Fig 2: Real time behavioral classification

(A) 1 hour of behavioral data illustrating variables used for classification of sleep/wake state, described in descending order. Top shows expanded raw EEG traces from the indicated periods; scale bar next to NREM example, 1 V, 0.5 s. Below expanded EEG is the sleep/wake classification expressed as a hypnogram, where colored bars indicate periods of wake (green), NREM (blue), or REM (magenta). Beneath the hypnogram is the full raw EEG trace; scale is 2 V. Spectral Frequency (Spect. Freq.) plots the full spectrogram of the EEG from 0-20 Hz, colored by power (normalized to maximum). Extracted features of the Spect. Freq. are plotted below: delta(0.5-4 Hz)/beta(20-35 Hz) power ratio (shown on log scale) is high during periods of NREM, while the theta(5-8 Hz)/delta(0.5-4 Hz) power ratio is high during periods of REM. Absolute EMG values were normalized and expressed as standard deviation. Finally, animal movement in pixels is plotted. The bottom 4 plots of vigilance state features each have an adjustable threshold for state classification, shown as dashed line; points above the threshold are shown as colored dots. (B) Schematic of real time classifier rig. (C) 3D plot of vigilance state features: delta ratio, theta ratio, and movement measures (zero movement assigned lowest observed value for log axis) colored by brain state. Data from example animal. (D) Accuracy of real time classifier compared to manual scoring by state: Overall (% time matching between all 3 states) = 93.3%; Wake = 94.2%; NREM = 93.2%; REM = 90.5%; Wake Dense (matching wake dense on and off times) = 96.0%; Sleep Dense = 96.0%.