Analysis of BOLD fMRI signal preprocessing pipeline on different datasets while reducing false positive rates

The technology of functional Magnetic Resonance Imaging (fMRI) based on Blood Oxygen Level Dependent (BOLD) signal has been widely used in clinical treatments and brain function researches. The BOLD signal has to be preprocessed before being analyzed using either functional connectivity measurements or statistical methods. Current researches show that data preprocessing steps may influence the results of analysis, yet there is no consensus on preprocessing method. In this paper, an evaluation method is proposed for analyzing the preprocessing pipeline of resting state BOLD fMRI (rs-BOLD fMRI) data under putative task experiment designs to cast some lights on the preprocessing stage, covering both first and second level analysis. The choices of preprocessing parameters and steps are altered to investigate preprocessing pipelines while observing statistical analysis results, trying to reduce false positives as reported by Eklund et al. in their 2016 PNAS paper. All of the experiment data are separated into 7 datasets, consisting of 220 healthy control samples and 136 patient data that are from 38 incomplete Spinal Cord Injury (SCI) patients and 16 Cerebral Stroke (CS) patients, including multiple scans of some patients at different time. These data were acquired from two different MRI scanners, which may cause difference in analysis results. The evaluation result shows that it has little effect to change parameters in each steps of the classical preprocessing pipeline, which consists head motion correction, normalization and smoothing. Removing time points and the following detrend step can reduce false positives. However, covariates regression and filtering has complicated effects on the data. Note that for single subject analysis, false positives declined consistently after filtering. The result of patient data and healthy controls data which are scanned under the same machine with the same acquisition protocol shows little difference. Yet data acquired with different scanner and protocol influences statistical analysis significantly. As a result, future research should pay more attention on the scanning machine and protocol used. This research is a preliminary investigation on rs-BOLD fMRI signal preprocessing. We hope the conclusions can be of value for other studies in the field of brain function research.