Optimizing Preprocessing and Analysis Pipelines for Single-Subject fMRI: 2. Interactions with ICA, PCA, Task Contrast and Inter-Subject Heterogeneity

Abstract

A variety of preprocessing techniques are available to correct subject-dependant artifacts in fMRI, caused by head motion and physiological noise. Although it has been established that the chosen preprocessing steps (or “pipeline”) may significantly affect fMRI results, it is not well understood how preprocessing choices interact with other parts of the fMRI experimental design. In this study, we examine how two experimental factors interact with preprocessing: between-subject heterogeneity, and strength of task contrast. Two levels of cognitive contrast were examined in an fMRI adaptation of the Trail-Making Test, with data from young, healthy adults. The importance of standard preprocessing with motion correction, physiological noise correction, motion parameter regression and temporal detrending were examined for the two task contrasts. We also tested subspace estimation using Principal Component Analysis (PCA), and Independent Component Analysis (ICA). Results were obtained for Penalized Discriminant Analysis, and model performance quantified with reproducibility (R) and prediction metrics (P). Simulation methods were also used to test for potential biases from individual-subject optimization. Our results demonstrate that (1) individual pipeline optimization is not significantly more biased than fixed preprocessing. In addition, (2) when applying a fixed pipeline across all subjects, the task contrast significantly affects pipeline performance; in particular, the effects of PCA and ICA models vary with contrast, and are not by themselves optimal preprocessing steps. Also, (3) selecting the optimal pipeline for each subject improves within-subject (P,R) and between-subject overlap, with the weaker cognitive contrast being more sensitive to pipeline optimization. These results demonstrate that sensitivity of fMRI results is influenced not only by preprocessing choices, but also by interactions with other experimental design factors. This paper outlines a quantitative procedure to denoise data that would otherwise be discarded due to artifact; this is particularly relevant for weak signal contrasts in single-subject, small-sample and clinical datasets.

The Stability of Behavioral PLS Results in Ill-Posed Neuroimaging Problems

“…The stability of the peak activations also allows us to identify reliable brain regions from a single split, which is not available to voxel-wise bootstrap estimation. The crossvalidation framework is therefore particularly useful when only limited f MRI data is available, and has been previously used to optimize preprocessing in brief task runs of less than 3 minutes in length (e.g., [18,19]). …”

“…Control blocks involved touching a fixation cross presented at random intervals of 1-3 s. The resulting 4D f MRI time series were preprocessed using standard tools from the AFNI package, including rigid-body correction of head motion (3dvolreg), physiological noise correction with RETROICOR (3dretroicor), temporal detrending using Legendre polynomials and regressing out estimated rigid-body motion parameters (3dDetrend, see [8] for an overview of preprocessing choices in f MRI). For the majority of results (see Sections 2.2 and 2.3), we preprocessed the data using a framework that optimizes the specific processing steps independently for each subject, as described in [18,19], within the split-half NPAIRS resampling framework [6]. In Section 2.4, we provide more details of pipeline optimization, and demonstrate the importance of optimizing preprocessing steps on an individual subject basis in the PLS framework.…”

“…However, taking our cue from the literature on split-half discriminant analysis in f MRI (see, e.g., [7,10,18,19,22]), we can regularize and de-noise the data space in which the analysis is performed, by first applying PCA to X, and then running a PLS analysis on a reduced PCA subspace.…”

“…For results in Sections 2.2 and 2.3, we preprocessed the f MRI data to correct for noise and artifact, as outlined in [18,19]. For this procedure, we included/excluded every combination of the preprocessing steps: (1) motion correction, (2) physiological correction, (3) regressing head-motion covariates and (4) temporal detrending with Legendre polynomial of orders 0 to 5, evaluating 2 3 × 6 = 48 different combinations of preprocessing steps ("pipelines").…”

A Novel Brain Networks Enhancement Model (BNEM) for BOLD fMRI Data Analysis With Highly Spatial Reproducibility

“…NPAIRS was widely used as the performance metrics in the optimal determination of preprocessing pipelines and dimensionality number, task contrast amplitude impacts investigation and physiological noise suppression, etc. [29][30][32][33][34][35][36][37][38]. The aforementioned researches revealed that enforcing reproducibility of the reproduced SPMs related to brain functional network was quite important in the neuroscience and clinical domain.…”