Limits...
The neural representation of Arabic digits in visual cortex.

Peters L, De Smedt B, Op de Beeck HP - Front Hum Neurosci (2015)

Bottom Line: In this study, we investigated how Arabic digits are represented in the visual cortex, and how their representation changes throughout the ventral visual processing stream, compared to the representation of letters.However, the activity in this region might have been confounded by string length-number words contain more characters than digits.We found an alteration in representations throughout the ventral processing stream from clustering based on amount of visual information in primary visual cortex (V1) towards clustering based on symbolic stimulus category higher in the visual hierarchy.

View Article: PubMed Central - PubMed

Affiliation: Parenting and Special Education Research Unit, KU Leuven Leuven, Belgium, Europe.

ABSTRACT
In this study, we investigated how Arabic digits are represented in the visual cortex, and how their representation changes throughout the ventral visual processing stream, compared to the representation of letters. We probed these questions with two functional magnetic resonance imaging (fMRI) experiments. In Experiment 1, we explored whether we could find brain regions that were more activated for digits than for number words in a subtraction task. One such region was detected in lateral occipital cortex. However, the activity in this region might have been confounded by string length-number words contain more characters than digits. We therefore conducted a second experiment in which string length was systematically controlled. Experiment 2 revealed that the findings of the first experiment were task dependent (as it was only observed in a task in which numerosity was relevant) or stimulus dependent (as it was only observed when the number of characters of a stimulus was not controlled). We further explored the characteristics of the activation patterns for digit and letter strings across the ventral visual processing stream through multi-voxel pattern analyses. We found an alteration in representations throughout the ventral processing stream from clustering based on amount of visual information in primary visual cortex (V1) towards clustering based on symbolic stimulus category higher in the visual hierarchy. The present findings converge to the conclusion that in the ventral visual system, as far as can be detected with fMRI, the distinction between Arabic digits and letter strings is represented in terms of distributed patterns rather than separate regions.

No MeSH data available.


Schematic presentation of a univariate (A) and multivariate correlational analysis (B). In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations over conditions. In multi-voxel correlational analyses, we correlated the patterns of activation of all conditions.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4585091&req=5

Figure 5: Schematic presentation of a univariate (A) and multivariate correlational analysis (B). In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations over conditions. In multi-voxel correlational analyses, we correlated the patterns of activation of all conditions.

Mentions: In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations (beta values) over conditions (see Figure 5A). In the multi-voxel correlational analyses, we divided the dataset into two halves, and correlated the patterns of activation of all conditions of the first half of the data with the second half, in the delineated regions of interest. This cycle of dividing data and correlating patterns was repeated 100 times; the correlations reported below are the average correlations over those repetitions, which were then transformed via a Fisher-z transformation, and were finally averaged over all subjects. In the multi-voxel patterns, the activation of each voxel for each condition was expressed in terms of the beta value of that condition subtracted by the mean beta value across all experimental conditions (“cocktail blank normalization”). Because of the normalization, positive correlations between their activity patterns in a certain brain region indicate more similarity between the corresponding conditions (see Figure 5B). The main advantage of multivariate analyses is that they can reveal differences between conditions that are possibly averaged out in univariate analyses (Norman et al., 2006). To visualize the results obtained from the multivariate correlational analyses, we performed multidimensional scaling (MDS) on the obtained averaged correlation matrices. MDS visualizes the similarity of conditions in 2D-space, with conditions that are represented similarly, and hence have higher correlated activation patterns, presented closer together. Conditions that are represented more distinctly (lower correlated activation patterns) will be shown further apart in the MDS visualization. We also determined the coordinates of the conditions in the MDS plots for each individual subject. These coordinates were then rotated using a Procrustes analysis, to fit the space of the MDS plots of the average correlation matrix. The rotated coordinates of each subject are used as error bars in the MDS plots.


The neural representation of Arabic digits in visual cortex.

Peters L, De Smedt B, Op de Beeck HP - Front Hum Neurosci (2015)

Schematic presentation of a univariate (A) and multivariate correlational analysis (B). In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations over conditions. In multi-voxel correlational analyses, we correlated the patterns of activation of all conditions.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4585091&req=5

Figure 5: Schematic presentation of a univariate (A) and multivariate correlational analysis (B). In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations over conditions. In multi-voxel correlational analyses, we correlated the patterns of activation of all conditions.
Mentions: In univariate analyses, we averaged the brain activation per condition over all the voxels in a certain region of interest, and compared these mean activations (beta values) over conditions (see Figure 5A). In the multi-voxel correlational analyses, we divided the dataset into two halves, and correlated the patterns of activation of all conditions of the first half of the data with the second half, in the delineated regions of interest. This cycle of dividing data and correlating patterns was repeated 100 times; the correlations reported below are the average correlations over those repetitions, which were then transformed via a Fisher-z transformation, and were finally averaged over all subjects. In the multi-voxel patterns, the activation of each voxel for each condition was expressed in terms of the beta value of that condition subtracted by the mean beta value across all experimental conditions (“cocktail blank normalization”). Because of the normalization, positive correlations between their activity patterns in a certain brain region indicate more similarity between the corresponding conditions (see Figure 5B). The main advantage of multivariate analyses is that they can reveal differences between conditions that are possibly averaged out in univariate analyses (Norman et al., 2006). To visualize the results obtained from the multivariate correlational analyses, we performed multidimensional scaling (MDS) on the obtained averaged correlation matrices. MDS visualizes the similarity of conditions in 2D-space, with conditions that are represented similarly, and hence have higher correlated activation patterns, presented closer together. Conditions that are represented more distinctly (lower correlated activation patterns) will be shown further apart in the MDS visualization. We also determined the coordinates of the conditions in the MDS plots for each individual subject. These coordinates were then rotated using a Procrustes analysis, to fit the space of the MDS plots of the average correlation matrix. The rotated coordinates of each subject are used as error bars in the MDS plots.

Bottom Line: In this study, we investigated how Arabic digits are represented in the visual cortex, and how their representation changes throughout the ventral visual processing stream, compared to the representation of letters.However, the activity in this region might have been confounded by string length-number words contain more characters than digits.We found an alteration in representations throughout the ventral processing stream from clustering based on amount of visual information in primary visual cortex (V1) towards clustering based on symbolic stimulus category higher in the visual hierarchy.

View Article: PubMed Central - PubMed

Affiliation: Parenting and Special Education Research Unit, KU Leuven Leuven, Belgium, Europe.

ABSTRACT
In this study, we investigated how Arabic digits are represented in the visual cortex, and how their representation changes throughout the ventral visual processing stream, compared to the representation of letters. We probed these questions with two functional magnetic resonance imaging (fMRI) experiments. In Experiment 1, we explored whether we could find brain regions that were more activated for digits than for number words in a subtraction task. One such region was detected in lateral occipital cortex. However, the activity in this region might have been confounded by string length-number words contain more characters than digits. We therefore conducted a second experiment in which string length was systematically controlled. Experiment 2 revealed that the findings of the first experiment were task dependent (as it was only observed in a task in which numerosity was relevant) or stimulus dependent (as it was only observed when the number of characters of a stimulus was not controlled). We further explored the characteristics of the activation patterns for digit and letter strings across the ventral visual processing stream through multi-voxel pattern analyses. We found an alteration in representations throughout the ventral processing stream from clustering based on amount of visual information in primary visual cortex (V1) towards clustering based on symbolic stimulus category higher in the visual hierarchy. The present findings converge to the conclusion that in the ventral visual system, as far as can be detected with fMRI, the distinction between Arabic digits and letter strings is represented in terms of distributed patterns rather than separate regions.

No MeSH data available.