Linear Phase Sharp Transition BPF to Detect Noninvasive Maternal and Fetal Heart Rate

Figures

• 

  Figure 1. Magnitude response H(ω) of the band-pass filter.

• 

  Figure 2. Illustration of the true reference FHR (direct scalp ECG) plotted with our algorithm computed FHRV for record r08 of adfecgdb (channel four) for one-minute trace. Blue dotted lines indicate ±10 bpm tolerance with respect to the reference FHR trace.

• 

  Figure 3. (a) Magnitude response of the proposed BPF with filter order N = 1001. (b) Linear plot. (c) Magnified view of the passband. (d) Magnitude response of the BPF LPST for various filter orders (N).

• 

  Figure 4. Average BPF transition width and stopband attenuation for various filter order (N).

• 

  Figure 5. Sensitivity, positive predictive value, and accuracy of record r08 (adfecgdb) with TP = 132 for various filter order (N).

• 

  Figure 6. (a) Direct fetal scalp ECG signal (channel one) as standard reference FECG. (b) Raw maternal aECG of adfecgdb database (channel 4—record r08).

• 

  Figure 7. (a) Frequency spectrum of the narrow sharp transition band-pass filtered signal. (b) FQRS time domain signal after band-pass filtering.

• 

  Figure 8. Noninvasive FHR detector for N = 5001 for adfecgdb database (channel 4—record r08). (a) Amplitude squaring of fetal R-peaks. (b) Moving window integration and adaptive threshold. (c) Fetal heart rate variability.

• 

  Figure 9. Illustration of the true reference FHR plotted with our algorithm computed FHRV for four sets of fetal frequency fiduciary edges of the BPF. The signal used is a one-minute trace of record r08, channel 4 of adfecgdb with filter order N = 5001 (ni is the time index corresponding to the ith computed fetal R-peak at the output of the FQRS detector). (a) Type 1: 27 Hz–53 Hz. (b) Type 2: 27 Hz–48 Hz. (c) Type 3: 35 Hz–53 Hz. (d) Type 4: 35 Hz–48 Hz. The dotted lines indicate the ±10 bpm tolerance assumed with respect to the reference FHR trace.

• 

  Figure 10. Sensitivity, positive predictive value, and accuracy for all the four channels of the 5 adfecgdb records.

• 

  Figure 11. Using LPST BPF for MHR detection with fiduciary edges, ωs1 = 10π and ωs2 = 40π for record r01 (channel 3) of adfecgdb. (a) Frequency spectrum of the narrow BPF signal. (b) Amplitude squaring of maternal R-peaks. (c) Moving window integration and adaptive threshold. (d) Histogram of the MHR. (e) The true reference MHR plotted with our algorithm computed MHR.

• 

  Figure 12. Percentage error of MHR detection compared with the MHR references for all the four channels of the 55 nifecgdb records.