Figure 7.

Eye metrics alone are an accurate predictor of arousal state. a, Arousal state predictions from three arousal state classification models (Eye, Physiologic, Combined) and manual scores for sample data. Data are the same as presented in Figure 1. b, Example trace of pupil diameter over time with respect to arousal state. For clarity, blinks are not shown. c, Position of the pupil changes when the animal goes into REM sleep, moving nasally and ventrally. d, Absolute change in the frame-by-frame position of the pupil indicates a high degree of movement during 'REM' sleep. e, Confusion matrix of a bagged random forest (70:30% training/testing) using only eye metrics such as pupil diameter, position, and motion (Eye Model). f, Confusion matrix of a bagged random forest (70:30% training/testing) using measurements of cortical and hippocampal LFP, electromyography, heart rate, and whisker motion (Physiologic Model). g, Confusion matrix of a model using both the eye and physiological data (Combined Model) for arousal scoring. h, Out-of-bag error estimation during model training for each model. Error bars (h) indicate SD. Bonferroni-corrected (3) paired t test, *α < 0.017, **α < 0.003, ***α < 0.0003, n.s.