Behavior Analysis
The completereaction time information of two subjects (one PD and one HC, sub-RC4114 and sub-RC4224) was unavailable. We first aggregated the mean reaction time and accuracy from the six combinations of cue type × flanker congruency conditions for the remaining subjects. Reaction times lower than 200 ms or higher than 2000 ms were excluded. Second, we calculated the three behavior attention network scores for each subject of each session. Third, we verified whether the alerting, orienting, and executive effects could be found at the behavior level and whether there were significant differences between PD and HC. Finally, we calculated the test-retest reliability estimates for the behavior attentional network scores.
Reproducibility across Sessions and Groups
We used the mixed-effects multilevel analysis (3dMEMA ) (Chen et al. , 2012) within an intersection mask to model the group mean activation for each session of the two groups. To make an intersection mask, all subjects’ functional masks were intersected with a threshold that at least 75% of the participants had valid values for each voxel. Consequently, we acquire four group-level statistical maps (2 groups × 2 sessions)for each contrast (alerting, orienting, executive) and subsequently derive the corresponding group-level statistical maps after applying post-thresholding criteria (employing a voxel-wise 1-sided threshold of 0.01 and a cluster extent of 40 voxels). Finally, the 3ddot command computed dice coefficients (Equation 1) to compare spatial overlap across populations and sessions. We also calculated the dice coefficients for varying statistical thresholds to examine the influence of threshold.
\(\mathbf{Dice=\ }\frac{\mathbf{2*|X\cap Y|}}{\left|\mathbf{X}\right|\mathbf{+|Y|}}\mathbf{,}\)(Equation 1)
In addition, to better illustrate the spatial overlap between sessions, we visualize the brain activations on the cerebral surface according to the brain parcellation atlas. The Schaefer-Yeo atlas of 200 parcellations (Schaefer et al. , 2018) was used to extract mean contrast estimates from each atlas region (3dROIstats ). Then, we used the ggseg package (Mowinckel & Vidal-Piñeiro, 2020) to visually compare the region-based activations across groups and sessions (with a liberal threshold of 0.05).
Because the present study focused on reproducibility and reliability, we did not report the group difference in brain activations.Moreover, the group differences have been examined by other researchers. Interested readers might refer to two published papers on the difference in functional activation between the patient and healthy group (Madhyastha et al. , 2015; Boord et al. , 2017).
Test-retest Reliability of fMRI measures
The voxel-wise test-retest reliability analysis was performed for each attention network contrast using 3dICC (Chen et al. , 2018). The ICC estimates for HC and PD were included in the supplementary material. For the region-level and network-level analyses , the mean contrast estimates for each atlas region were fed into the rptR package (Stoffel et al. , 2017). We reported the adjusted test-retest reliability by regressing out the systematic effect due to group, session, and their interaction. The ICC estimates were calculated for the patient and healthy group separately . The ICC map was visualized using ggseg (Mowinckel & Vidal-Piñeiro, 2020). In addition, we also plot the ICC distribution according to the seven networks defined by Yeo et al. (2011). Besides the region- and network-level analyses based on the Schafer-Yeo atlas, we also performed reliability analyses on basal ganglia (12 subregions) and thalamus (16 subregions) according to the Brainnetome atlas.