Autonomic Sleep Patterns with Polysomnography

Akane Sano*, Rosalind Picard, (Massachusetts Institute of Technology, Media Lab, Group) Robert Stickgold (Harvard Medical School, Center for Sleep and Cognition, Beth Israel Deaconess Medical Center) *

Introduction / Motivation Results What is Electrodermal activity (EDA)? Example: EDA with EEG spectrogram of channel C3, C4, O1 and O2 When EDA peaks happened, Electrodermal activity (EDA) provides a fine EDA Raw which stages did they occur in? measure of sympathetic nervous system arousal, Motion 0% 50% 100%

one of the main branches of the autonomic EDA Peaks 1 SWS

nervous system. EDA is a measure widely used Sleep stage NREM2 78% of EDA peaks in psychophysiology. 3 NREM1 occurred in Non- C3 REM REM2 and SWS. C4 5 EDA is interesting during sleep WAKE O1 Studies have shown that EDA is more likely to Participants 7 OTHER have high frequency peak patterns called “storms” O2 during deep sleep (Asahina, 1962). EDA has also been shown to have characteristic differences Delta power of EEG when EDA peaks are present Higher delta power EDA peaks associated with wake and sleep, although its 20 in each quarter of the night EDA Peak No EDA Peak usually occurs when patterns are not uniquely associated with EEG- 0% 50% 100% 15 EDA peaks based sleep stages (Koumans et al., 1968). 1 73% of EDA peaks

10 2 1Q occurred in the first Our work advances a quantitative characterization Higher EDA amplitude 3 2Q 5 and the second of the relationship between EDA and PSG. and more EDA peaks 4 3Q Delta Power Delta 5 quarters of the night

occur in deeper sleep 4Q

0 6

C4 C4 C4 C4 C4 C4

O1 O2 O1 O2 O1 O2 O1 O2 O1 O2 O1 O2 stages Participants 7

Data -5

Sub2 C3 Sub2 C3 Sub4 C3 Sub5 C3 Sub6 C3 Sub7 Sub1 C3 Sub1 8 Correlation Coefficient between EDA and One night of wrist EDA and polysomnography (PSG) Cross-Correlation delta power from 8 healthy adults in a sleep lab between EDA and delta power 1 0.6 EDA amplitude and delta power 1Q 2-4Q Actigraphy: sleep and wake detection with standard 0.8 0.5 1Q 2-4Q usually have a higher correlation zero-crossing and Cole's function 0.6 0.4 coefficient and a higher cross- EDA: a low-pass FIR filter (cutoff frequency 0.4 Hz, 0.4 0.3 correlation in the 1Q of sleep 32nd order) and peak detection with the slope 0.2 R 0.2 than in later quarters of sleep.

exceeding a value of 0.09 micro Siemens/s. 0.1

0

C4 C4 C4 C4 C4 C4 C4

O2 O1 O2 O1 O2 O1 O2 O1 O2 O1 O2 O1 O2

EEG: delta power was computed from the EEG (C3, O1

-0.2 0

C4 C4 C4 C4 C4 C4 C4

O2 O1 O2 O1 O2 O1 O2 O1 O2 O1 O2 O1 O2

C4, O1 and O2) O1

Sub1 C3 Sub1 C3 Sub2 C3 Sub3 C3 Sub4 C3 Sub5 C3 Sub6 C3 Sub7

Correlation and cross-correlation between EDA -0.4

Sub1 C3 Sub1 C3 Sub2 C3 Sub3 C3 Sub4 C3 Sub5 C3 Sub6 C3 Sub7 amplitude and delta power of EEG were also -0.6 computed. Full Disclosure: Picard is a full professor at MIT and also co-founder, chief Conclusions scientist, and chairman of , who made the Affectiva Q Sensors used to EDA high frequency peaking, measured over 8 nights in healthy adults, occurred in Non-REM2 and SWS, usually in the first half of the night. The EDA collect the EDA and actigraphy data in this study. The author participates fully in MIT’s monitoring of conflict-of-interest procedures. amplitude was most highly correlated with EEG delta power early in the night for most participants.