CiteULike: manabu-s's library [162 articles]

What we can do and what we cannot do with fMRI

Nikos Logothetis — 2008-06-11T21:05:36-00:00

Nature, Vol. 453, No. 7197. (12 June 2008), pp. 869-878.

A quantitative analysis of generation of saccadic eye movements by burst neurons.

JA Van Gisbergen — 2008-02-19T08:55:36-00:00

J Neurophysiol, Vol. 45, No. 3. (March 1981), pp. 417-442.

A model of the smooth pursuit eye movement system.

DA Robinson — 2008-05-14T00:49:28-00:00

Biological cybernetics, Vol. 55, No. 1. (1986), pp. 43-57.

Human, horizontal, smooth-pursuit eye movements were recorded by the search coil method in response to Rashbass step-ramp stimuli of 5 to 30 deg/s. Eye velocity records were analyzed by measuring features such as the time, velocity and acceleration of the point of peak acceleration, the time and velocity of the peaks and troughs of ringing and steady-state velocity. These values were averaged and mean responses reconstructed. Three normal subjects were studied and their responses averaged. All showed a peak acceleration-velocity saturation. All had ringing frequencies near 3.8 Hz and the mean steady-state gain was 0.95. It is argued that a single, linear forward path with any transfer function G(s) and a 100 ms delay (latency) cannot simultaneously simulate the initial rise of acceleration and ring at 3.8 Hz based on a Bode analysis. Also such a simple negative feedback model cannot have a steady-state gain greater than 1.0; a situation that occurs frequently experimentally. L.R. Young's model, which employs internal positive feedback to eliminate the built-in unity negative feedback, was felt necessary to resolve this problem and a modification of that model is proposed which simulates the data base. Acceleration saturation is achieved by borrowing the idea of the local feedback model for saccades so that one nonlinearity can account for the acceleration-velocity saturation: the main sequence for pursuit. Motor plasticity or motor learning, recently demonstrated for pursuit, is also incorporated and simulated. It was noticed that the offset of pursuit did not show the ringing seen in the onset so this was quantified in one subject. Offset velocity could be characterized by a single exponential with a time constant of about 90 ms. This observation suggests that fixation is not pursuit at zero velocity and that the pursuit system is turned on when needed and off during fixation.

The use of control systems analysis in the neurophysiology of eye movements.

DA Robinson — 2005-12-19T17:22:05-00:00

Annu Rev Neurosci, Vol. 4 (1981), pp. 463-503.

Hierarchical models of behavior and prefrontal function

Matthew Botvinick — 2008-05-02T06:40:39-00:00

Trends in Cognitive Sciences, Vol. 12, No. 5. (May 2008), pp. 201-208.

The recognition of hierarchical structure in human behavior was one of the founding insights of the cognitive revolution. Despite decades of research, however, the computational mechanisms underlying hierarchically organized behavior are still not fully understood. Recent findings from behavioral and neuroscientific research have fueled a resurgence of interest in the problem, inspiring a new generation of computational models. In addition to developing some classic proposals, these models also break fresh ground, teasing apart different forms of hierarchical structure, placing a new focus on the issue of learning and addressing recent findings concerning the representation of behavioral hierarchies within the prefrontal cortex. In addition to offering explanations for some key aspects of behavior and functional neuroanatomy, the latest models also pose new questions for empirical research.

Neuronal circuitry controlling the near response

Lawrence Mays — 2008-04-29T19:42:44-00:00

Current Opinion in Neurobiology, Vol. 5, No. 6. (December 1995), pp. 763-768.

Experiments in primates have contributed greatly to our understanding of the neural mechanisms involved in the eye movements required to view objects at different distances. Early work focused on the circuitry for generating horizontal vergence eye movements alone. However, vergence eye movements are associated with lens accommodation and are usually accompanied by saccadic eye movements, so more recent work has been directed at understanding the interactions between vergence and accommodation, and between vergence and saccades. A new model explains the neural basis for interactions between vergence and accommodation by a neural network in which pre-motor elements are shared by these two systems. The effects of saccades on vergence eye movements appear to be the result of shared pre-motor circuits as well. Current evidence suggests that pontine omnipause neurons, known to be crucial for the generation of saccades, play an important role in the vergence pre-motor circuitry.

On the predictive control of foveal eye tracking and slow phases of optokinetic and vestibular nystagmus.

S Yasui — 2008-04-29T18:39:14-00:00

The Journal of physiology, Vol. 347 (February 1984), pp. 17-33.

Smooth pursuit and saccadic components of foveal visual tracking as well as more involuntary ocular movements of optokinetic (o.k.n.) and vestibular nystagmus slow phase components were investigated in man, with particular attention given to their possible input-adaptive or predictive behaviour. Each component in question was isolated from the eye movement records through a computer-aided procedure. The frequency response method was used with sinusoidal (predictable) and pseudo-random (unpredictable) stimuli. When the target motion was pseudo-random, the frequency response of pursuit eye movements revealed a large phase lead (up to about 90 degrees) at low stimulus frequencies. It is possible to interpret this result as a predictive effect, even though the stimulation was pseudo-random and thus 'unpredictable'. The pseudo-random-input frequency response intrinsic to the saccadic system was estimated in an indirect way from the pursuit and composite (pursuit + saccade) frequency response data. The result was fitted well by a servo-mechanism model, which has a simple anticipatory mechanism to compensate for the inherent neuromuscular saccadic delay by utilizing the retinal slip velocity signal. The o.k.n. slow phase also exhibited a predictive effect with sinusoidal inputs; however, pseudo-random stimuli did not produce such phase lead as found in the pursuit case. The vestibular nystagmus slow phase showed no noticeable sign of prediction in the frequency range examined (0 approximately 0.7 Hz), in contrast to the results of the visually driven eye movements (i.e. saccade, pursuit and o.k.n. slow phase) at comparable stimulus frequencies.

Computational study on monkey VOR adaptation and smooth pursuit based on the parallel control-pathway theory.

H Tabata — 2008-04-29T18:26:06-00:00

Journal of neurophysiology, Vol. 87, No. 4. (April 2002), pp. 2176-2189.

Much controversy remains about the site of learning and memory for vestibuloocular reflex (VOR) adaptation in spite of numerous previous studies. One possible explanation for VOR adaptation is the flocculus hypothesis, which assumes that this adaptation is caused by synaptic plasticity in the cerebellar cortex. Another hypothesis is the model proposed by Lisberger that assumes that the learning that occurs in both the cerebellar cortex and the vestibular nucleus is necessary for VOR adaptation. Lisberger's model is characterized by a strong positive feedback loop carrying eye velocity information from the vestibular nucleus to the cerebellar cortex. This structure contributes to the maintenance of a smooth pursuit driving command with zero retinal slip during the steady-state phase of smooth pursuit with gain 1 or during the target blink condition. Here, we propose an alternative hypothesis that suggests that the pursuit driving command is maintained in the medial superior temporal (MST) area based on MST firing data during target blink and during ocular following blank, and as a consequence, we assume a much smaller gain for the positive feedback from the vestibular nucleus to the cerebellar cortex. This hypothesis is equivalent to assuming that there are two parallel neural pathways for controlling VOR and smooth pursuit: a main pathway of the semicircular canals to the vestibular nucleus for VOR, and a main pathway of the MST-dorsolateral pontine nuclei (DLPN)-flocculus/ventral paraflocculus to the vestibular nucleus for smooth pursuit. First, we theoretically demonstrate that this parallel control-pathway theory can reproduce the various firing patterns of horizontal gaze velocity Purkinje cells in the flocculus/ventral paraflocculus dependent on VOR in the dark, smooth pursuit, and VOR cancellation as reported in Miles et al. at least equally as well as the gaze velocity theory, which is the basic framework of Lisberger's model. Second, computer simulations based on our hypothesis can stably reproduce neural firing data as well as behavioral data obtained in smooth pursuit, VOR cancellation, and VOR adaptation, even if only plasticity in the cerebellar cortex is assumed. Furthermore, our computer simulation model can reproduce VOR adaptation automatically based on a heterosynaptic interaction model between parallel fiber inputs and climbing fiber inputs. Our results indicate that different assumptions about the site of pursuit driving command maintenance computationally lead to different conclusions about where the learning for VOR adaptation occurs. Finally, we propose behavioral and physiological experiments capable of discriminating between these two possibilities for the site of pursuit driving command maintenance and hence for the sites of learning and memory for VOR adaptation.

The local loop of the saccadic system closes downstream of the superior colliculus

R Kato — 2007-01-18T23:30:28-00:00

Neuroscience, Vol. 143, No. 1. (17 November 2006), pp. 319-337.

Models of the saccadic system differ in several respects including the signals fed back to their comparators, as well as the location and identity of the units that could serve as comparators. Some models place the comparator in the superior colliculus while others assign this role to the reticular formation. To test the plausibility of reticular models we stimulated electrically efferent fibers of the superior colliculus (SC) of alert cats along their course through the pons, in the predorsal bundle (PDB). Our data demonstrate that electrical stimulation of the PDB evokes saccades, even with stimuli of relatively low frequency (100 Hz), which are often accompanied by slow drifts. The velocity and latency of saccades are influenced by the intensity and frequency of stimulation while their amplitude depends on the intensity of stimulation and the initial position of the eyes. The dynamics of evoked saccades are comparable to those of natural, self-generated saccades of the cat and to those evoked in response to the electrical stimulation of the SC. We also show that PDB-evoked saccades are not abolished by lesions of the SC and that therefore antidromic activation of the SC is not needed for their generation. Our data clearly demonstrate that the burst generator of the horizontal saccadic system is located downstream of the SC. If it is configured as a local loop controller, as assumed by most models of the saccadic system, our data also demonstrate that its comparator is located beyond the decussation of SC efferent fibers, in the pons.

Models of the saccadic eye movement control system

David Robinson — 2008-04-29T17:32:29-00:00

Biological Cybernetics, Vol. 14, No. 2. (1 June 1973), pp. 71-83.

1.A sequence of four models is proposed for the saccadic eye movement control system. The models become increasingly complex as they are made to respond to increasingly more complicated target movements in accordance with experimental results. Compatibility with neurological structure and function is stressed in the formation of the models. In each case, the elements of the models are constructed to conform as closely as possible to neuroanatomical structures and behave in a way that has been established or suggested by neurophysiology.2.The dynamic behavior of the mechanics of the extraocular muscles and eyeball suspensory tissues has been established by recording from oculomotoneurons in alert monkeys. The transfer function of this mechanical system is used in these models.3.Recent experiments on the neural circuits in the brain stem that are responsible for saccadic eye movements suggest an arrangement of the premotor circuitry that contains two principal neural networks; an integrator and a pulse generator. This circuitry is used in the models.4.When the above modifications are made to existing models of the saccadic system, they remove the necessity of supposing that the visual information is sampled by the nervous system. The models do not include a sampler although the saccadic pulse generator still makes the overall system behavior similar to that of a sampled-data system.5.The basic model is modified to make its behavior agree with experimental eye movement responses to target ramps and step-ramps. This is done by using error and its rate of change to estimate the error that will exist one reaction time in the future.6.Parallel processing of data is a well recognized property of the nervous system. By utilizing it in combination with a random decision threshold, the model is extended to produce results in agreement with experiments for double-step target movements in which the second step occurs less than 0.2 sec after the first.7.Finally, a model is presented which incorporates a continuum of parallel processing to represent the retinotopic spatial organization of the visual system and the tecto-bulbar motor commands. The model is conceptual; it was not constructed or tested but is used to discuss more complex eye movement phenomena such as those that appear to occur when the decision process must shift between hemispheres and how the system might produce quick correcting saccades with latencies as short as 85 msec.

Recurrent cerebellar architecture solves the motor-error problem

John Porrill — 2008-04-29T17:22:53-00:00

Proceedings of the Royal Society B: Biological Sciences, Vol. 271, No. 1541. (22 April 2004), pp. 789-796.

Current views of cerebellar function have been heavily influenced by the models of Marr and Albus, who suggested that the climbing fibre input to the cerebellum acts as a teaching signal for motor learning. It is commonly assumed that this teaching signal must be motor error (the difference between actual and correct motor command), but this approach requires complex neural structures to estimate unobservable motor error from its observed sensory consequences. We have proposed elsewhere a recurrent decorrelation control architecture in which Marr-Albus models learn without requiring motor error. Here, we prove convergence for this architecture and demonstrate important advantages for the modular control of systems with multiple degrees of freedom. These results are illustrated by modelling adaptive plant compensation for the three-dimensional vestibular ocular reflex. This provides a functional role for recurrent cerebellar connectivity, which may be a generic anatomical feature of projections between regions of cerebral and cerebellar cortex.

The cerebellum and VOR/OKR learning models

Mitsuo Kawato — 2008-04-29T17:15:01-00:00

Trends in Neurosciences, Vol. 15, No. 11. (November 1992), pp. 445-453.

Although one particular model of the cerebellum, as proposed by Marr and Albus, provides a formal framework for understanding how heterosynaptic plasticity of Purkinje cells might be used for motor learning, the physiological details remain largely an enigma. Developments in computational neuroscience and artificial neural networks applied to real control problems are essential to understand fully how work-space errors associated with movement performances can be converted into motor-command errors, and how these errors can then be used as one kind of synaptic input by motor-learning algorithms that are based on biologically plausible rules involving heterosynaptic plasticity. These developments, as well as recent advances in the study of cellular mechanisms of synaptic plasticity, form the basis for the detailed computational models of cerebellar motor learning that have been proposed. These models provide hints toward resolving a long-standing controversy in the oculomotor literature regarding the sites of adaptive changes in the vestibulo-ocular reflex (VOR) and the optokinetic eye movement response (OKR), and suggest new experiments to elucidate general mechanisms of sensory motor learning.

Identifying natural images from human brain activity

Kendrick Kay — 2008-03-06T04:09:09-00:00

Nature (05 March 2008)

Portraits or People? Distinct Representations of Face Identity in the Human Visual Cortex

Gilles Pourtois — 2008-03-02T11:19:43-00:00

J. Cogn. Neurosci., Vol. 17, No. 7. (1 July 2005), pp. 1043-1057.

Humans can identify individual faces under different viewpoints, even after a single encounter. We determined brain regions responsible for processing face identity across view changes after variable delays with several intervening stimuli, using event-related functional magnetic resonance imaging during a long-term repetition priming paradigm. Unfamiliar faces were presented sequentially either in a frontal or three-quarter view. Each face identity was repeated once after an unpredictable lag, with either the same or another viewpoint. Behavioral data showed significant priming in response time, irrespective of view changes. Brain imaging results revealed a reduced response in the lateral occipital and fusiform cortex with face repetition. Bilateral face-selective fusiform areas showed view-sensitive repetition effects, generalizing only from three-quarter to front-views. More medial regions in the left (but not in the right) fusiform showed repetition effects across all types of viewpoint changes. These results reveal that distinct regions within the fusiform cortex hold view-sensitive or view-invariant traces of novel faces, and that face identity is represented in a view-sensitive manner in the functionally defined face-selective areas of both hemispheres. In addition, our finding of a better generalization after exposure to a 3/4-view than to a front-view demonstrates for the first time a neural substrate in the fusiform cortex for the common recognition advantage of three-quarter faces. This pattern provides new insights into the nature of face representation in the human visual system.

Location and spatial profile of category-specific regions in human extrastriate cortex.

M Spiridon — 2007-05-16T19:22:17-00:00

Hum Brain Mapp, Vol. 27, No. 1. (January 2006), pp. 77-89.

Subjects were scanned in a single functional MRI (fMRI) experiment that enabled us to localize cortical regions in each subject in the occipital and temporal lobes that responded significantly in a variety of contrasts: faces>objects, body parts>objects, scenes>objects, objects>scrambled objects, and moving>stationary stimuli. The resulting activation maps were co-registered across subjects using spherical surface coordinates [Fischl et al., Hum Brain Mapp 1999;8:272-284] to produce a "percentage overlap map" indicating the percentage of subjects who showed a significant response for each contrast at each point on the surface. Prominent among the overlapping activations in these contrasts were the fusiform face area (FFA), extrastriate body area (EBA), parahippocampal place area (PPA), lateral occipital complex (LOC), and MT+/V5; only a few other areas responded consistently across subjects in these contrasts. Another analysis showed that the spatial profile of the selective response drops off quite sharply outside the standard borders of the FFA and PPA (less so for the EBA and MT+/V5), indicating that these regions are not simply peaks of very broad selectivities spanning centimeters of cortex, but fairly discrete regions of cortex with distinctive functional profiles. The data also yielded a surprise that challenges our understanding of the function of area MT+: a higher response to body parts than to objects. The anatomical consistency of each of our functionally defined regions across subjects and the spatial sharpness of their activation profiles within subjects highlight the fact that these regions constitute replicable and distinctive landmarks in the functional organization of the human brain.

Face perception is mediated by a distributed cortical network.

A Ishai — 2006-09-09T01:56:37-00:00

Brain Res Bull, Vol. 67, No. 1-2. (30 September 2005), pp. 87-93.

The neural system associated with face perception in the human brain was investigated using functional magnetic resonance imaging (fMRI). In contrast to many studies that focused on discreet face-responsive regions, the objective of the current study was to demonstrate that regardless of stimulus format, emotional valence, or task demands, face perception evokes activation in a distributed cortical network. Subjects viewed various stimuli (line drawings of unfamiliar faces and photographs of unfamiliar, famous, and emotional faces) and their phase scrambled versions. A network of face-responsive regions was identified that included the inferior occipital gyrus, fusiform gyrus, superior temporal sulcus, hippocampus, amygdala, inferior frontal gyrus, and orbitofrontal cortex. Although bilateral activation was found in all regions, the response in the right hemisphere was stronger. This hemispheric asymmetry was manifested by larger and more significant clusters of activation and larger number of subjects who showed the effect. A region of interest analysis revealed that while all face stimuli evoked activation within all regions, viewing famous and emotional faces resulted in larger spatial extents of activation and higher amplitudes of the fMRI signal. These results indicate that a mere percept of a face is sufficient to localize activation within the distributed cortical network that mediates the visual analysis of facial identity and expression.

Reliability of functional localization using fMRI.

KM Swallow — 2008-03-02T11:07:46-00:00

Neuroimage, Vol. 20, No. 3. (November 2003), pp. 1561-1577.

Neuroimaging researchers increasingly take advantage of the known functional properties of brain regions to localize them and probe changes in their activity under different conditions. The utility of this approach depends in part on the reliability of the methods used to define these regions of interest. Two operations may affect the reliability of functionally identified regions: spatially normalizing data to a stereotactic atlas and statistically combining data across participants to form a composite region (as opposed to identifying individual regions for each participant). The effect of these two operations on reliability was evaluated for two functionally identifiable regions: the MT complex and the frontal eye fields. Spatial normalization had almost no effect on within-subject reliability, while grouping across participants negatively affected retest measures of the activation and location of regions defined on separate occasions. We conclude that, for typical sample sizes and numbers of observations per subject, functional localization is most reliable when performed for each individual using data in atlas space.

Separate face and body selectivity on the fusiform gyrus.

RF Schwarzlose — 2008-03-02T11:06:37-00:00

J Neurosci, Vol. 25, No. 47. (23 November 2005), pp. 11055-11059.

Recent reports of a high response to bodies in the fusiform face area (FFA) challenge the idea that the FFA is exclusively selective for face stimuli. We examined this claim by conducting a functional magnetic resonance imaging experiment at both standard (3.125 x 3.125 x 4.0 mm) and high resolution (1.4 x 1.4 x 2.0 mm). In both experiments, regions of interest (ROIs) were defined using data from blocked localizer runs. Within each ROI, we measured the mean peak response to a variety of stimulus types in independent data from a subsequent event-related experiment. Our localizer scans identified a fusiform body area (FBA), a body-selective region reported recently by Peelen and Downing (2005) that is anatomically distinct from the extrastriate body area. The FBA overlapped with and was adjacent to the FFA in all but two participants. Selectivity of the FFA to faces and FBA to bodies was stronger for the high-resolution scans, as expected from the reduction in partial volume effects. When new ROIs were constructed for the high-resolution experiment by omitting the voxels showing overlapping selectivity for both bodies and faces in the localizer scans, the resulting FFA* ROI showed no response above control objects for body stimuli, and the FBA* ROI showed no response above control objects for face stimuli. These results demonstrate strong selectivities in distinct but adjacent regions in the fusiform gyrus for only faces in one region (the FFA*) and only bodies in the other (the FBA*).

fMRI localizer technique: Efficient acquisition and functional properties of single retinotopic positions in the human visual cortex

Antje Kraft — 2008-03-02T10:00:34-00:00

NeuroImage, Vol. 28, No. 2. (1 November 2005), pp. 453-463.

Current fMRI retinotopic mapping procedures often use checkerboard stimuli consisting of expanding rings and rotating wedges to measure the topography within human visual areas. Efficient procedures are well described in the literature. For many experimental paradigms, e.g., visuo-spatial attention paradigms, the identification of task-relevant positions is the only mandatory prerequisite. To define these specific "regions-of-interest" (ROIs), spatially defined localizers are used. A precise evaluation of localizer techniques in regard to efficient scanning time, optimal BOLD (blood oxygenic level dependent) response, as well as quantification of the resulting ROIs within each visual area (size, overlap, surround effects) has not been studied to date. Here, we suggest a mapping procedure designed to quantify spatial and functional properties of single positions at close proximity in multiple human visual areas. During a passive viewing task, various stimuli (e.g., checkerboards or colored objects) subtending 1.4[degree sign] of visual angle were presented at one out of four positions in a randomized block design. We measured the degree of overlap between positions at different hierarchical levels of the visual system (V1-V4v) and quantified modulatory effects on a specific position by stimulation at neighboring (1.7[degree sign] spacing) or distant positions (5.1[degree sign] or 8.5[degree sign] spacing). Within each visual area, "mexican-hat" distributions of local signal intensity changes, which describe a particular combination of facilitatory and suppressive effects, were found. Cubic fitting revealed the most localized tuning effect in V1, which gradually decreased throughout the higher visual areas. Colored objects were most efficient in localizing circumscribed retinotopic positions in both early and higher areas.

A comparison of two computer-based face identification systems with human perceptions of faces.

PJ Hancock — 2008-02-21T03:53:41-00:00

Vision Res, Vol. 38, No. 15-16. (August 1998), pp. 2277-2288.

The performance of two different computer systems for representing faces was compared with human ratings of similarity and distinctiveness, and human memory performance, on a specific set of face images. The systems compared were a graph-matching system (Lades M, Vorbrüggen JC, Buhmann J, Lage J, von der Malsburg C, Würtz RP, Konen W. IEEE., Trans Comput 1993;42:300-311.) and coding based on principal components analysis (PCA) of image pixels (Turk M, Pentland A. J Cognitive Neurosci 1991;3:71-86.). Replicating other work, the PCA-based system produced very much better performance at recognising faces, and higher correlations with human performance with the same images, when the images were initially standardised using a morphing procedure and separate analysis of 'shape' and 'shape-free' components then combined. Both the graph-matching and (shape + shape-free) PCA systems were equally able to recognise faces shown with changed expressions, both provided reasonable correlations with human ratings and memory data, and there were also correlations between the facial similarities recorded by each of the computer models. However, comparisons with human similarity ratings of faces with and without the hair visible, and prediction of memory performance with and without alteration in face expressions, suggested that the graph-matching system was better at capturing aspects of the appearance of the face, while the PCA-based system seemed better at capturing aspects of the appearance of specific images of faces.

Human and automatic face recognition: a comparison across image formats.

AM Burton — 2008-02-21T03:52:19-00:00

Vision Res, Vol. 41, No. 24. (November 2001), pp. 3185-3195.

Human subjects perform poorly at matching different images of unfamiliar faces. When images are taken by different capture devices (cameras), matching is difficult for human perceivers and also for automatic systems. We test an automatic face recognition system based on principal components analysis (PCA) and compare its performance with that of human subjects tested on the same set of images. A number of variants of the PCA system are compared, using different matching metrics and different numbers of components. PCA performance critically depends on the choice of distance metric, with a Mahalanobis metric consistently outperforming a Euclidean metric. Under optimal conditions, the automatic PCA system exceeds human performance on the same images. We hypothesise that unfamiliar face recognition may be mediated by processes corresponding to rather simple functions of the inputs.

The distributed human neural system for face perception.

JV Haxby — 2008-02-21T03:46:33-00:00

Trends Cogn Sci, Vol. 4, No. 6. (June 2000), pp. 223-233.

Face perception, perhaps the most highly developed visual skill in humans, is mediated by a distributed neural system in humans that is comprised of multiple, bilateral regions. We propose a model for the organization of this system that emphasizes a distinction between the representation of invariant and changeable aspects of faces. The representation of invariant aspects of faces underlies the recognition of individuals, whereas the representation of changeable aspects of faces, such as eye gaze, expression, and lip movement, underlies the perception of information that facilitates social communication. The model is also hierarchical insofar as it is divided into a core system and an extended system. The core system is comprised of occipitotemporal regions in extrastriate visual cortex that mediate the visual analysis of faces. In the core system, the representation of invariant aspects is mediated more by the face-responsive region in the fusiform gyrus, whereas the representation of changeable aspects is mediated more by the face-responsive region in the superior temporal sulcus. The extended system is comprised of regions from neural systems for other cognitive functions that can be recruited to act in concert with the regions in the core system to extract meaning from faces.

Face processing in humans is compatible with a simple shape-based model of vision.

M Riesenhuber — 2008-02-20T06:23:35-00:00

Proc Biol Sci, Vol. 271 Suppl 6 (7 December 2004)

Understanding how the human visual system recognizes objects is one of the key challenges in neuroscience. Inspired by a large body of physiological evidence, a general class of recognition models has emerged, which is based on a hierarchical organization of visual processing, with succeeding stages being sensitive to image features of increasing complexity. However, these models appear to be incompatible with some well-known psychophysical results. Prominent among these are experiments investigating recognition impairments caused by vertical inversion of images, especially those of faces. It has been reported that faces that differ 'featurally' are much easier to distinguish when inverted than those that differ 'configurally'; a finding that is difficult to reconcile with the physiological models. Here, we show that after controlling for subjects' expectations, there is no difference between 'featurally' and 'configurally' transformed faces in terms of inversion effect. This result reinforces the plausibility of simple hierarchical models of object representation and recognition in the cortex.

Bayesian decoding of brain images

Karl Friston — 2007-11-21T09:36:40-00:00

NeuroImage, Vol. 39, No. 1. (1 January 2008), pp. 181-205.

This paper introduces a multivariate Bayesian (MVB) scheme to decode or recognise brain states from neuroimages. It resolves the ill-posed many-to-one mapping, from voxel values or data features to a target variable, using a parametric empirical or hierarchical Bayesian model. This model is inverted using standard variational techniques, in this case expectation maximisation, to furnish the model evidence and the conditional density of the model's parameters. This allows one to compare different models or hypotheses about the mapping from functional or structural anatomy to perceptual and behavioural consequences (or their deficits). We frame this approach in terms of decoding measured brain states to predict or classify outcomes using the rhetoric established in pattern classification of neuroimaging data. However, the aim of MVB is not to predict (because the outcomes are known) but to enable inference on different models of structure-function mappings; such as distributed and sparse representations. This allows one to optimise the model itself and produce predictions that outperform standard pattern classification approaches, like support vector machines. Technically, the model inversion and inference uses the same empirical Bayesian procedures developed for ill-posed inverse problems (e.g., source reconstruction in EEG). However, the MVB scheme used here extends this approach to include a greedy search for sparse solutions. It reduces the problem to the same form used in Gaussian process modelling, which affords a generic and efficient scheme for model optimisation and evaluating model evidence. We illustrate MVB using simulated and real data, with a special focus on model comparison; where models can differ in the form of the mapping (i.e., neuronal representation) within one region, or in the (combination of) regions per se.

Robust representations for face recognition: The power of averages

Mike Burton — 2008-02-17T13:29:11-00:00

Cognitive Psychology, Vol. 51, No. 3. (November 2005), pp. 256-284.

We are able to recognise familiar faces easily across large variations in image quality, though our ability to match unfamiliar faces is strikingly poor. Here we ask how the representation of a face changes as we become familiar with it. We use a simple image-averaging technique to derive abstract representations of known faces. Using Principal Components Analysis, we show that computational systems based on these averages consistently outperform systems based on collections of instances. Furthermore, the quality of the average improves as more images are used to derive it. These simulations are carried out with famous faces, over which we had no control of superficial image characteristics. We then present data from three experiments demonstrating that image averaging can also improve recognition by human observers. Finally, we describe how PCA on image averages appears to preserve identity-specific face information, while eliminating non-diagnostic pictorial information. We therefore suggest that this is a good candidate for a robust face representation.

Detecting faces in pure noise images: a functional MRI study on top-down perception.

H Zhang — 2008-02-17T08:01:30-00:00

NeuroReport, Vol. 19(2) (2008), pp. 229-233.

To assess the nature of top-down perceptual processes without contamination from bottom-up input, this functional MRI study investigated face detection in pure noise images. Greater activation was revealed for face versus nonface responses in the fusiform face area, but not in the occipital face area. Across participants, positive correlations were found for the degree of greater face-detection activation between the fusiform face area and bilateral inferior frontal gyri, suggesting a top-down pathway generating perceptual expectations. In contrast, the medial frontal, parietal, supplementary motor, parahippocampal, and striatal areas produced negative correlations between degrees of greater face-detection activation and behavioral responses, suggesting a possible role for these areas in selecting and executing appropriate responses that are based on the top-down expectations.

Expertise in object and face recognition

T Tanaka — 2008-02-16T11:19:03-00:00

(1997)

egorized for the community's nonlinguistic purposes or, to use his term, for the level of ##############. As Brown points out, the level of usual utility changes according to the demands of the linguistic community and this is especially true for expert populations. So, for example, while it is quite acceptable for most of us to refer to the object outside our office window as a "bird," if we were among a group of bird watchers, it would be important to specify whether the object was a...

What Is Special about Face Recognition?: Nineteen Experiments on a Person with Visual Object Agnosia and Dyslexia but Normal Face Recognition

Morris Moscovitch — 2008-02-16T11:00:12-00:00

J. Cogn. Neurosci., Vol. 9, No. 5. (1 September 1997), pp. 555-604.

人による認知特性と整合した顔表情認識のための特徴表現法

Akamatsu — 2008-02-16T08:36:08-00:00

(2003)

Representational capacity of face coding in monkeys.

LF Abbott — 2008-02-16T06:47:21-00:00

Cereb Cortex, Vol. 6, No. 3. (n 1996), pp. 498-505.

We examine the distributed nature of the neural code for faces represented by the firing of visual neurons in the superior temporal sulcus of monkeys. Both information theory and neural decoding techniques are applied to determine how the capacity to represent faces depends on the number of coding neurons. Using a combination of experimental data and Monte Carlo simulations, we show that the information grows linearly and the capacity to encode stimuli grows exponentially with the number of neurons. By decoding firing rates, we determine that the responses of the 14 recorded neurons can distinguish between 20 face stimuli with approximately 80% accuracy. In general, we find that N neurons of this type can encode approximately 3(2(04N)) different faces with 50% discrimination accuracy. These results indicate that the neural code for faces is highly distributed and capable of accurately representing large numbers of stimuli.

The development of the social brain in human infancy

Grossmann — 2007-02-28T13:07:32-00:00

European Journal of Neuroscience, Vol. 25, No. 4. (February 2007), pp. 909-919.

The functions of the orbitofrontal cortex.

ET Rolls — 2005-06-27T16:19:56-00:00

Brain Cogn, Vol. 55, No. 1. (June 2004), pp. 11-29.

The orbitofrontal cortex contains the secondary taste cortex, in which the reward value of taste is represented. It also contains the secondary and tertiary olfactory cortical areas, in which information about the identity and also about the reward value of odours is represented. The orbitofrontal cortex also receives information about the sight of objects from the temporal lobe cortical visual areas, and neurons in it learn and reverse the visual stimulus to which they respond when the association of the visual stimulus with a primary reinforcing stimulus (such as taste) is reversed. This is an example of stimulus-reinforcement association learning, and is a type of stimulus-stimulus association learning. More generally, the stimulus might be a visual or olfactory stimulus, and the primary (unlearned) positive or negative reinforcer a taste or touch. A somatosensory input is revealed by neurons that respond to the texture of food in the mouth, including a population that responds to the mouth feel of fat. In complementary neuroimaging studies in humans, it is being found that areas of the orbitofrontal cortex are activated by pleasant touch, by painful touch, by taste, by smell, and by more abstract reinforcers such as winning or losing money. Damage to the orbitofrontal cortex can impair the learning and reversal of stimulus-reinforcement associations, and thus the correction of behavioural responses when there are no longer appropriate because previous reinforcement contingencies change. The information which reaches the orbitofrontal cortex for these functions includes information about faces, and damage to the orbitofrontal cortex can impair face (and voice) expression identification. This evidence thus shows that the orbitofrontal cortex is involved in decoding and representing some primary reinforcers such as taste and touch; in learning and reversing associations of visual and other stimuli to these primary reinforcers; and in controlling and correcting reward-related and punishment-related behavior, and thus in emotion. The approach described here is aimed at providing a fundamental understanding of how the orbitofrontal cortex actually functions, and thus in how it is involved in motivational behavior such as feeding and drinking, in emotional behavior, and in social behavior.

The fusiform face area is tuned for curvilinear patterns with more high-contrasted elements in the upper part

Roberto Caldara — 2007-03-02T20:54:18-00:00

NeuroImage, Vol. 31, No. 1. (15 May 2006), pp. 313-319.

The ability to identify conspecifics from the face is of primary interest for human social behavior. Newborns' visual preference for schematic face-like stimuli has been recently related to a sensitivity for visual patterns with a greater number of elements in the upper compared to the lower part. At the adult level, neuroimaging studies have identified a network of cortical areas devoted to the detection and identification of faces. However, whether and how low-level structural properties of face stimuli contribute to the preferential response to faces in these areas remain to be clarified. Using functional magnetic resonance imaging (fMRI), here we investigated whether the adults' face-sensitive cortical areas show a preference for top-heavy patterns, similarly to newborns' preference. Twelve participants were presented with head-shaped and square patterns with either more elements in the upper or the lower vertical part. In the right fusiform gyrus (`fusiform face area', FFA), an area showing a preference for faces over other visual object categories, there was a larger activation for curvilinear patterns with more high-contrast elements in the upper part, even though these patterns were not perceived as face stimuli. These findings provide direct evidence that the FFA is tuned for geometrical properties fitting best with the structure of faces, a computational mechanism that might drive the automatic detection of faces in the visual world.

Facial expression decoding in early Parkinson's disease.

MD Pell — 2008-02-16T06:03:26-00:00

Brain Res Cogn Brain Res, Vol. 23, No. 2-3. (May 2005), pp. 327-340.

The ability to derive emotional and non-emotional information from unfamiliar, static faces was evaluated in 21 adults with idiopathic Parkinson's disease (PD) and 21 healthy control subjects. Participants' sensitivity to emotional expressions was comprehensively assessed in tasks of discrimination, identification, and rating of five basic emotions: happiness, (pleasant) surprise, anger, disgust, and sadness. Subjects also discriminated and identified faces according to underlying phonemic ("facial speech") cues and completed a neuropsychological test battery. Results uncovered limited evidence that the processing of emotional faces differed between the two groups in our various conditions, adding to recent arguments that these skills are frequently intact in non-demented adults with PD [R. Adolphs, R. Schul, D. Tranel, Intact recognition of facial emotion in Parkinson's disease, Neuropsychology 12 (1998) 253-258]. Patients could also accurately interpret facial speech cues and discriminate the identity of unfamiliar faces in a normal manner. There were some indications that basal ganglia pathology in PD contributed to selective difficulties recognizing facial expressions of disgust, consistent with a growing literature on this topic. Collectively, findings argue that abnormalities for face processing are not a consistent or generalized feature of medicated adults with mild-moderate PD, prompting discussion of issues that may be contributing to heterogeneity within this literature. Our results imply a more limited role for the basal ganglia in the processing of emotion from static faces relative to speech prosody, for which the same PD patients exhibited pronounced deficits in a parallel set of tasks [M.D. Pell, C. Leonard, Processing emotional tone from speech in Parkinson's disease: a role for the basal ganglia, Cogn. Affect. Behav. Neurosci. 3 (2003) 275-288]. These diverging patterns allow for the possibility that basal ganglia mechanisms are more engaged by temporally-encoded social information derived from cue sequences over time.

Is Face Processing Special?

Martin Tovee — 2008-02-16T05:57:07-00:00

Neuron, Vol. 21, No. 6. (December 1998), pp. 1239-1242.

Computational Modeling of Face Recognition Based on Psychophysical Experiments

Adrian Schwaninger — 2008-02-16T05:53:24-00:00

Swiss Journal of Psychology, Vol. 63, No. 3. (September 2004), pp. 207-215.

Recent results from psychophysical studies are discussed which clearly show that face processing is not only holistic. Humans do encode face parts (component information) in addition to information about the spatial interrelationship of facial features (global configural information). Based on these findings we propose a computational architecture of face recognition, which implements a component and configural route for encoding and recognizing faces. Modeling results showed a striking similarity between human psychophysical data and the computational model. In addition, we could show that our framework is able to achieve good recognition performance even under large view rotations. Thus, our study is an example of how an interdisciplinary approach can provide a deeper understanding of cognitive processes and lead to further insights in human psychophysics as well as computer vision.

Processing faces and facial expressions.

MT Posamentier — 2008-02-16T05:41:51-00:00

Neuropsychol Rev, Vol. 13, No. 3. (September 2003), pp. 113-143.

This paper reviews processing of facial identity and expressions. The issue of independence of these two systems for these tasks has been addressed from different approaches over the past 25 years. More recently, neuroimaging techniques have provided researchers with new tools to investigate how facial information is processed in the brain. First, findings from "traditional" approaches to identity and expression processing are summarized. The review then covers findings from neuroimaging studies on face perception, recognition, and encoding. Processing of the basic facial expressions is detailed in light of behavioral and neuroimaging data. Whereas data from experimental and neuropsychological studies support the existence of two systems, the neuroimaging literature yields a less clear picture because it shows considerable overlap in activation patterns in response to the different face-processing tasks. Further, activation patterns in response to facial expressions support the notion of involved neural substrates for processing different facial expressions.

Are you always on my mind? A review of how face perception and attention interact.

R Palermo — 2008-02-16T05:37:57-00:00

Neuropsychologia, Vol. 45, No. 1. (7 January 2007), pp. 75-92.

In this review we examine how attention is involved in detecting faces, recognizing facial identity and registering and discriminating between facial expressions of emotion. The first section examines whether these aspects of face perception are "automatic", in that they are especially rapid, non-conscious, mandatory and capacity-free. The second section discusses whether limited-capacity selective attention mechanisms are preferentially recruited by faces and facial expressions. Evidence from behavioral, neuropsychological, neuroimaging and psychophysiological studies from humans and single-unit recordings from primates is examined and the neural systems involved in processing faces, emotion and attention are highlighted. Avenues for further research are identified.

Processing of facial identity and expression: a psychophysical, physiological, and computational perspective.

A Schwaninger — 2008-02-16T05:33:39-00:00

Prog Brain Res, Vol. 156 (2006), pp. 321-343.

A deeper understanding of how the brain processes visual information can be obtained by comparing results from complementary fields such as psychophysics, physiology, and computer science. In this chapter, empirical findings are reviewed with regard to the proposed mechanisms and representations for processing identity and emotion in faces. Results from psychophysics clearly show that faces are processed by analyzing component information (eyes, nose, mouth, etc.) and their spatial relationship (configural information). Results from neuroscience indicate separate neural systems for recognition of identity and facial expression. Computer science offers a deeper understanding of the required algorithms and representations, and provides computational modeling of psychological and physiological accounts. An interdisciplinary approach taking these different perspectives into account provides a promising basis for better understanding and modeling of how the human brain processes visual information for recognition of identity and emotion in faces.

Behavioural and neurophysiological evidence for face identity and face emotion processing in animals.

AJ Tate — 2008-02-16T05:28:41-00:00

Philos Trans R Soc Lond B Biol Sci, Vol. 361, No. 1476. (29 December 2006), pp. 2155-2172.

Visual cues from faces provide important social information relating to individual identity, sexual attraction and emotional state. Behavioural and neurophysiological studies on both monkeys and sheep have shown that specialized skills and neural systems for processing these complex cues to guide behaviour have evolved in a number of mammals and are not present exclusively in humans. Indeed, there are remarkable similarities in the ways that faces are processed by the brain in humans and other mammalian species. While human studies with brain imaging and gross neurophysiological recording approaches have revealed global aspects of the face-processing network, they cannot investigate how information is encoded by specific neural networks. Single neuron electrophysiological recording approaches in both monkeys and sheep have, however, provided some insights into the neural encoding principles involved and, particularly, the presence of a remarkable degree of high-level encoding even at the level of a specific face. Recent developments that allow simultaneous recordings to be made from many hundreds of individual neurons are also beginning to reveal evidence for global aspects of a population-based code. This review will summarize what we have learned so far from these animal-based studies about the way the mammalian brain processes the faces and the emotions they can communicate, as well as associated capacities such as how identity and emotion cues are dissociated and how face imagery might be generated. It will also try to highlight what questions and advances in knowledge still challenge us in order to provide a complete understanding of just how brain networks perform this complex and important social recognition task.

Understanding the recognition of facial identity and facial expression

Andrew Calder — 2005-08-01T21:26:44-00:00

Nature Reviews Neuroscience, Vol. 6, No. 8. (01 August 2005), pp. 641-651.

Population Dynamics of Face-responsive Neurons in the Inferior Temporal Cortex

Narihisa Matsumoto — 2005-07-15T12:58:10-00:00

Cerebral Cortex, Vol. 15, No. 8. (August 2005), pp. 1103-1112.

Subspace projection approaches to classification and visualization of neural network-level encoding patterns.

R Oşan — 2008-02-15T12:26:55-00:00

PLoS ONE, Vol. 2, No. 5. (2007)

Recent advances in large-scale ensemble recordings allow monitoring of activity patterns of several hundreds of neurons in freely behaving animals. The emergence of such high-dimensional datasets poses challenges for the identification and analysis of dynamical network patterns. While several types of multivariate statistical methods have been used for integrating responses from multiple neurons, their effectiveness in pattern classification and predictive power has not been compared in a direct and systematic manner. Here we systematically employed a series of projection methods, such as Multiple Discriminant Analysis (MDA), Principal Components Analysis (PCA) and Artificial Neural Networks (ANN), and compared them with non-projection multivariate statistical methods such as Multivariate Gaussian Distributions (MGD). Our analyses of hippocampal data recorded during episodic memory events and cortical data simulated during face perception or arm movements illustrate how low-dimensional encoding subspaces can reveal the existence of network-level ensemble representations. We show how the use of regularization methods can prevent these statistical methods from over-fitting of training data sets when the trial numbers are much smaller than the number of recorded units. Moreover, we investigated the extent to which the computations implemented by the projection methods reflect the underlying hierarchical properties of the neural populations. Based on their ability to extract the essential features for pattern classification, we conclude that the typical performance ranking of these methods on under-sampled neural data of large dimension is MDA>PCA>ANN>MGD.

Face recognition: A literature survey

W Zhao — 2005-10-17T22:28:34-00:00

ACM Comput. Surv., Vol. 35, No. 4. (December 2003), pp. 399-458.

Psychological and Neural Perspectives on Human Face Recognition

Alice O’toole — 2008-02-15T12:07:07-00:00

Handbook of Face Recognition (2005), pp. 349-369.

Face recognition by independent component analysis.

MS Bartlett — 2008-02-15T12:00:40-00:00

IEEE Trans Neural Netw, Vol. 13, No. 6. (2002), pp. 1450-1464.

A number of current face recognition algorithms use face representations found by unsupervised statistical methods. Typically these methods find a set of basis images and represent faces as a linear combination of those images. Principal component analysis (PCA) is a popular example of such methods. The basis images found by PCA depend only on pairwise relationships between pixels in the image database. In a task such as face recognition, in which important information may be contained in the high-order relationships among pixels, it seems reasonable to expect that better basis images may be found by methods sensitive to these high-order statistics. Independent component analysis (ICA), a generalization of PCA, is one such method. We used a version of ICA derived from the principle of optimal information transfer through sigmoidal neurons. ICA was performed on face images in the FERET database under two different architectures, one which treated the images as random variables and the pixels as outcomes, and a second which treated the pixels as random variables and the images as outcomes. The first architecture found spatially local basis images for the faces. The second architecture produced a factorial face code. Both ICA representations were superior to representations based on PCA for recognizing faces across days and changes in expression. A classifier that combined the two ICA representations gave the best performance.

Study design in fMRI: Basic principles.

Edson Amaro — 2006-02-22T14:35:50-00:00

Brain Cogn (18 January 2006)

There is a wide range of functional magnetic resonance imaging (fMRI) study designs available for the neuroscientist who wants to investigate cognition. In this manuscript we review some aspects of fMRI study design, including cognitive comparison strategies (factorial, parametric designs), and stimulus presentation possibilities (block, event-related, rapid event-related, mixed, and self-driven experiment designs) along with technical aspects, such as limitations of signal to noise ratio, spatial, and temporal resolution. We also discuss methods to deal with cases where scanning parameters become the limiting factor (parallel acquisitions, variable jittered designs, scanner acoustic noise strategies).

Event-related fMRI contrast when using constant interstimulus interval: theory and experiment.

PA Bandettini — 2006-12-08T12:53:03-00:00

Magn Reson Med, Vol. 43, No. 4. (April 2000), pp. 540-548.

Event-related functional magnetic resonance imaging (ER-fMRI) involves the mapping of averaged hemodynamic changes resulting from repeated, brief (<3 sec) brain activation episodes. In this paper, two issues regarding constant-interstimulus interval ER-fMRI were addressed. First, the optimal interstimulus interval (ISI), given a stimulus duration (SD), was determined. Second, the statistical power of ER-fMRI relative to that of a blocked-design paradigm was determined. Experimentally, it was found that with a 2-sec SD, the optimal ISI is 12 to 14 sec. Theoretically, the optimal repetition interval (T(opt) = ISI + SD) is 12 to 14 sec for stimuli of 2 sec or less. For longer stimuli, T(opt) is 8 + 2 x SD. At the optimal ISI for SD = 2 sec, the experimentally determined functional contrast of ER-fMRI was only -35% lower than that of blocked-design fMRI. Simulations that assumed a linear system demonstrated an event-related functional contrast that was -65% lower than that of the blocked design. These differences between simulated and experimental contrast suggest that the ER-fMRI amplitude is greater than that predicted by a linear shift-invariant system.

Detection versus estimation in event-related fMRI: choosing the optimal stimulus timing.

RM Birn — 2006-12-08T12:57:20-00:00

Neuroimage, Vol. 15, No. 1. (January 2002), pp. 252-264.

With the advent of event-related paradigms in functional MRI, there has been interest in finding the optimal stimulus timing, especially when the interstimulus interval is varied during the imaging run. Previous works have proposed stimulus timings to optimize either the estimation of the impulse response function (IRF) or the detection of signal changes. The purpose of this paper is to clarify that estimation and detection are fundamentally different goals and to determine the optimal stimulus timing and distribution with respect to both the accuracy of estimating the IRF and the power of detection assuming a particular hemodynamic model. Simulated stimulus distributions are varied systematically, from traditional blocked designs to rapidly varying event related designs. These simulations indicate that estimation of the hemodynamic impulse response function is optimized when stimuli are frequently alternated between task and control states, with shorter interstimulus intervals and stimulus durations, whereas the detection of activated areas is optimized by blocked designs. The stimulus timing for a given experiment should therefore be generated with the required detectability and estimation accuracy.

Event-related functional magnetic resonance imaging: modelling, inference and optimization.

O Josephs — 2008-02-13T22:40:12-00:00

Philos Trans R Soc Lond B Biol Sci, Vol. 354, No. 1387. (29 July 1999), pp. 1215-1228.

Event-related functional magnetic resonance imaging is a recent and popular technique for detecting haemodynamic responses to brief stimuli or events. However, the design of event-related experiments requires careful consideration of numerous issues of measurement, modelling and inference. Here we review these issues, with particular emphasis on the use of basis functions within a general linear modelling framework to model and make inferences about the haemodynamic response. With these models in mind, we then consider how the properties of functional magnetic resonance imaging data determine the optimal experimental design for a specific hypothesis, in terms of stimulus ordering and interstimulus interval. Finally, we illustrate various event-related models with examples from recent studies.