Comparison of SMS-EPI and 3D-EPI at 7T in an fMRI localizer study with matched spatiotemporal resolution and homogenized excitation profiles.

Caroline Le Ster,Stanislas Dehaene,Nicolas Boulant,Franck Mauconduit,Antonio Moreno,Florent Meyniel,Evelyn Eger,Ruediger Stirnberg,Alexandre Vignaud,Benedikt A Poser,Vincent Gras

doi:10.1371/journal.pone.0225286

Abstract

The simultaneous multi-slice EPI (SMS-EPI, a.k.a. MB-EPI) sequence has met immense popularity recently in functional neuroimaging. A still less common alternative is the use of 3D-EPI, which offers similar acceleration capabilities. The aim of this work was to compare the SMS-EPI and the 3D-EPI sequences in terms of sampling strategies for the detection of task-evoked activations at 7T using detection theory. To this end, the spatial and temporal resolutions of the sequences were matched (1.6 mm isotropic resolution, TR = 1200 ms) and their excitation profiles were homogenized by means of calibration-free parallel-transmission (Universal Pulses). We used a fast-event “localizer” paradigm of 5:20 min in order to probe sensorimotor functions (visual, auditory and motor tasks) as well as higher level functions (language comprehension, mental calculation), where results from a previous large-scale study at 3T (N = 81) served as ground-truth reference for the brain areas implicated in each cognitive function. In the current study, ten subjects were scanned while their activation maps were generated for each cognitive function with the GLM analysis. The SMS-EPI and 3D-EPI sequences were compared in terms of raw tSNR, t-score testing for the mean signal, activation strength and accuracy of the robust sensorimotor functions. To this end, the sensitivity and specificity of these contrasts were computed by comparing their activation maps to the reference brain areas obtained in the 3T study. Estimated flip angle distributions in the brain reported a normalized root mean square deviation from the target value below 10% for both sequences. The analysis of the t-score testing for the mean signal revealed temporal noise correlations, suggesting the use of this metric instead of the traditional tSNR for testing fMRI sequences. The SMS-EPI and 3D-EPI thereby yielded similar performance from a detection theory perspective.

Highlights

With ultra-high fields (UHF), functional MRI benefits from increased signal-to-noise ratio (SNR) and sensitivity to BOLD contrast so that higher spatial resolutions can be reached
The Simultaneous multislice EPI (SMS-EPI) and 3D-EPI sequences were compared in terms of raw temporal SNR (tSNR), t-score testing for the mean signal [31], activation strength and activation accuracy. tSNR maps were computed for each subject as the mean over the standard deviation of the raw time-series of the functional scans for each voxel after realignment, distortion correction, detrending and normalization on the 2 mm resolution MNI brain template. tSNR maps were averaged over the population
The results presented were obtained from the general linear model (GLM) analysis performed with inclusion of the nuisance regressors and FAST noise whitening model

Summary

Introduction

With ultra-high fields (UHF), functional MRI (fMRI) benefits from increased signal-to-noise ratio (SNR) and sensitivity to BOLD contrast so that higher spatial resolutions can be reached. Recent studies at 3T [10, 11] and 7T [12,13,14] have shown that proper modelling of the physiological noise [15] can improve the tSNR of 3D-EPI so as to outperform SMS-EPI acquisitions. This is the case when the baseline SNR is sufficiently high to be in a physiological noise-dominated regime [16], i.e. when dealing with low to moderate spatial resolutions. 3D-EPI in principle outperforms SMS-EPI acquisitions because of the higher intrinsic SNR according to the theory [6]

Objectives

Methods

Results

Discussion

Conclusion