Fig. 1: Analytical pipeline for characterizing sleep quality and addressing subjective-objective sleep discrepancy (SOSD).
From: A continuous approach to explain insomnia and subjective-objective sleep discrepancy
A We analyzed PSG data from a population of Good Sleepers (GS), patients with insomnia without SOSD (SOSD−) and with SOSD (SOSD+ ). B Heterogeneous PSG recordings were harmonized via the estimation of hypnodensities, which enabled the quantification of the level of intrusion and instability of sleep dynamics. Finally, intrusions and instability were used as markers for predicting individual diagnoses and sleep quality measures. C PSG recordings (example of the first 4 h of sleep of a GS subject) were scored by experts in 30 s epochs (hypnogram) and then a set of features (catch22) were extracted to characterize each epoch in a multidimensional space. A machine learning algorithm was trained to predict the hypnogram. This also yielded, for each epoch, the probability of each stage, i.e., the hypnodensity (colors). Finally, the hypnodensities were analyzed with tools from information theory to measure the level of intrusions (entropy) and instability (DKL) of each epoch.
