Materials and methods
In this work, we report on resting-state fMRI data collected from 26 pilots on active duty and 24 education, handedness, and gender-matched healthy controls. The participants were recruited via online advertisements and given compensation. The exclusion criteria for all subjects included neurological illness, traumatic brain injury, substance-related disorders, and any metal implant in the body. All participants were fully informed about the procedure, and we obtained written informed consent. Detailed information of the participants can be found in our previous study [13].
This experiment was approved by the Ethics Committee of the University of Electronic Science and Technology of China (Chengdu, China) (No. 2018–042002).
Data acquisition
The data were collected using a 3-T MRI scanner (DISCOVERY MR 750, GE) at the University of Electronic Science and Technology of China.
A three-dimensional spoiled gradient echo pulse sequence was used to collect axial high-spatial-resolution structural images (TR = 5.972 ms, TE = 1.972 ms, flip angle = 9°, matrix = 256 × 256, slice thickness = 1 mm, FOV = 25.6 cm × 25.6 cm, number of slices = 154).
The resting-state functional images were collected using a gradient-echo echo-planar imaging sequence. During the 510-s scanning, the subjects were lying quietly with their eyes closed while remaining awake (TR = 2000 ms, TE = 30 ms, flip angle = 90°, matrix = 64 × 64, slice thickness = 4 mm with no gap, FOV = 24 cm × 24 cm, number of slices = 35/volume).
Data preprocessing
The SPM12 (Statistical Parametric Mapping 12, http://www.fil.ion.ucl.ac.uk/spm/) toolbox was used to conduct image pre-processing. The structural images were segmented into white matter, gray matter, and cerebrospinal fluid.
The first five scans of the resting-state fMRI data were discarded. The remaining images were slice-time- and head-motion-corrected, realigned, and coregistered to the structural images and transformed into a standard MNI space (3 mm × 3 mm × 3 mm). Subsequently, an 8-mm full-width at half-maximum Gaussian kernel was used to smooth the images.
The participants with excessive head motion (2 mm displacement in any direction or 2° spin in any direction) were excluded. The group head motion differences were compared. In addition, the head motion of each subject was analyzed as follows: M represents the total time point number; i is the certain time point; x1, y1, and z1 represent the translation parameters at a certain time point; and x2, y2, and z2 are the rotations.
Group independent component analysis
Spatial ICA was conducted using GIFT software (Version 1.4b, http://icatb.sourceforge.net/). The minimum description length criterion was used to determine the number of independent components of the datasets from the 50 participants. Ultimately, 35 independent components were determined. Principal component analysis was conducted on the concatenated functional data of all subjects. The Infomax algorithm was used to decompose the data by independent component estimation. This operation was repeated for 20 times in ICASSO to achieve reliable results. Subsequently, dual-regression was conducted to back-reconstruct individual subject components into a single-subject space. Finally, each map was scaled to a Z-score.
We inspected the aggregate spatial maps and average power spectra of the independent components [22] to identify the relevant networks. Three experts rated the components from zero (certain artifact) to one (definite brain network) based on the principles that the components should exhibit primary activations in gray matter, which reflect networks of interest [22, 23], and low spatial overlap with known vascular, ventricular, and susceptibility artifacts. In addition, the power spectra of the components should be dominated by low-frequency fluctuations. The ratio of integral spectral power below 0.1 Hz to power between 0.15. and 0.25 Hz was used. This ratio should be greater than 3.
Dynamic functional connectivity
Several methods can be used to map the dynamic functional connectivity, and the sliding window correlation is the most classical among them [24]. First, a fixed-length time window is used to divide the original fMRI data into a continuous series of fragments. Next, the Pearson correlation is computed between the three networks within each window. In the current study, we conducted a sliding window dynamic functional connectivity analysis using the Dynamic FNC Toolbox (v1.0a) in GIFT. In line with a previous study [25], the resting state data were divided using sliding windows of 20 repetition times (40 s) size with step by one repetition time (2 s). An additional L1 norm constraint on the inverse covariance matrix to promote sparsity in the graphical LASSO framework with 100 repetitions was imposed. Thus, each subject had 230 functional connectivity matrices. All functional connectivity matrices were then normalized using Fisher’s Z transformation.
Clustering analysis
A k-means clustering algorithm with square Euclidean distance, 500 iterations, and 150 replicates was used on the functional connectivity matrices for all subjects. The centroids of these clusters, which can be treated as “states,” present the average patterns during scanning. The optimal number of clusters was estimated using the elbow criterion. Finally, a k of four was obtained using this method in a search window of k from two to ten [26].
We then conducted two sample t-tests [27] to identify the differences between the groups in each state. The results were corrected for multiple comparisons using false discovery rate (FDR) (p < 0.05).
CC BY 4.0. "Increased functional dynamics in civil aviation pilots: Evidence from a neuroimaging study" by Chen X, Wang Q, Luo C et al., PloS one (2020). https://doi.org/10.1371/journal.pone.0234790. Licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Peer-reviewed and open access; full text from PubMed Central (PMC7302522). Headings and layout adapted for reading; figures and tables are in the original.