Complex Motion Processing and its Deficits

We are using psychophysics, computational modeling and fMRI to understand the mechanisms underlying complex motion perception and their modulation by learning.
 
Cortical activity elicited by discrimination of shift in complex motion patterns
Psychophysics and modeling for complex motion processing
Brain regions active during the hMT+ localization task. Stimuli were expanding dot patterns vs. static random dots, and activation is projected onto a rendered 3D brain surface. Shown are back, left and right views respectively. Data were collected in a 3T MRI scanner.

Contents

Back to Research

Two stages of processing complex motion. Using a psychophysical summation technique we studied the characteristics of global motion detectors.The stimuli consisted of radial, circular or translating RDKs in which a variable proportion of the dots contained the motion signal and the reminder provided motion noise. Stimuli were displayed in a circular aperture spatially curtailed into symmetrically opposed sectors. We measured performance for signal-to-noise sensitivity, defined by the inverse of the minimal proportion of coherent dots required to reliably discriminate motion direction, contrast sensitivity for motion discrimination, and contrast sensitivity for pattern detection. The results showed that for all types of motion, sensitivity increased with the stimulus area and performance was not dependent on contrast. Using again a summation technique to explore the properties of detectors tuned to optic flow patterns, we investigated the extent of the summation. The results showed that summation can occur over very large areas, consistent with the existence of optic-flow detectors with large receptive fields in physiology.

Top of page

Modeling Optic Flow Selectivity in MSTd. We studied the development of optic flow selectivity in the hidden layer of a two layer back-propagation network and tested motion sensitivity in a modified version of the trained network using simulated psychophysical stimuli. The results were compared with those we have obtained previously from human subjects. The network is described in [9]. The simulations produced hidden units whose position invariance and motion selectivity were consistent with MSTd responses to visual motion components of optic flow stimuli. Across different network complexities and training conditions, units in the hidden layer developed a continuum of optic flow selectivity independent of any biases associated with the specification of the motion selectivity in the output layer. Consistent with our previous psychophysical results, the network motion sensitivity to radial and circular motion under the no-mask condition suggests an ideal integrator model of complex motion detection as a function of signal area.

Top of page

Development of preferred motion responses across complex stimuli. Using the same network structure and methods, we examined the development of preferred motion responses in the hidden units across a wider range of complex motion stimuli (radial, circular, spiral, and planar motions), and compared these properties with physiology reported in the dorsal division of the medial superior temporal area (MSTd). Across trained networks, the hidden layer developed preferred responses to complex global patterns of motion spanning a continuum in the stimulus space formed by radial, circular, and planar motions. When hidden unit responses were classified using a multi-component scheme the distribution of units was consistent with results reported in MSTd (shown below).

Examination of the input-hidden layer weights identified single-component units that exhibited the center-surround activation consistent with previous results. Multi-component units including preferences for planar motions exhibited opposing zones of excitatory and inhibitory input activation, consistent with the overlapping gradient hypothesis. The significance of these results is that they: 1) are consistent with reported properties in MST, 2) suggest that multi-component units use subregions of their receptive fields to process complex patterns of motion, and 3) imply that "classical" receptive fields mapped in higher visual motion areas such as MST may depend on the type of motion.

Top of page

Dissociations of deficits in the 3D Motion subsystem. In 3 patients with deficits of speed and direction discrimination we compared their ability to infer the 3D structure of a rigid object from motion information (SFM) and their ability to judge heading (HFM). We found that patients impaired on the SFM task made normal judgments of straight-line heading. The data suggests that scene reconstruction is not essential for straight line heading judgments.

Impaired 3D Motion with good 2D motion perception. In 5 patients with posteriorparietal lobe lesions (4 unilateral and one bilateral) we compared theirperformance on perception of heading, radial motion, and 3D SFM with theirperformance on discriminating speed, direction and 2D form from motion.All five p atients were selectively impaired on 3D motion and had normalperformance on 2D motion, which supports the hierarchical view of motionprocessing.

Top of page

Speed Discrimination. Random dot kinematograms were used to simulateradial, circular and spiral optic flow. We proposed two novel experiments. Experiment 1 measured the perceived speed of a range of optic flow patterns against a circular comparison motion. The results showed that the perceived speed of complex motion stimuli depends upon the pattern of global motion, with radial motions found to appear faster than rotations by approximately 10% and a smaller but significant effect for spirals. Experiment 2 measured discrimination thresholds for pairs of similar optic flow stimuli identical in all respects except average dot speed. The speed discrimination thresholds measured did not vary as a function of the optic flow stimulus (data shownin Clifford CWG, Beardsley SA, Vaina LM "The perceptiuon and discriminationof speed in complex motion", Vis Res (accepted)). We suggest that a model satisfying simultaneous constraints on motion-in-depth and object rigidity most easily accounts for the perceived speed results. Speed discrimination thresholds for complex motions containing a global pattern of direction information and a radial speed gradient were found to be comparable for stimuli with identical speed profiles but no global pattern of direction information. This suggests that speed discrimination is based on the pooled responses of elementary motion detectors.

Top of page

Complex motion deficits. The subjects were normal control subjects (n=115) and patients (n=65) with unilateral ischemic vascular lesions (in almost all cases embolic). For the normal subjects thresholds were: RVF=9.75 (SD=1.24), LVF=10.50 (SD=1.69). The slope of the regression line predicting RVF from LVF was 1 (b1=1;p<0.05 and the intercept was close to 0, b0 =0.9%). In patients a significant relationship was found between the side of the lesion (L or R) and coherence threshold values (c2=4.46 p<0.03). Patients performed worst for stimuli presented in the visual field contralateral to the lesion. To determine whether the site of the lesion made a significant difference in performance we used neuroimaging information from CT and MRI to divide patients into three groups according to the anatomical areas involved in the lesion: I medial occipital and posterior temporal, II posterior parietal, parietal temporal lesions and III anterior parietal, and parietal temporal.

Analysis with Duncan's multiple comparison test for pairwise differences showed that all patient groups, except Group I, differed significantly from the normal controls (p< 0.05). A between group comparison indicated that the normal controls and Group I differed significantly from Group II and III for stimuli presented in either visual field (p<0.05). Since several patients had some visual field loss, we did a lesion group x field loss ANOVA which showed that none of the effects were significant for the test results in either visual field. A significant difference between the normal population and patients by lesion type (except Group I) was found even when we included in the analysis only the patients without any visual field loss.

The results suggest that direction discrimination in global motion is severely affected by lesions (in L or R hemisphere) along the dorsal route (Group II) which presumably involves the human homologue of monkey MT and MST. Patients with ventral occipital or posterior temporal lesions (Group I) had normal performance. Group III patients were impaired for stimuli presented in either visual field, in spite of having no visual field defects. A possible explanation for this bilateral deficit is that beyond MT, which is still retinotopic, receptive field sizes become larger straddling the vertical meridian, and farther in the anterior temporal or parietal lobes they encompass the entire visual field.

Top of page

Monte Carlo simulation of global motion direction discrimination mechanisms. We carried out Monte Carlo simulations of vector summation and winner-take-all mechanisms for computing global motion direction based on simulated responses of direction-selective neurons to stochastic random dot stimuli. We studied the performance of the mechanisms by varying the following parameters: receptive field size (normally distributed mean size range 0.1 deg2 to 9.0 deg2), number of neurons (25 to 100), stimulus dot density (1 to 8dots/deg2), and number of directions to which neurons were selective (8 or 16). The main result was that the receptive field size of a neuron was a critical parameter. Large receptive fields favored better performancein the global motion direction discrimination. The performance of the two mechanisms (vector summation and winner-take-all) was similar, and it improved with increasing dot density. The additional parameters did not have a significant effect on performance.

Top of page


PUBLICATIONS

1. Beardsley S, Vaina LM "A laterally interconnected neural architecture in MST accounts for psychophysical discrimination of complex motion patterns." J Comput Neurosci., 2001; 10(3): 255-280.
2. Clifford CW, Beardsley SA, Vaina LM "The Perception and Discrimination of Speed in Complex Motion", Vision Research, 1999; 39(13): 2213-2227.
3. Beardsley S, Vaina LM "Computational Modeling of Optic Flow Selectivity in MSTd Neurons," Network: Comput. Neural Syst., 1998; 9: 467-493.
4. Beardsley S, Clifford CWG, Vaina LM "Discrimination of shifted centers of motion in complex stimuli", Investigat. Opthalmology & Visual Science 1998; 39(4): S621.
5. Burr D, Morrone MC, Vaina LM "Large Receptive Fields for Optic Flow Detectors in Humans", Vision Research, 1998; 38(12): 1731-1743.
6. Vaina LM "Complex Motion Perception and its Deficits", Current Opinions in Neurobiology, 1998; 8: 494-502.
7. Beardsley SA, Vaina LM "Computational modeling of optic flow selectivity in MSTd neurons", Investigative Ophthalmology & Visual Science, 1997; 38(4): S80.
8. Beardsley S, Vaina LM, Poggio T "The development of optic flow selectivity in MSTd neurons using back-propagation networks", Soc. Neurosci. Abstr., 1996; 22(1): 1619.
9. Vaina LM, Royden C, Bienfang DC, Makris N, Kennedy D "Normal perception of heading in a patient with impaired structure from motion", Invest. Opth. Vis. Sci., 1996; 37: 4890.
10. Morrone MC, Burr DC, Vaina LM "Two stages of visual processing for radial and circular motion", Nature, 1995; 376(6540): 507-509.

Top of page