Neuropsychologia 41 (2003) 1769���1784 Normal and anomalous development of visual motion processing: motion coherence and ���dorsal-stream vulnerability��� Oliver Braddick a,���, Janette Atkinson b, John Wattam-Bell b a Department of Experimental Psychology, University of Oxford, Oxford, UK b Visual Development Unit, Department of Psychology, University College London, London, UK Received 3 June 2003 accepted 20 June 2003 Abstract Directional motion processing is a pervasive and functionally important feature of the visual system. Behavioural and VEP studies indicate that it appears as a cortical function after about 7 weeks of age, with global processing, motion based segmentation, and the use of motion in complex perceptual tasks emerging shortly afterwards. A distinct, subcortical motion system controls optokinetic nystagmus (OKN) from birth, showing characteristic monocular asymmetries which disappear as binocular cortical function takes over in normal development. Asymmetries in cortical responses are linked to this interaction in a way that is not yet fully understood. Beyond infancy, a range of developmental disorders show a deficit of global motion compared to global form processing which we argue reflects a general ���dorsal-stream vulnerability���. �� 2003 Elsevier Ltd. All rights reserved. Keywords: Optokinetic nystagmus (OKN) Nucleus of the optic tract (NOT) Visual evoked potential (VEP) 1. Introduction The detection of motion is one of the most pervasive fea- tures of the visual system. In every species where neurosci- entists have looked for visual motion sensitivity���from flies (Egelhaaf & Borst, 1993), beetles (Hassenstein & Reichardt, 1956), crabs (Zeil & Zanker, 1997) and frogs (Barlow, 1953), to rabbits (Barlow & Hill, 1963a), cats (Hubel & Wiesel, 1962) and primates (Hubel & Wiesel, 1968)���it has been found. Motion detection is basic to perception, cognition and action. It contributes to a great range of visual functions, in- cluding scene segmentation, depth perception, postural and oculomotor stabilisation, recognition of characteristic kine- matic events such as the actions of other individuals, and the control of actions in dynamic situations. Moving targets have a high salience in attracting attention in the peripheral visual field. Motion is continually present in the visual im- age, through eye movements, self-motion, and the motion of external objects. We expect, therefore, that the development of visual motion processing will be a very important part of the overall development of vision in infancy. We might also expect, given the range of processes depending on motion detection, that it would develop early. ��� Corresponding author. E-mail address: o.braddick@psy.ox.ac.uk (O. Braddick). To examine this question, we need to have a clear un- derstanding of what constitutes evidence for visual motion processing. Moving stimuli produce local changes in lumi- nance and contrast as they pass over the retina, and so will excite any neurons that are responsive to change. Infants in the first month of postnatal life are sensitive to remarkably high rates of flicker, reflected in a behavioural preference for a large flickering field over a static field of matched mean luminance (Regal, 1981). This general sensitivity to dynamic stimuli can explain the fact that very young in- fants will orient to a moving target (Volkmann & Dobson, 1976). In terms of contrast detection and acuity, which de- velop rapidly in the first 3 months of life, motion benefits infants��� sensitivity to low spatial frequency gratings, but only to the same degree as for adults (Atkinson, Braddick, & Moar, 1977). However, infants��� sensitivity and prefer- ence for dynamic stimuli do not necessarily imply that the infant���s visual system can extract motion information. The perceptual tasks dependent on motion���depth from optic flow, recognising biological motion, control of dynamic actions���need the visual system to represent the direction and speed of local motion. In visual neurophysiology, the classic criterion that a neuron is concerned with motion processing is directional selectivity���a differential response to the same target moving with equal speed in opposite di- rections. To state that an infant is processing visual motion, 0028-3932/$ ��� see front matter �� 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0028-3932(03)00178-7

1770 O. Braddick et al. / Neuropsychologia 41 (2003) 1769���1784 we shall similarly require evidence that the visual system is responding to directional information. Evidence that a directional system is operating at birth is provided by the eye movements of optokinetic nystagmus (OKN) which follow the direction of large-field movement. The direction of movement must be registered somewhere in the infant���s visual system, but we shall discuss the ev- idence in Sections 11���12 that this is a reflex, subcortical system that is distinct from the cortical directional sys- tems which provide the basis for motion discrimination and movement-based perceptual skills. We will discuss cortical motion systems first, going from simple relative motion perception to global processing, and then compare these systems with the subcortical OKN sys- tem. Cortical and subcortical systems must operate in a unified way, linking top���down and bottom���up and feeding through to motor control and frontal executive systems. In- fants start using cortical systems around 2���3 months of age, but are not tuned as highly as the adult system until much later in life. Adult motion coherence thresholds, for exam- ple, are not reached until around 8���10 years of age. We trace this development, reviewing research under a number of subheadings: ��� Neural systems of motion processing from physiological and anatomical evidence. ��� Development of directional sensitivity in infants from vi- sual evoked potential (VEP) and behavioural studies, us- ing both uniform and non-uniform motion, and the case of second-order motion. ��� Development of global motion processing, an indicator of ���dorsal��� stream development, which may be compared with global form processing. ��� Infants��� use of motion information in complex perceptual discriminations ��� Developmental links between the processing of disparity and global motion within the dorsal stream. ��� The development of the optokinetic response (OKN), monocular asymmetries of OKN as a signature of sub- cortical processing, and their relationship to cortical asymmetries. ��� The relative vulnerability of dorsal-stream development in developmental disorders, indicated by comparison of global motion and form sensitivity ��� Current neurobiological models of the relation between dorsal and ventral-stream development. 2. Neural systems for motion processing In primates the retinal and thalamic neurons that form the main pathway to striate cortex are not directionally selec- tive, but directional neurons are common in area V1. The development of motion processing, therefore, is one aspect of the wider development of the selective properties of vi- sual cortex, which begins in human infants in the early post- natal months (Atkinson, 1984, 2000 Braddick, Atkinson, & Wattam-Bell, 1989). Beyond V1, the extrastriate area V5 (also known as MT), which has homologues in human be- ings and macaques, has a high density of motion-sensitive neurons (Maunsell & Newsome, 1987 Tootell, Dale, Sereno, & Malach, 1996 Watson et al., 1993 Zeki, 1974). V1 neurons have small receptive fields that can detect local motion. The properties of V5 neurons, include larger recep- tive fields than V1 (Mikami, Newsome, & Wurtz, 1986a, 1986b), centre-surround interactions (Allman, Miezin, & McGuinness, 1985 Xiao, Raiguel, Marcar, & Orban, 1997), integration of different directions (Movshon, Adelson, Gizzi, & Newsome, 1986), and sensitivity to motion coherence (Britten, Shadlen, Newsome, & Movshon, 1992). All these suggest that V5 integrates local motion signals from V1 into the more global representations of motion needed as a basis for perceptual performance. Projections from V5 into the neighbouring area MST and into parietal lobe seem to pro- vide a good basis for the use of visual motion in the control of eye movements and other actions. For all these reasons, V5 is often considered the key to extrastriate motion processing. However, human functional neuroimaging studies (Braddick et al., 2001 Sunaert, Van Hecke, Marchal, & Orban, 1999) show a much wider network of cortical areas that respond strongly to directional motion. In particular, there is such an area some way posterior and superior to V5 which may be topographically homologous to the monkey area V3A. We do not know the functional role and connectivity of the various extrastriate motion areas even in the adult human. It would be rash to assume that the development of perceptual functions based on motion information necessarily involves the transmission of information only through V5. V5, V3A, and parietal lobe areas that they project to, are key parts of the dorsal cortical stream which is believed to process the visual information needed for understanding spa- tial relationships and controlling spatially directed actions (Milner & Goodale, 1995 Mishkin, Ungerleider, & Macko, 1983). In contrast, the ventral stream leading to the temporal lobe processes the information required for the recognition of objects, faces and colour. The two streams are distinct in terms of connectivity (Felleman & Van Essen, 1991 Young, 1992) and in the types of neural processing, with much higher proportions of directionally-selective neurons in dorsal-stream structures (De Yoe & Van Essen, 1988). Thus, the development of motion sensitivity, and its relationship to the shape processing needed for ventral-stream recognition performance, may serve as an indicator of the relative devel- opment of function in the two streams, and their vulnerability in anomalous brain development. The processing of binoc- ular disparity is another important function of the dorsal stream, and the developmental relation between motion and binocularity, and stereo sensitivity, will be discussed below. Motion processing is not purely a cortical issue. A small but significant projection from the retina goes to mid- brain nuclei involved in oculomotor control, in particular to the nucleus of the optic tract (NOT). NOT neurons are

O. Braddick et al. / Neuropsychologia 41 (2003) 1769���1784 1771 directionally selective (Hoffmann, 1981, 1986). The NOT appears to be a key structure in the pathway controlling the reflex response of optokinetic nystagmus which serves to stabilise gaze. As discussed below, this motion-based function has a quite different developmental course from cortical motion processing. The interaction of cortical and subcortical motion information in development may serve as a paradigm case for understanding the developmental in- terplay of cortical and subcortical structures more generally. One of the most obvious features of neural development in infancy is the progressive myelination of the visual path- way and of cerebral tracts more generally. There is a marked decrease in the latency of cortical responses over the first months of life (Moskowitz & Sokol, 1983 Porciatti, 1984) which presumably reflects this. An area now identified as V5 (Vaina, Cowey, & Kennedy, 1999) was described by Flechsig (1920) as heavily myelinated at birth. However, the myelination of the optic nerve and tract, and intracortical pathways to V5, will also be critical. Since the detection of motion depends on comparing the time-of-arrival of rapidly changing visual signals from different locations, slow trans- mission by an incompletely myelinated system might be ex- pected to be a significant constraint on the development of motion processing. However, the factors which would most obviously impair motion processing would be (a) variabil- ity in transmission time of visual signals from one fibre to another (b) poor transmission of high temporal frequency variations. It is not clear what might be the expected effects of immature myelination on these factors. 3. VEP evidence for directional selectivity Directionally-selective neurons should show a modula- tion of their responses with a stimulus that periodically re- verses its motion between their preferred and non-preferred directions. Directional selectivity in infants can therefore be looked for with scalp recordings of visual evoked potentials that are time-locked to direction reversals. In all tests for directionality, it is important to ensure that temporal inte- gration of the stimulus would not lead to distinctive features in the spatiotemporal array that could be detected without directional information. Such features would arise if a mov- ing, but otherwise unchanging, pattern was temporally in- tegrated around the moment of direction reversal. Our VEP method therefore uses a pattern of random pixels which is replaced by a fresh random array at each reversal, and also midway between reversals. Any response due to pixel re- placement then occurs at twice the reversal frequency, and a response at the reversal frequency must be specifically as- sociated with direction change. The first clear evidence for directional sensitivity in in- fants came from a longitudinal study using this method (Wattam-Bell, 1991). This showed a median age of 10 weeks for the first statistically reliable response to reversals with a velocity of 5���/sec, and 12 weeks for 20���/sec. The response to the pixel replacements was present at a consistently earlier age, implying that the absence of the directional response be- fore 10 weeks was not because the spatiotemporal properties of the pixel array were beyond the range of the infants��� visual systems. It is also striking that the direction-reversal VEP has a later onset than the orientation-reversal VEP which has been measured using an analogous approach (Braddick, 1993 Braddick, Wattam-Bell, & Atkinson, 1986). This was confirmed by studies of the two measures on the same in- fants (Atkinson, Wattam-Bell, & Braddick, 2002). These also showed that direction-reversal continued to show a later onset even when (a) both responses were measured with grating stimuli of the same spatial frequency and (b) the re- versal rate of the directional stimulus was slowed to allow for possible longer temporal integration required for motion processing. These results suggest that directional selectivity is a prop- erty of human cortical neurons that arises postnatally, later than orientation selectivity but earlier than selectivity to binocular correlation, which can also be tested by a simi- lar method (Braddick, 1996 Braddick, Wattam-Bell, Day, & Atkinson, 1983). The specific cortical area where the direction-selective VEP originates is not yet known, either in infant or in adult subjects. In future, the spatial resolution of MEG may help to answer this question. 4. Behavioural evidence for directional selectivity Any developmental argument depends on negative results���a capability which can be demonstrated at age x cannot be demonstrated at some earlier age. The absence of a detectable VEP might arise for a number of reasons and so converging evidence from behavioural techniques is important. One approach is through preferential looking (PL). In our work, one side of the display contains a uniform pattern of random pixels in oscillatory motion. The other side contains a similar pixel array, but the motion of a central strip is al- ways opposite to that of the strips above and below. Any non-directional system would see equivalent dynamic events in all regions of the display. Only a directional system will detect the shearing between opposite motions that produces strong segmentation in one side of the display for an adult observer. From about 7 to 8 weeks of age, infants��� looking patterns show a marked preference for the latter side, pre- sumably because it contains interesting structure beyond the random dots (Wattam-Bell, 1992, 1994, 1996a). Younger in- fants show no significant preference. Like the VEP response, the preference first becomes ap- parent for modest speeds between 5 and 10���/sec, and higher speeds at progressively later ages (Wattam-Bell, 1992). The speed range of motion responses depends on both spatial and temporal properties of neurons the results imply that the improvement from 12 weeks onwards depends primarily on the spatial parameters. Most aspects of the development