Computational models of V1 and MT neurons for estimation of visual motion direction

© 2016 Dr. Parvin Zarei EskikandThe processing of motion information in the brain initiates in the primary area of visual cortex (V1). V1 neurons transmit initial estimates of the motion of stimuli to neurons in higher areas of the cortex. However, V1 neurons measure only the component of the motion that is perpendicular to the edge of the stimulus because of their small receptive fields. A computational neural network model based on area MT is developed to estimate the correct direction of motion from the ambiguous information supplied by V1 neurons. The neural model processes the motion information of the stimulus through two stages. Complex V1 neurons at the first stage are spatiotemporal filters that represent ambiguous motion information along the edge of the stimulus but correct motion information at the end points (terminators) of the stimulus. The neurons responding to the terminators of the stimulus provide an unambiguous estimation of the direction of motion because of the two-dimensional structure of the corners. End-stopped neurons in V1 exclusively respond to the motion of terminators, which are modeled by inhibitory interconnections between the neighboring neurons. The incoming motion signals provided by complex V1 neurons and end-stopped V1 neurons proceed to MT neurons at the second stage. The excitatory and inhibitory interconnections between MT neurons result in the propagation of unambiguous motion information from the terminators to the other regions of the stimulus to achieve responses to the correct direction of motion. The information provided by end-stopped neurons is essential for the MT neurons to distinguish ambiguous motion information of the edges from the unambiguous information of the terminators. Although, end-stopped neurons provide a correct estimation of the direction of motion at the end-points (intrinsic terminators) of the stimulus, their represented local motion signals at extrinsic terminators (formed at the intersection of two overlapping stimuli moving in different directions) conflicts with the global motion direction of the stimuli. The neural model explains how interactions between form and motion information may assist neurons in the motion-specific regions of primate cortex to differentiate intrinsic from extrinsic terminators. In the proposed model, MT neurons additionally receive form information from neurons in the V1 area sensitive to the luminance of the stimulus with suppressive surrounds. As these neurons receive stronger inhibition from their surrounds at the extrinsic terminators, the excitatory inputs from these V1 neurons assist unambiguous motion signals at the intrinsic terminators to dominate over the local motion signals generated at the extrinsic terminators. The results show that, despite the inability of end-stopped neurons to distinguish two different types of terminators, center-surround V1 neurons have higher activity at the intrinsic terminators resulting in an accurate representation of motion by MT neurons. The proposed model also shows that the strength of the excitatory connections from center-surround V1 neurons, which supply initial form information to MT neurons, determines the pattern or component selectivity of MT neurons. A strong input from the center-surround V1 neurons inhibits the local motion information of the extrinsic terminators. Therefore, MT neurons reflect the component motion information of the stimuli received from V1 neurons in the absence of the intrinsic terminators. However, in the case of weak excitatory connections from center-surround V1 neurons, the pattern motion information propagates from extrinsic terminators to other regions over time and, after a temporal delay, MT neurons represent the pattern motion of the input stimulus. The proposed model of the V1 and MT neurons suggest the key role of terminators and the neurons enhancing or suppressing the motion information of these regions in the perceived motion of the stimulus. The findings are summarized in three main components: The results show that resolving ambiguity of the motion information along the edges is a two-stage process that is initiated by end-stopped V1 neurons and finalized by interconnections between MT neurons. - The model also suggests the necessity of form information for suppressing the effects of extrinsic terminators in the case of overlapping stimuli. - The results of the model question the generally believed hypothesis that the hierarchical process of pattern motion computation of MT neurons is based on the integration of component motions. The model suggests that the pattern or component motion preference of MT neurons is highly dependent on the strength of the input from center-surround V1 neurons