Optimization and validation of multi-echo, multi-contrast SAGE acquisition in fMRI

Abstract

The purpose of this study was to optimize and validate a multi-contrast, multi-echo fMRI method using a combined spin- and gradient-echo (SAGE) acquisition. It was hypothesized that SAGE-based blood oxygen level-dependent (BOLD) functional MRI (fMRI) will improve sensitivity and spatial specificity while reducing signal dropout. SAGE-fMRI data were acquired with five echoes (2 gradient-echoes, 2 asymmetric spin-echoes, and 1 spin-echo) across 12 protocols with varying acceleration factors, and temporal SNR (tSNR) was assessed. The optimized protocol was then implemented in working memory and vision tasks in 15 healthy subjects. Task-based analysis was performed using individual echoes, quantitative dynamic relaxation times T₂ ^* and T₂, and echo time-dependent weighted combinations of dynamic signals. These methods were compared to determine the optimal analysis method for SAGE-fMRI. Implementation of a multiband factor of 2 and sensitivity encoding (SENSE) factor of 2.5 yielded adequate spatiotemporal resolution while minimizing artifacts and loss in tSNR. Higher BOLD contrast-to-noise ratio (CNR) and tSNR were observed for SAGE-fMRI relative to single-echo fMRI, especially in regions with large susceptibility effects and for T₂-dominant analyses. Using a working memory task, the extent of activation was highest with T₂ ^*-weighting, while smaller clusters were observed with quantitative T₂ ^* and T₂. SAGE-fMRI couples the high BOLD sensitivity from multi-gradient-echo acquisitions with improved spatial localization from spin-echo acquisitions, providing two contrasts for analysis. SAGE-fMRI provides substantial advantages, including improving CNR and tSNR for more accurate analysis.

Keywords: BOLD contrast; BOLD contrast-to-noise ratio (CNR); functional MRI (fMRI); multi-echo fMRI; spin-echo fMRI.

Fig. 1.

Single-subject acquisition for each TE across acceleration factor combinations. T1-weighted (T1-w) sagittal image is included with each maximum brain coverage achievable per protocol listed in the legend.

Fig. 2.

Representative SAGE images for minimum and maximum acceleration protocols per TE. Arrows highlight visible acceleration-related artifacts, apparent in TE₁as bands and in TE₅as loss of definition and blurring.

Fig. 3.

Group-level mean tSNR across acceleration factor combinations for TE_1-5in gray (A) and white matter (B). There were no significant differences in tSNR in gray or white matter across TEs between protocols with feasible acquisition parameters (i.e., full brain coverage ≥120 mm, TE₁< 10 ms, TE₅< 100 ms).

Fig. 4.

TE₂per acquisition and corresponding g-factor analysis in a representative subject. Increasing g-factor values may correspond to acceleration-associated artifacts seen in TE₂. Histograms exhibiting the distribution of whole-brain voxelwise g-factor values are provided.

Fig. 5.

Group-level tSNR maps for the working memory task across macro- (TE₂, SAGE T₂*, and SAGEwT₂^*) and microvascular (TE₅, SAGE T₂, and SAGEwT₂) acquisitions. Anatomical T₁-w image is included for reference.

Fig. 6.

Group-level CNR maps for the working memory task across macro- (TE₂, SAGE T₂*, and SAGEwT₂^*) and microvascular (TE₅, SAGE T₂, and SAGEwT₂) acquisitions. Anatomical T₁-w image with task-related regions of interest (ROIs) is included for reference.

Fig. 7.

Group-levelt-maps for the working memory task across macro- (TE₂, SAGE T₂*, and SAGEwT₂^*) and microvascular (TE₅, SAGE T₂, and SAGEwT₂) acquisitions. # vox = voxel count for significant functional activation.

Fig. 8.

N-back task-based activation histograms for ROIs with negative (A-C; G) and positive (D-G) activation. Each histogram shows the distribution of statistically significant voxels (p < 0.001, cluster size corrected) for each single-echo or SAGE-based method, with the significant voxel count in the legend. FP had both positive (dorsolateral prefrontal cortex) and negative (dorsomedial prefrontal cortex) activation. Voxel clusters of less than 100 were removed for visualization. FMC = Frontal Medial Cortex; CGa = Cingulate Gyrus anterior division; CGp = Cingulate Gyrus posterior division; IFG = Inferior Frontal Gyrus; MFG = Middle Frontal Gyrus; PL = Parietal Lobe; FP = Frontal Pole.

Fig. 9.

Task-based activation (corrected for multiple comparisons, FDRq< 0.05) in a representative subject for all SAGE macro- (TE₂, SAGE T₂*, and SAGEwT₂^*) and microvascular (TE₅, SAGE T₂, and SAGEwT₂) analyses. The VENAT probabilistic atlas, co-registered into native space, is provided for reference corresponding to large draining veins in the brain. VENAT and activation maps are overlaid on T₁-w anatomical image co-registered into native space. Macrovascular-weighted activations overlap with the straight, transverse, and sagittal sinuses, whereas microvascular-weighted activations appear to be largely constrained to gray matter of the visual cortex. P(VENAT) = VENAT atlas probability estimate.