Limits...
Effective connectivity during animacy perception--dynamic causal modelling of Human Connectome Project data.

Hillebrandt H, Friston KJ, Blakemore SJ - Sci Rep (2014)

Bottom Line: Predictions about animate motion - relative to inanimate motion - should result in prediction error and increase signal passing from lower level sensory area MT+/V5, which is responsive to all motion, to higher-order posterior superior temporal sulcus (pSTS), which is selectively activated by animate motion.We found that forward connectivity from V5 to the pSTS increased, and inhibitory self-connection in the pSTS decreased, when viewing intentional motion versus inanimate motion.These prediction errors associated with animate motion may be the cause for increased attention to animate stimuli found in previous studies.

View Article: PubMed Central - PubMed

Affiliation: 1] Institute of Cognitive Neuroscience, University College London, London, WC1N 3AR, United Kingdom [2] Moral Cognition Laboratory, Department of Psychology, Harvard University, Cambridge, MA, 02138, United States.

ABSTRACT
Biological agents are the most complex systems humans have to model and predict. In predictive coding, high-level cortical areas inform sensory cortex about incoming sensory signals, a comparison between the predicted and actual sensory feedback is made, and information about unpredicted sensory information is passed forward to higher-level areas. Predictions about animate motion - relative to inanimate motion - should result in prediction error and increase signal passing from lower level sensory area MT+/V5, which is responsive to all motion, to higher-order posterior superior temporal sulcus (pSTS), which is selectively activated by animate motion. We tested this hypothesis by investigating effective connectivity in a large-scale fMRI dataset from the Human Connectome Project. 132 participants viewed animations of triangles that were designed to move in a way that appeared animate (moving intentionally), or inanimate (moving in a mechanical way). We found that forward connectivity from V5 to the pSTS increased, and inhibitory self-connection in the pSTS decreased, when viewing intentional motion versus inanimate motion. These prediction errors associated with animate motion may be the cause for increased attention to animate stimuli found in previous studies.

Show MeSH
The left graph shows the range of log-posterior probabilities of all possible models examined.The right graph shows the posterior probabilities of all models. Model 1024 had the highest posterior probability of (almost) 1. This graph shows data for the first session and the right hemisphere, results for other sessions and hemispheres were similar.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4150124&req=5

f5: The left graph shows the range of log-posterior probabilities of all possible models examined.The right graph shows the posterior probabilities of all models. Model 1024 had the highest posterior probability of (almost) 1. This graph shows data for the first session and the right hemisphere, results for other sessions and hemispheres were similar.

Mentions: We first assessed the model with the best evidence (a metric in which model fit is traded off against model complexity). Comparisons of the evidence for all possible 1024 models showed that the winning (optimal) model with the highest probability had a probability of (almost) 1 (Figure 5). The winning model in all four cases (2 (sessions) × 2 (hemispheres)) was always the full model that had all connections and all modulations (Figure 2A). This model has 10 free parameters describing the extrinsic and intrinsic connections and how these connections change with perceptual set. The profile of model (log) evidences over the ensuing 1024 models for one of the four cases is shown in Figure 5 (other plots were similar), suggesting that the full model had more evidence than any reduced variant (the probability was almost 1; the log-probability was almost 0). The next most probable model's probability was very low (almost 0; the log-probability was −70.9). The resulting Bayes factor, which can be obtained by dividing the winning model's probability (almost 1) by the next probable model's probability (almost 0), is considered decisive evidence for the winning model (corresponding to a highly significant difference)43. For comparison, even a Bayes Factor of 3:1 would still be considered positive evidence43. Additionally, this full connectivity was confirmed using several family-level inferences33, where families of models with certain parameters (such as connections or modulations existing) were compared with families of models without those parameters. This also showed that all (self) connections and their modulation by animacy were evident with a posterior probability of (almost) 1. The fact that we obtained strong evidence for all effects reflects the large sample size and high signal to noise ratio. Although there was very high evidence for the existence of all (self) connections and their modulations, this evidence cannot speak to the relative strength of forward versus backward connection (effect sizes). Therefore, we compared the relative strength of effective connectivity – under the winning model – in order to address specific hypotheses about the locus of animacy effects in quantitative terms.


Effective connectivity during animacy perception--dynamic causal modelling of Human Connectome Project data.

Hillebrandt H, Friston KJ, Blakemore SJ - Sci Rep (2014)

The left graph shows the range of log-posterior probabilities of all possible models examined.The right graph shows the posterior probabilities of all models. Model 1024 had the highest posterior probability of (almost) 1. This graph shows data for the first session and the right hemisphere, results for other sessions and hemispheres were similar.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4150124&req=5

f5: The left graph shows the range of log-posterior probabilities of all possible models examined.The right graph shows the posterior probabilities of all models. Model 1024 had the highest posterior probability of (almost) 1. This graph shows data for the first session and the right hemisphere, results for other sessions and hemispheres were similar.
Mentions: We first assessed the model with the best evidence (a metric in which model fit is traded off against model complexity). Comparisons of the evidence for all possible 1024 models showed that the winning (optimal) model with the highest probability had a probability of (almost) 1 (Figure 5). The winning model in all four cases (2 (sessions) × 2 (hemispheres)) was always the full model that had all connections and all modulations (Figure 2A). This model has 10 free parameters describing the extrinsic and intrinsic connections and how these connections change with perceptual set. The profile of model (log) evidences over the ensuing 1024 models for one of the four cases is shown in Figure 5 (other plots were similar), suggesting that the full model had more evidence than any reduced variant (the probability was almost 1; the log-probability was almost 0). The next most probable model's probability was very low (almost 0; the log-probability was −70.9). The resulting Bayes factor, which can be obtained by dividing the winning model's probability (almost 1) by the next probable model's probability (almost 0), is considered decisive evidence for the winning model (corresponding to a highly significant difference)43. For comparison, even a Bayes Factor of 3:1 would still be considered positive evidence43. Additionally, this full connectivity was confirmed using several family-level inferences33, where families of models with certain parameters (such as connections or modulations existing) were compared with families of models without those parameters. This also showed that all (self) connections and their modulation by animacy were evident with a posterior probability of (almost) 1. The fact that we obtained strong evidence for all effects reflects the large sample size and high signal to noise ratio. Although there was very high evidence for the existence of all (self) connections and their modulations, this evidence cannot speak to the relative strength of forward versus backward connection (effect sizes). Therefore, we compared the relative strength of effective connectivity – under the winning model – in order to address specific hypotheses about the locus of animacy effects in quantitative terms.

Bottom Line: Predictions about animate motion - relative to inanimate motion - should result in prediction error and increase signal passing from lower level sensory area MT+/V5, which is responsive to all motion, to higher-order posterior superior temporal sulcus (pSTS), which is selectively activated by animate motion.We found that forward connectivity from V5 to the pSTS increased, and inhibitory self-connection in the pSTS decreased, when viewing intentional motion versus inanimate motion.These prediction errors associated with animate motion may be the cause for increased attention to animate stimuli found in previous studies.

View Article: PubMed Central - PubMed

Affiliation: 1] Institute of Cognitive Neuroscience, University College London, London, WC1N 3AR, United Kingdom [2] Moral Cognition Laboratory, Department of Psychology, Harvard University, Cambridge, MA, 02138, United States.

ABSTRACT
Biological agents are the most complex systems humans have to model and predict. In predictive coding, high-level cortical areas inform sensory cortex about incoming sensory signals, a comparison between the predicted and actual sensory feedback is made, and information about unpredicted sensory information is passed forward to higher-level areas. Predictions about animate motion - relative to inanimate motion - should result in prediction error and increase signal passing from lower level sensory area MT+/V5, which is responsive to all motion, to higher-order posterior superior temporal sulcus (pSTS), which is selectively activated by animate motion. We tested this hypothesis by investigating effective connectivity in a large-scale fMRI dataset from the Human Connectome Project. 132 participants viewed animations of triangles that were designed to move in a way that appeared animate (moving intentionally), or inanimate (moving in a mechanical way). We found that forward connectivity from V5 to the pSTS increased, and inhibitory self-connection in the pSTS decreased, when viewing intentional motion versus inanimate motion. These prediction errors associated with animate motion may be the cause for increased attention to animate stimuli found in previous studies.

Show MeSH