Sleep / Wakefulness from
Actigraphy
Pedro Pires1,3, Teresa Paiva2 and João Sanches1,3
1Institute
for Systems and Robotics
2Faculdade de Medicina da Universidade de Lisboa
3Instituto Superior Técnico
IbPRIA, Póvoa do Varzim, 2009
1
Sleep Disorders Diagnosis
Polysomnography
• tests performed on patients during the sleep to evaluate sleep
disorders:
–
–
–
–
–
–
–
monitoring of the patient's airflow through the nose and mouth
blood pressure,
heartbeat as measured by an electrocardiograph (ECG)
blood oxygen level
brain wave patterns (EEG)
eye movements (EOG)
movements of respiratory muscles and limbs.
• it is accurate but it is complex and difficult in practice.
IbPRIA, Póvoa do Varzim, 2009
2
Motivation
A lot of typical sleep disorders involve movement and the actigraphy is a
complementary diagnostic tool that may be used to detect these disorders,
such as:
•
•
•
•
•
•
•
Narcolepsy – the condition of falling asleep spontaneously and unwillingly
Periodic limb movement disorder (PLMD) – sudden involuntary movement
of arm and/or legs during sleep
Circadian rhythm sleep disorder - jet lag and shift work sleep disorder
(SWSD)
Obstructive sleep apnea – the patient can’t get enough deep sleep
Sleepwalking
Sleep paralysis – temporary paralysis of the body shortly before or after
sleep
Delayed sleep syndrome (DSPS) – inability to awaken and fall asleep at
socially acceptable times
IbPRIA, Póvoa do Varzim, 2009
3
Actigraphy
• Three-axis accelerometer
IbPRIA, Póvoa do Varzim, 2009
4
Movement characterization
• The purposeness movements during the daytime are
intrinsically different from the purposeless nature of the
ones during the night.
• Different statistical distributions are associated to each
one; Maxwell during the day and Poisson [Gimeno et
al.,1999] during the night
p(r , (t ))   (t ) pd (r , M (t ))   (t ) pn (r , P (t ))
where    ; M ;  ; P 
T
IbPRIA, Póvoa do Varzim, 2009
5
Daytime distribution model
• Acceleration Magnitude
Maxwell Distribution
az
r
ay
ax
r  a x2  a y2  a z2
IbPRIA, Póvoa do Varzim, 2009
6
Mixture
ciclo.mpg
Poisson
(Normal)
IbPRIA, Póvoa do Varzim, 2009
Maxwell
7
Distribution Mixture
h ( n )    (c M ,  M )   N (c N ,  N )

ParameterL estimation
ˆ(n)  arg min  hn (k )  p( xk , ) 2

k 1
IbPRIA, Póvoa do Varzim, 2009
8
Sleep/Wakefulness (S/W) state
• SW estimation form : sw(n)   (n)   (n)
0 if sleep
SW (n)  
1 otherwise
IbPRIA, Póvoa do Varzim, 2009
9
Sleep/Wakefulness (S/W) state
estimation
• Binarization with Graph-Cuts
SW  arg min  sw(n)(1  2SW (n))    SW (n)  SW (n  1)
SW
n
n

 


Binarization
IbPRIA, Póvoa do Varzim, 2009
Regularization
10
Real data
IbPRIA, Póvoa do Varzim, 2009
11
Estimated parameters
Exercise
IbPRIA, Póvoa do Varzim, 2009
12
An insomnia case
IbPRIA, Póvoa do Varzim, 2009
13
SW
IbPRIA, Póvoa do Varzim, 2009
14
Conclusions
• Purposeness and purposeless movement
components of the human activity
• Different distributions to describe each
component: Maxwell for the vigil and
Poisson for the sleep state.
• Estimation of the parameters of the
mixture, and
• Estimation of the Sleep/Wakefulness state
using graph-Cuts
IbPRIA, Póvoa do Varzim, 2009
15
Download

Sleep/Wakefulness from Actigraphy