Skip to main content
. 2019 Aug 4;19(15):3420. doi: 10.3390/s19153420

Table 1.

Features calculated from PPG on a per-cycle basis.

Domain    Features
Temporal
  • Tc. Cycle duration time

  • Ts. Time from cycle start to systolic peak

  • Td. Time from systolic peak to cycle end

  • Tsteepest. Time from cycle start to first peak in PPG’ (steepest point)

  • TdiaNotch. Time from cycle start to second peak in PPG’ (dicrotic notch)

  • TSysToDiaNotch. Time from systolic peak to dicrotic notch

  • TdiaToEnd. Time from dicrotic notch to cycle end

  • Ratio. Ratio between systolic and diastolic amplitude

Frequency
  • Three largest magnitudes. Three peaks with the largest magnitude from the PSD were considered. These tell us the dominant frequencies in the cycle. Both the magnitude values and the frequencies (in Hz) were taken as features.

  • Energy. Calculated as the sum of the squared fast Fourier transform (FFT) component magnitudes. The energy was then normalized by dividing it with the cycle length.
    energy=1Nn=0N1|x(n)|2, (1)
    where x(n) is the n-th FFT component and N is the parameter specifying the number of FFT components to compute.
  • Entropy. Calculated as the information entropy of the normalized FFT component magnitudes.
    entropy=n=0N1x(n)log(x(n)) (2)
  • Binned distribution. A normalized histogram, which is essentially the distribution of the FFT magnitudes into 10 equal sized bins ranging from 0 Hz to 62.5 Hz.

  • Skewness and kurtosis. These describe the shape of the cycle. More precisely, skewness tells us about the symmetry while kurtosis tells us about the flatness.