Henriette Lambers^{1}, Martin Segeroth^{1}, Franziska Albers^{1}, Lydia Wachsmuth^{1}, Timo Mauritz van Alst^{1}, and Cornelius Faber^{1}

For a reliable analysis of BOLD fMRI data, a suitable model of the hemodynamic response is essential. Therefore, an accurate model of the BOLD response of small animals is required in preclinical studies. Commonly used analysis tools like SPM or BrainVoyager have implemented HRFs optimized for humans by default. Since the BOLD responses of rats proceed faster than those of humans, we have determined a generic rat HRF based on 98 BOLD measurements of 35 rats which can be used for statistical parametric mapping. Statistical analysis of rat data showed a significantly improved detection performance using this rat HRF.

Experiments
were performed on SD and Fischer rats under medetomidine or isoflurane anesthesia
at 9.4 T with single-shot GE-EPI (TR/TE 100/18ms, 350x325μm² or 375x375μm², 8-14
1.2mm thick slices) upon electrical paw, mechanical paw or optogenetic
stimulation using block design paradigms. Measurements were assigned to 13
different groups according to their experimental conditions (e.g. strain,
anesthesia, stimulation) (Table 1).
Positive BOLD responses of realigned (SPM8) datasets were extracted and
examined using MATLAB: A U-test determined voxelwise whether the
signal during stimulation and rest period differed significantly for the S1
region on the activated side of the brain. Time courses of the BOLD responses
were calculated by summing up the signal of all voxels, which showed a
significant and positive signal change, and subsequently averaging over all
stimulation cycles. The convolution of the stimulation paradigm and the
canonical HRF was fitted to the time course of the BOLD responses. The
canonical HRF was defined as in SPM^{1}, expanded by an amplitude parameter
A: $$ A \cdot e^{-bt}\left( \frac{b^{p_1}}{\Gamma \left( p_1\right) }
\cdot t^{p_1-1} - \frac{b^{p_2}}{V \cdot \Gamma \left( p_2\right) }
\cdot t^{p_2-1}\right)$$ A least squares fit of this equation to the measured BOLD response was
performed.
Time courses of the normalized HRFs for the different groups were
compared pairwise, using a customized functional t-test^{2}. Resulting
p-values were Bonferroni corrected. All HRFs were normalized and averaged
across all groups that showed no substantial differences. The canonical HRF
(without amplitude A) was fitted to
the resulting time course of the rat HRF. The resulting parameters (*b*, *p _{1}*,

1. Friston, K.J., 2017. SPM12. Wellcome Trust Centre for Neuroimaging.

2. Ramsay, J., Hooker, G., Spencer, G., 2009. Functional Data Analysis with R and MATLAB, 1st ed. Springer, Dordrecht, Heidelberg, London, New York.

Table
1: Measurements were assigned
to 13 different groups according to their experimental conditions (strain, gender,
anesthesia, ventilation, stimulation, paradigm (length of stimulation and rest
period), pain).

Figure 1: Calculated
HRFs for 1s and 5s stimulation and their confidence interval. The
functional t-test showed no significant differences (p=0.06), but
the HRFs tended to show differences in timing of maximum and undershoot.

Figure
2: The generic rat HRF and its standard deviation obtained from 98 individual
data sets is shown in green. It proceeds significantly faster than the human
HRF (black), which is implemented in SPM by default.

Figure
3: Statistical analysis was performed on 20 measurements with either 5s
or 10s electrical paw stimulation, using either the generic rat or the
human HRF.
BOLD maps showed clusters in the same regions but differed in cluster
size and
t-values (A). A U-test showed, that BOLD
clusters (B) and t-values (C) were significantly smaller when using the
human
HRF than using the generic rat HRF.