Multiband multi-echo simultaneous ASL/BOLD for task-induced functional MRI

Abstract

Typical simultaneous blood oxygenation-level dependent (BOLD) and arterial spin labeling (ASL) sequences acquire two echoes, one perfusion-sensitive and one BOLD-sensitive. However, for ASL, spatial resolution and brain coverage are limited due to the T1 decay of the labeled blood. This study applies a sequence combining a multiband acquisition with four echoes for simultaneous BOLD and pseudo-continuous ASL (pCASL) echo planar imaging (MBME ASL/BOLD) for block-design task-fMRI. A multiband acceleration of four was employed to increase brain coverage and reduce slice-timing effects on the ASL signal. Multi-echo independent component analysis (MEICA) was implemented to automatically denoise the BOLD signal by regressing non-BOLD components. This technique led to increased temporal signal-to-noise ratio (tSNR) and BOLD sensitivity. The MEICA technique was also modified to denoise the ASL signal by regressing artifact and BOLD signals from the first echo time-series. The MBME ASL/BOLD sequence was applied to a finger-tapping task functional MRI (fMRI) experiment. Signal characteristics and activation were evaluated using single echo BOLD, combined ME BOLD, combined ME BOLD after MEICA denoising, perfusion-weighted (PW), and perfusion-weighted after MEICA denoising time-series. The PW data was extracted using both surround subtraction and high-pass filtering followed by demodulation. In addition, the CBF/BOLD response ratio and CBF/BOLD coupling were analyzed. Results showed that the MEICA denoising procedure significantly improved the BOLD signal, leading to increased BOLD sensitivity, tSNR, and activation statistics compared to conventional single echo BOLD data. At the same time, the denoised PW data showed increased tSNR and activation statistics compared to the non-denoised PW data. CBF/BOLD coupling was also increased using the denoised ASL and BOLD data. Our preliminary data suggest that the MBME ASL/BOLD sequence can be employed to collect whole-brain task-fMRI with improved data quality for both BOLD and PW time series, thus improving the results of block-design task fMRI.

Fig 1. MBME ASL/BOLD pulse sequence design.

The sequence consists of an unbalanced pCASL labeling train, followed by a PLD, and finally an ME EPI readout. The first echo train was acquired after the acquisition of navigator echoes through the center of k-space. The phase was then rewound to the start of k-space, and the next echo train was acquired. This was repeated three times for a total of four echoes. MB imaging was implemented by replacing the single-band excitation pulse with a MB excitation pulse. Finally, blipped-CAIPI was utilized to shift the FOV of aliased slices and reduce g-factor penalties associated with MB imaging. Reprinted from [31] under a CC BY license, original copyright 2017.

Fig 2. Schematic showing the processing pipeline for the ASL and BOLD echoes.

The first and second echoes were processed separately to yield the PW_none and E2 data, respectively. Echoes were combined using a T2*-weighted approach to generate the MEC dataset. This dataset was further denoised using MEICA, resulting in the MECDN dataset. No additional regression was performed in the GLM for the PW and MECDN datasets. Example activation curves and model fits are shown for the different datasets.

Fig 3. Schematic showing the perfusion-weighted denoising procedure.

Echoes were first despiked, volume registered, and coregistered to MNI space. Each echo was then individually low-pass filtered at f ≤ 0.09 Hz. Echoes were then combined using a T2*-weighted approach. This low-pass filtered, multi-echo combined dataset was fed into the MEICA algorithm, which extracted independent components and classified them as artifact, BOLD, or indeterminate. The BOLD and artifact components were regressed from the unfiltered first-echo data resulting in a denoised first-echo dataset. Surround subtraction and high-pass filtering followed by demodulation were performed on this data leading to denoised PW datasets. A GLM was employed on this data to determine activation.

Fig 4. Representative PW and BOLD datasets.

(A) Mean PW_ss,none (top) and PWDN_ss,art+BOLD data (bottom) images. These images were created by averaging and subtracting the label images from the control images. MB imaging allows for the collection of whole-brain images in a relatively short readout time reducing T1 effects. (B) Example individual echo, MEC, and MECDN images from a single time point from one subject. Image quality improves with echo combination.

Fig 5. Group tSNR maps.

The tSNR significantly increased from the E2 to MECDN data (p<0.001). For the PW data, tSNR maps are shown for the SS data. The tSNR for the PWDN_ss,art+BOLD significantly increased compared to the PWDN_ss,art and PW_ss,none data (p<0.001).

Fig 6. Finger-tapping task BOLD results.

(A) Robust bilateral activation was seen in the motor cortex, including the pre and postcentral gyrus, medial frontal gyrus, and cerebellum for all datasets. Increased activation was observed for the MEC data compared to E2 data and for the MECDN data compared to the MEC and E2 data. Activation was also observed in subcortical areas for the MECDN data. All maps were thresholded at p<0.005 and cluster corrected with a minimum cluster size of 182 voxels (α<0.05). (B) Average BOLD time-series extracted from a mask of voxels active for all BOLD datasets.

Fig 7. Finger-tapping task PW results.

(A) For the SS results (left), bilateral activation was observed in the motor cortex for the PW_ss,none and PWDN_ss,art+BOLD data. An increased activation area was seen for the PWDN_ss,art+BOLD data compared to the PW_ss,none data. The HD data (right) showed increased activation compared to the SS data, however no differences were seen between the denoised and non-denoised data the. All maps were thresholded at p<0.005 and cluster corrected with a minimum cluster size of 131 voxels (α<0.05). (B) Average SS PW signal from one representative subject (left) and average HD PW signal from the same subject (right). All PW signal was extracted from a mask of voxels active for all PW datasets. The denoised SS time-series appear less noisy with less variance compared to the non-denoised time-series. This effect is less apparent for the HD data.

Fig 8. CBF/BOLD relationship.

(A) The mean ratio of CBF to BOLD signal is plotted across subjects in active voxels (CBF/BOLD response ratio) for SS (left) and HD data (right). The response ratio was examined for non-denoised (PW/E2) and denoised data (PWDN/MECDN). No significant difference was observed between the non-denoised and denoised response ratios averaged across the middle five activation TRs (TR #s 8–12). (B) CBF/BOLD response ratio plotted against baseline CBF for SS (left) and HD data (right). A significant negative correlation was observed between the CBF/BOLD ratio and mean baseline CBF for all processing schemes except for HD, PW/E2 which trended toward significance.

Fig 9. BOLD/CBF coupling.

(A) Group averaged correlation between CBF and BOLD time-series for non-denoised (PW,E2) and denoised (PWDN,MECDN) data for SS (left) and HD data (right). In general, CBF-BOLD correlation increased with denoising. This was confirmed quantitatively (B) where the mean correlation extracted from a mask of significantly correlated voxels was highest for fully denoised datasets (PWDN,MECDN). *** = p<0.001, ** = p<0.01, * = p<0.05, Bonferroni-corrected.