Introduction

Fast, quantitative and high-quality imaging is the common pursuit of structural and functional MRI. Recent studies suggest that T₂* mapping can significantly increase BOLD sensitivity in fMRI. This observation has led to a renewed interest in continued exploration of quantitative BOLD (qBOLD) fMRI signals.^1,2 However, multi-echo EPI commonly used in qBOLD fMRI pipeline delivers echo image through additional echo train, which reduces achievable temporal resolution and increases sensitivity to subject motion. Besides, conventional EPI can only acquire a small number of echoes and suffers from geometric distortion and image blurring, which limits its application in clinics. In this work, we propose a method that utilizes overlapping-echo detachment (OLED) technique^3-6 to improve acquisition efficiency, termed gradient-echo multiple overlapping-echo detachment (GRE-MOLED) method, and apply it in quantitative structural MRI (sMRI, including T₂* mapping and quantitative susceptibility mapping (QSM)) and functional MRI (qBOLD). We demonstrate that the GRE-MOLED enables high-quality quantitative sMRI and high BOLD sensitivity in fMRI.

Methods

Pulse sequence: Figure 1(a) shows the diagram of GRE-MOLED sequence with multiple echo trains. The four excitation pulses with a same flip angle α = 30° are used to prepare gradient echoes with different evolution time. The echo-shifting gradients G₁, G₂, G₃ and G₄ are used to shift the four echoes from the k-space center at the same k-space data collection.
In vivo experiments: In vivo data were acquired from 5 healthy volunteers, on a 3T whole-body MRI system with a 20-channel head coil (MAGNETOM Prisma, Siemens Healthineer, Erlangen, Germany). A conventional 3D GRE sequence and the proposed 2D GRE-MOLED were acquired for comparison and quantitative analysis in structural MRI scans. All functional MRI scans were acquired using the GRE-EPI sequence followed by the GRE-MOLED sequence. The acquisition parameters were as follows: FOV = 22×22 cm², voxel size = 1.7×1.7 mm², slice thickness = 3.5 mm (fMRI) and 1.7 mm (sMRI), TR = 3.2 s (fMRI) and 12 s (sMRI), slice number = 32 (fMRI) and 64 (sMRI), GRAPPA acceleration factor = 2, echo spacing (ESP) = 0.93 ms, and fMRI measurements = 120. The TE was 30 ms for GRE-EPI and 11.8, 32.7, 53.1, 76.5 ms (four overlapping echoes) for GRE-MOLED. After a resting state scan, a conventional block-design visual stimulation task with a 30 s “checkerboard” followed by a 30 s black field on-off blocks and motion task with fist exercise were performed for the fMRI acquisitions. A T₁-weighted anatomical image was also acquired using T₁-MPRAGE sequence with resolution of 1×1×1 mm³.
Parameters reconstruction and data analysis: As shown in Figure 1(b), a convolutional neural network (CNN) consisting of encoder sub-network and decoder sub-network was applied to perform end-to-end T₂* and tissue phase mapping from overlapping-echo images. QSM reconstruction of the tissue phase was performed in open-source STI Suite with iLSQR algorithm⁷. Totally 7000 pairs of synthetic images were generated for network training through Bloch simulation. As for fMRI data, preprocessing and statistical analysis were performed on conventional GRE-EPI data and MOLED T₂* maps using the statistical parametric mapping (SPM) software. A first level general linear model was used for regression analysis and the p value = 0.001 was used in the statistical maps.

Results

Figure 2 shows the visual and quantitative results of reconstructed T₂* maps and QSM of 1.7 mm isotropic GRE-MOLED and GRE acquisition. Single-shot GRE-MOLED provides high-quality images in 12 s, close to the results of multi-shot GRE in 6 min, and achieves high accuracy in ROI analysis. Figure 3 shows the distortion correction capability of GRE-MOLED with CNN reconstruction. Compared with GRE-EPI and original MR image from GRE-MOLED, the reconstructed T₂* maps are free from geometric distortion as the brain boundaries are identical to those from distortion-free GRE. The temporal signal-to-noise ratio (tSNR) maps calculated from resting-state scans are provided in Figure 4. It is evident that the T₂* maps of GRE-MOLED recover some signal lost and show increases in tSNR in the prefrontal regions. However, in other regions (especially in gray matter), the tSNR of GRE-EPI is much higher compared with both original overlapping-echo images and T₂* maps. Figure 5 illustrates the results of fMRI to vision task and motion task from T₂*-weighted images and T₂* maps using conventional method and GRE-MOLED. The percentage signal change (PSC) for BOLD sensitivity representation were calculated by parameters estimates and contrast maps, as described in previous work⁸. The results show a substantial increase in the T₂* map time series (from GRE-MOLED) in all tasks and subjects.

Discussion and conclusion

We have demonstrated that GRE-MOLED can provide simultaneous T₂* mapping and QSM comparable to traditional multi-shot GRE within a single shot. As quantitative T₂* is introduced with high temporal resolution, we see a significantly increased BOLD sensitivity compared with conventional method. T₂* map time series achieves higher BOLD sensitivity with lower tSNR, therefore, we believe that the separation of T₂* and M₀ can obviously reduce system-level effects (e.g. subject-motion and inflow) from the original T₂* signals, which has been reported previously¹. These results show the potential of GRE-MOLED for quantitative structural and functional MRI, which provides a means of detecting BOLD signals with higher sensitivity under standard fMRI pipeline.

Acknowledgements

This work was supported by the National Natural Science Foundation of China under grant numbers 82071913 and 11775184, and Science and Technology Project of Fujian Province 2019Y0001.