4/30/2007
1
Monitoring and Control in
Anaesthesia: an Implementation
Example
Catarina S. Nunes
Department of Applied Mathematics
Faculty of Sciences, University of Porto, Portugal
Research
Main Investigation topics
Modelling, identification and forecasting methods
Data analysis, clustering and knowledge extraction
Neural and decision support networks
Mathematical aspects of control systems design
Fault detection and alarm systems
Development and implementation of hybrid control algorithms
Software development
Medical Applications
Automatic control of drug delivery systems
Decision support and control in anaesthesia
Modelling and control of physiologic variables
http://pharmaria.fc.up.pt
4/30/2007
2
• Anaesthesia includes paralysis, unconsciousness (i.e. DOA) and
analgesia;
• Balanced anaesthesia, i.e. use of 3 drugs;
• DOA direct measurements are not available;
• The Bispectral Index (BIS) of the Electroencephalogram (EEG), a
multivariable regression model combining different features into a
linear numeric index, ranging from 0 (isoelectric EEG) to 100 (fully
awake).
• Auditory evoked potentials (AEP), i.e. evoked brain potentials, are
used to measure DOA (AEP are EEG responses to clicks applied to
both ears).
• State Entropy: EEG and EMG processing using Spectral Entropy.
Case Study: Depth of Anaesthesia (DOA)
Awareness
• “Anesthesia awareness is the phenomenon of being mentally alert (and terrified) while supposedly under full general anesthesia. The patient is paralyzed, unable to speak, and totally helpless to communicate his/her awareness. Actual cutting pain may or may not be present. “
• “The mission of the Anesthesia Awareness Campaign is to prevent patients (even one) from experiencing anesthesia awareness and its consequences through education, prevention, and empowerment by replacing ignorance or fear with knowledge.”
• “Anesthesia awareness continues to be reported between 100-200 times daily in the United States, in addition to an unknown number worldwide. Researchers believe that anesthesia awareness is under-reported by 50%-100% and occurs between 4-6 times more often in pediatric surgeries.”
http://www.anesthesiaawareness.com/
4/30/2007
3
Model Variability
ControlReference Error Action Unconsciousness
Analgesia
DEPTH OF ANAESTHESIA
Drug Interaction
AIM
Research Team
Faculty of Sciences
(Department of Applied
Mathematics)
Catarina S. Nunes
Teresa Feio Mendonça
Scholarships:
Nadja Bressan
Ana Castro
PhD Students:
Susana Brás
Hospital Geral de Santo António
(Department of Anaesthesiology)
Pedro Amorim
Francisco Lobo (also PhD Student)
Neurosurgery:
Isabel Alexandra Santos
Manuela Casal
Leónia Ferreira
Urology:
Eduarda Amadeu
Paula Sá Couto
Fátima Martins
4/30/2007
4
Research Team
University of Trás-os-Montes &
Alto Douro
(Department of Veterinary)
Luís Antunes
David A. Ferreira
PhD Students:
Heber Alves
Animal Hospital of Porto
Lénio Ribeiro (also PhD Student)
Institute of Molecular and Celular
Biology (Animal Welfare)
Ana Olsen,
Ana MariaValentim (Scholarship)
University Fernando Pessoa – Porto
(Pharmacology)
Pedro Barata
Institute of Oncology (Genetics)
Rui Medeiros
Case Study: Depth of Anaesthesia (DOA)
induced effectsensor
time varying
clinician
control
algorithm
computer infusion
pump
intravenous
administration
disturbances
observation noise
patient
Global main features:
!high interindividual variability
!input constraints in the control action
!level noise in measurements
!high degree of reliability
!good performance
!time varying parameters
Control objective: tracking a reference level
4/30/2007
5
Pharmacodynamic Drug Interactions
• Zero-interaction: the effect of the combination is the sum of the effects of the individual agents;
• Synergistic: the effect of the combination is greater than that expected as based on the concentration effect relationships of the individual agents;
• Antagonistic: the effect of the combination is less than the sum of the effects of the individuals;
• Propofol (anaesthetic) and Remifentanil (analgesic) have a synergistic interaction.
Anaesthesiologist
Initial Patient Response
!!!! 15 minutes (non-steady state)
Goal: Extracting from initial data Patient individual features
Control Modelling
Nunes CS, Mendonça TF, Antunes L, Ferreira DA, Lobo F, Amorim P. Modelling Drug’s Pharmacodynamic Interaction during General
Anaesthesia: the choice of Pharmacokinetic Model. Modelling and Control in Biomedical Systems 2006 (Including Biological Systems).
D. D. Feng, O. Dubios, J. Zaytoon and E. Carson, eds. Oxford, United Kingdom, Elsevier Science Ltd: 447-452, 2006.
4/30/2007
6
Potency
of the
mixture
Number of units (U)associated with 50%
of maximum effect
at ratio "
Normalised
concentrations
to the respective
potencies
Effect =
Interaction Model
Bruhn, J.; et al., 2003, Anesthesiology, vol. 98, pp. 621-627.
Real BIS values – 45 PatientsModelled BIS
0 5 10 1510
20
30
40
50
60
70
80
90
100
minutes
BIS
0 5 10 1510
20
30
40
50
60
70
80
90
100
minutes
BIS
4/30/2007
7
Predictive Adaptive Control of the Bispectral Index of the
EEG (BIS) - using the intravenous anaesthetic drug propofol
Mendonça T, Nunes CS, Magalhães H, Lemos JM, Amorim P: Predictive Adaptive Control of Unconsciousness –
Exploring Remifentanil as an Accessible Disturbance. Proceedings of the IEEE International Conference on Control Applications,
CCA06, Munich, Germany, October 4-6, pp. 205-210, 2006.
MUSMAR
4/30/2007
8
LNAI 4253 (Springer), Part 2: pp. 148-1455, 2006.
Fuzzy Logic – ANFIS – Modelling BIS
Journal of Intelligent and Fuzzy Systems 16(1):15-22, 2005
4/30/2007
9
Trainning Data
Testing Data