0364

Reconstruction of highly-accelerated multi-echo fMRI using parametric POCS based multiplexed sensitivity-encoding (POCSMUSE)
Shihui Chen1, Mei-Lan Chu2, Queenie Chan3, and Hing-Chiu Chang1

1The University of Hong Kong, Hong Kong, Hong Kong, 2National Central University, Taiwan, Taiwan, 3Philips Healthcare, HKSAR, China, Hong Kong, Hong Kong

Synopsis

Multi-echo fMRI is an emerging technique that can improve the fidelity and interpretability of fMRI, such as differentiating BOLD and non-BOLD signals. In multi-echo fMRI acquisition, high acceleration factor for parallel imaging (i.e., R = 3) was used to achieve reasonable TE interval and maintain the same resolution as single-echo acquisition. The accelerated multi-echo data are reconstructed using SENSE, which may suffer from undesired noise amplification. In this study, we proposed a multi-echo multi-segment EPI (MEMS-EPI) technique to acquire multi-echo fMRI with high acceleration factor, and then used parametric POCSMUSE algorithm to reconstruct the data.

Introduction

Multi-echo fMRI is an emerging technique that can improve the fidelity and interpretability of fMRI, such as differentiating BOLD and non-BOLD signals1. In multi-echo fMRI acquisition, high acceleration factor for parallel imaging (i.e., R = 3) was used to achieve reasonable TE interval and maintain the same resolution as single-echo acquisition. The accelerated multi-echo data were then subsequently reconstructed individually using parallel imaging reconstruction (e.g. SENSE)2. However, the SENSE-produced multi-echo image could suffer from undesired noise amplification. In addition, the feasible acceleration factor for parallel imaging can also limit the imaging matrix, so it is difficult to achieve better spatial resolution. In this study, we proposed a multi-echo multi-segment EPI (MEMS-EPI) technique to acquire multi-echo fMRI with high acceleration factor and used parametric POCSMUSE algorithm to reconstruct the data with reduced noise amplification.

Materials and methods

Data acquisition: Two sets of resting state fMRI (rsfMRI) data with different spatial resolution were acquired from one healthy subject using a 1.5T MRI scanner (GE HDxt). A modified multi-echo multi-segment EPI (MEMS-EPI) pulse sequence (Figure 1) was used to acquire multi-echo rsfMRI data with following scan parameters: TR = 2000ms, matrix size = 64×64/96×96, FOV = 256mm, number of slice = 24, slice thickness = 4mm, number of echo = 4, acceleration factor = 4 for each echo, TE64x64 = [9.5, 18.6, 27.8, 36.9] ms, and TE96x96 = [12.6, 28.3, 44.0, 59.7] ms.

Data reconstruction: First, the conventional SENSE reconstruction was performed on each echo data. Second, the accelerated multi-echo data was jointly reconstructed using parametric POCSMUSE algorithm as shown in Figure 2. At each iteration, the initial image was modulated by T2* decay weighting, coil sensitivity, and phase variations to generate multi-echo data. After applying inverse 2D FFT, the data projection was performed, and then transformed updated data back to image domain. Finally, all multi-echo images were demodulated by T2* decay weights, coil sensitivity, and phase variations, to generate initial image for the next iteration.

Data evaluation: The quality of multi-echo fMRI images reconstructed with either conventional SENSE or parametric POCSMUSE were evaluated by calculating temporal fluctuation noise image and signal-to-fluctuation ratio map. The combination of multi-echo data was performed before rsfMRI analysis. Assuming constant noise for each echo, the optimal weights of nth echo can be approximated as3

wn= TEn*S(n)/∑n TEn*S(n)

where is the average intensity across the time series of the nth echo.

Afterward, the echo-combined rsfMRI data were subsequently analysed using MELODIC implemented in FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl ). The default mode network (DMN) derived from images reconstructed with either conventional SENSE or parametric POCSMUSE were evaluated.

Results

Figures 3 and 4 show the representative multi-echo images, temporal fluctuation noise image and signal-to-fluctuation ratio map from the data with different resolution (i.e., 64×64 and 96×96) reconstructed with either conventional SENSE or parametric POCSMUSE. Figure 5 shows the DMN identified in rsfMRI data produced from different reconstruction pipeline.

Discussion

In our preliminary test, the parametric POCSMUSE reconstruction can improve the image quality compared with conventional SENSE reconstruction (i.e., less temporal fluctuation noise and higher signal-to-fluctuation ratio). It is because SENSE reconstruction suffers from undesired noise amplification that may affect the subsequent analysis of the data. With proposed method, we can increase the image matrix from 64×64 to 96×96 for improving the spatial resolution. The DMN component identified from the rsfMRI data produce by parametric POCSMUSE shows better approximation to standard DMN established from other studies4, 5. Further investigation is needed to evaluate the use of proposed method for different multi-echo fMRI researches.

Conclusion

In conclusion, the parametric POCSMUSE framework can reconstruct highly-accelerated multi-echo fMRI data with reduced noise amplification that may enable more reliable data analysis in fMRI study.

Acknowledgements

The work was in part supported by grants from Hong Kong Research Grant Council (GRF HKU17138616 and GRF HKU17121517).

References

1. Kundu, P., V. Voon, P. Balchandani, et al. Multi-echo fMRI: A review of applications in fMRI denoising and analysis of BOLD signals. Neuroimage. 2017; 154: 59-80.

2. Pruessmann, K.P., M. Weiger, M.B. Scheidegger, and P. Boesiger. SENSE: sensitivity encoding for fast MRI. Magnetic resonance in medicine. 1999; 42(5): 952-962.

3. Bhavsar, S., M. Zvyagintsev, and K. Mathiak. BOLD sensitivity and SNR characteristics of parallel imaging-accelerated single-shot multi-echo EPI for fMRI. Neuroimage. 2014; 84: 65-75.

4. Di, X. and B.B. Biswal. Identifying the default mode network structure using dynamic causal modeling on resting-state functional magnetic resonance imaging. Neuroimage. 2014; 86: 53-59.

5. Uddin, L.Q., A.C. Kelly, B.B. Biswal, F.X. Castellanos, and M.P. Milham. Functional connectivity of default mode network components: correlation, anticorrelation, and causality. Human brain mapping. 2009; 30(2): 625-637.

Figures

Figure 1 The multi-echo multi-segment EPI (MEMS-EPI) pulse sequence used to acquire multi-echo rs-fMRI data.

Figure 2 The flowchart of parametric POCSMUSE algorithm.

Figure 3 Evaluation of SENSE-produced resting state fMRI images and parametric POCSMUSE produced images with the resolution 64×64: (a) is the representative multi-echo images; (b) and (c) are temporal fluctuation noise images and signal-to-fluctuation-noise ratio maps for different echoes, respectively. The left column is derived from SENSE-produced resting state fMRI images and the right column is derived from parametric POCSMUSE produced images.

Figure 4 Evaluation of SENSE-produced resting state fMRI images and parametric POCSMUSE produced images with the resolution 96×96: (a) is the representative multi-echo images; (b) and (c) are temporal fluctuation noise images and signal-to-fluctuation-noise ratio maps for different echoes, respectively. The left column is derived from SENSE-produced resting state fMRI images and the right column is derived from parametric POCSMUSE produced images.

Figure 5 Default mode network (DMN) component identified in the resting state fMRI data from a single subject with different spatial resolution. The data were reconstructed with either conventional SENSE or parametric POCSMUSE algorithm.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
0364