Fig 1.
(a) The hemodynamic response (HRF) to a brief 1-second stimulation at three ages (Arichi et al, 2012). In adults the HRF is dominated by a positive peak, while at 38 weeks gestational age (GA) neonates have positive and negative peaks of similar magnitude. At 32 weeks GA the HRF is dominated by a positive peak, but it is much delayed. (b) The form of the HRF affects the power of different stimulation designs. To illustrate this, the response to a 30s-long cycle of stimulation (yellow) and rest was calculated by convolving the HRFs with a boxcar. For this design, at all three ages, there was substantial modulation of the BOLD signal through time. The signal in adults and 38 week infants was highly correlated, but at 32 weeks the signal has a different phase. (c) In contrast, for 45s of stimulation is followed by 45s of rest, the 38 week infants only have small peaks of modulation in the BOLD signal, and so much reduced power would be expected.
Table 1.
Demographic and clinical details of the neonates.
Fig 2.
To formally assess the efficacy of different designs, it is necessary to establish the characteristics of the noise in the fMRI signal.
In both adult (a) and neonatal (b) participants, spectra were well fitted by flat-frequency spectrum component combined with a 1/f component. In 2/5 neonates the overall level of noise was similar to adults, but in 3/5 neonates, the noise level was elevated. This increased level of noise was associated with larger movement in these neonates (c).
Fig 3.
Simulation was used to assess the power of block designs varying in stimulation/rest cycle duration (x-axis of each subplot).
Higher values (y-axis of each subplot) correspond to greater statistical power. For example, in the top left subplot, peak statistical power was obtained with a block design of total cycle length 24 s (i.e., 12 s stimulation, then 12 rest). These calculations were repeated for matched or mismatched HRFs during analysis (three rows–true HRF used in simulation; four columns–HRF used for modeling). For the first three HRF columns, the SPM-T statistic is reported. For the flexible HRF model in the fourth column, the square root of an adjusted F statistic is displayed so that the corresponding p-value will match that of the T statistics. The mean +/- one standard deviation is shown. The grey bars show the distribution of fits to null data.
Fig 4.
If the HRF for a participant is unknown, a flexible basis set might be useful.
(a) We followed the “FLOBs” procedure (see text). The HRFs were parameterized, to allow interpolation of shape between 32 and 38 weeks GA, and between 38 weeks and adults. The black curves show the original HRFs, and the red curves some illustrative interpolated values. (b) Principal components analysis was then used to find a basis set that captured the variance in the interpolated set. Three components captured 99.7% of the variance.
Fig 5.
The flexible basis set can be used to estimate the form of the HRF.
Here we show the accuracy of the estimation of the HRF for adult, 32 week GA and 38 week GA, as a function of the block design timing (stimulation/rest cycle, x-axis). Correlation was used to assess the similarity of the true (simulated) HRF to the HRF estimated from the flexible model. The mean +/- one standard deviation is shown (calculated in Fisher-transformed space, and then back-transformed).