The fractional amplitude of low frequency fluctuations (fALFF) of resting-state BOLD fmri (rs-fmri) frequency spectra comprises both neuronal and vasomotor fluctuations. Reductions in fALFF have been reported in certain diseases and normal ageing^1,2. However, overlap of cardio-respiratory frequency components with the fALFF range, and inter-subject variations in the noise baseline limit its suitability as a biomarker for individual subjects. Here, we acquired whole-brain rs-fmri data at 7 Tesla with short-TR (328ms) using multiband EPI^3,4. Established correction methods, such as RETROICOR⁵ minimise for cardio-respiratory components, but not for baseline noise and its implementations need further adaptions to handle mutiband data⁶. We propose an alternative method to correct multiband frequency spectra using a GLM post-processing in the frequency-domain that minimises cardio-respiratory and baseline noise. We used the previously reported fALFF decline with age to test the effectiveness of the post-processing corrections.An acquisition of TR=328ms is able to characterise frequencies up to 1.5Hz, reducing the risk of aliasing of cardio-respiratory frequencies into the fALFF range. However, physiological contributions, particularly respiration, still overlap with the fALFF range. Variations in cardiac and respiration rates between age groups or different health conditions might bias group study results. Minimisation of the baseline and physiological noise components is expected to increase statistical significance and is further necessary for fALFF to serve as a biomarker in individual subjects.

Data: 12 minutes of rs-fmri multiband EPI were acquired in 10 “younger” (mean age 23) and 10 “older” (mean age 63) subjects on a 7T whole body scanner (Siemens, Germany).

Sequence Parameters: TR=328ms, TE=26ms, FA=33º, voxel=3mm isotropic, 28 slices, FOV=200mm, GRAPPA=3, multiband=4. A T1-MPRAGE scan was acquired for registration and masking.

Physiological Recording: Cardio-respiratory signals were recorded at 1kHz (Acknowledge, BioPac, USA).

Pre-Processing: Spatial smoothing (6mm), high-pass filtering (100s), motion correction and regression of motion parameters were performed in FSL FEAT.

Spectral Post-Processing: Voxel-wise normalised power spectra were created by Fourier-transforming the time series:⁷ $$$\sqrt{a_{f_i}^2+b_{f_i}^2}\Large/ \{\normalsize \sum_{f_k = 0Hz}^{f_k = 1.5Hz} \sqrt{a_{f_k}^2+b_{f_k}^2}\Large\}$$$. Subsequently a voxel-wise GLM was applied in the frequency domain: We assume that the spectrum above 0.25Hz is free of neurovascular fluctuations and is solely modulated by a baseline offset ($$$\alpha$$$), and cardio-respiratory fluctuations ($$$\beta_{c}$$$ and $$$\beta_{r}$$$). The physiological recording was down-sampled to TR=328ms and the power spectra were derived to yield the explanatory variables ($$$X_{c}$$$ and $$$X_{r}$$$) for a GLM of the form $$ Y(f_{i}) = \alpha + \beta_{c}X_{c}(f_i) + \beta_{r} X_{r}(f_i). $$ $$$\alpha, \beta_c$$$ and $$$\beta_r$$$ were estimated from the spectrum between f=0.25Hz and the maximum resolvable frequency f=1.5Hz, see Fig.1. Voxel spectra $$$Y(f_i)$$$ with significant ($$$p<0.01$$$) baseline, cardiac and/or respiratory modulations were subsequently corrected over the entire spectrum, minimising noise in the fALFF range, see Fig.2.

Analysis: For each subject voxel spectra were processed with and without GLM minimisation. The average spectrum in the grey matter mask was calculated. Within that mask the fALFF, defined as the sum between 0.01Hz and 0.1Hz, was calculated, i.e. it quantifies the percentage of BOLD signal power within that frequency band. We compared the corrected and uncorrected fALFF between the younger and older cohort. We further calculated the fALFF for six-fold down-sampled data to compare our results with a more typical rs-fmri TR of 2 seconds. Two-sample t-tests were used for statistical analysis.

Across all subjects cardiac spectral components overlap the fALFF range in the short-TR spectra by 4± 2% and respiration components overlap by 12± 6%. Cardiac and respiratory regression results in an average fALFF reduction of 1± 0.8% and 5± 6% respectively. The baseline regression leads to an average fALFF reduction of 27±10%. Figure 3 shows the fALFF results in the younger and older groups. Short- and long-TR derived fALFF values show a decrease with age as expected. The long-TR fALFF difference of the group mean is 8%, which increases for the short-TR fALFF to 27% and 49% for the uncorrected and corrected values respectively. The statistical significance of the fALFF change for the corrected short-TR fALFF is a factor of 20 higher than the uncorrected short-TR fALFF (p=5.1x10^-5 vs. p=1.3x10^-3) and improves by a factor of 200 when compared to the down-sampled TR of 2 seconds (p=0.0126).

1. Baseline and physiological noise components in multiband rs-fmri frequency spectra can be minimised with a GLM fit when cardio-respiratory signals are recorded.

2. High temporal resolution data increase statistical significance of the fALFF group difference compared to conventional TRs of 2 seconds.

3. On average 33% of the short-TR fALFF was due to baseline and physiological components. Removing these components further improves statistical significance of the fALFF group difference.