Presentation is loading. Please wait.

Presentation is loading. Please wait.

Pattern Classification of Attentional Control States S. G. Robison, D. N. Osherson, K. A. Norman, & J. D. Cohen Dept. of Psychology, Princeton University,

Similar presentations


Presentation on theme: "Pattern Classification of Attentional Control States S. G. Robison, D. N. Osherson, K. A. Norman, & J. D. Cohen Dept. of Psychology, Princeton University,"— Presentation transcript:

1 Pattern Classification of Attentional Control States S. G. Robison, D. N. Osherson, K. A. Norman, & J. D. Cohen Dept. of Psychology, Princeton University, Princeton, NJ 08544 Background and Research Objectives Guided Activation Theory Guided Activation Theory of prefrontal function posits a key role for the frontal lobes in goal-directed behavior (Miller & Cohen, 2001). Claims of Theory PFC (a) represents task information, (b) guides posterior structures to process task-relevant stimuli, & (c) maintains representations in the face of distraction until task completion. Current Objectives Use multi-voxel pattern classification (e.g., Polyn et al., 2005) to detect and track the timecourse of distinct task states. Assess whether patterns corresponding to top-down control states are maintained during task execution. Long-term Objectives Compare role of PFC in attention to simple stimuli to more complex, ethologically-relevant dimensions (e.g., age, race, & expression of faces). Experimental Paradigm Overview Participants completed 6 runs of a delayed match to sample task. Participants viewed stimuli with distinct color, shape, & motion values. Matching Task: Selective Attention Color: Did you see an exact color match? Shape: Did you see the same shape of the configuration of dots? Motion: Did you see a stimulus that moved in the same direction? Dependent Variables: whole-brain event-related fMRI images, behavioral accuracy, & reaction times. Procedure Dimension Cue Feature Cue Search Epoch Response Trials consisted of dimension maintenance (i.e., remember task), feature maintenance (i.e., remember specific value), and search task epochs. Participants completed 18 pseudo-randomized trials per run. Each trial consisted of a cue stimulus & 5 probe stimuli (shown for 500ms) 3 of the probes matched the cue on one or both irrelevant dimensions. Real matches were less frequent than these “distractor” matches. Importance Maps: Feature Maintenance Which voxels did the classifier use to determine task? Distributed Voxel Pattern Color Shape Motion Color: Middle frontal gyrus, left BA 44, precuneus, insula. Shape: Inf. parietal cortex/BA40, (-) lft. precuneus, LOC/BA19 Motion: Sup. temporal gyrus, inf. parietal & frontal cortex, MT. Classification of Color, Shape, & Motion Tasks Research Question Determine whether the classifier can dissociate color, shape, and motion tasks during different task epochs. Performance is measured by correlating task and active output units. Results Feature Maintenance Search Epoch Open Questions and Future Directions What do these results suggest about nature of frontal task representations? Is PFC only necessary for ignoring distractions and learning new tasks? How can we tease apart top-down (task) and bottom-up (probe) activity? Can we characterize interactions between frontal and parietal attentional networks? Next steps: collect more data, introduce distraction during delay. Compare task representations for more complex dimensions (e.g., faces). Task Maintenance Over Search Epoch Hypothesis and Procedure The guided activation theory claims that the frontal cortex maintains task- related representations during task execution. Test this claim by using the feature maintenance classifier to predict search epoch activity for data from frontal and posterior structures. Results Acknowledgements Special thanks to the participants in this study as well members of the Norman and Cohen labs and Sabine Kastner at Princeton University for their help with paradigm development and analyses. Multi-Voxel Pattern Classification Pre-Classification Procedure Data were preprocessed using AFNI, and classification analyses were conducted using the Princeton Multi-Voxel Pattern Analysis Toolkit (currently in public beta testing: www.csbmb.princeton.edu/mvpa) An ANOVA was applied to individual voxels to identify those that showed variance associated with attention to color, shape, and motion tasks. Training and Testing the Network Train a neural network classifier using the backpropagation algorithm to discriminate between brain volumes that correspond to different cognitive states (e.g., Polyn et al., 2005) Subject-by-subject analysis; significance of correlations assessed with non-parametric stats Assess classifier’s performance at determining whether participants paid attention to color, shape, or motion dimensions for individual time points (TR = 2 sec) & compare results for classifiers trained on data from the whole-brain, frontal and posterior structures. N-1 generalization procedure: to avoid training and testing on the same data only a subset of data is used to train the classifier, & the withheld portion is used for testing. Funding sources: R01MH052864, NSF graduate fellowship to SGR References Corbetta, M. et al. (1990). Attentional modulation of neural processing of shape, color, and velocity in humans. Science, 248, 1556-1559. Kastner, S., & Ungerleider, L. G. (2000). Mechanisms of visual attention in the human cortex. Annual Reviews Neuroscience, 23, 315-341. Miller E. K., & Cohen, J. D. (2001). An integrated theory of prefrontal cortex function. Annual Reviews Neuroscience, 24, 167-202. Polyn, S. M., Natu, V. S., Cohen, J. D, & Norman, K. A. (2005). Category-specific cortical activity precedes retrieval during memory search. Science, 310, 1963-1966. SfN Program Number: 367.2/II11; Presentation Time: Monday, Oct. 16, 9-10am Did you see a color match? Variable Interval Variable Interval Whole BrainFrontal CortexPosterior CortexWhole BrainFrontal CortexPosterior Cortex Participant 10.6796, p < 0.01 0.3678, p < 0.01 0.7283, p < 0.01 0.6488, p < 0.01 0.2480, p < 0.01 0.6771, p < 0.01 Participant 20.2211, p > 0.05 -0.0979, p > 0.05 0.2977, p < 0.05 0.2120, p > 0.05 0.0244, p > 0.05 0.2851, p < 0.05 Participant 30.6917, p < 0.01 0.3060, p < 0.01 0.7301, p < 0.01 0.6067, p < 0.01 0.2815, p < 0.01 0.6211, p < 0.01 Participant 40.0922, p > 0.05 -0.0933, p > 0.05 0.1420, p > 0.05 0.1629, p > 0.05 -0.0886, p > 0.05 0.1722, p > 0.05 Whole BrainFrontal CortexPosterior Cortex Participant 10.4186, p < 0.01 0.1617, p < 0.05 0.4393, p < 0.01 Participant 20.1140, p > 0.05 -0.0315, p > 0.05 0.1668, p > 0.05 Participant 30.5311, p < 0.01 0.2481, p < 0.05 0.5339, p < 0.01 Participant 40.1107, p > 0.05 0.1079, p > 0.05 0.0787, p > 0.05 This is a color trial - 2 classifiable cases - No difference between frontal and posterior analyses - Event-related avgs. from classifiable subjects reveal maintenance through- out search task epoch


Download ppt "Pattern Classification of Attentional Control States S. G. Robison, D. N. Osherson, K. A. Norman, & J. D. Cohen Dept. of Psychology, Princeton University,"

Similar presentations


Ads by Google