Optical flow is a rich source of information about the three-dimensional motion and structure of the visual environment. Little is known of how the brain derives this information. One possibility is that it analyzes first-order elementary components of optical flow, such as expansion, rotation, and shear. Using a combination of physiological recordings and modeling techniques, we investigated the contribution of the middle superior temporal area (MST), a third-order cortical area in the dorsal visual pathway that receives inputs from the medial temporal area (MT Optical flow, which can be defined as the apparent motion of the image brightness on the retina, has long been considered as a useful representation of visual motion information (1). According to a fundamental theorem of kinematics, or Helmholtz theorem (2), optical flow, within a small area of the visual field, can be seen as the sum of a translation with four elementary flow components (EFCs): a rotation (circular motion), an expansion (radial motion), and two components of shear (deformations) (3). This result has been successfully used in computer and computational vision for the analysis of motion (3-7). In addition, EFCs have been shown to be biologically relevant (8,9). Therefore, as suggested by recent studies in which middle superior temporal area (MST) cells of the monkey brain were reported to be selective for rotation and expansion (10-15), one may propose that the brain explicitly represents EFCs. MST is a third-order cortical area in the dorsal pathway leading to the parietal cortex (16). It receives inputs from the middle temporal area (MT) (17, 18), which has been implicated in motion analysis (19)(20)(21)(22), and is mainly driven by magnocellular input (23). However, there is disagreement about the degree of selectivity of MST cells: MST cells have been reported to be selective for a single EFC (10, 24) but also for several components (12, 13). Furthermore, there is no information about whether or not complex motion patterns are decomposed into the (linear) superposition of EFCs. Finally, it is not clear how the MST selectivity for EFCs is generated.Therefore, we quantitatively compared responses of cortical cells to translation and EFCs in both areas (MST and MT). To clearly distinguish between responses to local flow vectors and to global arrangements of vectors characteristic of EFCs, we systematically tested for the position invariance of the selectivity. Once selectivity for an EFC was established, we then studied the responses of MST cells to a combination of components, keeping one component constant. The results of the physiological study were used to model the transformations effected by the MST cells on MT input.
METHODSWe have recorded from single MST and MT neurons of anesthetized and paralyzed monkeys (Macaca fascicularis) to obtain stable recordings over several hours and precisely localize the recorded units with respect to area and layer. The monkeys were anesthetized with sufentanil (5 pg-kg-1lh-1) and paralyze...