predictive adaptation to dynamic environments and ... · impedance adaptation the impedance and...
TRANSCRIPT
Koji ItoRitsumeikan University, Kyoto, Japan
Predictive Adaptation to Dynamic Environments and Application to Motor Rehabilitation
University of Naples Federico II: March 11, 2013
2
1. Introduction- Redundancy among Brain-Body-Environment
2. Experiments of arm reaching movements -Adaptation to dynamic environments-
3. EEG – FES system for stroke patients 4. Conclusion
Contents
3
Basic controlmechanism
Neural mechanisms for motor control
Cerebellum
Motor learning & adaptation
Fine tuning
Motor program
Selection
Basal ganglia
Somatosensory / visual feedback
Somatosensory / visual feedback
Brainstem
Spinal cord
MusculoskeletalSystem
Cerebral Cortex
4
External Dynamics
Internal Dynamics
Dynamic interaction among Brain-Body-Environment
BodyMotorOutput
SensoryInput
Environment
Brain
Internal dynamics has much redundancy.
It is required to adjust the internal dynamics before beginning voluntary movements.
How to reduce redundancy ofsensorimotor mapping?
5
Controller
Identification of external dynamics
Feedback Controller
Gc
yd (t)Goal
Sensory feedback
+
FeedforwardController
GF
+ y (t)
MotoroutputMotor
commandu(t)
BodyGp
Environment
Identification
External Dynamics
6Motor control
Internal Model
Motor Command
Sensory Feedback
Brain
Body & Environments
1) Simulation of body-environment dynamics2) Prediction of sensory feedback
7
Internal Dynamics
Controller
)(ˆ ty
FeedforwardController
+
Prediction
+- y (t)
u(t)Body
Environment
Arm Impedance
External Dynamics
Internal ModelInternal ModelEfference copyu(t)
Sensory feedbacksTime Delay
)( tt Δy
Conceptual control model of human movements
Targetyd(t) Feedback
Controller+
Time Delay)( tt Δy
Sensory feedbacks
Feedforward & Feedback
8
Experiments of Arm Reaching Movements -Adaptation to dynamic environments-
9Adaptation to dynamic environments
Arm Reaching MotionUnknown environments
A B C
Computational Model of Adaptation to EnvironmentsModeling of motor adaptation and learning
Neuro-physiological Approach- Brain imaging during reaching motions- Development of new equipments
10Velocity-dependent force field (VF)
yx
yx
FF
y
x
13181813
32B
(by Shadmehr, 1994)
11Hand trajectories and hand force
Initial phase After learning After-effect
Hand trajectories Hand force
After learning
12Hand trajectories and EMG during reaching motion
1 st time
72 nd time
Initial position
(c)Triceps Time(sec)
EMG (average of 24 subjects)
EMG
(b)Biceps Time(sec)
ModifiedFirst 8 timesLast 8 times
Free motion
(a) Hand trajectories
Target (45゚)
Force field
(by Shadmehr, 1999)
13Unstable (divergent) force fields (DF)
yx
kk
00
F
0dx
FF
y
x β>0
x [m]
y[m
]
0.1
0.2
0- 0.1 0.1
(E. Burdet et al, Nature, 2001)
14Hand trajectories and forces in DF
Hand trajectories Hand force
15Muscle activities (EMG) during reaching motions
* : P<0.05, ** : P<0.01, *** : P<0.001
NF: Free motionDF: Unstable force fieldAE: After-effects
Posteriordeltoid
Triceps long head
Tricepslateral head
Pectoralis major
Bicepsbrachii
Brachio radialis
(Franklin, Exp Brain Res., 2003)
The subject plans the muscle activities before beginning the reaching motion.
EMG under After-effects is similar to the unstable force field.
msec msec msec
16
Question?Can we program the internal modelcontrol and impedance control in afeedforward manner?
1) We plans Internal Model Control before beginningthe reaching movements under the dynamicenvironment.
2) We presets Impedance Control before beginningthe reaching movements.
17Switched force fields
(a) SF1 (VF→DF)
x(0, 0)
(0, 0.2)
y
VF
DF
(b) SF2 (DF→VF)
DF
VF
(0, 0)
(0, 0.2)
y
x
VF→DF (SF1)
Hand forceBefore learning
After learning
DF→VF (SF2)
Hand forceBefore learning
After learning
Impedance Adaptation
The impedance and internal-modelcontrols are programmed in afeedforward manner in adaptation to thecontexts of dynamic environments.
Program Adaptation0.2 m
Internal Model Adaptation
Internal Model Adaptation
Impedance Adaptation
18
19
Experimental results suggests that the predictive adaptationto the environment dynamics is composed of three levels:
Three adaptation levels of motor control
[1] Impedance adaptation→Parameter adaptation
[2] Internal model adaptation→State dynamics adaptation
[3] Program adaptation→Context adaptation
20
Motor Controller
Lateralpart
Intermediatepart & vermis
CerebellumInternal model adaptation
Programadaptation
Limb/Body
Muscleviscoelasticity
Spinalreflex system
+ +
Environment
External dynamics
Cerebral cortex
Prefrontalcortex
Supplementarymotor area
Premotorcortex
Parietalcortex
Motorcortex
Basalganglia
Brain stem-Spinal system
Impedance adaptation
Motor adaptation mechanisms
Electroencephalogram (EEG) –Functional Electrical Stimulation (FES) System
for Stroke Patients
22Motor/Neuro rehabilitation
FESPhysical therapy Practice for walking
It is not strictly to recover the motor performance as it was, but reconstruct optimal motor performance in the new situation
→ Re-optimization
It is essential to give the opportunity for motivated and intensivepractice and exercise in a stimulating environment.
23Motor intention and sensory feedback for motor rehabilitation
Visual & Auditory feedbacks
Prefrontalmotor cortex
Prefrontalmotor cortex
Parietalcortex
Parietalcortex
Brain stem-Spinal system
Brain stem-Spinal system
Primarymotorcortex
Primarymotorcortex
Primarysomatosensory
cortex
Primarysomatosensory
cortex
Basal ganglia&
Cerebellum
Basal ganglia&
CerebellumMotor Intention Sensory feedbacks
Proprioceptive feedbacksMotor command
24Reconstruction of motor function
EEG-BCI EEG-FES EMG-FES Leg powered wheelchair
Target Spinal injury (paraplegia)
Spinal injury / stroke (paraplegia)
Stroke(hemiplegia)
Stroke(hemiplegia)
Object Transmission of intention
Transmission of motor intention
Reconstruction of motor function
Reconstruction of motor function
How to
Detect the motor intention from EEG at the motor imagery of arm or leg
Regain movements based on the motor intention from EEG at the motor imagery of arm or leg
Activate the muscles by FES based on the motor intention from EMG
Realize the somato-sensory feedback bypedaling the wheelchair
Advantage
Voluntary EMG is not required
Learning effects of voluntary somato-sensory feedbacks
Learning effects of somatosensory feedbacks & enhancement of motivation
Weak points
Not motor learning Difficult to detect motor intention from EEG
Voluntary EMG activity is required.
Effect on gait disorder is not clear.
Proposed EEG-FES system for motor rehabilitation
Cerebral cortex
Motor cortex
Somatosensoryarea
Premotor Cortex &
Supplementarymotor area
Parietalcortex
Basal ganglia
Brain stem-Spinal system
Environment
Sensory feedback
Motor command
EEG FES
-
+
Cerebellum
Proprioceptivefeedback
Musculo-Skeletal system
Muscle ActuationMotor Intention
25
26Event Related Desynchronization(ERD) The characteristics of ERD are as follows.
Decreasing of electric potential in specific frequency band (alpha (8-15Hz), beta (15-35Hz) band)
Observed during motion and motor imagery Activation of cortico-thalamus loop?
ERD detection. Band pass filtered (25-30Hz) Full wave rectified and
50 times sum averaged
(Lopes da Silva, 1991)
0 2 4 6 8 100
1
2
3
[μV
]
[sec]
Motor Imagery
ERD
ERS10-20 methods
Foot motor area (Cz)
27
Basic Experiments
28Experimental setup
Multi telemeter system
Computer screenEEG
FES
Cue presented (3 seconds)
Blank screen(3 seconds)……
EEG measurement during FES activation on both quadriceps. 17 healthy subjects participated in the experiment. EEG from 7 electrodes (around Cz area) were measured. One task consisted of 50 trials of FES.
Foot motor area (Cz)
Motorimagery
29Frequency analysis
500 msec
Elec
tric
pote
ntia
l[μ
V]
Time [sec]3
Pow
er
spec
trum
24 26 2822 frequency [Hz]20
2 Hz
Ensemble average
The data for each 3 seconds is divided into 500 msec data. After converting to frequency space, the ensemble average is
calculated. To quantify the amount of ERD, r2 value was calculated for the
frequency band for 24 – 26 Hz.
30Power spectrum before and after motor imagery training
No motor imagery
Motor imagery
0 10 20 30 40 50Frequency [Hz]
10
10
10
Pow
er sp
ectru
m
1
2
3
(a) Before training (b) After 3 days training
0 10 20 30 40 50Frequency [Hz]
10
10
10
Pow
er sp
ectru
m
1
2
3
No motor imagery
Motor imagery
31r2 value
frequency [Hz]24 26
Absence of FES
Presence of FES
x group
y group
x
y
yx ,
Pow
er sp
ectru
m r2 value is used to calculate the amount of ERD (This method is
popular in BCI researches). r2 value uses the within- and between- variance with two groups. Larger r2 value mean the clearer detection of ERD.
yx
yx
nnyxb
byx
bny
nx
yxr
22
22
22
2
)()(
),(
32Training effects of r2 value
Before training 1st day 2nd day 3rd day
0.3
0.2
0.1
0
-0.1
r2va
lue
* * p<0.05
33
Training results of a stroke patient
34Experiment set-up
FES
Telemeter(sender)
OptotrackPC
Telemeter(reciver)
main PC
Optotrack camera
Subject:Brain stem infarction. 30 months after stroke. Paralyzed on the left side of the body (Foot-pat test:0/6)
35Experiment
5 min. 5 min.30 min.
36Training by EEG-FES system
Intend to move left
foot
FES
outp
ut [m
A]
10
0
FES
500 msec
ERD
Sensory feedback
37Before training
38After training
39Leg movements
Paralyzed side Normal side
RightLeft
Ankle joint
Left Right
Knee joint
40
Statistical significance of ankle joint movements
(a) Paralyzed side (b) Normal side
41
70
60
50
40
Ang
le [d
eg]
0 2 4 6 8 10Time [sec]
70
60
50
40
Ang
le [d
eg]
0 2 4 6 8 10Time [sec]
Bef
ore
train
ing
Ankle joint movements
Paralyzed side Normal side
70
60
50
40
Ang
le [d
eg]
0 2 4 6 8 10Time [sec]
70
60
50
40
Ang
le [d
eg]
0 2 4 6 8 10Time [sec]
Afte
r tra
inin
g
42
0 2 4 6 8 10Time [sec]
80
60
40
20
Elec
tric
pote
ntia
l [μV
]
0
100
0 2 4 6 8 10Time [sec]
80
60
40
20
Elec
tric
pote
ntia
l [μV
]
0
100B
efor
e tra
inin
g
0 2 4 6 8 10Time [sec]
80
60
40
20
Elec
tric
pote
ntia
l [μV
]
0
100
0 2 4 6 8 10Time [sec]
80
60
40
20El
ectri
c po
tent
ial [μV
]
0
100
Afte
r tra
inin
g
Muscle activity levels
Paralyzed side Normal side
43Statistical significance of muscle activities
(a) Paralyzed side (b) Normal side
44Brain image
Brain stem
45Conclusion
EEG FES
EEG-FESSystem
Sensory feedbacks
Motor Intention
New non-invasive brain activity measurement
Sensory FeedbacksRobotics
Haptic devicesVariable impedance
46Motor intention and sensory feedback for motor rehabilitation
Visual & Auditory feedbacks
Prefrontalmotor cortex
Prefrontalmotor cortex
Parietalcortex
Parietalcortex
Brain stem-Spinal system
Brain stem-Spinal system
Primarymotorcortex
Primarysomatosensory
cortex
Basal ganglia&
Cerebellum
Basal ganglia&
CerebellumMotor Intention Sensory feedbacks
Proprioceptive feedbacksMotor command
47
Mitsuru Takahashi (Terumo corporation)
Kotaro Takeda (ATR Computational Neuroscience Laboratories)
Rieko Osu (ATR Computational Neuroscience Laboratories)
Kotaro Otaka (Keio University, Medical School Hospital)
Takashi Hanakawa (National Center of Neurology and Psychiatry)
Toshiyuki Kondo (Tokyo University of Agriculture and Technology)
Joint researches with
M. Takahashi, K. Takeda, Y. Otaka, R. Osu, T. Hanakawa, M. Gouko and K. Ito: Event related desynchronization-modulated functional electrical stimulation system for stroke rehabilitation: A feasibility study,Jr. of NeuroEngineering and Rehabilitation, 9:56 (16 Aug 2012)
Journal paper
48
Grazie mille per attenzione!Thanks a lot!
どうもありがとう(Domo Arigato)!