SLIDE 11 6.1. The BOLD Signal
... comes with a delay after neuronal activity because it reveals oxygenated blood that moves to neurons that have “worked” - it reflects metabolic activity rather than neuronal activity (Another way to put this: it reflects activity of astrocytes rather than neurons) The delay is called “hemodynamic response” ... and depends on the brain regions (but that we usually ignore in the analysis) Fortunately, the effect of sustained neuronal activity is LINEAR - see picture (rCBF = regional Cerebral Blood Flow)
21
Characterizing the Dynamic Perfusion Response to Stimuli of Short Duration
- K. L. Mikrt,
- W. M. Luh, T. T. Liu, A. Martinez, T. Obata, E. C. Wang, L. R. Frank and R. B. Buxton
University of California ut San Diego, and +Stanford University, California
Introduction: Recent advances in perfusion imaging en- able fMRl studies of the regional cerebral blood flow (rCBF) response to brain activation. One important char- acteristic of this response is its relationship to stimulus
- duration. In particular, a linear time-invariant relation-
ship would indicate that the flow response is a convo- lution of the presented stimulus with a hemodynamic (impulse) response. The accuracy of such a model has implications for the general characterization of the cere- bral response, as well as for the design and analysis of fMRI studies [l]. Using a linearity analysis, we compare the rCBF response to visual stimuli of short duration to previously reported data in the motor cortex [Z]. Methods: In each of two experiments (motor and visual stimuli) arterial spin labeling (ASL) data was collected in 3 subjects using PICORE QUIPSS II [3] with an EPI acquisition (TR=Zs, TE=30ms, T11=700ms, T12=1400ms, FOV=24 cm, slice thickness=8mm, matrix=64x64). Data for the motor experiment was collected on a GE Signa 1.5T scanner fitted with a local gradient head coil; data for the visual experiment was collected on a Siemens Vi- sion 1.5T scanner fitted with a receive-only surface coil centered over the occipital cortex. Stimulus presenta- tion consisted of 8 cycles of either 2,6 or 18 s of stimulus (finger tapping or radial checkerboard flicker) followed by 19 s without stimulus. Two runs of each stimulus du- ration were collected in each subject. An additional run
- f the 18 s stimulus pattern was used for identification
- f activated voxels. The rCBF response was calculated
by subtracting magnetically tagged images from control images acquired at the same timepoint in the stimulus cycle, resulting in a flow measurement at each second of the stimulus. This response to a single cycle of stimu- lus presentation was then temporally smoothed to a 2 s time resolution and expressed as a percent change from
- baseline. Response linearity was analyzed by summing
shifted replicas of the measured rCBF response to short stimuli to match the duration of longer stimuli. Results: The results of the linearity analysis are shown in Figure 1. As previously reported [2], the flow response in the motor cortex appears to be fairly consistent with a linear relationship to stimulus duration, although the 2 s response makes a slight overprediction of the 6 s re-
- sponse. In contrast, the flow response in the visual cor-
tex exhibits a strong, consistent nonlinearity: an over- prediction of the long duration response by the short duration response. This nonlinearity is very similar to previously reported findings of the BOLD response [4]. Modeling: We tested whether the observed nonlineari- ties are consistent with a simple nonlinear model for the neural response to a block stimulus followed by a linear transformation from the neural response to rCBF response (see Figure 2). This model takes into account habituation effects of neural firing rates [4]. The neural response model takes parameters TV, a decay time con- stant; td, an onset delay; and a, the amount the initial response overshoots the steady-state response. To test
2s “s. 6s 2s YS. 18s 6s YS. 18s
Figure 1: Linearity analysis of motor and visual flow data.
Titles indicate the short-term response used in prediction (gray) and measured long-term responses (black).
if the measured flow responses were consistent with a linear transformation of such a neural response, we con- volved the neural response model with a simple model for the hemodynamic response (a gamma-variate func- tion with width parameter (FWHM) wh). We found that a single set of parameters was able to describe all 3 du- ration responses for each stimulus type: in the motor
COl-tt?X,
Tn=o.%,td=o.65S, P0.75, wh=5.&; in the ViSUti
cortex (see Figure 3), Tn=0.5S,td=1.5S, a=3, wh=6.2s. Figure 2: Stimulus response model. Displayed models fit
motor data (dashed) and visual cortex (solid).
Figure 3: Flow model fit (black) for visual data (gray). Conclusion: The rCBF response to brief stimuli exhibits different dynamics in the motor and visual cortices. Flow in the motor cortex is fairly consistent with a lin- ear relationship to stimulus presentation pattern; flow in the visual cortex appears nonlinear. Both rCBF re- sponses are consistent with a linear transformation of a simple nonlinear neural response model. References: 111 Dale, A. et al., EZBM,
5:329 (1997).
[21 Miller, K.L et al., Proc., 7th ISMRM, 381 (1999). [31 Wong, E.C. et al., MRM, 39:855 (1998). [41 Boynton, et al., J. Neuroscience, 16(13):4207 (1996).
- Proc. Intl. Sot. Mag. Reson. Med. 8 (2000)
500
6.2. Statistical Analysis: GLM
22
