Multiband multi-echo imaging of simultaneous oxygenation and flow timeseries for resting state connectivity

Abstract

A novel sequence has been introduced that combines multiband imaging with a multi-echo acquisition for simultaneous high spatial resolution pseudo-continuous arterial spin labeling (ASL) and blood-oxygenation-level dependent (BOLD) echo-planar imaging (MBME ASL/BOLD). Resting-state connectivity in healthy adult subjects was assessed using this sequence. Four echoes were acquired with a multiband acceleration of four, in order to increase spatial resolution, shorten repetition time, and reduce slice-timing effects on the ASL signal. In addition, by acquiring four echoes, advanced multi-echo independent component analysis (ME-ICA) denoising could be employed to increase the signal-to-noise ratio (SNR) and BOLD sensitivity. Seed-based and dual-regression approaches were utilized to analyze functional connectivity. Cerebral blood flow (CBF) and BOLD coupling was also evaluated by correlating the perfusion-weighted timeseries with the BOLD timeseries. These metrics were compared between single echo (E2), multi-echo combined (MEC), multi-echo combined and denoised (MECDN), and perfusion-weighted (PW) timeseries. Temporal SNR increased for the MECDN data compared to the MEC and E2 data. Connectivity also increased, in terms of correlation strength and network size, for the MECDN compared to the MEC and E2 datasets. CBF and BOLD coupling was increased in major resting-state networks, and that correlation was strongest for the MECDN datasets. These results indicate our novel MBME ASL/BOLD sequence, which collects simultaneous high-resolution ASL/BOLD data, could be a powerful tool for detecting functional connectivity and dynamic neurovascular coupling during the resting state. The collection of more than two echoes facilitates the use of ME-ICA denoising to greatly improve the quality of resting state functional connectivity MRI.

Fig 1. MBME ASL/BOLD pulse sequence design.

The sequence consists of a pCASL labeling train, followed by a post-label delay, and a multi-echo EPI readout. The phase is rewound to the start of k-space after each echo and the next echo is acquired. Multiband imaging was also utilized by inserting a multiband excitation pulse in place of the single band pulse. Finally, blipped-CAIPI was employed to reduce g-factor penalties associated with MB imaging.

Fig 2. Schematic showing the resting state functional connectivity pipeline for the ASL and BOLD echoes.

The first and second echoes were processed separately to yield the PW and E2 data respectively. Echoes were also combined using a T2*-weighted approach to generate the MEC dataset. This dataset was further denoised using ME-ICA resulting in the MECDN dataset. Each echo was despiked, volume registered using the transformation matrix from the first echo, detrended, registered to the anatomical image, and transformed to MNI space. Additional preprocessing steps differ for each dataset and are shown in Row 2. For all datasets, after preprocessing, functional connectivity was assessed with seed-based and dual-regression approaches (Row 3).

Fig 3. Representative perfusion-weighted, individual echo, and multi-echo images.

(A) Example individual echo, MEC, and MECDN images from one subject. Image SNR decreases with echo time. Image quality improves with echo combination and signal in the inferior portions of the brain is recovered. All images share the same color scale. (B) Perfusion weighted images. Images were produced by averaging and subtracting the label images from the control images. High-resolution whole-brain images were collected in a relatively short amount of time with reduced signal loss caused by T1-relaxtion of the labeled blood.

Fig 4. MEICA performance.

A. Curves of κ and ρ for all subjects. Both curves display the characteristic “L” shape expected from the ME-ICA algorithm. κ describes the goodness of fit to the TE dependence of each component and ρ described the fit to a ΔS₀ model. In general, components with κ above and ρ below the elbow are kept in the denoised timeseries. B-E display example networks from one representative subject. B. An example accepted BOLD component (DMN). C. A rejected component classified as an R2* artifact. D. A rejected non-BOLD component. E. A rejected PW component.

Fig 5. Group seed-based functional connectivity maps.

Connectivity maps are displayed for PCC, L/R motor cortex, L/R insula, and L/R hippocampus seeds. Maps are the result of a one-sample t-test on the Fisher-transformed z-scores and were thresholded at P < 0.005 with minimum cluster size = 46, α = 0.05. For all seed regions, connectivity was markedly increased in terms of network size and correlation strength for the MECDN datasets compared to the others. Limited insula and hippocampus connectivity was observed for the E2 and PW datasets. Bilateral insular connectivity was seen for the MEC dataset. The MECDN data produced significant bilateral connectivity with long-range connections for both the insula and hippocampus seeds. Robust connectivity was detected with the PW data for the PCC and motor network seeds. Some bilateral connectivity was seen for the insula seeds.

Fig 6. Quantitative results.

Mean correlation in overlapping significant voxels (Top), mean correlation in significant voxels for the E2, MEC, MECDN, and PW data separately (Middle), and network size, displayed as a fraction of intracranial voxels (Bottom). Voxels with P < 0.005 and minimum cluster size = 46, α = 0.05 were considered significant. Parameters were extracted on an individual subject basis. * = P < 0.05; ** = P < 0.01; *** = P < 0.001, Bonferroni-corrected.

Fig 7. Group dual-regression based functional connectivity maps.

Connectivity maps are displayed for the default mode network, motor network, and salience network. Maps are the result of a one-sample t-test on the z-scores and were thresholded at P < 0.005 with minimum cluster size = 46, α = 0.05. Additional clusters were seen for the MEC data compared to the E2 data and for all networks, and for the MECDN data compared to the MEC data for the DMN and salience networks. Existing clusters also tended to be larger for the MECDN data for these networks. Motor network connectivity was similar between the MEC and MECDN datasets. Bilateral, long range connectivity was seen for all datasets, including the PW data.

Fig 8. Group CBF/BOLD coupling.

Results are shown for the E2, MEC and MECDN datasets. Widespread coupling was observed and increased coupling was seen within well-known brain networks including the DMN, and visual networks. Stronger, more widespread coupling was seen for the MECDN data compared to the MEC and E2 data.