Pattern Decoding Research Articles

Geometric classification of brain network dynamics via conic derivative discriminants

BackgroundOver the past decade, pattern decoding techniques have granted neuroscientists improved anatomical specificity in mapping neural representations associated with function and cognition. Dynamical patterns are of particular interest, as evidenced by the proliferation and success of frequency domain methods that reveal structured spatiotemporal rhythmic brain activity. One drawback of such approaches, however, is the need to estimate spectral power, which limits the temporal resolution of classification. New methodWe propose an alternative method that enables classification of dynamical patterns with high temporal fidelity. The key feature of the method is a conversion of time-series data into temporal derivatives. By doing so, dynamically-coded information may be revealed in terms of geometric patterns in the phase space of the derivative signal. ResultsWe derive a geometric classifier for this problem which simplifies into a straightforward calculation in terms of covariances. We demonstrate the relative advantages and disadvantages of the technique with simulated data and benchmark its performance with an EEG dataset of covert spatial attention. We reveal the timecourse of covert spatial attention and, by mapping the classifier weights anatomically, its retinotopic organization. Comparison with existing methodWe especially highlight the ability of the method to provide strong group-level classification performance compared to existing benchmarks, while providing information that is complementary with classical spectral-based techniques. The robustness and sensitivity of the method to noise is also examined relative to spectral-based techniques. ConclusionThe proposed classification technique enables decoding of dynamic patterns with high temporal resolution, performs favorably to benchmark methods, and facilitates anatomical inference.

Dual-modality endoscopic probe for tissue surface shape reconstruction and hyperspectral imaging enabled by deep neural networks.

Surgical guidance and decision making could be improved with accurate and real-time measurement of intra-operative data including shape and spectral information of the tissue surface. In this work, a dual-modality endoscopic system has been proposed to enable tissue surface shape reconstruction and hyperspectral imaging (HSI). This system centers around a probe comprised of an incoherent fiber bundle, whose fiber arrangement is different at the two ends, and miniature imaging optics. For 3D reconstruction with structured light (SL), a light pattern formed of randomly distributed spots with different colors is projected onto the tissue surface, creating artificial texture. Pattern decoding with a Convolutional Neural Network (CNN) model and a customized feature descriptor enables real-time 3D surface reconstruction at approximately 12 frames per second (FPS). In HSI mode, spatially sparse hyperspectral signals from the tissue surface can be captured with a slit hyperspectral imager in a single snapshot. A CNN based super-resolution model, namely "super-spectral-resolution" network (SSRNet), has also been developed to estimate pixel-level dense hypercubes from the endoscope cameras standard RGB images and the sparse hyperspectral signals, at approximately 2 FPS. The probe, with a 2.1 mm diameter, enables the system to be used with endoscope working channels. Furthermore, since data acquisition in both modes can be accomplished in one snapshot, operation of this system in clinical applications is minimally affected by tissue surface movement and deformation. The whole apparatus has been validated on phantoms and tissue (ex vivo and in vivo), while initial measurements on patients during laryngeal surgery show its potential in real-world clinical applications.

Prospective motion correction improves the sensitivity of fMRI pattern decoding.

We evaluated the effectiveness of prospective motion correction (PMC) on a simple visual task when no deliberate subject motion was present. The PMC system utilizes an in‐bore optical camera to track an external marker attached to the participant via a custom‐molded mouthpiece. The study was conducted at two resolutions (1.5 mm vs 3 mm) and under three conditions (PMC On and Mouthpiece On vs PMC Off and Mouthpiece On vs PMC Off and Mouthpiece Off). Multiple data analysis methods were conducted, including univariate and multivariate approaches, and we demonstrated that the benefit of PMC is most apparent for multi‐voxel pattern decoding at higher resolutions. Additional testing on two participants showed that our inexpensive, commercially available mouthpiece solution produced comparable results to a dentist‐molded mouthpiece. Our results showed that PMC is increasingly important at higher resolutions for analyses that require accurate voxel registration across time.

A continuous time-resolved measure decoded from EEG oscillatory activity predicts working memory task performance

Objective. Working memory (WM), crucial for successful behavioral performance in most of our everyday activities, holds a central role in goal-directed behavior. As task demands increase, inducing higher WM load, maintaining successful behavioral performance requires the brain to work at the higher end of its capacity. Because it is depending on both external and internal factors, individual WM load likely varies in a continuous fashion. The feasibility to extract such a continuous measure in time that correlates to behavioral performance during a working memory task remains unsolved. Approach. Multivariate pattern decoding was used to test whether a decoder constructed from two discrete levels of WM load can generalize to produce a continuous measure that predicts task performance. Specifically, a linear regression with L2-regularization was chosen with input features from EEG oscillatory activity recorded from healthy participants while performing the n-back task, . Main results. The feasibility to extract a continuous time-resolved measure that correlates positively to trial-by-trial working memory task performance is demonstrated (r = 0.47, p < 0.05). It is furthermore shown that this measure allows to predict task performance before action (r = 0.49, p < 0.05). We show that the extracted continuous measure enables to study the temporal dynamics of the complex activation pattern of WM encoding during the n-back task. Specifically, temporally precise contributions of different spectral features are observed which extends previous findings of traditional univariate approaches. Significance. These results constitute an important contribution towards a wide range of applications in the field of cognitive brain–machine interfaces. Monitoring mental processes related to attention and WM load to reduce the risk of committing errors in high-risk environments could potentially prevent many devastating consequences or using the continuous measure as neurofeedback opens up new possibilities to develop novel rehabilitation techniques for individuals with degraded WM capacity.

Dynamic pattern decoding of source-reconstructed MEG or EEG data: Perspective of multivariate pattern analysis and signal leakage

Recently, an increasing number of studies have employed multivariate pattern analysis (MVPA) rather than univariate analysis for the dynamic pattern decoding of event-related responses recorded with a MEG/EEG sensor. The use of the MVPA approach for source-reconstructed MEG/EEG data is uncommon. For these data, we need to consider the source orientation information and the signal leakage among brain regions. In the present study, we evaluate the perspective of the MVPA approach in the context of source orientation information and signal leakage in source-reconstructed MEG data. We perform face vs. tool object category decoding (FvsT-OCD) of event-related responses from single or multiple voxels from a brain region using a univariate analysis approach and/or the MVPA approach. We also propose and perform symmetric signal leakage correction of source-reconstructed data using an independent component analysis-based approach. FvsT-OCD using single voxel information shows higher sensitivity for the MVPA approach than univariate analysis, as the MVPA approach efficiently utilizes information on all three dipole orientations and is less affected by inter-subject variability. The MVPA approach shows higher sensitivity for FvsT-OCD when considering information from multiple voxels than for a single voxel in a brain region. This finding suggests that the MVPA approach captures the latent multivoxel distributed pattern. However, the results may be partly or entirely attributable to signal leakage between brain regions, as the sensitivity is substantially reduced after signal leakage correction. A consideration of signal leakage is therefore essential during the evaluation of MVPA outcomes.

Robust pattern decoding in shape-coded structured light

Decoding is a challenging and complex problem in a coded structured light system. In this paper, a robust pattern decoding method is proposed for the shape-coded structured light in which the pattern is designed as grid shape with embedded geometrical shapes. In our decoding method, advancements are made at three steps. First, a multi-template feature detection algorithm is introduced to detect the feature point which is the intersection of each two orthogonal grid-lines. Second, pattern element identification is modelled as a supervised classification problem and the deep neural network technique is applied for the accurate classification of pattern elements. Before that, a training dataset is established, which contains a mass of pattern elements with various blurring and distortions. Third, an error correction mechanism based on epipolar constraint, coplanarity constraint and topological constraint is presented to reduce the false matches. In the experiments, several complex objects including human hand are chosen to test the accuracy and robustness of the proposed method. The experimental results show that our decoding method not only has high decoding accuracy, but also owns strong robustness to surface color and complex textures.

Sequential Test Decompressors with Fast Tester Bits Wide-Spreading

Usually, test pattern decompressors with dynamic reseeding are reset before starting a new test pattern decoding. The first few scan chain slices are then filled with test vectors that have lower decoding ability as the number of free variables is limited by the test access mechanism bandwidth. We have found that even within this limitation, it is possible to improve the decodability by creating fast and wide-spreading as many as possible independent linear combinations of the tester bits and using them for the scan chain loading. We evaluated features influencing the decompression quality and the hardware overhead for different decompressor principles. According to the evaluation results, we proposed a decompressor combining a XOR network and a linear feedback shift register (LFSR)-like automaton; we place the XOR network on the LFSR inputs. We demonstrate that due to this arrangement, the combined decompressor can be used without any phase shifter or state skipping ability of the LFSR. We have experimentally verified that adopting the proposed decompressor structure improves test coverage, saves the hardware resources and shortens the test application time.

Tracking Deceased-Related Thinking With Neural Pattern Decoding of a Cortical-Basal Ganglia Circuit

BackgroundDeceased-related thinking is central to grieving and potentially critical to processing of the loss. Self-report measurements might fail to capture important elements of deceased-related thinking and processing. Here, we used a machine learning approach applied to functional magnetic resonance imaging—known as neural decoding—to develop a measure of ongoing deceased-related processing. MethodsTwenty-three subjects grieving the loss of a first-degree relative, spouse, or partner within 14 months underwent two functional magnetic resonance imaging tasks. They first viewed pictures and stories related to the deceased, a living control person, and a demographic control figure while providing ongoing valence and arousal ratings. Second, they performed a 10-minute Sustained Attention to Response Task with thought probes every 25 to 35 seconds to identify deceased, living, and self-related thoughts. ResultsA conjunction analysis, controlling for valence/arousal, identified neural clusters in basal ganglia, orbital prefrontal cortex, and insula associated with both types of deceased-related stimuli versus the two control conditions in the first task. This pattern was applied to functional magnetic resonance imaging data collected during the Sustained Attention to Response Task and discriminated deceased-related but not living or self-related thoughts, independently of grief severity and time since loss. Deceased-related thoughts on the Sustained Attention to Response Task correlated with self-reported avoidance. The neural model predicted avoidance over and above deceased-related thoughts. ConclusionsA neural pattern trained to identify mental representations of the deceased tracked deceased-related thinking during a sustained attention task and also predicted subject-level avoidance. This approach provides a new imaging tool to be used as an index of processing the deceased for future studies of complicated grief.

EEG pattern decoding of rhythmic individual finger imaginary movements of one hand

The results of four-class classification of the motor imagery EEG patterns corresponding to the right hand finger movements (little finger, thumb, index and middle fingers) of eight healthy subjects are presented in this study. The motor imagery of individual right-hand finger movements was executed by the subjects in a prescribed rhythm and the trials contained no external stimuli. Classification was performed by means of a specially developed two-level committee of classifiers on the basis of support vector machine and artificial neural networks at the first level and by generalizing an artificial neural network at the second level. The area under the EEG signal curve and the curve length calculated in a sliding time window for sites F3, C3, and Cz of the International 10?20 System were selected as the key features of signals from the sensorimotor and adjoining frontal cortical areas contralateral to the movements. The average accuracy of four-class singletrial classification for all subjects was 50 ± 7 [SD] (maximum, 58%) for the pair of sites F3–C3 and 46 ± 11% [SD] (maximum 62%) for the pair of sites C3–Cz with a theoretical guessing level 25%.

Near‐field communication method using smartphone LCD luminescence

In this Letter, a near-field communication method using smart-phone liquid-crystal-display luminescence is proposed. The proposed system is composed of three parts: colour pattern generation, smartphone detection and colour pattern decoding. When the user inputs digital data such as text or numbers, the sender section converts the digital data into RGB colour pattern images using a previously prepared pattern generation method. The receiver section detects these colour pattern images using a camera, extracts the colour values and decodes them into original digital data. The data transfer accuracy to show that this system can work in real situations is tested.

Probing the reaching-grasping network in humans through multivoxel pattern decoding.

The quest for a putative human homolog of the reaching-grasping network identified in monkeys has been the focus of many neuropsychological and neuroimaging studies in recent years. These studies have shown that the network underlying reaching-only and reach-to-grasp movements includes the superior parieto-occipital cortex (SPOC), the anterior part of the human intraparietal sulcus (hAIP), the ventral and the dorsal portion of the premotor cortex, and the primary motor cortex (M1). Recent evidence for a wider frontoparietal network coding for different aspects of reaching-only and reach-to-grasp actions calls for a more fine-grained assessment of the reaching-grasping network in humans by exploiting pattern decoding methods (multivoxel pattern analysis--MVPA). Here, we used MPVA on functional magnetic resonance imaging (fMRI) data to assess whether regions of the frontoparietal network discriminate between reaching-only and reach-to-grasp actions, natural and constrained grasping, different grasp types, and object sizes. Participants were required to perform either reaching-only movements or two reach-to-grasp types (precision or whole hand grasp) upon spherical objects of different sizes. Multivoxel pattern analysis highlighted that, independently from the object size, all the selected regions of both hemispheres contribute in coding for grasp type, with the exception of SPOC and the right hAIP. Consistent with recent neurophysiological findings on monkeys, there was no evidence for a clear-cut distinction between a dorsomedial and a dorsolateral pathway that would be specialized for reaching-only and reach-to-grasp actions, respectively. Nevertheless, the comparison of decoding accuracy across brain areas highlighted their different contributions to reaching-only and grasping actions. Altogether, our findings enrich the current knowledge regarding the functional role of key brain areas involved in the cortical control of reaching-only and reach-to-grasp actions in humans, by revealing novel fine-grained distinctions among action types within a wide frontoparietal network.

Decoding actions at different levels of abstraction.

Brain regions that mediate action understanding must contain representations that are action specific and at the same time tolerate a wide range of perceptual variance. Whereas progress has been made in understanding such generalization mechanisms in the object domain, the neural mechanisms to conceptualize actions remain unknown. In particular, there is ongoing dissent between motor-centric and cognitive accounts whether premotor cortex or brain regions in closer relation to perceptual systems, i.e., lateral occipitotemporal cortex, contain neural populations with such mapping properties. To date, it is unclear to which degree action-specific representations in these brain regions generalize from concrete action instantiations to abstract action concepts. However, such information would be crucial to differentiate between motor and cognitive theories. Using ROI-based and searchlight-based fMRI multivoxel pattern decoding, we sought brain regions in human cortex that manage the balancing act between specificity and generality. We investigated a concrete level that distinguishes actions based on perceptual features (e.g., opening vs closing a specific bottle), an intermediate level that generalizes across movement kinematics and specific objects involved in the action (e.g., opening different bottles with cork or screw cap), and an abstract level that additionally generalizes across object category (e.g., opening bottles or boxes). We demonstrate that the inferior parietal and occipitotemporal cortex code actions at abstract levels whereas the premotor cortex codes actions at the concrete level only. Hence, occipitotemporal, but not premotor, regions fulfill the necessary criteria for action understanding. This result is compatible with cognitive theories but strongly undermines motor theories of action understanding.

Planning Ahead: Object-Directed Sequential Actions Decoded from Human Frontoparietal and Occipitotemporal Networks.

Object-manipulation tasks (e.g., drinking from a cup) typically involve sequencing together a series of distinct motor acts (e.g., reaching toward, grasping, lifting, and transporting the cup) in order to accomplish some overarching goal (e.g., quenching thirst). Although several studies in humans have investigated the neural mechanisms supporting the planning of visually guided movements directed toward objects (such as reaching or pointing), only a handful have examined how manipulatory sequences of actions-those that occur after an object has been grasped-are planned and represented in the brain. Here, using event-related functional MRI and pattern decoding methods, we investigated the neural basis of real-object manipulation using a delayed-movement task in which participants first prepared and then executed different object-directed action sequences that varied either in their complexity or final spatial goals. Consistent with previous reports of preparatory brain activity in non-human primates, we found that activity patterns in several frontoparietal areas reliably predicted entire action sequences in advance of movement. Notably, we found that similar sequence-related information could also be decoded from pre-movement signals in object- and body-selective occipitotemporal cortex (OTC). These findings suggest that both frontoparietal and occipitotemporal circuits are engaged in transforming object-related information into complex, goal-directed movements.

A Robust Surface Coding Method for Optically Challenging Objects Using Structured Light

Though the structured light measurement system has been successfully applied to the profi le measurement of diffuse ob- jects, it is still a challenge to measure shiny objects due to the mix of both specular and diffuse reflections. To this end, we propose a robust encoding and decoding method in this paper. First, the monochromatic stripe patterns are utilized to eliminate the effect of texture and color of objects. Second, an intensity mask, dynam- ically adjusting the intensity of a projected pattern, is applied to avoid overexposure without any pre-knowledge of the workpiece. Thus, it is more flexible and efficient, compared with the existing methods. Third, to solve the internal reflection of the shiny part, an extrapolation model, combined with the intensity mask, is de- veloped to detect the stripe edge for pattern decoding, resulting in accurate and robust3Dreconstruction. Comparedwith traditional polarization based methods, it does not need to readjust for a new part. The experimental results show that the proposed method is capable of measuring various parts without surface pretreatment. Note to Practitioners—In the manufacturing industry, there is a demand to measure workpiece for quality control. A Coordi- nate Measurement Machine (CMM) is usually used. However, its pointwise measurement makes the inspection time consuming and cannotdo100%inspectionforalltheparts.Vision-based3Drecon- struction method could measure the parts quickly. But the method isweakagainstopticallychallengingobjects.Tothisend,thispaper proposed a robust surface coding method to measure the optically challenging object via a vision-based method.

Multiclass fMRI data decoding and visualization using supervised self-organizing maps

When multivariate pattern decoding is applied to fMRI studies entailing more than two experimental conditions, a most common approach is to transform the multiclass classification problem into a series of binary problems. Furthermore, for decoding analyses, classification accuracy is often the only outcome reported although the topology of activation patterns in the high-dimensional features space may provide additional insights into underlying brain representations. Here we propose to decode and visualize voxel patterns of fMRI datasets consisting of multiple conditions with a supervised variant of self-organizing maps (SSOMs). Using simulations and real fMRI data, we evaluated the performance of our SSOM-based approach. Specifically, the analysis of simulated fMRI data with varying signal-to-noise and contrast-to-noise ratio suggested that SSOMs perform better than a k-nearest-neighbor classifier for medium and large numbers of features (i.e. 250 to 1000 or more voxels) and similar to support vector machines (SVMs) for small and medium numbers of features (i.e. 100 to 600voxels). However, for a larger number of features (>800voxels), SSOMs performed worse than SVMs. When applied to a challenging 3-class fMRI classification problem with datasets collected to examine the neural representation of three human voices at individual speaker level, the SSOM-based algorithm was able to decode speaker identity from auditory cortical activation patterns. Classification performances were similar between SSOMs and other decoding algorithms; however, the ability to visualize decoding models and underlying data topology of SSOMs promotes a more comprehensive understanding of classification outcomes. We further illustrated this visualization ability of SSOMs with a re-analysis of a dataset examining the representation of visual categories in the ventral visual cortex (Haxby et al., 2001). This analysis showed that SSOMs could retrieve and visualize topography and neighborhood relations of the brain representation of eight visual categories. We conclude that SSOMs are particularly suited for decoding datasets consisting of more than two classes and are optimally combined with approaches that reduce the number of voxels used for classification (e.g. region-of-interest or searchlight approaches).

Real-time fMRI pattern decoding and neurofeedback using FRIEND: an FSL-integrated BCI toolbox.

The demonstration that humans can learn to modulate their own brain activity based on feedback of neurophysiological signals opened up exciting opportunities for fundamental and applied neuroscience. Although EEG-based neurofeedback has been long employed both in experimental and clinical investigation, functional MRI (fMRI)-based neurofeedback emerged as a promising method, given its superior spatial resolution and ability to gauge deep cortical and subcortical brain regions. In combination with improved computational approaches, such as pattern recognition analysis (e.g., Support Vector Machines, SVM), fMRI neurofeedback and brain decoding represent key innovations in the field of neuromodulation and functional plasticity. Expansion in this field and its applications critically depend on the existence of freely available, integrated and user-friendly tools for the neuroimaging research community. Here, we introduce FRIEND, a graphic-oriented user-friendly interface package for fMRI neurofeedback and real-time multivoxel pattern decoding. The package integrates routines for image preprocessing in real-time, ROI-based feedback (single-ROI BOLD level and functional connectivity) and brain decoding-based feedback using SVM. FRIEND delivers an intuitive graphic interface with flexible processing pipelines involving optimized procedures embedding widely validated packages, such as FSL and libSVM. In addition, a user-defined visual neurofeedback module allows users to easily design and run fMRI neurofeedback experiments using ROI-based or multivariate classification approaches. FRIEND is open-source and free for non-commercial use. Processing tutorials and extensive documentation are available.

A Dendritic Mechanism for Decoding Traveling Waves: Principles and Applications to Motor Cortex

Traveling waves of neuronal oscillations have been observed in many cortical regions, including the motor and sensory cortex. Such waves are often modulated in a task-dependent fashion although their precise functional role remains a matter of debate. Here we conjecture that the cortex can utilize the direction and wavelength of traveling waves to encode information. We present a novel neural mechanism by which such information may be decoded by the spatial arrangement of receptors within the dendritic receptor field. In particular, we show how the density distributions of excitatory and inhibitory receptors can combine to act as a spatial filter of wave patterns. The proposed dendritic mechanism ensures that the neuron selectively responds to specific wave patterns, thus constituting a neural basis of pattern decoding. We validate this proposal in the descending motor system, where we model the large receptor fields of the pyramidal tract neurons — the principle outputs of the motor cortex — decoding motor commands encoded in the direction of traveling wave patterns in motor cortex. We use an existing model of field oscillations in motor cortex to investigate how the topology of the pyramidal cell receptor field acts to tune the cells responses to specific oscillatory wave patterns, even when those patterns are highly degraded. The model replicates key findings of the descending motor system during simple motor tasks, including variable interspike intervals and weak corticospinal coherence. By additionally showing how the nature of the wave patterns can be controlled by modulating the topology of local intra-cortical connections, we hence propose a novel integrated neuronal model of encoding and decoding motor commands.

Structured-light stereo: Comparative analysis and integration of structured-light and active stereo for measuring dynamic shape

In computer vision, two major active range imaging methods have been frequently employed for rapid and efficient shape recovery: (a) conventional active stereo vision and (b) conventional structured-light vision. This paper presents a comparative analysis and an integration of the two active approaches, namely, a structured-light stereo approach for the acquisition of dynamic shape. We first investigate the strengths and weaknesses of the two approaches in terms of accuracy, computational cost, field of view, depth of field, and color sensitivity. Based on this analysis, we propose a novel integrated method, the structured-light stereo, to recover dynamic shapes from a wider view with less occlusion by taking most of the benefits of the two approaches. The main idea is as follows. We first build a system composed of two cameras and a single projector (just a basic setup for conventional active stereo), and the projector projects a single “one-shot” color-stripe pattern. The next step is to estimate reliable correspondences between each camera and the projector via an accurate and efficient pattern decoding technique, and some remaining unresolved regions are explored by a stereo matching technique, which is less sensitive to object surface colors and defocus due to the projector's short depth of field, to estimate additional correspondences. We demonstrate the efficacy of the integrated method through experimental results.

Pattern classification using functional magnetic resonance imaging.

Over the past decade, pattern classification methods have become widespread in functional magnetic resonance imaging (fMRI). These methods, typically referred to as multivoxel pattern analysis (MVPA) or multivariate pattern decoding, are now applied to a wide range of neuroscientific questions. There has been particular interest in applying these approaches, e.g., in detecting deception or for diagnostic purposes. In this review, we will focus on what can be achieved by pattern classification analyses of fMRI data; the strengths and weaknesses of this approach; and the biological processes giving rise to the signals measured by this method. Finally, we will discuss how these multivariate approaches are starting to be applied to the analysis of anatomical magnetic resonance imaging (MRI) and magnetoencephalographic (MEG) data. WIREs Cogni Sci 2011 2 568-579 DOI: 10.1002/wcs.141 For further resources related to this article, please visit the WIREs website.

Decoding of coherent but not incoherent motion signals in early dorsal visual cortex

When several scattered grating elements are arranged in such a way that their directions of motion are consistent with a common path, observers perceive them as belonging to a globally coherent moving object. Here we investigated how this coherence changes the representation of motion signals in human visual cortex using functional magnetic resonance imaging (fMRI) and multivariate voxel pattern decoding, which have the potential to reveal how well a stimulus is encoded in different contexts. Only during globally coherent motion was it possible to reliably distinguish fMRI signals evoked by different directions of motion in early visual cortex. This effect was specific to the retinotopic representation of the visual field quadrant in V1 traversed by the coherent element path and could not simply be attributed to a general increase in signal strength. Decoding was more reliable for cortical areas corresponding to the lower visual field. Because some previous studies observed poorer speed discrimination when motion was grouped, we also conducted behavioural experiments to investigate this with our stimuli, but did not reveal a consistent relationship between coherence and perceived speed. Taken together, these data show that neuronal populations in early visual cortex represent information that could be used for interpreting motion signals as unified objects.