Transcript Complex Biomedical Signals

DIMACS April, 2002
Nonlinear Dynamics, Chaos, and
Complexity in Bedside Medicine
Ary L. Goldberger, M.D.
Harvard Medical School
NIH/NCRR Research Resource for
Complex Physiologic Signals (PhysioNet)
A Time Series Challenge:
Heart Rate Dynamics in Health and Disease
Which time series is normal?
Heart Failure
Heart Failure
Normal
Atrial Fibrillation
Cardiac Electrical System
How is Heart Rate Dynamics Regulated?
Coupled Feedback Systems Operating Over Wide Range of
Temporal/Spatial Scales
Three Themes
• Healthy systems show complex dynamics, with
long-range (fractal) correlations and multiscale
nonlinear interactions.
• Life-threatening pathologies and aging are
associated with breakdown of fractal scaling and
loss of nonlinear complexity.
• Open-source databases and software tools are
needed to catalyze advances in complex signal
analysis.
Hallmarks of Complexity
Healthy Heart Rate Dynamics
• Nonstationarity
• Statistics change with time
• Nonlinearity
• Components interact in unexpected ways ( “cross-talk” )
• Multiscale Variability
• Fluctuations may have fractal properties
Is the Physiologic World Linear or Nonlinear?
• Linear World:
• Things add up
• Proportionality of input/output
• High predictability, no surprises
• Nonlinear World:
• Whole  sum of parts (“emergent” properties)
• Small changes may have huge effects
• Low predictability, anomalous behaviors
What’s Wrong with this Type of
Signal Transduction Picture?
Answer: No feedback; No nonlinearity
Complicated! but …Complex dynamics missing!
*** Danger ***
Linear Fallacy: Widely-held assumption that biological
systems can be largely understood by dissecting out
micro-components and analyzing them in isolation.
“Rube Goldberg physiology”
Nonlinear/Fractal Mechanisms in Physiology
• Bad news: your data are complex!
• Good news: there are certain generic
mechanisms that do not depend on details
of system (universalities)
Wonderful World of Complexity:
Sampler of Nonlinear Mechanisms in Physiology
• Abrupt changes
• Bifurcations
• Bursting
• Bistability
•
•
•
•
• Nonlinear waves: spirals;
scrolls; solitons
• Stochastic resonance
• Time irreversibility
• Complex networks
• Emergent properties
Hysteresis
Nonlinear oscillations
Multiscale (fractal) variability
Deterministic chaos
Ref: Goldberger et al. PNAS 2002 99 Suppl. 1: 2466-2472.
Six Examples of
Spiral Waves in Excitable Media
From: J. Walleczek, ed. Self-Organized Biological Dynamics and Nonlinear Control
Cambridge University Press, 2000.
Multiscale Complexity and Fractals
Fractal: A tree-like object
or process, composed of
sub-units (and sub-subunits, etc) that resemble the
larger scale structure.
This internal look-alike
property is known as
self-similarity or
scale-invariance.
Fractal Self-Organization:
Coronary Artery Tree
Fractal Self-Organization:
His-Purkinje Conduction Network
Fractal Self-Organization:
Purkinje Cells in Cerebellum
Multiscale Complexity and Fractals
Fractal: A tree-like object
or process, composed of
sub-units (and sub-subunits, etc) that resemble the
larger scale structure.
This internal look-alike
property is known as
self-similarity or
scale-invariance.
Loss of Multiscale Fractal Complexity
with Aging & Disease
Healthy Dynamics: Multiscale Fractal Variability
Two Patterns of
Pathologic Breakdown
Single Scale Periodicity
Uncorrelated Randomness
Lancet 1996; 347:1312
Nature 1999; 399:461
Fractal Analysis of Nonstationary Time Series
Fractal Scaling in Health and Disease
Why is it Healthy to be Fractal?
• Healthy function requires capability to cope
with unpredictable environments
• Fractal systems generate broad repertoire of
response  adaptability
• Absence of characteristic time scale helps
prevent mode-locking (pathologic
resonances)
Concept of
DE-COMPLEXIFICATION OF DISEASE
• The output of many systems becomes more
regular and predictable with pathologic
perturbations
• Clinical medicine not feasible without such
stereotypic, predictable behaviors – clinicians look
for characteristic patterns/scales
• Healthy function: multi-scale dynamics/scale-free
behavior harder to characterize
Loss of Fractal Complexity
Resolves Clinical Paradox
Patients with wide range of disorders often display strikingly
predictable (ordered) dynamics
Reorder vs. Disorder
Examples:
Parkinsonism / Tremors
Obsessive-compulsive behavior
Nystagmus
Cheyne-Stokes breathing
Obstructive sleep apnea
Ventricular Tachycardia
Hyperkalemia  “Sine-wave” ECG
Cyclic neutropenia
etc., etc.
Warning!
Excessive Regularity is Bad For Your Health
Example: Photic (Stroboscopic) Stimulation and Seizures
What’s the Cure?
Finding and Using Hidden Information
• Physiologic dynamics exhibit an extraordinary
range of complexity that defies:
• Conventional statistics
• Homeostatic models
• Important information hidden in
complex signal fluctuations relating to:
• Basic signaling mechanisms
• Novel biomarkers
The Bad News for Complex Signal Analysis
• Databases are largely unavailable
or incompletely documented
• Investigators use different, undocumented
software tools on different databases
“ Babel-ography ”
NCRR Research Resource for
Complex Physiologic Signals - “PhysioNet”
www.physionet.org
Start date: September 1, 1999
100,000+ visits to date
1 terabyte of data downloaded!
Design of the PhysioNet Resource
PhysioNet
• Dissemination portal
• Tutorials
• Discussion Groups
Design of the PhysioNet Resource
PhysioBank
• Reference Datasets
• Multi-Parameter (e.g. sleep
apnea; intensive care unit)
• ECG
• Gait
• Other Neurological
• Images
• Data supporting publications
• 30+ gigabytes currently online
• 1+ terabytes online in 2003
Design of the PhysioNet Resource
PhysioToolkit
• Open source software
• Data analysis packages
• Physiologic models
• Software from publications
PhysioNet Signal Analysis Competitions
• Challenge 2001:
Can you forecast an imminent
cardiac arrhythmia (atrial fibrillation)
during normal cardiac rhythm?
• Challenge 2002:
Can you simulate/model complex healthy
heart rate variability?
• Future:
Seizure forecasting; Biomedical image processing, etc.
Conclusions
• Homeostasis revisited:
Physiologic control
Complex (fractal/nonlinear) dynamics
• Loss of fractal/nonlinear complexity:
New markers of life-threatening pathology/aging
• Needed: Open-source data and software for basic
mechanisms and bedside diagnostics
Welcome to PhysioNet!
www.physionet.org
Please visit and contribute