Fig 1. Multi-scale permutation entropy as useful feature for neonatal sleep stage classification.

(a) Six possible ordinal patterns are identified (pi), their distribution is formed and Shannon entropy is computed. (b) Simulated time series prior (left) and after the coarse-graining procedure (right) are shown. Note that the granulation removes the fast-varying changes, which allows estimating the entropy on a slower temporal scale. (c) PSDs of individual epochs for a single recording (week-5) and corresponding visual/manual sleep staging, log-transformed power is shown for better visibility. (d) Corresponding MSPE values for the same recording depicted separately for the original signal (left) and after coarse-graining (right). Note that these three classes or stages are distinguishable on both scales, yet different patterns are observed in the original and coarse-grained computation.