Therapeutic Implications of
Computer Models of Brain Activity
for Alzheimer Disease.
|
Włodzisław Duch
|
|
Computational Intelligence Laboratory,
Department of Informatics,
Nicolaus Copernicus University,
Grudziądzka 5, 87-100 Toruń, Poland.
WWW: http://www.fizyka.umk.pl/~duch
Recent Projects of the
Department of Informatics
Computational Intelligence (CI)
- New Computational Intelligence algorithms: Feature Space Mapping (neurofuzzy), D-MLPs (generalization of MLP), SSV (decision tree), ontogenic IncNet network, MDS-based visualization, transfer functions ...
- Integration of neural, fuzzy, pattern recognition, statistical and machine learning methods, based on
analysis of similarity and searching for models for a given data.
- Data mining, data understanding, scientific discovery.
Cognitive Science
- Collaboration with neurobiology/education/psychology and philosophy
departments, cognitive science journal.
- Theories between neurosciences and cognitive psychology:
space for mind events.
- Philosophy of mind.
Applications
- Ghostminer project: Intelligent Decision Support System for medical,
psychometric, chemical, molecular biology
and commercial data analysis (with Fujitsu).
- Cognitive toys (with Fujitsu).
- fMRI image analysis (with MPI Neuropsychology Institute).
Some papers are in our on-line archive:
WWW: http://www.fizyka.umk.pl/kmk/publications.html
Plan
- Computational psychiatry
- Methods and applications
- Alzheimer disease
- Competing hypothesis
- Therapeutic implications
- Future plans
Computational psychiatry
- Medicine and psychiatry - experimental, phenomenological.
- Understanding of causal interactions is difficult using the stimulus-response paradigm.
- Neurochemistry/psychopharmacology: answers only to what happens if, not how.
- Different levels of organization.
- Different questions at different levels.
History.
- Neurodynamics - N. Rashevski 1938.
- W. McCulloch and W. Pitts 1947-49, D. Hebb 1949.
- A.L. Hodgkin and A.F. Huxley, biophysical neural model 1952.
- S. Grossberg, "Studies of mind and brain" 1982.
- Hopfield, dynamical model 1982; Kohonen, self-organized mapping 1982.
- PDP group (D. Rumelhart, J. McClleland, G.Hinton) 1985, backpropagation 1986.
- Cognitive computational neuroscience - 1992 book "Computational brain", P. Churchland, T. Sejnowski.
- D. Amit "Modeling brain function", 1989 book.
- NIMH sponsors computational simulations since 1988, first conference in 1995.
- Review "Neural Modeling of Psychiatric Disorders", E. Ruppin, Network 1995.
“Neural network models offer a better chance of rescuing the study of human psychological responses to drugs than anything else currently available”
(Callaway, Halliday, Naylor, Yano, Herzig, Neuropsychopharmacology 1994).
Methods and applications
Psychiatric models must first perform basic cognitive functions, than pathology is introduced.
Lesioning connections disrupts dynamic mechanisms.
Similar - neurological disorders, such as epilepsy, cortical reorganization after stroke, frontal lobe syndromes ...
Neuropsychological and cognitive disturbances.
Review: E. Ruppin, Network 6 (1995) 635-656
Some psychiatric and neurological problems that has been modeled so far.
Hopfield model (1982)
Active nodes (neurons)
(V=+1, dark)
Non-active (V=-1, light)
All nodes are connected.
Symmetric synaptic connections Wij=Wji
Stable patterns of excitations = memory traces.
|
|
Simplest threshold dynamics:
Point attractors correspond to the minima of:
More sophisticated Hebbian networks have cyclic attractors.
Associative memory:
- content addressable recall, pattern completion;
- timing independent of the number of patterns stored;
- robust against network damage;
- interference of similar patterns is more frequent than different;
- storing too many patterns leads to memory break-down.
Alzheimer disease
Global dysfunction at the behavioral level:
- memory impairment,
- thinking and judgment impairment,
- poverty of associations,
- paranoid delusions (in half of the patients),
- sometimes slow, sometimes rapid deterioration, always fatal.
Etiology unknown; some forms due to genetic risk.
10% of people over 70 have it.
Biochemical level - responsible for degeneration, but neural network level for behavior.
Drugs slowing down the memory impairment - increase acetylocholine realease.
At the neuron level:
- senile amyloid plaques and neuritic plaques,
- neurofiblirary tangles, microtubule tau protein,
- brain atrophy, especially in frontal and temporal regions;
- total neuron loss up to 20%,
- plaque and tangle formation accounts only partially for synaptoc patology.
- AD confirmed by brain biopsy since other
types of dementia may give similar symptoms.
Neural mechanisms.
Synaptic deletion - reduction of the number of synapses;
Synaptic compensation - increase in the size of the remaining ones.
Types of questions:
How deletion/compensation influences memory deterioration?
What strategies of compensation could maintain memory capacity?
Model of associative memory based on Hopfield-like attractor network (D. Horn et/al. 1993).
Assume that:
- connections are randomly deleted;
- remaining connections adjust to preserve memory capacity.
Estimation of associative memory capacity - methods developed in statistical physics.
Each pattern has a basin of attraction, patterns similar to those memorized should be correctly recognized.
Memory deterioration is delayed if the remaining connections grow by:
d - level of random synaptic deletion,
k=k(d) is a compensation-strategy parameter, fitted to experimental data.
Different k=k(d) functions lead to a different time course of the disease.
Horn and Ruppin, new hypothesis for the appearence of parkinsonian symptoms in AD patients.
Problem with Hopfield networks - non-local learning, not plausible from neurobiological point of view.
Local Hebbian synaptic storage, attractor network model, Willshaw model (Ruppin & Reggia 1995):
Simulations with 1500 neurons, 75 memory patterns, activity 0.05.
Local compansation, different for each neural module: CiWij
In reality: changes of firing thresholds.
Left to right: no compensation, global and local compension.
Compensation normalizes sizes of the basins of attractors.
Hebbian synaptic modification - learning, memory.
Neuronal activity-dependent compensation mechanism - maintaining.
This model explains:
- temporal gradients of memory decline,
- predicts relations between neuroanatomical degradation and clinical manifestation of AD,
Local compensation is history dependent
⇒ broad variability in magnitude of degradation for the same cognitive competencies.
AD: failure of normal regulatory (compensatory) mechanism.
Neural tangles result from disruption of cellural transportantion system, hence deficient synaptic compensation.
AD patients without tangles - probably due to excessive synaptic loss.
Coupling between metabolic cellular degeneration and system level effects.
Alternative hypothesis: Hasselmo (1992-1995).
Storing new patterns guided by interference with many old patterns, too many explicit combinations to store.
If external strength is large enough or if internal inhibition strong enough it may be prevented, but beyond critical storage capacity it is unavoidable.
Alzheimer Disease - due to the synaptic runaway?
Decrease of the cortical inhibition
⇒ excess memory storage requirements
⇒ reduced synaptic decay
⇒ pathological growth of synaptic connections
⇒ excessive metabolic demands
⇒ neuronal degradation.
Synaptic loss, lowering synapse/neuron ratio: large cognitive deficits with little structural damage.
Neural loss - large structural damage, faster degradation.
This model explains:
- why enthorinal regions, lacking internal inhibition, are more damaged than cortical,
- why loss of cholinergic innervation leads to low internal inhibition,
- ACh does not influence external (between modules) inhibition; sprouting of cholinergic innervation in dentate gyrus in AD patients may reflect the attempt to stop synaptic runaway.
Some evidence of neuro-regulatory processes: pyramidal neurons scale the overall strength of their synaptic connections as a function of their activity.
Search for more experimental evidence in favour of these models.
More detailed models needed but ...
Therapeutic implications
Some suggestions resulting from computational models:
- Minimize new memory load.
Use simple and regular daily routine; minimize the number of new facts remembered.
Avoid following visual, auditory or printed stories, TV news, series.
Sedatives may have positive effect on the memory overload.
- Strengthen the old, well-established memory patterns.
Recal stories and facts of patient's life.
Maintain 'self', cf. "Self-Maintenance-Therapy" + use emotional arousal to deepen basic facts/memories.
- Simplify brain dynamics to avoid memory interference.
Synaptic runaway processes overheating the brain - cool the "brain temperature".
Use 'alpha relaxation state' biofeedback, yoga meditation, deep relaxation techniques.
Other attempts to make some therapeutic suggestions: neurological damage.
For example ischemic tissue damage represented by Cortical Spreading Depression Waves with high concentration of K+ ions, during acute stroke.
What needs to be done
- More sophisticated neural models are needed!
- System level: inclusion of hipppocampal formation and enthorinal/peririhnal cortex.
- Interactions with neuromodulatory systems.
This model explains many forms of amnesia.
- Neural level: specific responses to specific drugs.
- Spiking neurons - necessary to make connections to neurophysiology.
- Connection between changes of firing thresholds and model parameters.
- Metabolic/neural activity coupling poorly understood.
- Finding optimal simplification of neural description for this type of models.
Computational simulations of more sophisticated models - only quite recently (J. Murre, private information).
Włodzisław Duch