Effects of phase-encoding on BOLD data with a positive control task

1361 

Abstract Submission 

Céline Provins¹, Alexandre Cionca¹, Elodie Savary¹, Patric Hagmann¹, Oscar Esteban¹

¹Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland

Céline Provins  
Lausanne University Hospital and University of Lausanne
Lausanne, Switzerland

Alexandre Cionca  
Lausanne University Hospital and University of Lausanne
Lausanne, Switzerland

Elodie Savary  
Lausanne University Hospital and University of Lausanne
Lausanne, Switzerland

Patric Hagmann  
Lausanne University Hospital and University of Lausanne
Lausanne, Switzerland

Oscar Esteban  
Lausanne University Hospital and University of Lausanne
Lausanne, Switzerland

Quality assessment and quality control (QA/QC) checkpoints layered throughout the dataflow are essential to ensure the reliability of neuroimaging analyses (Niso et al. 2022). In the case of functional MRI, best practices recommend collecting a 'positive control' task with which the different layers of QA/QC can be validated. These are short and simple tasks designed to elicit robust and precisely located brain activation patterns, permitting the diagnosis of potential issues in the workflow. Here, we examine how the phase-encoding direction (PE) choice in echo-planar imaging (EPI) blood-oxygen-level-dependent (BOLD) fMRI influences the resulting activation maps using a positive control task that includes visual and motor paradigms.

Data. 36 fMRI images were extracted from a dense sampling dataset, called the Human Connectome Phantom (HCPh), acquired as part of a registered report (Provins et al. 2023). A single male subject underwent repeated scans over four weeks. fMRI was acquired in a 3T Siemens Magnetom PrismaFit using a multi-echo EPI BOLD sequence varying PE across sessions in the four possible directions: anterior-posterior (AP), PA, left-right (LR), and RL. The echo times were TE=(12.60/33.04/53.48/73.92)ms. The other parameters, unchanged across sessions, were: 99 volumes with TR=1.6s, FA=64°, 2.2×2.2×2.2[mm3] resolution, distance factor 0%, 60 slices, 96×96 matrix, FOV=211mm, GRAPPA factor 2, SMS factor 4.
Task. Our quality control task (QCT), implemented with PsychoPy (Peirce et al. 2019), was adapted from the 'eye-movement' variant of the tasks proposed by (Harvey et al. 2018). It consists of four paradigms: a central fixation dot (blank), gaze movement, visual grating patterns, and a finger-tapping (left/right) block. The presentation order and realization of these paradigms (e.g. the coordinates in the gaze movement and the hand in fingertapping) were randomized.
Preprocessing. Data were preprocessed with fMRIPrep (Esteban et al. 2019) and further denoised by regressing out the 6 motion parameters, the WM and CSF mean signal, censoring frames with a framewise displacement above 0.4mm and smoothed to an estimated 4mm Gaussian kernel. No susceptibility distortion correction was applied to preserve the impact of different PE directions. Four QCT fMRI scans were excluded, three due to failed fMRIPrep runs and one for incorrectly defined events.
Task activation analysis. We used an event-based first-level model to estimate which voxels are significantly active during the tasks. Second-level models were then constructed using PE as confounds. The models were implemented with nilearn ("Nilearn" 2023).
Data & code availability. The HCPh dataset and the fMRIPrep derivatives will be publicly released with the Stage 2 culmination of the corresponding registered report. The task implementation is openly available at https://github.com/TheAxonLab/HCPh-fMRI-tasks. Our analysis is openly available as an educational notebook at https://www.axonlab.org/hcph-sops/analysis/qct-activation/.

Our QCT produces precisely-located brain activations. Confirming that the responses align with the expected activations is an integral component of the quality control protocol for our HCPh project (Figure 1).
Different PE caused activation disparities beyond susceptibility distortion-prone areas. We constrasted several pairs of PE and report the most relevant one in Figure 2. The latter highlights significant differences in fingertapping-induced activation within a segment of the primary motor cortex, a region that is usually not considered impacted by susceptibility distortion.

Quality control tasks, acquired in just a few minutes, can easily be integrated into any acquisition protocol, serving as a potent tool to assess the integrity of analysis workflows. We showcased its potential by examining how PE affects task activation maps. Next steps include inspecting maps in subject space and further interpreting the differences among PE directions.

Activation (eg. BOLD task-fMRI) ¹

Workflows ²

BOLD fMRI

Data analysis

Design and Analysis

FUNCTIONAL MRI

Modeling

Open Data

Open-Source Code

Open-Source Software

Pre-registration

Other - Quality Control; positive control task;