2. Sleep
• Approximately 1/3 of the human lifespan is spent in sleeping.
• An 8 hour sleep comprises 4 or 5 sleep cycles.
• Each cycle lasts approximately 90 minutes and comprises
different stages including
• ght sleep (Stages 1 & 2), deep sleep (Slow Wave Sleep),
rapid eye movement (REM).
Wake
S1
REM REM REM REM REM
S2
SWS
1 2 3 4 5 6 7 8
NCKU 神經運算與腦機介面實驗室
3. Sleep Problems
• A considerable portion of population in
the world have sleep problems, including
insomnia (~30%) and sleep apnea (2-4%).
• Sleep diseases seriously affect a
patient’s quality of life such as causing
daytime sleepiness, irritability,
depression, unexpected accidence, etc.
• To deal with these problems, the first
step is to do effective and efficient
sleep diagnosis. NCKU 神經運算與腦機介面實驗室
4. Sleep Diagnosis
• All-night polysomnographic (PSG) recordings
– electroencephalograms (EEGs),
– electrooculograms (EOGs),
– electromyograms (EMGs),
are usually acquired from patients in hospitals or sleep
centers.
• Problems
– First night effect in in an unfamiliar environment.
– Disturbance from multiple recording wires of PSG
affects sleep quality.
– Visual sleep scoring is a time-consuming and subjective
process.
-
NCKU 神經運算與腦機介面實驗室
7. Sleep Monitoring for Homecare
• The Actiwatch and Portable PSG were
developed for rough and detailed sleep
monitoring.
Polysomnography(PSG)
Actiwatch
Sleep scoring for Actiwatch Sleep scoring for PSG
NCKU 神經運算與腦機介面實驗室
8. Actiwatch
• The actiwach measure the
movement activities of the
user during sleep and the
developed scoring method can
analyze the recordings and
report the sleep quality of
the user.
http://www.wantchinatimes.com/news-subclass-cnt.aspx?cid=1204&MainCatID=12&id=20110706000001
NCKU 神經運算與腦機介面實驗室
10. Consistency Comparison
(a) Result of sleep efficiency of Sadeh’s method
(b) Result of sleep efficiency of Jean-Louis’s method
(c) Result of sleep efficiency of Sazonova’s method
(d) Result of sleep efficiency of Tilmanne’s method
(e) Result of sleep efficiency of our method
NCKU 神經運算與腦機介面實驗室
12. User Interface
An example of bad sleep quality (sleep efficiency:46.97%)
原始訊號
動作能量
判讀結果
各 睡 使用者
睡 眠 年齡層
眠 /
參 清
醒 睡眠品質
數
時
間
比 睡眠問題
NCKU 神經運算與腦機介面實驗室
13. Portable PSG for Homecare
• A modularized and distributed PSG system
that is more convenient and has potential
for recording at home.
• It is composed of multiple tiny, low-cost
and wireless-synchronized signal acquisition
nodes, and each node acquires specific
physiological signals including, EEG, EOG
EMG, airflow, respiratory bands, and blood
oxygen saturation.
NCKU 神經運算與腦機介面實驗室
15. Novelty
• Each modualized node acquires specific
physiological signals within a small body region.
• Novel wireless-synchronization technology is
utilized to reduce sleep disturbance.
• The developed system has better comfortableness
performance in terms of several objective and
subjective sleep indices.
NCKU 神經運算與腦機介面實驗室
18. R&K Sleep Staging
Rechtschaffen and Kales (1968) Sleep Staging Criteria
Sleep Stage Scoring Criteria
>50% of the page (epoch) consists of alpha (8-13 Hz) activity or low voltage,
Waking
mix (2-7 Hz) frequency activity.
50% of the page (epoch) consists of related low voltage mixed (2-7 Hz)
Stage 1 activity. Slow rolling eye movements lasting several seconds often seen in
early stage 1.
Appearance of sleep spindles and/or K complexes and <20% of the epoch may
Stage 2 contain high voltage (>75 μV, <2 Hz) activity. Sleep spindles and K
complexes each must last >0.5 seconds.
20%-50% of the epoch consists of high voltage (>75 μV), low frequency <2
Stage 3
Hz activity.
Stage 4 >50% of the epoch consists of high voltage (>75 μV), <2 Hz delta activity.
Relatively low voltage mixed (2-7 Hz) frequency EEG with episodic rapid eye
Stage REM
movements and absent or reduced chin EMG activity.
NCKU 神經運算與腦機介面實驗室
26. 法則式自動睡眠判讀系統
Preprocessing Feature Extraction Classification
Movement epochs
Input: EEG (C3-A2), Downsampling detection
EOG, EMG (256Hz)
Spectral / temporal
feature extraction
Staging with a rule
(12 Features)
based decision
Band-pass filtering tree (14 rules)
(EEG/EOG 0.5-30Hz,
EMG 5-100Hz)
Contextual rule
Feature smoothing
Segmented into Normalization
30-s epochs Movement epochs Scoring
elimination Result
26
•“A Rule-based Automatic Sleep Staging Method,” Journal of Neuroscience Methods, vol. 205, no. 1, pp. 169-176,
2012.
NCKU 神經運算與腦機介面實驗室
27. 特徵分析 (Features)
No. Type Feature Source Label
1 PS Total power of 0-30 Hz EEG 0-30 E
2 PS Total power of 0-30 Hz EMG 0-30 M
3 PR 0-4 Hz/0-30 Hz EEG 0-4 E
4 PR 8-13 Hz/0-30 Hz EEG 8-13 E
5 PR 22-30 Hz/0-30 Hz EEG 22-30 E
6 PR 0-4 Hz/0-30 Hz EOG 0-4 O
7 SF Mean frequency of 0-30 Hz EEG Mean(fre.) E
8 SF Mean frequency of 0-30 Hz EMG Mean(fre.) M
9 DR Alpha ratio EEG Alpha E
10 DR Spindle ratio EEG Spindle E
11 DR SWS ratio EEG SWS E
12 EMG energy Mean amplitude EMG Amp M
* PS(=Power spectrum), PR(=Power ratio), SF(=Spectral frequency), DR(=Duration ratio)
NCKU 神經運算與腦機介面實驗室
28. 決策樹(Decision Tree)
E: EEG Features
O: EOG
M: EMG
1
Wake, S1, S2, REM SWS, S1, S2, REM
Alpha E
8-13 E
2 3
Wake, S1, S2 REM, S1, S2 SWS, S2 REM, S1, S2
Alpha E 0-4 E
0-30 M 22-30 E
4 5 6 7
0-30 E
S2, S1 Wake, S1 REM, S1 0-30 E S2, S1 REM, S1 0-30 E S2, S1
Spindle E
0-4 E Spindle E Spindle E
SWS E
SWS E SWS E
0-4 O
SWS S2
8 9 10 11 (9) (10) 12 13
0-4 E Mean(fre.) E 0-4 E 0-4 E
Amp M Amp M
Spindle E Mean(fre.) M Spindle E Spindle E
S1 S2 Wake S1 REM S1 S1 S2 REM S1 S1 S2
(1) (2) (3) (4) (5) (6) (7) (8) (11) (12) (13)
28
(14)
NCKU 神經運算與腦機介面實驗室
29. 效能評估
• 資料包含17位受試者所量測的14,391 30-s epochs PSG 訊號。
• 與專家判讀結果一致性超過82% 方可接受 (Norman et al., 2000;
Whitney et al., 1998).
Method Wake S1 S2 SWS REM Overall
Our method 88.43% 35.12% 87.01% 90.8% 90.51% 86.68%
Schaltenbrand et al. 91.73% 4.67% 90.61% 86.86% 79.96% 84.75%
Hae-Jeong et al. 90.79% 3.04% 87.38% 69.12% 53.76% 73.95%
NCKU 神經運算與腦機介面實驗室
30. Hypnogram (睡眠結構圖)
Mov
Wake
REM
S1
S2
SWS
23:40 01:00 03:00 05:00 07:00 hr
(a)
Mov
Wake
REM
S1
S2
SWS
23:40 01:00 03:00 05:00 07:00 hr
(b)
Mov
Wake
REM
S1
S2
SWS
23:40 01:00 03:00 05:00 07:00 hr
(c)
(a) the original manually scored hypnogram, (b) the automatic staging without
smoothing hypnogram, and (c) the automatic staging with smoothing hypnogram.
NCKU 神經運算與腦機介面實驗室
32. Novel Tech. and Applications
• EEG, EOG, EMG EEG EOG?
• IF EOG is ok for sleep scoring, what is a
good design for EOG measurement?
• In addition to measurement, can the
system provide active feedback to users?
• In addition to patients, can the system
benefit normal users?
NCKU 神經運算與腦機介面實驗室