Self-navigated prospective motion correction of repeated 3D-EPI acquisitions for functional MRI applications

Bayih, Samuel Getaneh

Self-navigated prospective motion correction of repeated 3D-EPI acquisitions for functional MRI applications

Thesis / Dissertation

2024

Publisher

University of Cape Town

Department

Department of Human Biology

Faculty

Faculty of Health Sciences

Abstract

Functional magnetic resonance imaging (fMRI) typically employs one of the fastest pulse sequences – known as echo-planar imaging (EPI) – to repeatedly image the whole brain as stacks of two-dimensional slices. Two-dimensional EPI (2D-EPI) is preferred over threedimensional (3D)-EPI as it is more motion robust. However, 3D-EPI offers various advantages over 2D-EPI for functional MRI (fMRI), including the absence of spin-history artifacts, potential for faster acceleration, and better signal quality at higher spatial resolution. As such, 3D-EPI would be beneficial for fMRI if it could be made more motion robust. Various prospective and retrospective motion correction techniques have been developed to minimize subject motion or its effects during MR imaging. The predominant retrospective motion correction techniques, which aim to re-align imaging volumes after acquisition, are incorporated as pre-processing steps in functional MRI analysis packages. However, these typically do not perform intra-volume motion correction. There are also advanced methods that aim to generate artifact-free volumes by using inverse modelling and motion estimates. An alternative approach, termed prospective motion correction (PMC), aims instead to track and correct motion in real time during scanning. PMC techniques typically require either additional external hardware or additional gradient and radio-frequency (RF) pulses inserted into the scanning sequence to track subject motion, thereby increasing scan costs or time. This work demonstrates the application of a novel self-navigated PMC (𝑠𝑛𝑃𝑀𝐶) approach in repeated 3D-EPI acquisitions to track subject motion and readjust the imaging FOV in real time while a volume is still being acquired without the need for any external hardware or additional gradient and RF pulses. The technique utilizes the fact that a subset of the partitions acquired to construct an entire 3D volume can be used to re-construct a low-resolution volume image, termed a volumetric self-navigator (𝑣𝑆𝑁𝑎𝑣), that will contain the same general features as the entire volume and can be used to detect motion while remaining partitions are still being acquired. In our first implementation, we tracked subject motion once per volume by registering each volumetric self-navigator (𝑣𝑆𝑁𝑎𝑣), constructed from 24 of 52 partitions, to a reference (𝑣𝑆𝑁𝑎𝑣𝑅𝑒𝑓) acquired and constructed during the first volume acquisition. The estimated motion parameters are sent to the sequence and the sequence updates its imaging field of view (FOV) once per volume. The 𝑠𝑛𝑃𝑀𝐶 3D-EPI sequence was validated without and with intentional motion in phantoms and in vivo on a 3T Skyra (Siemens, Erlangen, Germany) MRI. For the in vivo scans, motion estimates from the 𝑣𝑆𝑁𝑎𝑣’s were compared to retrospective motion estimates obtained using FLIRT (FMRIB’s Linear Image Registration Tool). Both phantom and in vivo data demonstrated accurate and stable motion estimates in the absence of motion. For phantom acquisitions with intentional motion, estimated residual motion after motion correction were within acceptable thresholds (i.e., < 10% of the slice thickness and < 0.2⁰ rotation). For in vivo acquisitions, motion estimates using 𝑣𝑆𝑁𝑎𝑣 and FLIRT agreed to within 0.23 mm (< 10% of the slice thickness) and 0.14⁰ in all directions. The performance of our 𝑣𝑆𝑁𝑎𝑣 3D-EPI sequence for fMRI data acquisition was compared to the widely used 2D-EPI sequence with prospective acquisition correction (PACE). Four healthy volunteers were scanned with both the 3D- and 2D-EPI sequences while performing a block design finger tapping task. Except for flip angles, which were 16⁰ for 3D-EPI and 90⁰ for 2D-EPI, respectively, imaging parameters (i.e., spatial resolution 3.3𝗑3.3𝗑3.1 mm3) of the two sequences were almost identical. Intentional and unintentional head motions were induced during the experiments while real-time motion detection and correction – 𝑠𝑛𝑃𝑀𝐶 for 3D-EPI and standard PACE for 2D-EPI – were active. After applying identical pre-processing steps and statistical analysis pipelines on the 3D-EPI and 2D-EPI data, it was found that the 3D-EPI data had a greater number of voxels with higher inherent temporal SNR than the 2D-EPI data. While the temporal SNR of the BOLD signal is expected to be higher with 3D-EPI than 2DEPI at high spatial resolutions, this result was unexpected at this spatial resolution. However, the application of 3.3 mm, 6.6 mm, 8 mm, 9.9 mm and 12 mm FWHM Gaussian spatial filters enhanced the 2D data more than the 3D data, resulting in similar findings following statistical analyses, except that fewer spurious activations were evident in the 3D-EPI data. To fully exploit the superior quality of the 3D-EPI BOLD signal, it will be critical to develop an optimized pre-processing pipeline for 3D-EPI data. Finally, we aimed to increase the temporal resolution of 𝑠𝑛𝑃𝑀𝐶 by acquiring and registering two (double) volumetric self-navigators (𝑑𝑣𝑠𝑁𝑎𝑣1 and 𝑑𝑣𝑠𝑁𝑎𝑣2) constructed from two different subsets of the partitions during each volume acquisition. Each 𝑑𝑣𝑠𝑁𝑎𝑣 is registered to its respective reference acquired during the first volume acquisition (𝑑𝑣𝑠𝑁𝑎𝑣1𝑅𝑒𝑓 or 𝑑𝑣𝑠𝑁𝑎𝑣2𝑅𝑒𝑓). After the motion parameters estimated by either 𝑑𝑣𝑠𝑁𝑎𝑣1 or 𝑑𝑣𝑠𝑁𝑎𝑣2 are sent back to the 3D-EPI sequence, the sequence immediately adjusts its imaging FOV and acquires the subsequent partitions (including those used to construct the next self-navigator) using the updated FOV. The performance of our double volumetric self-navigated (𝑑𝑣𝑠𝑁𝑎𝑣) 3D-EPI sequence was validated both in a phantom and in vivo, and the in vivo motion estimates were compared to retrospective estimates from FLIRT. The phantom data demonstrated successful motion estimation and correction; induced motions were accurately detected and completely corrected before accumulation of the partitions for the next 𝑑𝑣𝑠𝑁𝑎𝑣. However, the in vivo data demonstrated above threshold (i.e., > 10% of slice thickness or > 0.2⁰) differences between motion estimates from the 𝑑𝑣𝑠𝑁𝑎𝑣 sequence and FLIRT. It appears that effects of continuously pulsating motions, which are likely due to physiological noise, limit the accuracy of motion estimation. The detection of continuously pulsating motions was unexpected. These data could potentially be included in motion models to better account for physiological processes. In conclusion, our results demonstrate that self-navigated 3D-EPI is feasible for functional MRI applications. Further work is needed to accelerate the temporal resolution of motion tracking, to increase both the temporal and spatial resolution of our self-navigated 3D-EPI sequence, and to optimize pre-processing pipelines for 3D-EPI data.

Keywords

Human Biology

Reference:

Collections

PhD / Doctoral

Full item page