracted using a temporal model. The morphological feature vector and the temporal feature vector are concatenated to generate a fused feature vector; The fused feature vector is input into the classifier to obtain the detection result.

2. The method for fusing morphological and temporal features according to claim 1, characterized in that, Multiscale decomposition includes two-dimensional wavelet transform.

3. The method for fusing morphological and temporal features according to claim 1, characterized in that, Multiple sub-images contain approximate components (LL), horizontal detail components (HL), vertical detail components (LH), and diagonal detail components (HH).

4. The method for fusing morphological and temporal features according to claim 3, characterized in that, The main morphological features cA, cH, cV and cD of LL, HL, LH and HH are extracted using a convolutional network, and the feature maps are reconstructed by inverse wavelet transform.

5. The method for fusing morphological and temporal features according to claim 1, characterized in that, Features are extracted from each subgraph in the subgraph sequence using a convolutional network to form a feature sequence.

6. The method for fusing morphological and temporal features as described in claim 1, characterized in that, Classifiers include linear classifiers or support vector machines.

7. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes a computer program, it implements the method of fusing morphological and temporal features as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method of fusing morphological and temporal features as described in any one of claims 1-6.

Images