Evaluating the performance of pre-processing methods for laminar applications

4th May 2022

Patricia Pais-Roldán, Seong Dae Yun and N. Jon Shah

Investigating functional dynamics in the brain in terms of cortical depth rather than cortical region is increasing in interest due to the development of new fMRI sequences capable of delivering diverse contrasts with high spatial resolution. Of these, gradient-echo (GE) sequences used to measure blood-oxygen-level-dependent (BOLD) signals are most advantageous due to their lower specific absorption rate, higher signal-to-noise ratio and faster acquisitions. However, despite providing robust functional images with unprecedented coverage and resolution, these sequences are generally more biased to field inhomogeneities, and pre-processing is required to substantially diminish signal contamination if the resulting images are to be analysed in a laminar context.

In this work, the performance of high-resolution GE BOLD fMRI pre-processed with twelve different pipelines was evaluated, principally in terms of signal localisation, both in resting-state and task conditions. Rather than using a novel method to pre-process, a series of evaluations were made in order to better understand how the existing approaches benefit or hinder GE-data for high-resolution applications.

The analysis demonstrates that GE data acquired with sub-millimetre spatial resolution is particularly sensitive to pre-processing and that routine smoothing-free fMRI cleaning methods with combined phase regression (to reduce vein bias) significantly improve the definition of depth-dependent activation patterns.

These findings will be used to further advance research into brain connectivity from a laminar perspective.

Original publication: Pre-processing of sub-millimeter GE-BOLD fMRI data for laminar applications