. 2018 Jan 10;41(3):zsy006. doi: 10.1093/sleep/zsy006

Table 2.

Arousal detection features

Type	Feature	Number
Time domain	$\bar{R R}$ , ¹⁸ SDNN, ¹⁸ SDSD, ¹⁸ RMSSD ¹⁸	1–4
	Range(RR), ^18,19 MAD(RR) ^18,19	5–6
	$ℚ (R R, [0.10, 0.25, 0.50, 0.75, 0.90])$ ^18,19	7–11
	MSLD^short(RR), MSLD^long(RR), LR, ¹⁰ LR^{back, 10}	12–15
Frequency domain	HF, ¹⁸ LF, ¹⁸ VLF, ¹⁸ TP, ¹⁸ all from CTW	16–19
Hjorth parameters	Activity, ²⁰ mobility, ²⁰ complexity²⁰	20–22
Sleep stage	p(wake), p(NREM), p(REM)	23–25

Features used for arousal detection. Features 1–11 were computed for 30 s windows. Features 12–13 were computed for local windows of length 5 s and 15 s and global windows of length 30 s and 180 s, respectively. Features 14–15 were computed on a beat-to-beat basis. Features 16–25 were calculated in 1 s bins.

RR = RR interval; $\bar{(\cdot)}$ = mean; SDNN = standard deviation of RR; SDSD = standard deviation of RR differences; RMSSD = root mean square of successive RR differences; MAD = mean absolute difference; MSLD = median signed local difference; LR = likelihood ratios; HF = high frequency; LF = low frequency; VLF = very low frequency; TP = total power; CTW = continuous wavelet transformation; NREM = non-rapid eye movement; REM = rapid eye movement.