Non-neural BOLD variability in block and event-related paradigms

Abstract

Block and event-related stimulus designs are typically used in fMRI studies depending on the importance of detection power or estimation efficiency. The extent of vascular contribution to variability in block and event-related fMRI-BOLD response is not known. With scaling, the extent of vascular variability in the fMRI-BOLD response during block and event-related design tasks was investigated. Blood oxygen level-dependent (BOLD) contrast data from healthy volunteers performing a block design motor task and an event-related memory task requiring performance of a motor response were analyzed from the regions of interest (ROIs) surrounding the primary and supplementary motor cortices. Average BOLD signal change was significantly larger during the block design compared to the event-related design. In each subject, BOLD signal change across voxels in the ROIs had higher variation during the block design task compared to the event-related design task. Scaling using the resting state fluctuation of amplitude (RSFA) and breath-hold (BH), which minimizes BOLD variation due to vascular origins, reduced the within-subject BOLD variability in every subject during both tasks but significantly reduced BOLD variability across subjects only during the block design task. The strong non-neural source of intra- and intersubject variability of BOLD response during the block design compared to event-related task indicates that study designs optimizing for statistical power through enhancement of the BOLD contrast (for, e.g., block design) can be affected by enhancement of non-neural sources of BOLD variability.

1. Introduction

Functional magnetic resonance imaging (fMRI) using the blood oxygen level-dependent (BOLD) contrast has emerged as a preferred tool in studies of brain function in health and disease due to its noninvasive nature. In BOLD-fMRI, neural activity is inferred by measuring the change in the oxy- and deoxyhemoglobin ratio in the brain tissue due to a sensory or motor activity. Hence, during brain activation, the BOLD signal response is a convolution of neural and vascular components. The weighting of the neural component in BOLD signal change is the highest from voxels representing microvascular structures (arterioles, capillaries, etc.,) whereas the weighting of the vascular component is the highest from voxels representing venous structures and large draining veins in the brain. Thus brain function, or neural activity, as inferred from the BOLD contrast can be influenced by vascular structures that can vary regionally in the brain or vascular functional differences between subjects. To account for the vascular contribution to neural activation-induced BOLD signal change, scaling or normalization can be used [1]. Scaling refers to the division of the task-induced BOLD response by mild hypercapnia-induced response in each voxel, wherein hypercapnia is induced by either administration of carbon dioxide or a brief breath-hold (BH) performed for a few seconds. Scaling using BH and resting state fluctuation of amplitude (RSFA) decreases vascular contributions to BOLD variability and is thus believed to reflect more accurate neural activity changes [2–6].

The extent of vascular weighting in typically used fMRI tasks, namely, “block” and “event related,” is not known. There are some indications from prior studies that the extent of vascular contribution to BOLD variability may depend on the stimulus design [2,4]. However, these studies affected different cortical systems in the brain making definite conclusions difficult. Furthermore, variables such as subject cohort and the type of task used to induce hypercapnia can affect comparisons [5]. To compare the extent of vascular contribution, it is necessary to obtain measurements in a single group of subjects performing a block design and event-related paradigms in the same session and study their response in the same cortical system. We hypothesized that, in a block design fMRI response, where BOLD contrast change is relatively larger than event related, the contribution to variability in the BOLD signal response from non-neural sources should also increase. To test this hypothesis, we used scaling of BOLD signal responses from the motor regions of 24 subjects who performed a motor task in block and event-related design tasks in the same scan session. Scaling was performed to mitigate vascular-related variation in the BOLD signal response. The extent of mitigation of BOLD variability indicated that the block design task had a relatively stronger non-neural contribution to intersubject BOLD signal variability than the event-related task. Thus subject-wise differences in baseline physiological variables that lead to variability in BOLD responses are larger during a block design paradigm.

2. Methods

2.1. Subjects

Twenty-four healthy subjects (11 males and 13 females; mean age: 41 years; range: 19–71 years) with no history of head trauma and neurological diseases were scanned. All experimental procedures were approved by the Institutional Review Board of the University of Texas at Dallas. Written informed consent was obtained from all subjects and paid on an hourly basis during the study.

2.2. Experimental tasks

Each subject performed a BH, bilateral fingertapping (FTAP) and digit–symbol substitution task (DSST) paradigm. Resting state fMRI scans were also obtained for all the subjects.

2.2.1. Breath-hold

The BH experiment consisted of a 40-s normal breathing period followed by three repetitions of alternate epochs of 20 s of BH and 40 s of normal breathing. Subjects performed an end-inspirational BH, inhaling a volume of air comparable to the volume they would inhale when breathing normally [7,8]. Subjects were trained on the BH technique a few minutes prior to the actual scanning session, which helped them avoid inhaling larger than normal volumes of air during BH task performance. A white circle remained centered on the screen during the normal respiration periods, and to signal the BH periods, the circle changed color to cyan and began flashing at 0.5 Hz. Participants were instructed to take a normal breath when the circle changed colors and began flashing and to hold their breaths until the circle stopped flashing.

2.2.2. Block design task (bilateral fingertapping)

A white circle remained centered on the screen during the rest periods, and to signal the finger-tapping periods, the circle changed color to cyan and began flashing at 0.5 Hz. Participants sequentially touched each finger of each hand to its respective thumb making one touch and release, as best they could, in synchrony with the flashing circle. The FTAP paradigm consisted of an initial 20-s rest period followed by four repetitions of alternate epochs of 20 s of bilateral fingertapping and 20 s of rest.

2.2.3. Event-related task (DSST)

Adapted from the Wechsler Adult Intelligence Scale [9], the modified DSST consisted of a code table that represents pairings of digits and nonsense symbols. The code table containing digit–symbol pairs and a single digit–symbol probe was projected simultaneously on a screen for 4 s viewed by subjects inside the MR scanner [6,10]. If the probe pair matched one of those in the table, subjects pressed a right-thumb button; otherwise, they pressed a left-thumb button. There were a total of 52 trials in a single scanning run. On half the trials, the probe pair matched one of the digit–symbol pairs in the code table; on the other half, the probe pair did not match one of the pairs in the code table. The stimuli stayed on the screen for 4 s followed by variable intertrial intervals of 0, 4, 8 or 12 s. As the subjects had to perform a motor activity in the form of a button press with the thumb on either hand, this event-related paradigm was expected to activate motor cortical areas similar to the bilateral fingertapping task in the participating subjects.

2.2.4. Resting state

Subjects remained relaxed with their eyes closed during the acquisition of the resting-state MR images.

2.3. MRI Acquisition

MR Imaging was performed on a 3-T Philips Achieva scanner. The imaging system was equipped with an eight-element, SENSE, receive-only head coil and a fixed asymmetric head gradient coil. Subjects were positioned in a supine position on the gantry with head in a midline location in the coil. Foam padding and a pillow were used to minimize head motion. High-resolution T1-weighted anatomical images were obtained from all subjects. Gradient echo-EPI images were subsequently obtained during rest, BH, block design and the event-related tasks. Thirty-two slices were obtained in the axial plane covering the entire brain. Imaging parameters were as follows: FOV of 220 mm, matrix size of 64×64, TR/TE=2000/30 ms and slice thickness of 4 mm. Ninety EPI images were obtained during the block design task, and 110 EPI images were obtained during BH and the event-related task. Imaging parameters were kept the same for all four runs. High-resolution anatomical images were obtained using an MPRAGE sequence with 1 mm isovoxel, sagittal, TE=3.7 ms and flip angle=12°.

2.4. Data analysis

All fMRI data sets were preprocessed using AFNI [11]. The EPI images were corrected for motion using a rigid-body volume registration algorithm available in AFNI. The motion correction algorithm calculated motion in six directions of rotation and translation throughout each run. After considering motion in all six directions, single maximal displacement value (D) for each volume was estimated for every subject [12]. EPI data sets with D>2 mm were omitted from further processing. Analysis was done only on the voxels that represented the brain tissue, and all voxel time series were detrended to remove quadratic trends.

To determine activated areas during each task, a gamma-variate function was convolved with the task reference function and cross correlated with the BOLD signal on a voxel-wise basis. Parameters of the gamma-variate function were derived according to Ref. [13]. During BH, the reference function was shifted by 12 s to take into account the large hemodynamic delay during the BH response [8].A threshold of P<.01, Bonferroni corrected, was considered for determining active voxels [14]. Group activation maps were determined by converting each subject’s functional map to standard stereotaxic space based on the Talairach and Tournoux [15] atlas using a linear transformation. The correlation coefficients (r) from each individual subject’s functional maps were z-transformed {z=0.5*log[(1+r)/(1−r)]} and averaged. The group-averaged z-maps were transformed back to obtain the group correlation coefficient map. Three regions of interest (ROIs) were defined covering the left and right primary motor cortices and the supplementary motor area using significantly active voxels (P<.01) from the group map in response to the block design motor task. All further analysis was carried out within the ROIs defined in Fig. 1A.

Fig. 1.

(A) Regions of interest enclosing active voxels after a logical union of block design motor task-induced activity over all subjects. The ROI included the primary, supplementary motor cortices and parietal areas. The ROIs in the axial plane completely covered the precentral gyrus. (B) Group activation maps during the block design motor task and (C) event-related tasks. A voxel-wise cross-correlation of the BOLD signal time course was performed with the reference function representing the task paradigm for each subject convolved with a gamma-variate function. The subject-wise correlation coefficient maps representing activation were z-transformed and averaged. The group activation map was determined after an inverse z-transform and using a correlation coefficient threshold corresponding to a Bonferroni corrected P<.01. Primary and supplementary motor activation is observed during both block design and event-related motor tasks.

2.4.1. Scaling of task-induced response

Scaling was accomplished by dividing the BOLD signal response amplitude during the motor task with the BH-induced BOLD response amplitude or the RSFA obtained for every voxel in the brain [2–4,6,7]. BOLD signal amplitude change during task was obtained from the temporal standard deviation of the BOLD signal time series during the motor task. RSFA and BOLD signal amplitude change during the BH task were obtained from the temporal standard deviation of the BOLD signal time series during the resting state scan and the BH scan, respectively. The BOLD signal amplitude change defined here is the ΔBOLD and not a fractional percent change.

3. Results

Spatial extent of activation varied significantly in several subjects (particularly higher ages) with activity in the parietal areas during the performance of the block design motor task. Thus an anatomical prior hypothesis was not possible in the selection of the ROIs for further analysis. Hence, a logical union map of active voxels from all subjects performing the block design motor task was obtained using a threshold of P<.01. Functional ROIs were drawn enclosing all active voxels in the logical union map (Fig. 1A) [16]. These ROIs included the primary and supplementary motor cortices and parts of the parietal areas of the brain, maximizing the significantly active regions during the block design task from all subjects. Analyses of BOLD data from the healthy volunteers performing the block and event-related tasks were performed from the same ROIs. As observed from the group activation maps, the spatial extent of activation in the primary motor cortex was substantially higher during the block design paradigm compared to event related. The supplementary motor cortex was, however, activated more during the event-related task (Fig. 1B and C).

Average BOLD signal change estimated as the mean from all active voxels in the ROIs over all subjects was 1.85±0.42% during the block design task and was significantly larger than the 0.97±0.20% observed during the event-related task (Fig. 2A; Student’s t test; P<9×10⁻¹²). Scaled fMRI responses were obtained by dividing the task-induced response in each voxel by the corresponding BH response or RSFA. Scaling with RSFA significantly reduced the average block design BOLD signal change to 1.46±0.28% (compared with unscaled; Student’s t test; P<3×10⁻⁴), while the event-related response scaled with RSFA significantly reduced to 0.85±0.18% (compared with unscaled; Student’s t test; P<.02) (Fig. 2B). Scaling with BH significantly reduced the average block design BOLD signal change to 1.03±0.19% (compared with unscaled; Student’s t test; P<5×10⁻¹⁰), while the event-related response scaled with BH reduced to 0.75±0.17% (compared with unscaled; Student’s t test; P<2×10⁻⁴) (Fig. 2C).

Fig. 2.

(A) Mean BOLD signal change in the ROI in all subjects during the block design and event-related tasks. (B) Mean BOLD signal change after scaling with RSFA and (C) mean BOLD signal change after scaling with BH.

The BOLD signal variability within each subject was calculated as the ratio of the standard deviation and mean BOLD signal change within the ROIs, defined as the intrasubject coefficient of variation (CV_intra). During the block design task, the CV_intra prior to scaling was 1.23, which reduced to 1.14 and 0.39 after scaling with RSFA and BH, respectively (Fig. 3A). Although CV_intra reduced after scaling with both RSFA and BH, only the reduction after scaling with BH was significant. During the event-related task, CV_intra was 0.39 prior to scaling, which did not change significantly after scaling with RSFA and BH (Fig. 3B). Thus scaling with BH, which significantly decreased the spatial variability in the BOLD signal during the block design task, indicated a strong non-neural contribution to BOLD signal variability during the response to the block design motor task.

Fig. 3.

(A) Spatial variability of BOLD signal change prior to and after scaling with RSFA and BH during the block design task. (B) Spatial variability of BOLD signal change prior to and after scaling with RSFA and BH during the event-related task. (*Significantly different compared to the unscaled condition, P<1×10⁻¹¹ paired t test.)

In order to determine how the BOLD signal response scaled in every subject during the block design and event-related tasks, the distributions of the BOLD signal change in every activated voxel were obtained before and after scaling. Fig. 4 shows the distributions of the BOLD signal change from the ROIs of five randomly chosen subjects from the group. In every subject, a higher median BOLD signal change (between 2% and 3%) was observed during the block design compared to the event-related task (between 1% and 2%; Fig. 4A and C). The median of the distributions after scaling with RSFA was reduced to a greater extent during the block design compared to the event-related task (Fig. 4B and E). A similar trend was observed after scaling with BH (Fig. 4C and F). Taken together with results from Fig. 3, the block design BOLD signal change showed greater variability within each subject compared to the event-related task. In other words, intrasubject BOLD variability in the motor cortex was greater during the block design task than during the event-related task.

Fig. 4.

BOLD signal distributions during the block design (A–C) and event-related task (D–F). (A, C) Unscaled, (B, D) scaled with RSFA and (C, F) scaled with BH. (Note the different x-axis scale in the scaled distributions for figure clarity.)

To assess the intersubject variability, BOLD signal variability between subjects was calculated as the coefficient of variation, i.e., the ratio of the group standard deviation of the task-induced BOLD signal change and the group mean (CV_inter). Significant difference between the scaled and unscaled conditions was tested using the Bartlett’s test of homogeneity of variance. CV_inter during the block design task was 0.22, which reduced to 0.19 after scaling with RSFA and BH (Table 1). For the block design task-induced BOLD responses, the variance was significantly reduced after scaling with RSFA and BH (Table 1; Bartlett’s test, P<.05 and P<.001, respectively). During the event-related task, the CV_inter was 0.21, which did not change significantly after scaling with RSFA and BH (Table 1; Bartlett’s test, P<.6 and P<.1, respectively).

Table 1.

BOLD signal change in the ROIs during the block design and event-related design tasks before and after scaling with scaling variables RSFA and BH

Subject	Block design			Event-related design
	Unscaled	Scaled_RSFA	Scaled_BH	Unscaled	Scaled_RSFA	Scaled_BH
1	1.99	na	0.90	1.25	na	0.71
2	1.65	1.40	1.00	0.81	0.74	0.59
3	1.53	1.36	0.96	0.82	0.73	0.55
4	1.51	1.57	1.25	0.94	0.74	0.58
5	1.73	1.52	0.79	0.87	0.54	0.61
6	2.51	1.90	1.08	0.93	1.02	0.85
7	1.73	1.29	0.98	0.94	0.99	0.94
8	1.66	1.41	0.74	0.60	0.64	0.51
9	1.93	1.36	1.21	1.07	0.85	0.80
10	2.01	1.68	1.14	1.23	1.02	1.01
11	1.66	1.65	1.02	0.72	0.99	0.68
12	2.48	1.33	1.02	0.88	0.82	0.68
13	1.75	1.00	1.21	1.09	1.23	1.11
14	1.13	0.96	0.84	1.12	0.61	0.79
15	1.42	1.00	0.95	0.78	0.93	0.56
16	2.07	1.47	0.72	0.80	0.90	0.90
17	2.83	1.74	1.44	1.33	1.08	1.04
18	1.52	1.32	0.97	0.81	0.64	0.52
19	2.32	1.89	1.27	1.10	0.92	0.81
20	2.33	1.73	1.22	1.14	0.79	0.72
21	1.34	1.30	1.24	1.35	0.65	0.69
22	1.85	1.89	0.81	0.78	1.01	0.87
23	1.70	1.42	0.90	0.87	0.96	0.91
Mean±S.D.	1.85±0.42	1.46±0.28 *	1.03±0.19 **	0.97±0.20	0.85±0.18ns	0.76±0.17ns
CVinter	0.22	0.19	0.19	0.21	0.21	0.22

^nsNot significant compared to the unscaled condition in the event-related design.

^*Significantly different compared to the unscaled condition, P<.05, Bartlett’s test.

^**Significantly different compared to the unscaled condition, P<.001, Bartlett’s test.

4. Discussion

fMRI stimuli are typically presented in block fashion (i.e., stimuli and/or motor task performance is repeated regularly over finite periods of time) or event related (i.e., stimuli and/or motor task performance has no finite period repetition), depending on the study design and objectives. While block designs have larger power to detect activation due to large changes in the BOLD signal, event-related designs are useful for determining the hemodynamic response function [17]. Event-related paradigms are often preferred in studies that aim to account for learning effects and habituation in subjects. As the fMRI-BOLD signal is a convolution of neural and vascular components, it can be hypothesized that block design fMRI responses, where the BOLD contrast changes are relatively larger than event related, BOLD variability from non-neural sources should also increase. We tested the above hypothesis by measuring the extent of mitigation of BOLD variability after scaling of BOLD signal responses from the motor regions when subjects performed a motor task in block and event-related designs.

Scaling decreases vascular contributions to BOLD variability and is thus believed to reflect signal changes due to neural activity after accounting for variability due to vascular differences [1–5,7]. Scaling refers to the division of the task-induced BOLD response by the BOLD response to mild hypercapnia in each voxel, whereas RSFA scaling refers to the division of the task-induced BOLD response by the BOLD response obtained from the resting state fluctuation of amplitude [4], or a brief BH task performed for a few seconds. Both RSFA and BH used for scaling on a voxel-wise manner performed comparably in this study in reducing intersubject variability.

Intrasubject variation in the BOLD response (i.e., variation across voxels within each subject) after scaling decreased by a larger percent during the block design compared to the event-related task. Thus, BOLD signal change during the block design task had a relatively larger variability in the non-neural contribution within the motor cortical region than the event-related task. This suggests that voxels associated with draining veins and large vessels that have a higher BOLD weighting are likely to contribute more to BOLD signal variability in a cortical region during a block design task than during an event-related task. This relative BOLD difference gets enhanced as BOLD signal change elicited by a stimulus gets relatively stronger such as that elicited by the block design task.

Subject-wise differences in baseline physiological variables such as resting-state cerebral blood volume, blood flow, vascular reactivity and hematocrit levels can lead to differences in BOLD amplitude variability. The intersubject BOLD response variability reduced by 14% during the block design task and did not change significantly during the event-related task. These results indicate that intersubject variability gets enhanced with enhanced BOLD signal change during a task. Thus BOLD responses from block study designs that optimize statistical power through enhancement of the BOLD contrast are affected by non-neural sources of BOLD variability such as the subject’s baseline physiological state and vascular compliance that can compromise statistical power in a group analysis. BOLD responses from event-related designs are less affected by non-neural sources of variability and thus might be more sensitive to the neural basis of BOLD signal change. Alternative methods of mitigating non-neural sources of variability in block task designs are, however, presently available in the form of RSFA scaling without the necessity to perform any additional tasks (e.g., breathing carbon dioxide or breath-holding) beyond those performed in a typical fMRI study [4,6].