Hemodynamic nonlinearities affect BOLD fMRI response timing and amplitude

Abstract

The interpretation of functional magnetic resonance imaging (fMRI) studies based on blood oxygen-level dependent (BOLD) contrast generally relies on the assumption of a linear relationship between evoked neuronal activity and fMRI response. While nonlinearities in this relationship have been suggested by a number of studies, it remains unclear to what extent they relate to the neurovascular response and are therefore inherent to BOLD fMRI. Full characterization of potential vascular nonlinearities is required for accurate inferences about the neuronal system under study. To investigate the extent of vascular nonlinearities, evoked activity was studied in humans with BOLD fMRI (n=28) and magnetoencephalography (MEG) (n=5). Brief (600-800 ms) rapidly repeated (1 Hz) visual stimuli were delivered using a stimulation paradigm that minimized neuronal nonlinearities. Nevertheless, BOLD fMRI experiments showed substantial remaining nonlinearities. The smallest stimulus separation (200-400 ms) resulted in significant response broadening (15-20% amplitude decrease; 10-12% latency increase; 6-14% duration increase) with respect to a linear prediction. The substantial slowing and widening of the response in the presence of preceding stimuli suggest a vascular rather than neuronal origin to the observed nonlinearity. This was confirmed by the MEG data, which showed no significant neuro-electric nonlinear interactions between stimuli as little as 200 ms apart. The presence of substantial vascular nonlinearities has important implications for rapid event-related studies by fMRI and other imaging modalities that infer neuronal activity from hemodynamic parameters.

Figure 1

(a) M-sequence response kernels (cf. Fig. 3) for the 200-ms gap MEG experiment for 1 volunteer, derived from the average signal over 10 detectors. The primary kernel is shown in red, the first three second-order kernels in blue, green and pink, respectively, and the first third-order kernel in black. The yellow line shows the stimulus timing and is derived from the output of an optical sensor in the projector beam. For clarity an arbitrary baseline offset is used for all but the primary response kernel. (b) The response to an isolated stimulus (respS) and the responses to a stimulus that closely follows on a preceding identical stimulus (resp1 and resp2, corresponding to inter-stimulus intervals of 0.2 s and 1.2 s, respectively). These responses were derived using the pri, sec1 and sec2 kernels shown in (a) (see text and Appendix A for details). The yellow line again shows the stimulus timing.

Figure 2

Example of the observed average BOLD-fMRI hemodynamic response (first order kernel pri) in a representative slice in one of the volunteers during a full-contrast experiment with 200-ms gap. In the left-most pane, a t-score map (derived from the same experiment) is superimposed on an anatomical image (first image of the EPI time-series data) of the slice. The first 10 s of the response in the bottom-half of the slice are shown in the remaining 10 images. The time in s relative to stimulus onset is indicated in the top right-hand corner of each image. Since this specific slice was number 7 out of 10 its acquisition timing was delayed 0.3 s relative to stimulus onset.

Figure 3

ROI and volunteer-averaged response time-courses (m-sequence response kernels) obtained from BOLD fMRI experiments. (a) full-contrast 200 ms gap (n=18); (b) 400-ms gap (n=13); and (c) low-contrast 200-ms gap (n=13). The mean first-order kernel (red) and the first 3 second-order kernels (respectively blue, green and pink) are shown, as well as the first third-order kernel (black). Error bars indicate inter-subject standard error.

Figure 4

The pri, sec1 and sec2 kernels (see Fig. 3) were used to compute the hemodynamic response to a single stimulus (respS) on a volunteer-by-volunteer basis. Also computed were the response that occurred when the current stimulus was closely preceded by another, identical stimulus, either in the bin directly preceding it (resp1, 0.2-0.4 s stimulus separation) or in the bin before that (resp2, 1.2-1.4 s stimulus separation). (a) full-contrast 200 ms gap (n=18); (b) 400-ms gap (n=13); and (c) low-contrast 200-ms gap (n=13). Error bars indicate inter-subject standard error.