Functional Neuroimaging of Place Learning
in Computer Generated Space
Principal Investigator: W. Jake Jacobs, 626-4825, wjj@u.arizona.edu
Designated Project Coordinator: Ming Hsu, mhsu@u.arizona.edu
Click on Image for Full Map
Directions
[Overview][Procedure][Behavioral data][MRI Data][Cluster Analysis][Discussion][Refereces]


Overview
According to the model of memory we have been developing, the hippocampus may be centrally involved in encoding spatiotemporal aspects of memory. We have been using a Computer-Generated Arena (C-G Arena) task patterned after the Morris Water Maze (MWM) to test predictions made by cognitive mapping theory.

One of our earlier studies with the C-G Arena indicated that humans are capable of place learning through observation alone.  In the experiment, participants who received observational learning trials showed more direct movement to an invisible target than those participants not receiving such trials.

The goal of this project is, therefore, two-fold. The first, and immediat7e goal is to employ brain imaging techniques in order to understand the neuroanatomical correlates of learning and memory performance, in this case observational learning. The second, more distant goal is to find ways to enhance learning in memory-impaired patients by taking advantage of memory processes or cognitive systems that might be spared in such patients. We will evaluate the utility of functional magnetic resonance imaging (fMRI) as a tool for mapping the changes in patients’ learning strategies (and, presumably, brain function) that occur through the course of cognitive rehabilitation.


 
Procedure

Stage 1: Each participant was prepared and packed into the magnet with a pair of 3D goggles covering their eyes. Visual stimuli were fed through the goggles.  The participants viewed:


(Click to Enlarge)

a) 4 trials (30sec each) of successful searches for an invisible target. Screenshot
b) 4 trials (30sec each) of successful searches for a visible target. Screenshot
c) 8 trials (30sec each) of pseudo-random color patterns resembling a kaleidoscope like patterns.
note: kaleidoscope patterns are created through the shareware screensaver Kaleidoscope by Syntrillium. Download to see what the patterns look like.

Stage 2: Upon completion of the visual stimuli recording, participants were taken out of the magnet and given a layout confidence questionnaire. 

Stage 3: They were then told to find the same target in the computer program, the participants were given:

  1. 6 searches for an invisible target.
  2. 6 searches for a visible target.
  3. 1 probe trial (in which the target was, unbeknownest to the participant, absent.)
Stage 4: Participants were then administered the Arena Reconstitution Task (ART) Example 50K. They were told to put reconstruct the arena using jigsaw puzzle-like pieces which resembled the walls and objects of the arena.  (Finished Example 56K)

Stage 5: Participants were given a spatial knowledge questionnaire and a target location questionnaire.

Stage 6: Participants were given a full debriefing and dismissed.

Screenshot 50K
A shortened two-trial version of the recording
 
 


 
Data Analyses: Behavioral
Click for full graph Latency (participant's time to target) of naive participants.  The latency is used as a measure of the participant's knowledge of the target. We took this measure after the subjects have seen the recording in the MRI, therefore we use it as a measure of the extent of "observational learning" that the participant achieved.  Although there are 6 trials in this task, it is the first three that are of particular interest, since they reflect the subject's knowledge learned from the recording while they were in the arena. We are, therefore, using performance in this task as a measure of learning in the observational stage.
  • Good performers: MF, DS
  • Moderately good performers: TL, JJ
  • Poor performers: AP, DC, AH
  This is the latency graph of the experienced participants. There are only 2 plots at the time, one of the experienced participant's data remain to be graphed.  Unlike the naive group, the experienced particpants were specially trained on the specific arena prior to the scanning. In fact, they were all from this very lab, so it is little surprise that their latency is so low.
  • Good performers: KT, MH, and CB
  • Moderately good performers: none
  • Poor performers: none

 
Data Analyses: Perception Models
Correlational analysis in MRI data was done under the cognitive subtraction model.  In the analysis, the experimental stage was subtracted from the control stage.  Data analysis was performed on a PC running Linux 2.2.x using Afni, developed by Dr. Robert Cox of the University of Wisconsin.
This, the first of the data analysis models, compared Experimental Condition 1 against the control--invisible against kaleidoscope.  It is designed to detect areas of the brain activated during perception and visual integration.

Summary:

  • Regions consistently activated (i.e., in every subject) in these models are: Parietal lobes, bilateral; Occipital lobes, bilateral; Frontal Eye Fields; cerebellum.
  • Regions sporadically activated (i.e., in around 1/2 of subjects) are: Hippocampal Formation.
Models:
  1. inv_kal: invisible - kaleidoscopesubtracted kaleidoscope trials from trials with an invisible target. (Probably a better model than model 2).
  2. inv_all: invisible - (visible + kaleidoscope) subtracted trials with a visible target and the kaleidoscope trials from trials with an invisible target.
  3. vis_kal: visible - kaleidoscope subtracted kaleidoscope trials from trials with a visible target. (Probably a better model than model 4).

  4. vis_all: visible - (invisible + kaleidoscope)subtracted trials with an invisible target and the kaleidoscope trials from trials with a visible target.
inv_kal   Naïve         Experienced
Location X MF AP AH JJ CM RD KT MH CB
1. Frontopolar Left + +   -          
  Right + +   -          
2. Frontal  Left   + + + +        
  Right + + + + +       +
3. Precentral  Left + + + + +   + + +
  Right + + + + +   +    
4. Postcentral Left                  
  Right                  
5. Parietal Left + + + + + + + + +
  Right + + + + + + + + +
6. Temporal Left +     + +        
  Right   + + +     +    
7. MT Left + + + +          
  Right   + + + + + +    
8. Occipital Left - + + +       + +
  Right - + + +     + + +
9. Cerebellum Left + +   + + + +   +
  Right + +   + +     +  
Summary: Activity centers on precentral gyrus, intraparietal sulcus, occipital gyri, and cerebellum

Activation Curves: this shows what happens inside those activations within their scan time.  Here are some selected graphs from subjects

Note: It may take me sometime to put up the links here, but the graphs themselves are usually uploaded to the website each working day. If you don't mind waddling through some strange looking names, here are the graphs for the inv_kal and vis_kal models.  They are labeled by subject initials and location of activation.

 
Data Analysis: Learning & Memory Models
The following models are intended to discover the neuronal substraits that underly spatial learning (easier said than done). 
 
2inv_2inv   Naïve         Experienced
Location X MF AP AH JJ CM RD KT MH CB
1. Frontopolar Left + + + + +     + +  
  Right + + +   +     + +  
2. Frontal  Left + +                
  Right + +   +            
3. Precentral  Left + +     +          
  Right   +     + +       +
4. Postcentral Left                    
  Right               +    
5. Parietal Left + +   +       + +  
  Right + + + +            
6. Temporal Left + +   + +          
  Right + + +   +     + + +
7. MT Left + +                
  Right + +           + +  
8. Occipital Left +     +         +  
  Right +     + +     + + +
9. Cerebellum Left + +   + + +   + + +
  Right + + + +   +   + + +
Summary: Activity centers on right temporal lobe and frontopolar areas.

 
Preliminary Discussion

Hippocampus

Although hippocampal activation has been spotted only sporadically, we should not conclude that the hippocampus is inactive or in any way “dormant” during the experiment. Evidence from non-human primates suggest that hippocampus in these non-human primates fire intensely for but a brief interval of time, when it is scanning the environment, either novel or experienced (Feigenbaum & Rolls, 1991) although there are conflicting reports. Studies of the hippocampus in rats found neurons in the rat hippocampus fire more strongly when the rat determined its own movement than when it is being passively guided by a human hand (Feigenbaum & Rolls, 1991). Goldman-Rakic et al. Speculated this to be one possible reason for the lack of correlation between the rat hippocampal neuronal recording studies and those of the monkeys—that the monkey studies employed primarily passive behavioral tasks, whereas rats were used in such tasks as the radial maze and the MWM (Goldman-Rakic, et al. 1989).

Even more so than the monkey studies, behavioral tasks in neuroimaging consists of primarily passive tasks. Participants at most push buttons to indicate a decision on a particular task. In perhaps the only spatial memory and navigation studies in which the participant himself controlled navigation via manipulation of a joystick, Aguirre 1996 detected significant activation in the hippocampal formation. But as even Aguirre himself admitted, the free navigation of space resulted in such complex data that is difficult to analyze in detail (Aguirre 1996). This is but one possible explanation for the scarcity of hippocampal activity detected in this experiment. Numerous other possibilities exist, from the low spatial resolution (relative to single cell recordings) of fMRI to the imprecision of the “cognitive subtraction” models. Aguirre perhaps put it best when the author remarked the hippocampus to have been “a recalcitrant target of neuroimaging” (Aguirre 1996).

Rolls et al noted that during passive viewing, approximately 10% of the “space” cells in the hippocampus fire. In addition, the peak-firing rate is low, and the topographical dispersion is sparse (Rolls et al 1998). It is, therefore, dangerous to draw conclusions about cognitive mapping and spatial functions in the hippocampus in neuroimaging studies. Indeed, most of the activity in the hippocampal formation has been found in the parahippocampus, which has many functions besides spatial cognition. The parahippocampus serves as a connection between the hippocampus and a number of cerebral cortex structures. This, along with the other cognitive functions the hippocampus subserves, could mean that activation’s in the hippocampal formation in neuroimaging studies do not reflect spatial processing or mapping at all but rather a neural pathway or other cognitive functions that occur during spatial cognition, much to the same extent that the hippocampal formation appear activated in verbal non-spatial tasks (e.g. Fernandez et al. 1998). 

Parietal Lobe

With or without detected hippocampal activity, the extensive activation of the parietal lobe is of great interest in itself. The parietal cortex has been long regarded as a center for sensorimotor integration. The exact nature of this integration, as well as the possibilities of other cognitive functions, however, remains fuzzy.
One influential theory regarding the functions of the parietal lobe is the “What-where” theory introduced by Mishkin and Ungleider in 1985. In this theory, the parietal areas serve as the dorsal “where” stream, and the occipital cortex is the location of the ventral “what” stream. These two streams then pass into the temporal lobes and into the hippocampus. The current results are within the predictions made by the what-where theory. As our experiment was not designed to parcel out the dissociation between the two systems—both object and spatial property are most intense during the same period of time (when the video is panning)—it is impossible to pinpoint which is “what” and which is “where.”

This system would explain the extensive activations in the parietal and occipital areas, but it would still be missing, although not completely, how. Recent recordings from monkey, however, may shed light on the process. Theories about parietal functions have ranged from Mountcastle’s generation of motor command, to visual attention, to motor planning, and the generation of visual maps (Colby et al. 1995). Andersen et al. 1990b first reported evidence of “dynamic maps” among parietal neurons. Much like hippocampal maps, these maps contain relationship and distance information. Unlike hippocampal maps, however, maps in the parietal structure holds integrity only when the monkey makes a saccade, and whatever images and relations are enveloped into the fovea by the saccade are mapped onto the parietal neurons. With each new saccade, a new map among the neurons is formed.

Preliminary results from cluster analysis support the foveal maps proposed by Andersen et al. During presentations of invisible trials, MR signals in the parietal lobe, in the same region as the lateral intraparietal area identified in Andersen et al, showed peak activity during the last 10-15 sec of the trials, when the video pans 360º, and provides, for the first time in that particular trial, a full scope of the arena, whereas during the first 10-15 sec, they were shown a relatively straight path to the target. Therefore, the amount of information that subjects must encode is inordinately greater in the latter half of the trial. 

ET Rolls in 1996 introduced a similar idea. Rolls proposed that the spatial computation center was located in the neocortex, perhaps in the parietal area. And because of the highly developed primate visual systems and of the neocortex, the hippocampus is required only to store or retrieve certain information in the performance of certain tasks. According to Rolls’s theory, the hippocampus is operating as an “on-the-fly” memory storage area, which, because of the convergence of the numerous pathways from the different regions of the brain, the neural signals are combined, whereupon later, they are retrieved by the neocortex, most likely the parietal cortex, and gradually “binded” into richer, semantic representations. (Rolls 1996)  This would account for the paucity of hippocampal activity in neuroimaging, as hippocampal activity by this theory would be transient and dispersed. There is, however, to our knowledge, no substantial data supporting the “binding” and backpropagation of information back to the neocortex. 

This, of course, does not preclude the theories of motor command and planning. The activity in the parietal areas may simply be a function of parietal projections into the cerebellum and the motor cortex, since why else would the cerebellum be consistently activated during both invisible and visible stages? Cerebellar activity, however, may not be related to motor planning at all. Indeed, cerebellar activity has appeared in numerous neuroimaging studies, including some that seemingly involves neither motor movement nor planning, such as verbal working memory.

Cerebellum

Indeed, there has been many experiments showing cerebellar activity during cognitive tasks that seemingly do not, and should not, involved motor activity. This has led to some researchers to speculate on the possibilities of the cerebellar involvement in a cognitive capacity. Experiments that been run to show cerebellar activation by subtracting MR signals of certain cognitive tasks from the baseline control, whereas subtraction of the control experiment condition from the baseline control yielded no activation or less activation (Ackermann et al. 1998).

Single cell recordings from subhuman primates and cats, however, offer a simpler, and perhaps more parsimonious explanation. Kakei et al. 1997 has shown two different areas of the parietal cortex that receive cerebellar outputs. They speculate that the one of the functions of the projection may be to acts as a filter that may either enhance or suppress feedback signals from ongoing movement (Kakei et al. 1997). This suppression/enhancement could also include control of saccades (e.g. Keele & Ivry 1990, Ron & Robinson 1973), that has been demonstrated to be within the domain of parietal functions (e.g., Andersen 1995, Heide & Kompf 1997).  Therefore, the cerebellar activations seen in neuroimaging studies may involve the controlling of saccadic movements via the parietal cortex, but it need not be involved in any explicit “cognitive processing.” But there exists numerous projections between the parietal lobe and the cerebellum, as well as between parietal, motor, and cerebellum, and, as the nonhuman primate literature demonstrates, a number, if not most, of these cortical connections are bi-directional (Schmahmann & Pandya, 1997). Therefore, even if the cerebellum activations in these cognitive tasks are not related to saccadic control, it may be due to any of projections from the aforementioned connection. Thus, cerebellar activity during purely cognitive tasks need not be interpreted as a sign of cerebellar involvement in cognitive processing.

Another theory, by Keele & Ivry, suggests that part of cerebellum functions as a “temporal computational clock,” responsible for motor and perceptual timing. Therefore, it may be also that the cerebellar activity observed here has more to do with perceptual timing than motor planning or other functions.
A most interesting idea was proposed by JM Bower. In it, Bower proposed that the cerebellar activity in neuroimaging studies often do not reflect cerebellar contributions to “cognition,” or “attention,” or “perception” per se, but rather the cerebellum serves as a facilitator for these processes, much like its well-established facilitation of fine motor movements. In sardonic wit, Bower dismissed conclusions made in previous neuroimaging experiments as being simple-minded and methodologically unsound with the observation that no one has ever thought of the fovea as contribution to visual processing (Bower 1997).

Temporal Lobes

The temporal cortex has been implicated in visual learning and memory in a number of studies across various methodologies, including functional neuroimaging and lesion studies in humans (e.g., Vandenberghe et al., 1995), single cell recording in primates and rats (e.g., Nakamura 1996, Li et al., 1993). The temporal cortex has been theorized to be the endpoint of the visual pathways, and as such, subserves much of what we call learning and memory. The temporal cortex, especially the inferior temporal cortex, has been implicated object learning. In a study on novelty effects of stimuli, it was found that a novel object acting as a stimulus activates the inferior temporal cortex more than does the same object once it becomes familiar (Li et al., 1993).

The dissociation between the temporal and parietal activations is of particular interest. In the “perception models,” only the dorsal and ventral areas in the parietal and occipital cortices were seen activated; the temporal areas, except a few areas near the hippocampal formation, remained largely devoid of activity. In the “learning models,” very little of the dorsal and ventral systems were seen activated; but the temporal areas were consistently and extensively activated. In line with much of the existing neuroimaging literature, temporal lobe activity appears to be associated with familiarity of stimulus.

Of even more interest is the dissociation between the invisible and the visible trials. Whereas the “perception models” show no significant differences between the invisible and the visible trials, activations in the frontal and the temporal areas do reveal some differences. This is perhaps an indication of the attention or learning. Although the perceptual stage of spatial cognition does not differentiate between the invisible and visible trials, since they are relatively similar, the encoding and attention stages may differentiate between the two conditions. Whereas the invisible trials contained a target in a fix relationship with its surroundings, the visible trials contained a target in different relationships to its surroundings. Therefore, spatial relationships in the visible trials will need to be reassessed with each trial, whereas in the invisible trials do not require further encoding once the spatial relationships have been learned.

In addition, whereas the intensity of the activation in the parietal cortex is relatively similar between experienced and naive subjects, it shows great disparities in the learning models. Temporal cortex activity is extensive in naive subjects, but the intensity is much lower than activations in experienced subjects, which otherwise have fewer voxels of activation. This may be the result of the temporal lobe’s role in retrieval of memory. Whereas the experienced subjects have much more localized and efficient areas of encoding and retrieval, the naive subjects, have more extensive but less efficient areas. Therefore, it is possible that efficient encoding and retrieval may be the difference between the two groups.

Fronto-Parieto-Temporal Network

That the frontal, along with the parietal and temporal areas of the brain constitute a network within the brain for spatial memory and attention has been a staple in the neuroimaging literature in recent years (e.g., Gitelman et al. 1999, Vallar et al., 1999, Corbetta et al., 1998). Gitelman et al summarized this network particularly well in stating that: 

(S)patial representation is coordinated by a network of interconnected cortical areas, the epicentres of which are located in cingulate gyrus…  The frontal eye field has a relative specialization for the overt motor-exploratory aspect of spatial attention, the posterior parietal cortex for the sensory representational and sensorimotor aspect, and the cingulate gyrus for the limbic-motivational aspect.
If there be such a network, this study shows the close temporal relationship between the precentral and parietal areas. The precentral and parietal areas show extremely similar activation patterns—both areas are activated most when there is a most amount of stimuli to be processed. The temporal cortex, however, does not appear to share in this similarity. Perhaps the neuronal activity ceases after the first few trials, but this conclusion is compromised by the absence of temporal activity in visible trials. Further research clearly is needed to discern the temporal cortex’s role in the network, as well as the network as a whole.
 

 
References
  • • Ackermann H., Wildgruber D., Daum I., Grodd W. (1998)  Does the cerebellum contribute to cognitive aspects of speech production? A functional magnetic resonance imaging (fMRI) study in humans. Neurosci Lett, 247, 187-90.

  • • Andersen, R.A., Bracewell R.M., Barash S., Gnadt J.W., & Fogassi L. (1990) Eye position effects on visual, memory, and saccade-related activity in areas LIP and 7a of macaque. Journal of Neuroscience, 10, 1176-1796.
    • Baker, C. B., Hsu, M., Ryan, T. L., Nadel, L., & Jacobs, W. J. (1998).  Functional Neuroimaging of Place Learning in Computer-Generated Space. Poster presented to the Robert S. Flinn Foundation Life Sciences and Biomedical Research.
    • Bower J.M. (1997). Is the cerebellum sensory for motor's sake, or motor for sensory's sake: the view from the whiskers of a rat?  Prog Brain Res, 114, 463-96.
    • Cohen, J. D., Perlstein, W. M., Braver, T. S., Nystrom, L.E., Noll, D. C., Jonides, J., & Smith, E. E. (1997).  Temporal dynamics of brain activation during a working memory task. Nature, 10, 604-608.
    • Corbetta, M., Shulman, G.L. (1998) Human cortical mechanisms of visual attention during orienting and search.  Philos Trans R Soc Lond B Biol Sci, 353, 1353-62
    • Gitelman, D. R., Nobre, A. C., Parrish, T. B., LaBar, K. S., Kim, Y. H., Meyer, J. R., & Mesulam, M. M. (1999). A large-scale distributed network for covert spatial attention: anatomical delineation based on stringent behavioural and cognitive controls. Brain, 122, 1093-1106.
    • Feigenbaum, J. D., & Rolls, E. T. (1991).  Allocentric and egocentric spatial information-processing in the hippocampal-formation of the behaving primate.  Psychobiology, 19, 21-40.
    • Fernandez G., Weyerts H., Schrader-Bolsche M., Tendolkar I., Smid H. G., Tempelmann C., Hinrichs H., Scheich H., Elger C. E., Mangun G. R., Heinze H. J. (1998) Successful verbal encoding into episodic memory engages the posterior hippocampus: a parametrically analyzed functional magnetic resonance imaging study.  J Neurosci, 18, 1841-7.
    • Jacobs, W. J., Laurance, H. E., & Thomas, K. G. F. (1997). Place learning in virtual space I: Acquisition, overshadowing, and transfer. Learning and Motivation, 28, 521-541. 
    • Jacobs, W. J., Thomas, K. G. F., Laurance, H. E., & Nadel, L. (1998). Place learning in virtual space II: Stimulus control. Learning and Motivation, 29, 288-308.
    • Laurance, H.E, Thomas, K.G.F,, Nadel, L. & Jacobs, W.J. (1998). Place learning in real and computer-generated space: Performance of younger and older adults. Journal of Cognitive Neuroscience, 16, Suppl. S MAR 18.
    • Laurance, H.E., Thomas, K.G.F., Nadel, L., & Jacobs, W.J. (in preparation). Place learning V: Old adults map computer-generated space but do not find locations within it. 
    • Li L., Miller E.K., & Desimone R. (1993).  The representation of stimulus familiarity in anterior inferior temporal cortex. J Neurophysiol, 69, 1918-29.
    • Morris, R. G. M. (1981).  Spatial location does not require the presence of local cues.  Learning and Motivation, 12, 239-260.
    • Morris R. G. M., Garrud, J., Rawlins, N. P., & O’Keefe, J. (1982).  Place navigation impaired in rats with hippocampal lesions.  Nature (London), 297, 681-683.
    • Nadel, L., Thomas, K.G.F., Laurance, H.E., Skelton, R., Tal, T. & Jacobs, W. J. (1998). Human spatial cognition in a virtual arena. In C. Freksa, C Habel, K. F. Wender (Eds.), Spatial Cognition - An interdisciplinary approach to representation and processing of spatial knowledge. (pp. 399-427). Springer-Verlag: Berlin.
    • Nakamura K., & Kubota K. (1996).  The primate temporal pole: its putative role in object recognition and memory. Behav Brain Res, 77, 53-77.
    • Rolls, E. T. (1996). A theory of hippocampal function in memory. Hippocampus, 6, 601-620.
    • Schmahmann J. D., & Pandya D. N. (1997). The cerebrocerebellar system. Int Rev Neurobiol, 41, 31-60.
    • Skelton, R., Bakach, C.M., Laurance, H.E., Thomas, K.G.F., & Jacobs, W.J. (submitted). Humans with traumatic brain injuries show place-learning deficits in computer-generated virtual space. Journal of Clinical and Experimental Neuropsychology
    • Thomas, K. G. F., Nadel, L., & Jacobs, W. J. (in preparation). Place learning in virtual space III: Perceptive learning.
    • Vallar G., Lobel E., Galati G., Berthoz A., Pizzamiglio L., & Le Bihan D. (1999). A fronto-parietal system for computing the egocentric spatial frame of reference in humans. Exp Brain Res, 124, 281-286.
    • Winstein, C. J., Grafton, S. T., & Pohl, P. S. (1997). Motor Task difficulty and Brain Activity: Investigation of Goal-Directed Reciprocal Aiming Using Positron Emission Tomography. Journal of Neurophysiology, 77, 1581-94.



This page, and all files included with it, broken links et al., are the sole responsibilities of Ming Hsu. Please forward all questions to <mhsu@u.arizona.edu>
.[Home] [Projects]