Multi-classification method of arrhythmia based on multi-scale residual neural network and multi-channel data fusion

Figures

• 

  FIGURE 1. V5 lead and MLII lead ECG.

• 

  FIGURE 2. Data pre-processing.

• 

  FIGURE 3. The overall architecture of a multiscale residual neural network.

• 

  FIGURE 4. The three modules used in the network structure proposed here: (A) represents the conv_block module, (B) represents the identity_block module, and (C) represents the multi-scale module.

• 

  FIGURE 5. Flowchart of diagnostic methods.

• 

  FIGURE 6. Experimental results on the MIT-BIH dataset: (A) is the experimental result of ResNet, and (B) is the experimental result of the proposed method.

• 

  FIGURE 7. Single-channel and multi-channel experimental results: (A) is the experimental result of V5 channel, (B) is the experimental result of MLII, and (C) is the experimental result of multi-channel data.

• 

  FIGURE 8. ResNet of the sudden cardiac death holter dataset and the experimental results of the method in this paper: (A) is the experimental results of ResNet, and (B) is the experimental results of the method in this paper.

• 

  FIGURE 9. ResNet and the t-SNE and confusion matrix diagram of the method presented herein.

• 

  FIGURE 10. Confusion matrix and cluster analysis results for single-channel and multi-channel data fusion.

• 

  FIGURE 11. ResNet of the sudden cardiac death holter dataset and the prediction results of this method: (A) is the network of ResNet, (B) is the method of this paper.

• 

  FIGURE 12. The prediction results of three categories in the sudden cardiac death holter dataset: (A) is the network of ResNet, (B) is the method of this paper.