A FUNCTIONAL MRI STUDY ON INTERNAL
MODELS OF DYNAMIC TRANSFORMATIONS DURING LEARNING A VISUOMOTOR TASK.
H. Imamizu*1, S. Miyauchi2,
Y. Sasaki2, R. Takino2, B. Putz2, and M. Kawato1 1
1ATR Human Information Processing Research Labs., Kyoto, Japan; 2Communications
Research Labs., Tokyo, Jap
Some recent computational studies
have stated that the central nervous system (CNS) acquires internal models
of the kinematics and/or dynamics of controlled objects and environments
for rapid and smooth control (e.g. Kawato & Gomi, 1992, TINS,
15, pp. 445-453). However, it is still unclear as to which part of the
CNS is involved in acquiring these models during the control of visually
guided limb movements. Using fMRI (1.5T, Siemens VISION, echo planar imaging)
we examined cortical and subcortical activity in normal human subjects
learning a visuomotor task under the dynamic alteration of visual feedback
of the hand position. The task for the subjects was to move a computer
mouse so that the cursor follows the target moving on the screen as closely
as possible (a modification of the task used in Flament et al. 1994, Soc.
Neuosci. Abst. p. 20). The position of the cursor was altered by a
dynamic transformation of the mouse movement, namely integration of the
previous mouse positions, in learning periods whereas no transformation
was imposed in control periods. The errors, measured by taking a distance
between the cursor and the target, decreased with practice in the learning
period. Images (8-10 slices, 7-10 mm thick, through the cerebellum, occipital,
parietal and part of the frontal lobes) scanned in both periods were compared
with cross correlation. We found that the MR signal intensity in the biventer
lobule of the cerebellum and parietal areas 5 and 7 was significantly higher
in the learning period. The results suggest that these areas are involved
in error correction by comparing visual information and kinesthetic information
and in acquiring the internal models of the imposed dynamic transformation