Limits...
Interpreting response time effects in functional imaging studies.

Taylor JS, Rastle K, Davis MH - Neuroimage (2014)

Bottom Line: However, even for contrasts designed to tap neural effort, activity remained after factoring out the RT-BOLD response correlation.This may reveal unpredicted differences in neural engagement (e.g., learning phonological forms for pseudowords>words) that could further the development of cognitive models of reading aloud.Our framework provides a theoretically well-grounded and easily implemented method for analysing and interpreting RT effects in neuroimaging studies of cognitive processes.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Royal Holloway University of London, Egham Hill, Egham TW20 0EX, UK. Electronic address: j.taylor@rhul.ac.uk.

Show MeSH

Related in: MedlinePlus

Plots showing four possible outcomes with regard to the relationship between RT and the BOLD response as measured by fMRI, in a hypothetical brain region specialized for processing faces. Graphs show response time and BOLD signal (scaling of a canonical HRF) for single trials and the interpretation of parameter estimates (β) from a general linear model fitted to this data. In all cases, two conditions differ in mean RT, and RT is correlated with BOLD signal. However, we can distinguish two sources of activation differences: β effort is the amount of change in BOLD signal per unit change in RT, calculated on the basis of all stimuli in the experiment. β engagement is the difference in BOLD signal between conditions, over and above the difference predicted from RT. We acknowledge that this approach presupposes that the relationship between RT and BOLD is the same in both conditions depicted. Inclusion of a condition-by-RT interaction term in statistical analysis allowed us to validate this assumption for the present data (reported in footnote ii), and would allow assessment of differential engagement even in the case of a significant interaction. In (A) and (C) β engagement is enhanced when response time differences are taken into account, in (B) and (D) β engagement is absent or reduced when response time differences are taken into account. These four profiles might be anticipated in the following circumstances: (A) faces engage/activate this brain region more than houses despite the fact that houses are more effortful/take longer to process. (B) Familiar and less familiar faces both engage this brain region but less familiar faces elicit greater activity entirely due to their longer RTs/greater processing effort. (C) Animal and human faces equivalently activate this brain region, but animal faces are more effortful to process, and accounting for these RT differences reveals the greater engagement of this region by faces. (D) Familiar and less familiar faces both engage this brain region, but less familiar faces elicit greater activity over and above that which would be expected on the basis of their longer RTs. This reveals that less familiar faces are both more effortful and engage an additional process relative to familiar faces.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4121088&req=5

f0010: Plots showing four possible outcomes with regard to the relationship between RT and the BOLD response as measured by fMRI, in a hypothetical brain region specialized for processing faces. Graphs show response time and BOLD signal (scaling of a canonical HRF) for single trials and the interpretation of parameter estimates (β) from a general linear model fitted to this data. In all cases, two conditions differ in mean RT, and RT is correlated with BOLD signal. However, we can distinguish two sources of activation differences: β effort is the amount of change in BOLD signal per unit change in RT, calculated on the basis of all stimuli in the experiment. β engagement is the difference in BOLD signal between conditions, over and above the difference predicted from RT. We acknowledge that this approach presupposes that the relationship between RT and BOLD is the same in both conditions depicted. Inclusion of a condition-by-RT interaction term in statistical analysis allowed us to validate this assumption for the present data (reported in footnote ii), and would allow assessment of differential engagement even in the case of a significant interaction. In (A) and (C) β engagement is enhanced when response time differences are taken into account, in (B) and (D) β engagement is absent or reduced when response time differences are taken into account. These four profiles might be anticipated in the following circumstances: (A) faces engage/activate this brain region more than houses despite the fact that houses are more effortful/take longer to process. (B) Familiar and less familiar faces both engage this brain region but less familiar faces elicit greater activity entirely due to their longer RTs/greater processing effort. (C) Animal and human faces equivalently activate this brain region, but animal faces are more effortful to process, and accounting for these RT differences reveals the greater engagement of this region by faces. (D) Familiar and less familiar faces both engage this brain region, but less familiar faces elicit greater activity over and above that which would be expected on the basis of their longer RTs. This reveals that less familiar faces are both more effortful and engage an additional process relative to familiar faces.

Mentions: Our framework suggests that a stimulus type that is represented by a particular brain region should engage that region more than another non-represented stimulus type. Here and throughout, we use the term “represents” to mean “represents some property of the stimulus”, for example, for written words this could be letters, phonemes, more basic visual or acoustic properties, as well as higher level conceptual information. These representations may be permanently instantiated in a neural system (e.g. specialized neurons that code for specific letters or words in posterior regions), or transient reading-related representations in frontal regions that serve related functions in other tasks (e.g. phonological output representations also used during object naming and spontaneous speech). This seems appropriate given that we are not committed to any particular representational system (e.g., localist versus distributed). As contrasts between represented and non-represented stimuli tap differences in engagement, clusters of activity revealed by such contrasts should survive correction for RT. However, if a brain region represents both stimulus types, then differential activity will be driven by processing effort and hence should positively correlate with RT. In such cases, correcting for RT should account for differential activity. Given these proposals, we can distinguish four possible outcomes in functional neuroimaging studies, as illustrated in Fig. 2, panels A to D.(A)


Interpreting response time effects in functional imaging studies.

Taylor JS, Rastle K, Davis MH - Neuroimage (2014)

Plots showing four possible outcomes with regard to the relationship between RT and the BOLD response as measured by fMRI, in a hypothetical brain region specialized for processing faces. Graphs show response time and BOLD signal (scaling of a canonical HRF) for single trials and the interpretation of parameter estimates (β) from a general linear model fitted to this data. In all cases, two conditions differ in mean RT, and RT is correlated with BOLD signal. However, we can distinguish two sources of activation differences: β effort is the amount of change in BOLD signal per unit change in RT, calculated on the basis of all stimuli in the experiment. β engagement is the difference in BOLD signal between conditions, over and above the difference predicted from RT. We acknowledge that this approach presupposes that the relationship between RT and BOLD is the same in both conditions depicted. Inclusion of a condition-by-RT interaction term in statistical analysis allowed us to validate this assumption for the present data (reported in footnote ii), and would allow assessment of differential engagement even in the case of a significant interaction. In (A) and (C) β engagement is enhanced when response time differences are taken into account, in (B) and (D) β engagement is absent or reduced when response time differences are taken into account. These four profiles might be anticipated in the following circumstances: (A) faces engage/activate this brain region more than houses despite the fact that houses are more effortful/take longer to process. (B) Familiar and less familiar faces both engage this brain region but less familiar faces elicit greater activity entirely due to their longer RTs/greater processing effort. (C) Animal and human faces equivalently activate this brain region, but animal faces are more effortful to process, and accounting for these RT differences reveals the greater engagement of this region by faces. (D) Familiar and less familiar faces both engage this brain region, but less familiar faces elicit greater activity over and above that which would be expected on the basis of their longer RTs. This reveals that less familiar faces are both more effortful and engage an additional process relative to familiar faces.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4121088&req=5

f0010: Plots showing four possible outcomes with regard to the relationship between RT and the BOLD response as measured by fMRI, in a hypothetical brain region specialized for processing faces. Graphs show response time and BOLD signal (scaling of a canonical HRF) for single trials and the interpretation of parameter estimates (β) from a general linear model fitted to this data. In all cases, two conditions differ in mean RT, and RT is correlated with BOLD signal. However, we can distinguish two sources of activation differences: β effort is the amount of change in BOLD signal per unit change in RT, calculated on the basis of all stimuli in the experiment. β engagement is the difference in BOLD signal between conditions, over and above the difference predicted from RT. We acknowledge that this approach presupposes that the relationship between RT and BOLD is the same in both conditions depicted. Inclusion of a condition-by-RT interaction term in statistical analysis allowed us to validate this assumption for the present data (reported in footnote ii), and would allow assessment of differential engagement even in the case of a significant interaction. In (A) and (C) β engagement is enhanced when response time differences are taken into account, in (B) and (D) β engagement is absent or reduced when response time differences are taken into account. These four profiles might be anticipated in the following circumstances: (A) faces engage/activate this brain region more than houses despite the fact that houses are more effortful/take longer to process. (B) Familiar and less familiar faces both engage this brain region but less familiar faces elicit greater activity entirely due to their longer RTs/greater processing effort. (C) Animal and human faces equivalently activate this brain region, but animal faces are more effortful to process, and accounting for these RT differences reveals the greater engagement of this region by faces. (D) Familiar and less familiar faces both engage this brain region, but less familiar faces elicit greater activity over and above that which would be expected on the basis of their longer RTs. This reveals that less familiar faces are both more effortful and engage an additional process relative to familiar faces.
Mentions: Our framework suggests that a stimulus type that is represented by a particular brain region should engage that region more than another non-represented stimulus type. Here and throughout, we use the term “represents” to mean “represents some property of the stimulus”, for example, for written words this could be letters, phonemes, more basic visual or acoustic properties, as well as higher level conceptual information. These representations may be permanently instantiated in a neural system (e.g. specialized neurons that code for specific letters or words in posterior regions), or transient reading-related representations in frontal regions that serve related functions in other tasks (e.g. phonological output representations also used during object naming and spontaneous speech). This seems appropriate given that we are not committed to any particular representational system (e.g., localist versus distributed). As contrasts between represented and non-represented stimuli tap differences in engagement, clusters of activity revealed by such contrasts should survive correction for RT. However, if a brain region represents both stimulus types, then differential activity will be driven by processing effort and hence should positively correlate with RT. In such cases, correcting for RT should account for differential activity. Given these proposals, we can distinguish four possible outcomes in functional neuroimaging studies, as illustrated in Fig. 2, panels A to D.(A)

Bottom Line: However, even for contrasts designed to tap neural effort, activity remained after factoring out the RT-BOLD response correlation.This may reveal unpredicted differences in neural engagement (e.g., learning phonological forms for pseudowords>words) that could further the development of cognitive models of reading aloud.Our framework provides a theoretically well-grounded and easily implemented method for analysing and interpreting RT effects in neuroimaging studies of cognitive processes.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Royal Holloway University of London, Egham Hill, Egham TW20 0EX, UK. Electronic address: j.taylor@rhul.ac.uk.

Show MeSH
Related in: MedlinePlus