|
|
|
|
Cortical Processes Underlying Visual Target Recognition
Background: Current models of visual object recognition propose information flows through a series of feed-forward processing stages in which low level features are extracted from a visual scene, then integrated under constraints imposed by adjacent and top-down connections. The true nature of cortical circuits responsible for perception and recognition remains a mystery. In fact there is much debate as to whether recognition relies on information flow through cortico-cortical feedback loops or rather one feed-forward sweep through the visual system.
While direct functional imaging of cortical circuits is not yet feasible, indirect evidence from single unit recordings, event related potential and psychophysical studies describe macroscopic cortical regions comprising the visual system in terms of both anatomical spatial constraints and functional temporal constraints. The challenge in any such study is designing experiments that tease apart cortical processing stages involved with object recognition and delineate the spatial extent and temporal order, latency, duration, and influence of each stage in response to specific classes of visual stimuli.
Approach: One experimental task that simulates natural saccadic scene acquisition is Rapid Serial Visual Presentation (RSVP) . During an RSVP task a continuous sequence of images is presented in a static location. We have recently reported single-trial detection of spatial signatures in EEG related to visual target recognition within 200 milliseconds of image onset during an RSVP task. This RSVP task differed from Thorpe’s original experiment in that subjects were asked to detect target images within sequences (“barrages”) of 100 images that had a 50% chance of containing a single target image. Target images consisted of a person/people comprising no more than 25% of a natural scene while distractor images were natural scenes. Subjects were instructed to press a button at the beginning of a sequence and release it if a target appeared. An example of an abbreviated RSVP sequence can be seen below.
Figure 1: An example of an RSVP sequence
Results: EEG from target and distractor trials was compared on a single-trial basis using linear discriminant analysis and a forward linear model was used to determine sensor projections of the discriminating source activity. This forward model indicated that discriminating activity began approximately 200 ms following image presentation moving anteriorly over sensory motor areas 300-400 ms following image presentation. Since these signatures are learned/detected single-trial, it is possible to analyze variability between trials as well as determine classification performance on new trials.
Figure 2: Neural activity becomes strongly response locked during transition from far-frontal to parietal activity. The animation shows the evolution of discriminating activity as the training window (outlined with red sigmoids) is translated. Each line in the image on the left corresponds to a trial. Trials are sorted by and locked to response time (vertical black line). Note the profile of the training window is determined by finding scaled stimulus times (first black sigmoid) that yield maximum discrimination performance, quantified as area under an ROC curve (Az). The translating black line is the projection of stimulus times onto single-trial peak latencies (black dots). Bars shown on either side of the scalp plot indicate degree of response locking of the component (RL%, left) and discrimination performance (Az, right). These bars change from red to green if the Az is significant (p < 0.01).
Conclusions: Our approach illustrates how a simple linear classifier based on spatial weighting of EEG suggests a forward model that enables approximate localization of discriminating activity. By sliding the window used to train a linear discriminator, we are able to study the temporal sequence of neuronal responses evoked by visual stimuli. Due to the high temporal resolution afforded by EEG this method provides an intuitive description of communication between visual and sensorimotor cortex.
Such reports provide vague evidence that object recognition is achieved through a series of activations in distinct cortical regions and places temporal limits on processing time in the absence of motor response. As of yet studies do not indicate the degree of interaction between these processing stages, or the spatial extent of each region, let alone the nature of the underlying cortical circuits. Our recent exploration with simultaneous EEG/fMRI should provide additional clues, however it is clear that more sophisticated experimental designs and imaging modalities are necessary to clarify the nature of neural activity responsible for human object recognition.
OTHER PROJECTS
DOWNLOADS
PUBLICATIONS A. Gerson, L. Parra, P. Sajda (2005) Cortical Origins of Response Time Variability During Rapid Discrimination of Visual Objects, NeuroImage, 28 (2) 342-353. L. C. Parra, C. D. Spence, A. D. Gerson and P. Sajda (2005) Recipes for the Linear Analysis of EEG, NeuroImage, 28 (2) 326-341. A. Gerson, L. Parra, P. Sajda (2006) Single-trial Analysis of EEG for Enabling Cognitive User Interfaces, IEEE Neural Engineering Handbook, Ed. Metin Akay, Wiley/IEEE, in press. A. Luo, A. Gerson, P. Sajda (2004) Comparison of Supervised and Unsupervised Linear Methods for Recovering Task-Relevant Activity in EEG, Proceedings of the IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Arlington, VA, April 15-18, 2004, pp. 1012-1015. A. C. Tang, C. J. McKinney, M. T. Sutherland, L. Parra, A. Gerson, and P. Sajda (2004) Extraction of Single-Trial ERPs from Frontal and Visual Cortex During Video Game Play Despite Continous Free Eye Movement, International Society for Brain Electromagnetic Tomography. P. Sajda, A. Gerson and L. Parra (2003) Spatial Signatures of Visual Object Recognition Events Learned from Single-trial Analysis of EEG, IEEE Engineering in Medicine and Biology Annual Meeting, Cancun, MEXICO, vol. 3, 2087-2090.
P.
Sajda, A. Gerson and L.C. Parra (2003) High-throughput Image Search
via Single-trial Event Detection in a Rapid Serial Visual
Presentation Task, IEEE Conference on Neural Engineering Capri
Island, Italy, March 20-22, 2003, 7-10.
L.
Parra, C. Spence, A. Gerson and P. Sajda, (2003) Response Error
Correction - A Demonstration of Improved Human-Machine Performance
Using Real-Time EEG Monitoring. IEEE Transactions on Neural
Systems and Rehabilitation Engineering, 11 (2) 173-177.
P.
Sajda, A. Gerson, K-R Mueller, B. Blankertz and L. Parra,(2003) A
Data Analysis Competition to Evaluate Machine Learning Algorithms
for use in Brain-Computer Interfaces.
IEEE Transactions on Neural Systems and Rehabilitation
Engineering, 11 (2) 184-185.
Abstracts A. Gerson, L. Parra and P. Sajda (2005) Real-time Image Triage based on Single-trial Detection of Visual Recognition and Discrimination Events in EEG, 3rd International Meeting on Brain-Computer Interface Technology, Rensselaerville, NY, June 2005. A. Gerson, D. Friedman, P. Sajda (2005) Imaging differences in cortical function between young and aging populations using single-trial analysis of EEG, Society for Neuroscience 2005, Washington D.C. R. Goldman, A. Gerson, M. Cohen, T. Brown and P. Sajda (2005) Simultaneous EEG and fMRI for Event Related Studies, Human Brain Mapping 2005, Toronto A. Gerson, L. Parra, P. Sajda (2004) Assessing Asymmetry in Behavioral Response and Associated Neural Activity for a Rapid Serial Visual Presentation Task, Society for Neuroscience 2004, San Diego. L. Parra, A. Gerson, P. Sajda (2004) Origins of Response Time Variability in a Rapid Serial Visual Presentation Task, Computational and Systems Neuroscience 2004, Cold Spring Harbor Laboratory. A. Gerson, L. Parra and P. Sajda (2003) Single-trial Event Detection of Visual Object Recognition in EEG, Human Brain Mapping 2003, New York. A. C. Tang, C. J. McKinney, M. T. Sutherland, L. Parra, B. C. Reeb, N. A. Malaszenko, A. Gerson, P. Sajda (2003) Source Localization from High Density EEG Data During a Real World Task, Society for Neuroscience 2003, New Orleans A. Gerson and P. Sajda (2002) Single-trial Denoising of EEG with a Wavelet Domain Hidden Markov Tree, 2nd International Workshop on Brain Computer Interfaces, Rensselaerville NY, June 2002.
|
