Fetal ECG Signal Extraction from Maternal Abdominal ECG Signals Using Attention R2W-Net

Figures

• 

  Figure 1. Block diagram of the proposed method for extracting FECG from maECG.

• 

  Figure 2. Comparison of pre-processing of maECG and FECG signals.

• 

  Figure 3. The FECG signal extraction model based on Attention R2W-Net.

• 

  Figure 4. (a) The recurrent residual convolutional unit and (b) the unfolded version of the recurrent convolutional unit.

• 

  Figure 5. Structure diagram of the AG module.

• 

  Figure 6. Examples of FECG extraction on the ADFECGDB database by the proposed method. (a–d) represent channels I, II, III, and IV, respectively. The positions of the FECG signal, MECG signal, and MECG signal overlapping with the FECG signal in the maECG signal are indicated by ‘F’, ‘M’, and ‘F + M’, respectively.

• 

  Figure 7. Examples of FECG extraction on the PCDB database by the proposed method. (a,b) represent channels I and II of data a03, respectively. The positions of the FECG signal, MECG signal, and MECG signal overlapping with the FECG signal in the maECG signal are indicated by ‘F’, ‘M’, and ‘F + M’, respectively.

• 

  Figure 8. The impact of loss functions on model performance.

• 

  Figure 9. The impact of the number of convolutional layers on model performance.

• 

  Figure 10. Structural diagrams of the forward convolution block (a), the recurrent convolution block (b), and the residual convolution block (c).

• 

  Figure 11. Comparison of the effectiveness of different algorithms in extracting FECG signals, (a) maternal abdominal ECG signal; (b) FECG extracted by FastICA; (c) FECG extracted by EKF; (d) FECG extracted by Attention R2W-Net.