Post-lesion lateralisation shifts in a computational model of single-word reading

The mechanisms underlying lateralisation of language are incompletely understood. Existing data is inconclusive, for example, in determining which underlying asymmetries in hemispheric anatomy/physiology lead to lateralisation, the precise role of interhemispheric connections in this process, and exactly how and why lateralisation can shift following focal brain damage. Although these issues will ultimately be settled by experimentation, it is likely that computational modelling can be used to suggest, focus, and even interpret such empirical work. We have recently studied the emergence of lateralisation in an artificial neural network model having paired cerebral hemispheric regions, as the model learned to generate the correct pronunciation for simple words. In this paper we extend this previous work by examining the immediate and longer-term changes in lateralisation that occur following simulated acute hemispheric lesions. Among other things, the results demonstrate that the extent to which the non-lesioned model hemispheric region contributes to recovery is a function of lesion size, prelesion lateralisation, and assumptions about the excitatory/inhibitory influences of the corpus callosum. The relevance of these results to the currently controversial suggestion that language lateralisation shifts following focal damage to language areas, and that the unlesioned hemisphere contributes to recovery from stroke-induced aphasia in adults, is discussed.

Cortical spreading depression and the pathogenesis of brain disorders: a computational and neural network-based investigation

This paper reviews our recent studies of the role of cortical spreading depression (CSD) in the pathogenesis of brain disorders. Our investigation is a computational one, involving the development and utilization of a complex neuro-metabolic model of the interactions assumed to occur in the cortex during the passage of multiple CSD waves. Incorporating these neuro-metabolic changes of CSD within a neural network model of normoxic cortex produces cortical activation patterns during the passage of a CSD wave that, projected onto the visual fields, resemble the visual hallucinations observed during the migraine aura. When focal ischemia is simulated with the model, the evoked CSD waves are found to affect the expansion of the infarction into the ischemic penumbra. Our findings support the hypothesis that CSD does play an important pathogenic role in these and other neurological disorders, and suggest additional experimental studies that may further substantiate it.

The callosal dilemma: explaining diaschisis in the context of hemispheric rivalry via a neural network model

It is often suggested that a major factor in diaschisis is the loss of transcallosal excitation to the intact hemisphere from the lesioned one. However, there is long-standing disagreement in the broader experimental literature about whether transcallosal interhemispheric influences in the human brain are primarily excitatory or inhibitory. Some experimental data are apparently better explained by assuming inhibitory callosal influences. Past neural network models attempting to explore this issue have encountered the same dilemma: in intact models, inhibitory callosal influences best explain strong cerebral lateralization like that occurring with language, but in lesioned models, excitatory callosal influences best explain experimentally observed hemispheric activation patterns following brain damage. We have now developed a single neural network model that can account for both types of data, i.e., both diaschisis and strong hemisphere specialization in the normal brain, by combining excitatory callosal influences with subcortical cross-midline inhibitory interactions. The results suggest that subcortical competitive processes may be a more important factor in cerebral specialization than is generally recognized.

Cortical map asymmetries in the context of transcallosal excitatory influences

There is long-standing disagreement among experimentalists about whether transcallosal interhemispheric influences are primarily excitatory or inhibitory. Past computational models exploring this issue have encountered a similar dilemma: inhibitory callosal influences best explain hemispheric functional asymmetries, but excitatory callosal influences best explain transcallosal diaschisis. We recently hypothesized that this dilemma might be resolved by assuming excitatory callosal influences and a subcortical mechanism for cross-midline inhibition. Here we explore the feasibility of this hypothesis by examining a model of map formation in corresponding left and right cortical regions. The results show for the first time that both map asymmetries and diaschisis-like changes can be produced in a single model, suggesting that subcortical inhibitory processes may contribute more to asymmetric cortical functionality than is generally recognized.

Conditions enabling the evolution of inter-agent signaling in an artificial world

In the research described here we extend past computational investigations of animal signaling by studying an artificial world in which a population of initially noncommunicating agents evolves to communicate about food sources and predators. Signaling in this world can be either beneficial (e.g., warning of nearby predators) or costly (e.g., attracting predators or competing agents). Our goals were twofold: to examine systematically environmental conditions under which grounded signaling does or does not evolve, and to determine how variations in assumptions made about the evolutionary process influence the outcome. Among other things, we found that agents warning of nearby predators were a common occurrence whenever predators had a significant impact on survival and signaling could interfere with predator success. The setting most likely to lead to food signaling was found to be difficult-to-locate food sources that each have relatively large amounts of food. Deviations from the selection methods typically used in traditional genetic algorithms were also found to have a substantial impact on whether communication evolved. For example, constraining parent selection and child placement to physically neighboring areas facilitated evolution of signaling in general, whereas basing parent selection upon survival alone rather than survival plus fitness measured as success in food acquisition was more conducive to the emergence of predator alarm signals. We examine the mechanisms underlying these and other results, relate them to existing experimental data about animal signaling, and discuss their implications for artificial life research involving evolution of communication.

Interhemispheric effects of simulated lesions in a neural model of letter identification

Experimental studies have produced conflicting results about the extent to which the intact, nonlesioned cerebral hemisphere is responsible for recovery from cognitive deficits following focal brain damage such as a stroke. To obtain a better theoretical understanding of interhemispheric interactions during recovery, we examined the effects of simulated lesions to a bihemispheric neural model of letter identification under various assumptions about hemispheric asymmetries, corpus callosum influence, and lesion size. Among other results, the model demonstrates that the intact hemispheric region's participation in the recovery process is a function of preexisting lateralization and lesion size, indicating that interpretation of experimental work should take these factors into account.

A computational model of lateralization and asymmetries in cortical maps

While recent experimental work has defined asymmetries and lateralization in left and right cortical maps, the mechanisms underlying these phenomena are currently not established. In order to explore some possible mechanisms in theory, we studied a neural model consisting of paired cerebral hemispheric regions interacting via a simulated corpus callosum. Starting with random synaptic strengths, unsupervised (Hebbian) synaptic modifications led to the emergence of a topographic map in one or both hemispheric regions. Because of uncertainties concerning the nature of hemispheric interactions, both excitatory and inhibitory callosal influences were examined independently. A sharp transition in model behavior was observed depending on callosal strength. For excitatory or weakly inhibitory callosal interactions, complete and symmetric mirror-image maps generally appeared in both hemispheric regions. In contrast, with stronger inhibitory callosal interactions, partial to complete map lateralization tended to occur, and the maps in each hemispheric region often became complementary. Lateralization occurred readily toward the side having a larger cortical region or higher excitability. Asymmetric synaptic plasticity, however, had only a transitory effect on lateralization. These results support the hypotheses that interhemispheric competition occurs, that multiple underlying asymmetries may lead to function lateralization, and that the effects of asymmetric synaptic plasticity may vary depending on whether supervised or unsupervised learning is involved. To our knowledge, this is the first computational model to demonstrate the emergence of topographic map lateralization and asymmetries.

Interhemispheric effects of simulated lesions in a neural model of single-word reading

A neural model consisting of paired cerebral hemispheric regions interacting via homotopic callosal connections was trained to generate pronunciations for 50 monosyllabic words. Lateralization of this task occurred readily when different underlying cortical asymmetries were present. Following simulated focal cortical lesions of systematically varied sizes, acute changes in the distribution of cortical activation were found to be most consistent with experimental data when interhemispheric interactions were assumed to be excitatory. During subsequent recovery, the contribution of the unlesioned hemispheric region to performance improvement was a function of both the amount of preexisting lateralization and the side and size of the lesion. These results are discussed in the context of unresolved issues concerning the mechanisms underlying language lateralization, the nature of interhemispheric interactions, and the role of the nondominant hemisphere in recovery from adult aphasia.

Interhemispheric effects on map organization following simulated cortical lesions

During recent years there has been increasing use of neural models to investigate the implications of hypotheses about brain and cognitive disorders. Here we systematically study the effects of sudden simulated lesions on cortical maps in a neural model consisting of left and right hemispheric regions connected by a corpus callosum. The model identifies conditions under which damage to one hemispheric region leads to reorganization of the contralateral, intact hemispheric region. The intact hemisphere's participation in the recovery process is found to be a function of pre-existing map lateralization/symmetry and lesion size, indicating that interpretation of future experimental work should take these factors into account.

A neural network model of lateralization during letter identification

The causes of cerebral lateralization of cognitive and other functions are currently not well understood. To investigate one aspect of function lateralization, a bihemispheric neural network model for a simple visual identification task was developed that has two parallel interacting paths of information processing. The model is based on commonly accepted concepts concerning neural connectivity, activity dynamics, and synaptic plasticity. A combination of both unsupervised (Hebbian) and supervised (Widrow-Hoff) learning rules is used to train the model to identify a small set of letters presented as input stimuli in the left visual hemifield, in the central position, and in the right visual hemifield. Each visual hemifield projects onto the contralateral hemisphere, and the two hemispheres interact via a simulated corpus callosum. The contribution of each individual hemisphere to the process of input stimuli identification was studied for a variety of underlying asymmetries. The results indicate that multiple asymmetries may cause lateralization. Lateralization occurred toward the side having larger size, higher excitability, or higher learning rate parameters. It appeared more intensively with strong inhibitory callosal connections, supporting the hypothesis that the corpus callosum plays a functionally inhibitory role. The model demonstrates clearly the dependence of lateralization on different hemisphere parameters and suggests that computational models can be useful in better understanding the mechanisms underlying emergence of lateralization.

Pathogenic mechanisms in ischemic damage: a computational study

The pathogenesis of penumbral tissue infarction during acute ischemic stroke is controversial. This peri-infarct tissue may subsequently die, or survive and recuperate, and its preservation has been a prime goal of recent therapeutic trials in acute stroke. Two major hypotheses currently under consideration are that penumbral tissue is recruited into an infarct by cortical spreading depression (CSD) waves, or by a non-wave self-propagating process such as glutamate excitotoxicity (GE). Careful experimental attempts to discriminate between these two hypotheses have so far been quite ambiguous. Using a computational metabolic model of acute focal stroke we show here that the spatial patterns of tissue damage arising from artificially induced foci of infarction having specific geometric shapes are inherently different. This is due to the distinct propagation characteristics underlying self-regenerating waves and non-wave diffusional processes. The experimental testing of these predicted spatial patterns of damage may help determine the relative contributions of the two pathological mechanisms hypothesized for ischemic tissue damage.

Spreading depression in focal ischemia: a computational study

When a cerebral infarction occurs, surrounding the core of dying tissue there usually is an ischemic penumbra of nonfunctional but still viable tissue. One current but controversial hypothesis is that this penumbra tissue often eventually dies because of the metabolic stress imposed by multiple cortical spreading depression (CSD) waves, that is, by ischemic depolarizations. We describe here a computational model of CSD developed to study the implications of this hypothesis. After simulated infarction, the model displays the linear relation between final infarct size and the number of CSD waves traversing the penumbra that has been reported experimentally, although damage with each individual wave progresses nonlinearly with time. It successfully reproduces the experimental dependency of final infarct size on midpenumbra cerebral blood flow and potassium reuptake rates, and predicts a critical penumbra blood flow rate beyond which damage does not occur. The model reproduces the dependency of CSD wave propagation on N-methyl-D-aspartate activation. It also makes testable predictions about the number, velocity, and duration of ischemic CSD waves and predicts a positive correlation between the duration of elevated potassium in the infarct core and the number of CSD waves. These findings support the hypothesis that CSD waves play an important causal role in the death of ischemic penumbra tissue.

Computational studies of lateralization of phoneme sequence generation

The mechanisms underlying cerebral lateralization of language are poorly understood. Asymmetries in the size of hemispheric regions and other factors have been suggested as possible underlying causal factors, and the corpus callosum (interhemispheric connections) has also been postulated to play a role. To examine these issues, we created a neural model consisting of paired cerebral hemispheric regions interacting via the corpus callosum. The model was trained to generate the correct sequence of phonemes for 50 monosyllabic words (simulated reading aloud) under a variety of assumptions about hemispheric asymmetries and callosal effects. After training, the ability of the full model and each hemisphere acting alone to perform this task was measured. Lateralization occurred readily toward the side having larger size, higher excitability, or higher-learning-rate parameter. Lateralization appeared most readily and intensely with strongly inhibitory callosal connections, supporting past arguments that the effective functionality of the corpus callosum is inhibitory. Many of the results are interpretable as the outcome of a "race to learn" between the model's two hemispheric regions, leading to the concept that asymmetric hemispheric plasticity is a critical common causative factor in lateralization. To our knowledge, this is the first computational model to demonstrate spontaneous lateralization of function, and it suggests that such models can be useful for understanding the mechanisms of cerebral lateralization.

Self-replicating structures: evolution, emergence and computation

Since von Neumann's seminal work around 1950, computer scientists and others have studied the algorithms needed to support self-replicating systems. Much of this work has focused on abstract logical machines (automata) embedded in two-dimensional cellular spaces. This research was motivated by a desire to understand the basic information-processing principles underlying self-replication, the potential long-term applications of programmable self-replicating machines, and the possibility of gaining insight into biological replication and the origins of life. We view past research as taking three main directions: early complex universal computer-constructors modeled after Turing machines, qualitatively simpler self-replicating loops, and efforts to view self-replication as an emergent phenomenon. We discuss our recent studies in the latter category showing that self-replicating structures can emerge from nonreplicating components, and that genetic algorithms can be applied to program automatically simple but arbitrary structures to replicate. We also describe recent work in which self-replicating structures are successfully programmed to do useful problem solving as they replicate. We conclude by identifying some implications and important research directions for the future.

Metrics for cortical map organization and lateralization

Cerebral lateralization refers to the poorly understood fact that some functions are better controlled by one side of the brain than the other (e.g. handedness, language). Of particular concern here are the asymmetries apparent in cortical topographic maps that can be demonstrated electrophysiologically in mirror-image locations of the cerebral cortex. In spite of great interest in issues surrounding cerebral lateralization, methods for measuring the degree of organization and asymmetry in cortical maps are currently quite limited. In this paper, several measures are developed and used to assess the degree of organization, lateralization, and mirror symmetry in topographic map formation. These measures correct for large constant displacements as well as curving of maps. The behavior of the measures is tested on several topographic maps obtained by self-organization of an initially random artificial neural network model of a bihemispheric brain, and the results are compared with subjective assessments made by humans.

Computational models for the formation of protocell structures

There have been various attempts to simulate the self-assembly process of lipid aggregates by computers. However, due to the computationally complex nature of the problem, previous simulations were often conducted with unrealistic simplifications of the molecules' morphology, intermolecular interactions, and the environment in which the lipid molecules interact. In this article, we present a new computational model in which each lipid is simulated by a more realistic amphiphilic particle consisting of a hydrophilic head and a long hydrophobic tail. The intermolecular interactions are approximated by a set of simple forces reflecting physical and chemical properties of lipids, for example, hydrophobicity and electrostatic forces, which are believed to be crucial for the formation of various aggregates. With a set of carefully selected parameters, this model is able to simulate successfully the formation of micelles in an aqueous environment and reversed micelle structures in an oil solvent from an initially randomly distributed set of lipidlike particles. This model can be used to study, at the microscopic level, the self-assembly of different protocell structures in the evolutionary process and the impact of environmental conditions on the formation of these structures. It may be further generalized to simulate the formation of other, more complex structures of amphiphilic molecules such as monolayer and bilayer aggregates.

A computational model of acute focal cortical lesions

BACKGROUND AND PURPOSE

Determining how cerebral cortex adapts to sudden focal damage is important for gaining a better understanding of stroke. In this study we used a computational model to examine the hypothesis that cortical map reorganization after a simulated infarct is critically dependent on perilesion excitability and to identify factors that influence the extent of poststroke reorganization.

METHODS

A previously reported artificial neural network model of primary sensorimotor cortex, controlling a simulated arm, was subjected to acute focal damage. The perilesion excitability and cortical map reorganization were measured over time and compared.

RESULTS

Simulated lesions to cortical regions with increased perilesion excitability were associated with a remapping of the lesioned area into the immediate perilesion cortex, where responsiveness increased with time. In contrast, when lesions caused a perilesion zone of decreased activity to appear, this zone enlarged and intensified with time, with loss of the perilesion map. Increasing the assumed extent of intracortical connections produced a wider perilesion zone of inactivity. These effects were independent of lesion size.

CONCLUSIONS

These simulation results suggest that functional cortical reorganization after an ischemic stroke is a two-phase process in which perilesion excitability plays a critical role.

Frequency-spatial transformation: a proposal for parsimonious intra-cortical communication

This work examines a neural network model of a cortical module, where neurons are organized on a 2-dimensional sheet and are connected with higher probability to their spatial neighbors. Motivated by recent findings that cortical neurons have a resonant peak in their impedance magnitude function, we present a frequency-spatial transformation scheme that is schematically described as follows: An external input signal, applied to a small input subset of the neurons, spreads along the network. Due to a stochastic component in the dynamics of the neurons, the frequency of the spreading signal decreases as it propagates through the network. Depending on the input signal frequency, different neural assemblies will hence fire at their specific resonance frequency. We show analytically that the resulting frequency-spatial transformation is well-formed; an injective, fixed, mapping is obtained. Extensive numerical simulations demonstrate that a homogeneous, well-formed transformation may also be obtained in neural networks with cortical-like "Mexican-hat" connectivity. We hypothesize that a frequency-spatial transformation may serve as a basis for parsimonious cortical communication.

Feature discovery and classification of Doppler umbilical artery blood flow velocity waveforms

Doppler umbilical artery blood flow velocity waveform measurements are used in perinatal surveillance for the evaluation of fetal condition. There is an ongoing debate on the predictive value of Doppler measurements concerning the critical effect of the selection of parameters for the interpretation of Doppler waveforms. In this paper, we describe how neural network methods can be used both to discover relevant classification features and subsequently to classify Doppler umbilical artery blood flow velocity waveforms. Results obtained from 199 normal and high risk patients' umbilical artery waveforms highlighted a classification concordance varying from 90 to 98% accuracy.

Alignment of coexisting cortical maps in a motor control model

How do multiple feature maps that coexist in the same region of cerebral cortex align with each other? We hypothesize that such alignment is governed by temporal correlations: features in one map that are temporally correlated with those in another come to occupy the same spatial locations in cortex over time. To examine the feasibility of this hypothesis and to establish some of its detailed implications, we studied a multilayered, closed-loop computational model of primary sensorimotor cortex. A simulated arm moving in three dimensions formed the external environment for the model cortical regions. Coexisting proprioceptive and motor maps formed and generally aligned in a fashion consistent with the temporal correlation hypothesis. For example, in simulated proprioceptive sensory cortex the map of elements responding strongly to stretch of a particular muscle matched the map of tension sensitivity in antagonist muscles. In simulated primary motor cortex the map of elements responding strongly to increased tension in specific muscles matched the map of output elements for the same muscles. These computational results suggest specific experimental measurements that can support or refute the temporal correlation hypothesis for map alignments.

Learning activation rules rather than connection weights

In the construction of neural networks involving associative recall, information is sometimes best encoded with a local representation. Moreover, a priori knowledge can lead to a natural selection of connection weights for these networks. With predetermined and fixed weights, standard learning algorithms that work by altering connection strengths are unable to train such networks. To address this problem, this paper derives a supervised learning rule based on gradient descent, where connection weights are fixed and a network is trained by changing the activation rule. It incorporates both traditional and competitive activation mechanisms, the latter being an efficient method for instilling competition in a network. The learning rule has been implemented, and the results from several test networks demonstrate that it works effectively.

A computational model of visual hallucinations in migraine

Cortical spreading depression is a wave of electrical and biochemical changes that spreads across the cerebral cortex. It has been hypothesized to be an important underlying cause of the visual disturbances occurring during the migraine aura, but this is difficult to test in animals or humans. We created a computational model of cortical spreading depression and found that during the wave of biochemical changes the spatial pattern of neural activity broke up into irregular patterns of lines and small patches of highly activated elements. The corresponding visual disturbances that would be produced by these patterns of neural activity resemble the hallucinations reported during the migraine aura, providing strong support for the cortical spreading depression hypothesis of migraine. The model also makes the testable prediction that these hallucinations move at an exponentially increasing speed across the visual field.

Pathogenesis of schizophrenic delusions and hallucinations: a neural model

We implement and study a computational model of Stevens' theory of the pathogenesis of schizophrenia. This theory hypothesizes that the onset of schizophrenia is associated with reactive synaptic regeneration in brain regions that receive degenerating temporal lobe projections. Concentrating on one such area, the frontal cortex, we model a frontal module as an associative memory neural network whose input synapses represent incoming temporal projections. Modeling Stevens' hypothesized pathological synaptic changes in this framework results in adverse side effects similar to hallucinations and delusions seen in schizophrenia: spontaneous, stimulus-independent retrieval of stored memories focused on just a few of the stored patterns. These could account for the delusions and hallucinations that occur in schizophrenia without any apparent external trigger and for their tendency to concentrate on a few central cognitive and perceptual themes. The model explains why the positive symptoms of schizophrenia tend to wane as the disease progresses, why delayed therapeutic intervention leads to a much slower response, and why delusions and hallucinations may persist for a long time when they do occur.

Patterns of functional damage in neural network models of associative memory

Current understanding of the effects of damage on neural networks is rudimentary, even though such understanding could lead to important insights concerning neurological and psychiatric disorders. Motivated by this consideration, we present a simple analytical framework for estimating the functional damage resulting from focal structural lesions to a neural network model. The effects of focal lesions of varying area, shape, and number on the retrieval capacities of a spatially organized associative memory are quantified, leading to specific scaling laws that may be further examined experimentally. It is predicted that multiple focal lesions will impair performance more than a single lesion of the same size, that slit like lesions are more damaging than rounder lesions, and that the same fraction of damage (relative to the total network size) will result in significantly less performance decrease in larger networks. Our study is clinically motivated by the observation that in multi-infarct dementia, the size of metabolically impaired tissue correlates with the level of cognitive impairment more than the size of structural damage. Our results account for the detrimental effect of the number of infarcts rather than their overall size or structural damage, and for the "multiplicative" interaction between Alzheimer's disease and multi-infarct dementia.

A neural model of memory impairment in diffuse cerebral atrophy

BACKGROUND

Computer-supported neural network models have been subjected to diffuse, progressive deletion of synapses/neurons, to show that modelling cerebral neuropathological changes can predict the pattern of memory degradation in diffuse degenerative processes such as Alzheimer's disease. However, it has been suggested that neural models cannot account for more detailed aspects of memory impairment, such as the relative sparing of remote versus recent memories.

METHOD

The latter claim is examined from a computational perspective, using a neural associative memory model.

RESULTS

The neural network model not only demonstrates progressive memory deterioration as diffuse network damage occurs, but also exhibits differential sparing of remote versus recent memories.

CONCLUSIONS

Our results show that neural models can account for a large variety of experimental phenomena characterising memory degradation in Alzheimer's patients. Specific testable predictions are generated concerning the relation between the neuraonatomical findings and the clinical manifestations of Alzheimer's disease.

High-specificity neurological localization using a connectionist model

Most previous connectionist models for diagnosis have been developed using error backpropagation. While these systems function reasonably well, they have been limited by their need for a large database of test cases, to situations where a single disorder is present, and by the large number of connections required between fully-connected sets of processing units. Here we describe a recently developed connectionist model that overcomes these limitations. This approach can reuse existing causal knowledge bases, works well in situations where multiple disorders can occur simultaneously, and does not require fully-connected sets of processing units. We demonstrate that the accuracy of this model is comparable to that of more conventional AI programs using the same knowledge base in determining precisely the site of brain damage in a group of 50 stroke patients. These results support the conclusion that connectionist models can effectively use pre-existing causal knowledge bases from AI systems, and that they can function accurately when handling actual clinical problems.

A neural model of cortical map reorganization following a focal lesion

Neural models based on fairly simple assumptions have been able to account for topographic map formation in sensory cortex and the map reorganization that occurs following repetitive stimulation and deafferentation. The spontaneous reorganization that follows an acute focal cortical lesion, however, has not been modeled successfully. We have developed a computational model of cortex based on the hypothesis that cortical activation is distributed competitively. This model exhibited spontaneous reorganization following a focal cortical lesion and makes a testable prediction about the time course of that reorganization. We describe our model and the hypotheses upon which it is based, and examine some of the factors which influence post-lesion reorganization. We also demonstrate that the extent of post-lesion reorganization can be greatly improved through selective repetitive stimulation, suggesting a clinical rehabilitation technique that can be tried in an experimental setting for patients suffering sensory loss due to focal brain damage.

Functional pronunciation units in English words

Two distinct factors limit the orthographic regularity of English words: (a) Most characters can correspond to several different sounds and (b) characters can either stand alone or be combined in various ways for pronunciation as a single phoneme. This study addresses the second of these issues through the analysis of a large corpus of English words. Data are presented describing the frequency that each character (or character cluster) functioned in the corpus as a correspondent of a single phoneme rather than being combined with other characters (or decomposed). Examples are provided regarding potential applications of these data in the construction of stimulus materials for cognitive studies, in neuropsychological investigations of dyslexia, and in computational models of word naming.

Map formation in proprioceptive cortex

Current understanding of feature maps in proprioceptive cortex is quite limited. To complement experimental studies, we developed a computational model of map formation in proprioceptive cortex. Muscle length and tension from six muscle groups controlling the position of a model arm in three-dimensional space served as input to the simulated cortex. The resultant feature map consisted of regularly spaced clusters of cortical columns representing individual muscle lengths and tensions. Cortical units became tuned to plausible combinations of tension and length, and multiple representations of each muscle group were present. The map was organized such that compact regions within which all muscle group lengths and tensions are represented could be identified. Most striking was the observation that, although not explicitly present in the input, the cortical map developed a representation of the three-dimensional space in which the arm moved. These findings represent testable predictions about proprioceptive cortex, and may also help clarify some organizational issues concerning primary motor cortex.

An extended cellular space method for simulating autocatalytic oligonucleotides

Self-replicating nucleotides and other self-replicating molecules are an active area of study today by organic chemists. Such studies are important for improving our understanding of the origins of life. Computational studies of self-replicating molecules could increase our insight into their properties, but existing computational techniques have been limited in their usefulness for such reactions (numerical simulation of differential equations requires reaction rate constants that are difficult to obtain, cellular automata models are too restrictive for modeling molecular movements and bindings, etc.). We have thus developed an efficient modified cellular automata method that supports the study of self-replicating oligonucleotides. We explain this method and illustrate its use with a specific self-replicating (autocatalytic) deoxyribohexanucleotide.

Modeling cortical spreading depression

Cortical spreading depression is a wave of electrical silence and biochemical changes that spreads across the cerebral cortex. Recently there has been a growing recognition that it may be an important pathophysiological event in a number of neurological disorders. In this paper, we describe a reaction-diffusion model of the extracellular potassium changes that are a central part of this process. Simulations with the model show that an appropriate stimulus evokes a moving wave of increased potassium with many similarities to that seen experimentally. The resultant model is a useful computational tool for future study of the effects of spreading depression on the cortex.

Modeling brain adaptation to focal damage

Determining how feature maps in the cerebral cortex adapt to sudden, focal damage is important for gaining a deeper understanding of neurological illnesses such as stroke. In this paper we describe a neural model of the region of primary sensory cortex related to upper extremity proprioception, and show how the feature map there reorganizes following a simulated lesion. A perilesion zone with decreased activity appears and then gradually expands with time. These results differ from those seen with previous models of cortical lesions, and offer an alternative mechanism to the "ischemic penumbra" seen in certain types of stroke.

Interpretation of Doppler blood flow velocity waveforms using neural networks

Doppler umbilical artery blood flow velocity waveform measurement is used in perinatal surveillance for the evaluation of pregnancy status. There is an ongoing debate on the predictive value of Doppler measurements concerning the critical effect of the selection of parameters for the evaluation of Doppler output. In this paper, we describe how neural network methods can be used both to discover relevant classification features and subsequently to classify patients. Classification accuracy varied from 92-99% correct.

Effects of normalization constraints on competitive learning

Implementations of competitive learning often use input and weight vectors "normalized" based on the sum of weight vector components. While it is realized that some distortion of results can occur with this procedure, it is generally not appreciated how dramatic the distortion can be, and that it compromises the dot product as a similarity measure. We show here that in some cases an input vector identical to an existing output node weight vector can be classified as belonging to a different output node. This contradicts the generally-accepted concept of weight vectors developing as prototypes during competitive learning. Ways to minimize this problem are also given.

Multiple disorder diagnosis with adaptive competitive neural networks

Backpropagation neural networks have repeatedly been used for diagnostic problem-solving, but have not been demonstrated to work well when multiple disorders are present. We hypothesized that letting nodes in a backpropagation neural network compete to be part of a diagnostic solution would produce better performance than the use of existing backpropagation methods. To test this hypothesis, we derived an error backpropagation learning rule that can be used with competitive units (competitive backpropagation). Artificial neural networks were then trained using both this new learning rule and standard error backpropagation on a specific medical diagnosis problem: identification of the location of damage in the brain given a set of examination findings. Training samples included solely 'prototypical' cases where a single location of damage is present. The trained networks were then tested with atypical cases where the manifestations of more than one disorder were present or only a single manifestation was present. Networks employing competition among units were found to perform qualitatively better with these multiple-disorder cases than standard networks and also to perform better on single-manifestation cases. The reasons for this are explained. The competitive backpropagation learning rule described here provides a promising new tool for adaptive diagnostic problem-solving.

Neural computation in medicine

During the last decade there has been a great revival of interest in neural modelling. Powerful new computational methods have resulted from work in this area and are being applied to an increasing range of medical problems. This paper briefly explains the nature of a neural model and then reviews work in neural computation involving problems in medical informatics (e.g. expert systems) and modelling of psychiatric and neurological phenomena. The state of the art is assessed, and speculation about future developments is given.

Simple Systems That Exhibit Self-Directed Replication

Biological experience and intuition suggest that self-replication is an inherently complex phenomenon, and early cellular automata models support that conception. More recently, simpler computational models of self-directed replication called sheathed loops have been developed. It is shown here that "unsheathing" these structures and altering certain assumptions about the symmetry of their components leads to a family of nontrivial self-replicating structures, some substantially smaller and simpler than those previously reported. The dependence of replication time and transition function complexity on initial structure size, cell state symmetry, and neighborhood are examined. These results support the view that self-replication is not an inherently complex phenomenon but rather an emergent property arising from local interactions in systems that can be much simpler than is generally believed.

Minimizing complexity in cellular automata models of self-replication

Understanding self-replication from an information processing perspective is important because, among other things, it can shed light on molecular mechanisms of biological reproduction and on prebiotic chemical evolution. Intuition, biological knowledge, and early computational models of self-replication all suggested that self-replication is an inherently complex process. In this paper we describe recent computational studies that challenge this viewpoint. We summarize our recent work with cellular automata models of simple yet non-trivial self-replicating structures called unsheathed loops. For example, one unsheathed loop consists of only six components and requires only 20 rules to specify the local intercomponent interactions needed to bring about replication. The implication of this work is that, when viewed as an emergent property of numerous local, concurrent interactions between components, self-replicating systems can be substantially simpler than is generally recognized.

Modelling acquired dyslexia: a software tool for developing grapheme-phoneme correspondences

In extending a computer model of acquired dyslexia, it has become necessary to develop a way to group printed characters in a word so that the character groups essentially have a one-to-one correspondence with the word's phonemes (speech sounds). This requires deriving a set of correspondences (legal character groupings, legal associations of character groups with phonemes, etc.) that yield a single grouping or "segmentation" of characters when applied to any English word. To facilitate and partially automate this task, a segmentation program has been developed that uses an interchangeable set of correspondences. The program segments words according to these correspondences and tabulates their success over large sets of words. The program has been used successfully to segment a 20,000 word corpus, demonstrating that this approach can be used effectively and efficiently.

The design and automated testing of an expert system for the differential diagnosis of acute stroke

Stroke is the third leading cause of death in the United States and a major source of morbidity. [1] Recent studies have shown a potential use for thrombolytic agents in the treatment of ischemic stroke (IS) but these agents are contraindicated in intracerebral hemorrhage (ICH). A computed tomographic scan is used to distinguish between these two stroke types prior to the use of thrombolytic agents, but may not be readily obtainable. Decision making aids such as algorithms developed at Guy's Hospital and Strong Memorial Hospital have been designed in an attempt to make this distinction on clinical grounds. We have constructed computerized medical decision-making (CMD) systems based upon these algorithms and compared their performance to a system we developed with the use of National Stroke Data Bank data. Relevant medical data for each of 337 patient cases in the Mount Sinai Hospital Stroke Data Bank were presented to each of the CMD systems. In consideration of the clinical task of using thrombolytic agents, we attempted to maximize the positive predictive value (PPV) for ischemic stroke. The CMD systems based upon the Guy's Hospital and Mount Sinai algorithms produced PPV's of 95% and 94% with sensitivities of 77% and 78% respectively compared to a PPV of 93% and sensitivity of 56% with the Strong Memorial CMD system. The Mount Sinai CMD system was judged more efficacious than the Guy's Hospital system in that it required less clinical information that could be more easily obtained to arrive at similar results.

Maryland MIRRORS/II: a connectionist model simulator

Developing and evaluating connectionist models (also called neural models) is a difficult and time-consuming task. To address this issue, we designed a software system called Maryland MIRRORS/II for the construction of connectionist models in biomedicine and other fields. Maryland MIRRORS/II is distinguished from previous and current related systems by its support of a high-level nonprocedural language, a general-purpose event-handling mechanism, and an indexed library of system resources. These features make Maryland MIRRORS/II a convenient software tool for use in biomedicine. This paper describes Maryland MIRRORS/II and provides a simple example in which it uses error back propagation learning to select the appropriate treatment for a given set of manifestations.