Online monitoring van
levende organismen
Prof. Dr. Ir. Daniel Berckmans
M3-BIORES
K.U.Leuven
Lessen voor de 21e eeuw - March 8 2010
Van intensive care tot wielrennen en
formule 1
Head: Daniel Berckmans Prof J. Aerts Ir K. Van Loon Ir G. De Bruyne Ir T. Leroy Ir J. Lefever Ir. S. A. Haredasht Ir. F. Borgonovo Dr V. Exadaktylos Ir A. Bulckaert Ir M. Silva Ir A. Aydin Dr. S. Ferrari Secretary W. Meulemans Technical Assistance Ing J. Lemaire L. Happaerts Private Companies BIORICS N.V. Ir F. Jansen Ir J. Bellinckx Ing L. Wollants Ing M. Milutinovic A. Goeman Dr C. Bahr Prof E. Vranken Ir O. Cangar Ir A. Pluk Ir A. Poursaberi Ir E. Bites Romanini Ir. Y Ir S. Eren Ozcan Ir N. Alban Ir S. De Boodt Ir A. Youssef Ir. X
Overview
• What is a bioresponse?
• Methodology
• A living organism is a CITD – system
• Modelling bioresponses
• Examples / Results
- Monitoring Bioresponses
- Controlling Bioresponses
• Applications in sports
Bioresponses
Micro-environment
Laboratory for Agricultural Buildings Research, K.U.Leuven, Belgium
Block diagram of a system
System
Process
Disturbance variables (Environment)
Input variables
(t)
Output variables
(t)
(Environment)
(t)
bio
system
bioresponses
Active monitoring/control of a
biological response??
DESIRED DIRECTION
PREDICTION
2
Direction Position and balance
on the board PREDICTION-BASED CONTROLLER
Methodology
1
FEEDBACK MEASURE MEASURE3
PREDICTION
2
MEASURE
DESIRED DIRECTION
Modern Process Control e.g. Automatic Pilot
Direction Process MANAGE
1
FEEDBACK MEASURE Steering gear3
DYNAMIC BIORESPONSE MICRO-ENVIRONMENT
Process
1
FEEDBACK PREDICTION MODEL2
DESIRED PROCESS OUTPUT
MONITOR/ REGELAAR
3
MEASURE MEASURE
1991: Modern control theory applicable to
living organisms?
A living organism is a
CITD system
A living organism:
ComplexA living organism:
ComplexComplex Individual Hea rt bea t (bpm) Time (s)
A living organism:
Complex Individual Time-Varying
A living organism:
TIME (HOURS) 0 1 2 3 4 5 HEA T PR OD UCTION (W /KG) 11 12 13 14 15 16 17 MEASURED MODELLED (1ST ORDER) MODELLED (2ND ORDER)5 days old
TIME (HOURS) 0 1 2 3 4 5 HE A T P RODU CT ION ( W /K G) 7 8 9 10 11 12 13 MEASURED MODELLED (1ST ORDER)30 days old
Example: Heat production of broiler chickens
1 0 1 1 2 0 1 2
( )
b
b z
y k
a
a z
a z
0 1 0 1( )
b
y k
a
a z
Complex Individual Time-Varying Dynamic
A living organism:
1. Measure
2. Model
3. Manage
In an on-line way
Living organism =
CITD
- system
Complex Individual Time-Varying Dynamic
Output variables
Swimming
activity
Light intensity
PH
Water temperature
Oxygen concentration
Input variables
Disturbances
Age, Cu-concentration
Example: Daphnia-monitor
*
Daphnia TIME (SECONDS) 0 20 40 60 80 100 VE RT ICAL PO SIT IO N (M M ) 0 10 20 30 40 50 60 70 MEASURED MODELLED (1ST ORDER) MODELLED (2ND ORDER) 1 2 4 5 6 7 8 9 10 3 10
Example: monitoring behaviour of laying hens*
Battery cages Furnished cages
Quad unit
Digital recorder
Experiments:
– 18 animals, 1 hour of recording each
– 4 Camera‟s on top of the compartments + digital video recorder
– Audio-visual scoring of behaviour by ethologist as reference
• Features:
– Position
– Orientation
– Shape
• Experiments:
– 4 Camera’s
– 18 hours of real-time video
– Audio-visual scoring of
behaviour by ethologist as
reference method
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 time Video input Image processing Posture parameters p1 pm p2 …
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 Video input Image processing Posture parameters time p1 pm p2 …
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 Video input Image processing Posture parameters time p1 pm p2 …
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 Video input Image processing Posture parameters time p1 pm p2 …
Step 2: dynamic modelling
Fitting a mathematical model to the posture
parameters in each time window
For example: modelling scratching behaviour
time p1 „scratching‟
posture parameters
p1,…,pm
p
1[k] = d
1* u[k] - d
1* p
1[k-1] -
… - d
1* p
1[k-n+1]
dynamic parameters:
d1,…, dn
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 Posture parameters Dynamic modelling Dynamic parameters 1 2 3 4 5 6 7 8 -1.3 -1.2 -1.1-1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 1 2 3 4 5 6 7 8 -0.4 -0.2 0 0.2 0.4 0.6 0.8 time p1 pm p2 … dn d1 …
1 2 3 4 5 6 7 8 1.8 1.9 2 2.1 2.2 2.3 1 2 3 4 5 6 7 8 30 32 34 36 38 1 2 3 4 5 6 7 8 14.5 15 15.5 16 16.5 17 1 2 3 4 5 6 7 8 Posture parameters Dynamic modelling Dynamic parameters 1 2 3 4 5 6 7 8 -1.3 -1.2 -1.1-1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 1 2 3 4 5 6 7 8 -0.4 -0.2 0 0.2 0.4 0.6 0.8 time p1 pm p2 … dn d1 … „scratching‟
Classification
Compare dynamic parameters to pre-learned
bounding box
– Result: 1 if behaviour occurs, 0 else
dl dk
10 hours of real-time video data
Today, standing, walking, scratching behaviour, producing
an egg can be classified in an on-line way
Example: model-based computer vision
of laboratory mice(*)
• 135 Swiss mice, 10 min. recording each
• Open field test setup:
(*) Collaboration with Laboratory of Biological Psychology (Prof. R. D’Hooge), Leuven 26 cm
53 cm 34.5 cm
Walk model fit to real video images
total model output
param. 1
param.
values
Model fit to walk patterns
• Reference mouse:
• Drugged mouse (injected with pentobarbital):
Calving monitor
Example: Calving monitor for cows*
X and Y coordinates of the
centre point
Orientation (degrees) Body width/length ratio
Hip length (m)
Back area (m2)
Walking trajectory
Distance walked (m) (*) In collaboration with TEAGASC (Ireland)
Example: lameness detection
*
Sleepiness monitoring
*
MODEL 2 1 MODEL-BASED PREDICTIVE MONITOR - Heart rate - Biorhythm - Heat balance - Driving performance Driver sleepiness Signs of sleepiness MEASURE
Driver sleepiness detection & prediction
based on continuous measurement of bio-responses from the driver’s bodyOn-line Pig Sound Analysis
*
Pig Cough Sounds*
Example Healthy cough sound Example Sick cough sound
RESPIRATORY PATHOLOGIES IN PIG FARMS Mortality, Production Use of antibiotics Am plitu de (dB ) F req ue nc y (Hz ) Time (s) A C 77777777 B Time (s) 0 0,2 0,2 0,4 0,4 1000 1500 0 500 Am plitu de (dB ) F req ue nc y (Hz ) Time (s) A C 77777777 B A C 7777777777777777 B Time (s) 0 0,2 0,2 0,4 0,4 1000 1500 0 500
Pig Cough Localization
Cough hazards Using the difference in time arrival between
several microphones in a stable, the location of the cough sounds was determined
Silva et al. Computers and Electronics in
Climate controller V(t) T Q(t) Antibiotics sound Therapeutic decision infection Sound analysis micro
Main future application: Reducing the
use of Antibiotics
• Chronic Obstructive
Pulmonary Disease (COPD)
• More than 680 000 patients
(40+) in Belgium
Cough is more than just a sound
Monitoring the health status of individual critically ill patients on the basis of on-line modelling approaches
Example : Intensive Care
Project together with UZ Leuven (Prof. G. Van den Berghe) and Computer Sciences (KULeuven, Prof. M. Bruynooghe)
Treatment
Health
Patient 728 (WBC) Time (days) 0 5 10 15 20 25 30 35 40 a-parameter -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Survivor (intensive insuline treatment)
Trained on 141 patients, validated on 58 patients
Stationarity criterion: a > -1
Patient 1276 (WBC) Time (days) 0 5 10 15 20 25 30 35 a-parameter -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5> 16h 8 -16h 0 – 8h
10/12 patients correct
Moment of extubation
DYNAMIC BIORESPONSE MICRO-ENVIRONMENT
Process
1
FEEDBACK PREDICTION MODEL2
DESIRED PROCESS OUTPUT
MONITOR/ REGELAAR
3
MEASURE MEASURE
1991: Modern control theory applicable to
living organisms?
Example : Control of the crawl trajectory of
larvae of Calliphora vicina
t t y(t)
u(t)
Response of crawl direction to variations in ligth
Light intensity
crawl trajectory
-45.00 0.00 45.00 90.00 135.00 180.00 225.00 270.00 315.00 360.00 0 3 6 9 12 15 18 21 24 27 30 33 36 Tim e C ra w l d ir e c ti o n ( °) 270.00 180.00 12.55 315.00 270.00 9.72 45.00 90.00 13.59 90.00 180.00 15.77 225.00 90.00 12.77 135.00 270.00 9.60 0.00 0.00 15.60 180.00 0.00 15.61
Responses of crawl direction to steps in light
direction
t t y(t)
u(t)
Response of crawl direction to variations in ligth
Light intensity
crawl trajectory
MEASURE MEASURE a1 a2 … an b0 b1 … bm A(z-1) = 1+a 1z-1+ a2 z-2+ … + anz-n B(z-1) = b 0 + b1z-1+ … + bmz-m
)
(
)
(
)
(
)
(
)
(
1 1k
d
k
u
z
A
z
B
k
MODEL
Light intensity
crawl trajectory
FEEDBACK
MODEL-BASED CONTROLLER
Active control of crawl direction of larvae
Active control of 2D crawl direction of 6 individual
larvae
X
Y Z
Running speed
Heart rate
Example : optimisation of physical
training of horses
v
Heart rate (bpm)
Sensors on the horse
Speed (km/h)GPS
(Garmin Forerunner 205)
Heart rate monitor (Polar S610i)
Experimental design
Methods and Results
Overview of the experiments
Pre-experimental
work
25
Accuracy of the
equipment
4
Step experiments
45
5 horses x 3 riders
x 3 repetitions
Controlling
experiments
6
2 horses x 1 riders
x 3 repetitions
TOTAL
80
Velocity and Heart rate - Kyrielle (Bert) 01-01 0 5 10 15 20 25 30 0:00:00 0:05:00 0:10:00 0:15:00 0:20:00 0:25:00 time (u:mm:ss) v e lo c it y ( k m /u ) 0 20 40 60 80 100 120 140 h e a rt r a te ( b p m ) Velocity Heart rate
Input: Speed Output: Heartrate
2
Model Model-based controller Objective ‘On-line’ Measurement1
Actuator3
Procedure of Model-Predictive heart rate
control for horses (MPC)
HR en speed PC with MPC controller
Auditory feedback from MPC controller
40 50 60 70 80 90 100 110 120 0:00:00 0:10:00 0:20:00 0:30:00 0:40:00 0:50:00 1:00:00 time(h:mm:ss) h e a rt r a te ( b p m )
Controlled heart rate Target heart rate
Time-constant of HR over time 0 10 20 30 40 50 60 70
Week 0 Week 1 Week 2 Week 3 Week 4
Time (date) T C ( s e c ) TC
Optimisation of physical training of race
horses: Heart rate recovery time
Training exercises
Performance
Body
What happens today in sports
Process
Reading &
Measuring
Trainers
experience
prediction
feedback
Physical training Physical performance
Process: physical monitoring
CITD
Model
Process
Heart Rate
Physical activity
Exercise type 1
Exercise type 2
……
Physical monitoring
Input
Output
The player
Polar heart rate belt Hosand HR module Inmotio position antenna Telemetry connection Abatec serverBase stations Inmotio/Abatec transponder
Wire connection 3D Accelerometer
Inmotio/Abatec Transponder Hosand HR module Antenna 1 Antenna 2
Method
• Player arrives at training
• Player is measured in Milan Lab
– Measurements of Physical condition
• Real-Time monitoring on training
field with BioRICS system
• Player in the Mind Room
Milanlab:
Dry test Physical/Mental conditionAllenamento:
Variable 1 Variable 2 Variable n . . . Activity(Inmotio) Heart Rate
Mathematical Model Physical/Mental condition Variable 1 Variable 2 Variable n . . . ??? ??? ???
Objective:
Milan Lab references during training
- Test installations in Milan Lab
Total of 255 experiments over 3 years
In 19 training sessions recorded for 9 Primavera players
(4 players/training = 76 recordings )
Algorithm to monitor
“Physical condition”
DYNA reference 3 4 5 6 7 8 9 20070731 20070906 20070911 20070913 20070914 20070920 20070921 20071122 20071123 20071126 20071129 20071130 20071203 20071204 DYNA reference
3 4 5 6 7 8 9 20070731 20070906 20070911 20070913 20070914 20070920 20070921 20071122 20071123 20071126 20071129 20071130 20071203 20071204
DYNA reference Algo Physical
“Physical algorithm” for Milan Lab
physical reference:
Results: numeric
• 50 training sessions with Milan Lab physical reference
• 5 Primavera players
• Algorithm correctness: exact Milan Lab physical reference
-> 40/50 training sessions or
80% correct
• Algorithm correctness: error > 1 point on Milan Lab physical
reference score:
-> 48/50 training sessions or
96% correct
on [-1 1]
interval of Milan Lab physical reference
Physical training Physical performance P erf o rman ce
Total performance = Mental performance + Physical performance
Mental performance Mental training
Process: mental monitoring
CITD
Heart Rate
Physical
Activity
Reference
mental status
??? TimeOn-line mental
monitor
Mental algorithmMental monitoring
MINDROOM
Method
• Player arrives at training
• Player is measured in Milan Lab
– Measurements of Physical condition
• Real-Time monitoring on training
field with BioRICS system
• Player in the Mind Room
Milanlab:
Dry test Physical/Mental conditionAllenamento:
Variable 1 Variable 2 Variable n . . . Activity(Inmotio) Heart Rate
Mathematical Model Physical/Mental condition Variable 1 Variable 2 Variable n . . . ??? ??? ???
Objective:
Milan Lab references during training
- Test installations in Milan Lab
Mindroom data (1133 and 1156) -1.5 -1 -0.5 0 0.5 1 1.5 2 7 /0 3 /2 0 0 6 3 1 /0 3 /2 0 0 6 3 /4 /2 0 0 6 7 /4 /2 0 0 6 1 3 /4 /2 0 0 6 1 9 /4 /2 0 0 6 2 4 /4 /2 0 0 6 2 7 /4 /2 0 0 6 1 /5 /2 0 0 6 2 /5 /2 0 0 6 3 /5 /2 0 0 6 4 /5 /2 0 0 6 1 0 /5 /2 0 0 6 2 7 /0 3 /2 0 0 6 3 1 /0 3 /2 0 0 6 7 /4 /2 0 0 6 1 1 /4 /2 0 0 6 Date (ddmmyy) Me n ta l s ta tu s [p o s , n e g o r n e u tr a l] MINDROOM 1156 1133
Milan Lab Mindroom vs. Mental algorithm
Mental algo vs Mindroom
-1.5 -1 -0.5 0 0.5 1 1.5 2 7 /0 3 /2 0 0 6 3 1 /0 3 /2 0 0 6 3 /4 /2 0 0 6 7 /4 /2 0 0 6 1 3 /4 /2 0 0 6 1 9 /4 /2 0 0 6 2 4 /4 /2 0 0 6 2 7 /4 /2 0 0 6 1 /5 /2 0 0 6 2 /5 /2 0 0 6 3 /5 /2 0 0 6 4 /5 /2 0 0 6 1 0 /5 /2 0 0 6 2 7 /0 3 /2 0 0 6 3 1 /0 3 /2 0 0 6 7 /4 /2 0 0 6 1 1 /4 /2 0 0 6 Date (ddmmyy) Me n ta l s ta tu s [p o s , n e g o r n e u tr a l] MINDROOM MENTAL ALGO 1156 1133
Results: numeric
• 30 training sessions with mental score (MR + Questionnaire)
• 5 Primavera players
• Algorithm correctness:
• Objective: Quantification of mental status (fear) of a
horse in a non-invasive and continuous way during
physical exercise
• Experimental design
5’ 5’ 1’ 4’ 1’
Input
5’ 5’ 1’ 4’ 1’
walking trotting walking
Mathematical Model Physical activity (ActiGraph GTM1) Measured Heart rate (Polar RS 800) Output
• Modelling
5’ 5’ 1’ 4’ 1’
walking trotting walking
Input
5’ 5’ 1’ 4’ 1’
walking trotting walking
Mathematical Model Modelled Heart rate (HRphysical) Physical activity (ActiGraph GTM1) Measured Heart rate (Polar RS 800) Model error (HRmental) Output
• Prediction
5’ 5’ 1’ 4’ 1’
walking trotting walking
• Results:
a) Blue = input signal
Green = part of input signal
used for modelling
b) Blue = measured heart rate Black = modelled heart rate c) Black = model error
Red = presence of stressor
Legend:
33/37 :
mathematical model describing the relationship
between Physical Activity and Heart Rate Response
could be built (R
2avg
=0.93).
33/33 :
presence of the stressor could be automatically
detected
• Remarkable result:
Jansen et al. 2009. The veterinary
Heart Rate
T
bodyVideo
images
Example: development of a real-time monitor
& controller for status F1-driver
Practice\Qualification\Race
G-forces
Breaking
Steering
Car tuning
Weather
conditions
Track
conditions
… Mathematical model Algorithm Physical Algorithm MentalKarting: experiments
(Jeffrey van Hooydonk)
Sensors
• EMG
• Heart rate (ECG)
• Acceleration
• VO2max
-Longitudinal
-Transverse
3D accelerometer & Unipro
700 750 800 850 900 950 1000 1050 1100 1150 1200 -1 -0.5 0 0.5 1 Time [s] A c c e le ra ti o n [ g ]Yannick de Brabander (Horensbergdam 23/02/2008; lap 2)
700 750 800 850 900 950 1000 1050 1100 1150 1200 50 60 70 80 90 100 110 120 Time [s] S p e e d [ k m h ] longitudinal transverse 1 2 3 4 5 6 7 8 9 10 1112 Right turns Left turns