Comparison of fMRI data from passive listening and active-response story processing tasks in children

Abstract

Purpose

To use functional MRI methods to visualize a network of auditory and language-processing brain regions associated with processing an aurally-presented story. We compare a passive listening (PL) story paradigm to an active-response (AR) version including on-line performance monitoring and a sparse acquisition technique.

Materials/Methods

Twenty children (ages 11−13) completed PL and AR story processing tasks. The PL version presented alternating 30-second blocks of stories and tones; the AR version presented story segments, comprehension questions, and 5s tone sequences, with fMRI acquisitions between stimuli. fMRI data was analyzed using a general linear model approach and paired t-test identifying significant group activation.

Results

Both tasks activated in primary auditory cortex, superior temporal gyrus bilaterally, left inferior frontal gyrus. The AR task demonstrated more extensive activation, including dorsolateral prefrontal cortex and anterior/posterior cingulate cortex. Comparison of effect size in each paradigm showed a larger effect for the AR paradigm in a left inferior frontal ROI.

Conclusion

Activation patterns for story processing in children are similar in passive listening and active-response tasks. Increases in extent and magnitude of activation in the AR task are likely associated with memory and attention resources engaged across acquisition intervals.

The ability to process an aurally-presented story is supported by a network of auditory and language-processing brain regions (2-5) that changes during development (6,7). Multiple aspects of language processing are engaged during this naturalistic speech task, including speech perception, word recognition, syntactic processing and discourse coherence. Consequently, this task has been cited as one method for effectively mapping eloquent cortical areas (3,4). In patients undergoing resective surgery, language lateralization information is particularly important, with the goal of using non-invasive functional MRI to replace the invasive and risky intracarotid amobarbital procedure (Wada test). A number of studies (8-10)have demonstrated the ability of fMRI to detect the expected left-lateralized activation pattern associated with in typically-developing children. In addition, fMRI has revealed changes in lateralization associated with neuropathology (11,12). However, a standard paradigm optimized for clinical use in the pediatric population has not been established. In the present study, we investigate two versions of the story processing task that are potentially effective procedures for determining language lateralization and localization in children.

Previous fMRI studies have examined the neural basis of story processing in children, using a block-periodic, passive listening (PL) design contrasting short stories with tone sequences (7,13,14). These studies were based on data from a large group of children (313) ages 5−18. Results revealed areas of activation in superior temporal gyrus and superior frontal gyrus (bilaterally), and left posterior superior temporal gyrus. Independent components analysis of the same data (ICA) revealed additional activation in angular gyrus together with the precuneus, posterior cingulate, left inferior parietal lobule and inferior frontal gyrus in addition to the regions mentioned above (7). ICA reveals additional areas of activation compared to standard fMRI analysis using model driven methods because it is a data-driven approach that does not rely on a model of the hemodynamic response function to detect changes in BOLD response(15). One drawback associated with the PL story comprehension task is that no on-line measure of task performance is collected. The studies described above (7,13,14) made use of a post-scan comprehension test on which 87% of participants performed above chance level, suggesting that a majority of children were actively engaged in story comprehension. However, this does not address the accuracy of participant's comprehension during the fMRI scan, nor their level of attention to the stimuli. Young pediatric participants, some of whom require sedation to undergo an MRI procedure, may not be capable of actively generating responses to language stimuli. A passive listening task has already been demonstrated to activate bilateral auditory processing areas in such patients (16). Therefore, examining whether passive listening and active-response tasks evoke similar activation patterns, especially in children, is relevant for application of fMRI in these patients.

Continuous fMRI scanning throughout the passive listening task generates acoustic noise that may interfere with participants’ comprehension of the stories. This may also make the identification of task-related activation patterns more difficult due to the presence of background noise (17-19). Schmidt et al. (19) compared BOLD activation in response to identical auditory sentence stimuli tested during continuous fMRI scanning versus sparse fMRI acquisition, and found increased levels of BOLD response in superior temporal regions with the sparse acquisition technique, suggesting that this technique may be more sensitive for detecting activation associated with auditory language stimuli.

The present study examines these issues by comparing the PL story paradigm to an active-response (AR) version that includes on-line performance monitoring and utilizes a clustered fMRI acquisition technique that eliminates scanner noise during auditory presentation (20,21). Our goal in developing the AR task was to potentially improve upon the PL version of the task as much as possible since it is regularly used in the clinical setting.

Methods

Participants

Participants were 20 native-English-speaking children ages 11−13 (8 females), right-handed, with no history of neurological or language disorders. These subjects were drawn from a longitudinal subgroup recruited from a larger cross-sectional sample of subjects previously included in our fMRI studies of language development (6,9). (NICHD – R10HD38578, P.I. Holland) using the PL story paradigm. Thirty children of the children who entered the cross-sectional study at age 5, 6 or 7 were enrolled in a longitudinal subgroup and agreed to undergo annual neuropsychological testing and fMRI scans for a period of five years; with approval of the Institutional Review Board at Cincinnati Children's Hospital Medical Center, and written informed consent was obtained from a parent/guardian of all participants, along with written assent from each child. Upon completion of the first 5 year longitudinal study, the longitudinal component of the study was extended for another 5 years and the AR Story Paradigm (described below) was added to the fMRI battery. IRB approval and informed consent were obtained for the continuation of the longitudinal study. The data in this report was obtained in year six of the longitudinal study, the first year in which the AR Story paradigm was performed. Twenty subjects successfully completed both the PL and AR version of the Story Paradigm during year six of the study. The remaining 10 subjects were either unable to participate in year 6 due to orthodontic braces or did not yield satisfactory data for one of the tasks.

Procedure

Participants completed two versions of a story processing task during a single fMRI scanning session. Task (1) always preceded task (2), with at least one unrelated task intervening.

1. Passive listening Design: This paradigm was based on a periodic 30s on–off, block design. A different story, read by an adult female speaker, was presented during each 30s on period (active). Each 30-second story contained 10 sentences with simple words (similar in word frequency across the stories) and a variety of syntactic constructions (e.g., conjoined sentences and center embeddings), and was designed by a Speech-Language pathologist to be appropriate for children 5 and up. The inclusion of complex syntactic structures was designed to increase the relative processing load during the comprehension task (audio tracks of these stories are available for download at http://www.irc.cchmc.org/software/pedaudio.php). During the control epochs, 1s duration tones were presented at random frequencies (400− 2500 Hz) and intervals (1−3 s), to control for sublexical auditory processing. Five cycles of active and control stimuli were presented for a total scan time of 5 minutes. An initial 30s control period was included at the beginning of the sequence to control for MRI T1 equilibrium effects and these images were discarded prior to statistical post-processing. Subjects were instructed to listen to the stories carefully so that they could answer story-related questions after completion of the MRI procedure. Post-scan performance data were obtained by asking the subject to answer ten multiple-choice questions (2 questions per story).

2. Active-response Design: This As shown in Figure 1, in this version of the Story Paradigm, a single story (differing in content, but similar in structure and word frequency to the stories in the PL version) is presented in short, 5 second segments, each containing two sentences. During this interval the MRI scanner was quiescent. Each story segment was followed by a 6s data acquisition during which three image volumes were acquired with 2s repetition time (TR=2s). A comprehension question based on the preceding sentences was presented during the next second 5s, no-scan interval. Participants answered each question with a yes/no button-press response to a visual cue during this interval, which was followed by another acquisition period. The third segment of each stimulus presentation consisted of 5s of random tones (again with no scanning) followed by a third 6s acquisition period (3 volumes, TR=2s). These three phases of the paradigm were presented for 15 cycles at 36 seconds/cycle for a total scan time of 9 minutes; all of the sentences, taken together, comprised a coherent story. Inclusion of comprehension questions allows for on-line performance monitoring; however it also disrupts the continuous flow of the story somewhat. The sparse acquisition approach (21) allows participants to clearly hear the presented story segments without interference from gradient noise.

Figure 1.

Diagram of the Active-response Story Processing task, including sparse acquisition timing.

Scanning was performed on a 3T Bruker Biospec 30/60, MRI scanner (fMRI parameters: TR/TE = 3000/38 ms, FOV 25.6 × 25.6 cm, matrix 64 × 64, slice thickness = 5 mm, resulting in a voxel size of 4×4×5mm, 25 axial slices. For the sparse acquisition method the TR was 2000ms). MRI data analysis was performed using routines written in IDL (Interactive Data Language, (ITT Visual Information Solutions, Boulder, CO). The EPI images were corrected for Nyquist ghosts and geometric distortion using the multi-echo reference method, and retrospectively corrected for motion using a pyramid coregistration algorithm (22) and then spatially normalized into Talairach space. A general linear model and a paired t-test were implemented to identify voxels activated by story processing (images associated with story listening relative those associated with tone listening in both the PL and AR tasks) for each participant. For the AR task we examine the contrast between the story segments and tones for comparison with the parallel contrast produced in the PL version of the task. After Talairach transformation, random-effects analysis was performed to determine significant group activations. Group activation maps for each task were thresholded at z>7.2, with a cluster size of at least 30 (this corresponds to p<.05, corrected for multiple comparisons via Monte Carlo simulation). Note that our previous work in the larger cross-sectional cohort from which these subjects were drawn has demonstrated that the Talairach reference is adequate for the coregistration of pediatric brain image data from multiple subjects over a range of ages (23).

Lateralization indices (LIs) were calculated in 2 functionally defined regions of interest (ROIs) based on a conjunction analysis of active voxels in the PL and AR group maps (specifically, the ROIs included the intersection of voxels active in the group random effects composite maps for both tasks). One ROI included activated voxels in the left inferior frontal gyrus; the second ROI included activated voxels in the left inferior, middle and superior temporal gyri. The extent of these ROIs was determined by union of the sets of intersecting voxels in the right and left hemispheres, resulting in an inferior frontal ROI of 68 voxels and a temporal ROI of 488 voxels. These combined ROIs were projected onto both hemispheres for purposes of laterality calculation. Voxels with z-scores greater than or equal to the median z-score (resulting from the stories>tones contrast in each task) within an ROI for each individual subject were used in the calculation of LIs. Voxels above this median z-score threshold were counted, and a LI was defined as the difference in the number of activated voxels, summed independently for the left and right ROIs, divided by the summed total of active voxels in the left and right regions of interest. This procedure yields LIs ranging from −1 (right) to 1 (left) (24), and we consider an |LI| > 0.1 as clearly defined lateralization (9,12,25). Conversely, a subject is considered bilateral if the −0.1<LI<0.1.

Effect Size Analysis

In order to compare effectiveness of the two paradigms for eliciting activation in language ROIs, we compared the difference in magnitude of the signal in the story listening > tone listening contrast in our selected regions of interest between the two paradigms. First, each active-response data set was truncated so that the scan time for two paradigms was equal. A total of 5 minutes of scanning time during story listening and tone listening contributed to the results for both paradigms. Note that we equated the scan in terms of time spent for each participant to listen to the story and tone segments of each task, including the no-scan intervals in the active-response version. We did not equate for number of EPI acquisitions. In order to statistically test for the effect size differences, the following procedure was applied to the differences between average t scores (across each ROI) for the two methods. The null distribution for the effect size difference was found via Monte Carlo simulation. The method repeatedly samples from a normal distribution assumed to be the same for both methods and calculates the effect size difference. A significant difference in the effect size from the null distribution would indicate a significant difference in effect size between the two methods.

Results

Performance Data

On the post-scan comprehension test, participants were able to correctly answer questions based on both the PL (mean percent correct, 75.1, SD 12.7) and AR (mean percent correct (79.1, SD 9.1) versions of the story processing task. Comprehension results from the two versions of the task did not differ (p>.2). Performance data was also recorded from the comprehension questions interspersed in the AR version of the task, though this data was only available for 16 participants due to problems with the response recording equipment. Mean percent correct on these questions was 89.7, SD 15.2. All fMRI data was included in the data analysis regardless of participant response, since error rates were so low.

Group Activation Maps

Areas of activation in the passive listening version of the story processing task (relative to tone listening) included left and right temporal regions as well as left inferior frontal gyrus (see Figure 2(a), Table 1).

Figure 2.

(a) Group activation map for the Passive Listening Story Processing Task. (b) Group activation map for the Active-response Story Processing Task, presented in radiological convention (left on the picture corresponds to the right hemisphere). Activated voxels have nominal Z = 7 (blue) to Z = 15 (red) are superimposed on an average T1 weighted anatomical image generated from all subjects/all sessions. All regions are significant with p < 0.05 corrected for multiple voxel comparisons. Fifteen axial slices selected for display (slice range: Z = −15 to + 55 mm, Talairach coordinates). For exact location of the BOLD signal changes see Table 1.

Table 1.

Areas of significant group activation in each version of the story processing task.

	Coordinates (center-of-mass)			Brodmann area
	X	Y	Z
Passive Listening
Left Inferior Frontal Gyrus	−47	20	−2	47
Left Superior Temporal Gyrus	−52	−24	1	22
Right Superior Temporal Gyrus	51	−15	−2	22
Active-Response
Left Inferior Frontal Gyrus	−43	15	16	44
Medial Frontal Gyrus/Anterior Cingulate	0	49	27	9
Left Middle Temporal Gyrus	−46	−46	7	22
Left Superior Temporal Gyrus	−55	−14	−3	21
Right Superior Temporal Gyrus	42	−53	24	39
Right Superior Temporal Gyrus	48	−15	−4	22
Posterior Cingulate Cortex	−1	−47	24	23
Cuneus	2	−77	6	18

Areas of activation in the active-response version of the story processing task (relative to tone listening) included all regions found in the passive listening task plus more extensive bilateral temporal regions and left inferior frontal gyrus extending into dorsolateral prefrontal cortex, as well as anterior cingulate cortex and posterior cingulate cortex and cuneus (see Figure 2(b), Table 1). While the centers-of-mass of the activations in left inferior frontal gyrus (listed in Table 1) differed between the two tasks, the activations were largely overlapping, with the active area in the AR task extending more dorsally than that for the PL task.

Lateralization Indices

The PL and AR versions of the story processing task showed little difference in degree of lateralization in the selected regions of interest (see Figure 3); a paired t-test was not significant (p>.2). Lateralization indices in both paradigms in both ROIs differed significantly from zero [Frontal ROI: PL task t(18) = 2.21, p<.05, AR task t(18) = 3.57, p<.01; Temporal ROI: PL task t(18) = 3.38, p<.01, AR task t(18) = 3.64, p<.01], indicating that both paradigms evoke consistent left-lateralized activation. Furthermore, In the PL task, in IFG, 79% of participants had a positive (leftward) LI, all of these greater than .1; in the temporal ROI, 79% had a positive LI, 74% greater than .1.

Figure 3.

Mean lateralization indices in frontal and temporal ROIs for the Active-response (AR) and Passive listening (PL) versions of the story processing task.

In the AR task, in IFG, 74% of participants had a positive (leftward) LI, 63% greater than .1; in the temporal ROI, 79% had a positive LI, 68% greater than .1.

Effect Size Analysis

The AR task showed a larger effect in the frontal ROI than the PL task [t(18) = 3.39, p<0.05] when comparing story processing to tone listening, equated for scan time. The two tasks did not differ in effect size the temporal ROI (p>.3).

Discussion

Both passive listening and active-response story processing tasks result in similar patterns of frontal and temporal activation supporting multiple aspects of language processing in children. We found a great deal of overlap in the extent of activation for the two versions of the task in these regions. The activation differences observed can be attributed to cognitive strategic aspects of the two versions of the task as well as differences in background noise characteristics. Specifically, the introduction of comprehension questions interspersed with story segments in the AR task is likely to engage memory and attention faculties not required in the PL version, where no overt responses are generated during the task. Consequently, increased activation in dorsolateral prefrontal cortex during the AR task was likely associated with maintenance and manipulation of sentence segments in working memory across acquisition intervals in order to construct the complete story and answer interspersed questions (26). A number of fMRI studies of attention (27), stimulus/response compatibility (28) and self-monitoring(28) have shown activation in cingulate cortex, so increased activation during the AR task may relate to these functions, as attention must be shifted between story listening and responding to questions, and children may be self-monitoring their responses to questions in an effort to perform accurately. Interestingly, cingulate cortex has recently been suggested to play a major role in the brain's “default mode” (1,29). Specifically, it is an area often seen active during a resting state relative to a stimulated behavior. In our results we see increased activation in the posterior cingulate gyrus during the language condition. This suggests that the cingulate may have more of a self-monitoring and attention function in these tasks. Deactivation of the default-mode network elements in this region during the speech stimulus would tend to attenuate the positive activation associated with the stimulus and we are not able to deconvolve these influences in our data. However, the self-monitoring role of the cingulate has been suggested in these studies of the default mode well (the notion that individuals may focus their attention internally during rest) so this may be consistent with the cingulate being active in some cognitive tasks as well as during rest.

An additional area of activation in the cuneus is likely related to visual attention (30) to yes/no response cue for the questions – no visual stimulus was included in the PL version of the task.

Differences in overall activation patterns were observed, but the two versions of the story processing task did not differ significantly in their degree of left-lateralization (quantified using lateralization indices) in the frontal and temporal lobe language processing areas that we selected as regions of interest. (Qualitatively, the AR task showed more leftward lateralization than the PL task in the frontal region, but this difference did not reach statistical significance). Our regions of interest were those frontal and temporal regions where both tasks showed significant activation, and we found consistent left-lateralization for both tasks.

A concern in the interpretation of our results is that the speech sounds making up the stories are a much more complex auditory stimulus than the tones in the control condition. Therefore, the areas of activation resulting from the stories > tones contrast may be partially related to increased primary auditory processing, particularly in temporal regions where auditory information is processed. This might make it more difficult to detect a left-lateralized pattern related to language processing. However, we do observe strong left-lateralization in both versions of the task. A more acoustically complex control task, such as backward speech (5) may have revealed narrower ROIs, and possibly revealed more subtle differences in degree of left-lateralization between the two tasks, though the present results make no prediction about such differences.

In frontal and temporal regions activated by both versions of the task, we examined possible differences in effect size (story listening > tone listening) between the two paradigms. The AR version of the task showed a significantly larger effect size in the frontal ROI only, which suggests that it may be a more effective method for eliciting activation in this region. This may also be due to cognitive strategic factors, since left inferior frontal gyrus has been suggested to be involved in maintenance of semantic information in verbal memory and semantic retrieval (31) which may be more actively engaged when comprehension questions are included in the task.

Alternatively, a greater effect size in the AR task may be due to the reduction in scanner noise, leading to more effective processing of the speech stimuli. However, the effect size was similar between the AR and PL tasks in temporal regions, where differences in effect size due to auditory processing might be most likely to occur. In fact, when directly contrasting continuous scanning and clustered acquisition techniques using the same stimuli, Schmidt et al. (19)found increases in activation limited to temporal areas. This lack of an effect size difference between the PL and AR tasks in the temporal ROI may also be a result of increases in primary auditory processing of the more complex speech sounds relative to the simpler control tones: this stories>tones contrast was employed in both versions of the task. A more acoustically complex control condition may have resulted in a temporal ROI that would be more specific to story processing give a more detailed picture of the potential difference between PL and AR story-processing tasks in this brain region.

In general, both PL and AR story processing paradigms are effective methods for activating eloquent cortex in children. Including comprehension questions as an on-line performance measure during the AR task leads to increased engagement of memory and attention networks in the brain, resulting in additional areas of activation and a greater effect size in left inferior frontal gyrus. For children that are able to perform the AR task, being able to map this more extensive network may be beneficial. However, a passive story listening task also reveals brain regions involved in language processing, producing the same degree of lateralization and an equally large effect size in the temporal language area. This paradigm, therefore, may be appropriate for younger children or others who cannot perform the comprehension task. Given that the lateralization indices computed for ROIs in the classical language areas of the frontal and temporal lobes do not differ significantly between the two versions of the task, we can have confidence in the lateralization pattern and corresponding LI derived from the passive story listening paradigm when no responses are recorded, as previously published as a function of age(6). This provides further support for the use of fMRI for passive studies in children where responses may not be feasible.