Impact of global normalization in FMRI acupuncture studies

Abstract

Global normalization is often used as a preprocessing step for dispelling the "nuisance effects." However, it has been shown in cognitive and emotion tasks that this preprocessing step might greatly distort statistical results when the orthogonality assumption of global normalization is violated. The present study examines this issue in fMRI acupuncture studies. Thirty healthy subjects were recruited to evaluate the impacts of the global normalization on the BOLD responses evoked by acupuncture stimulation during De-qi sensation and tactile stimulation during nonpainful sensations. To this end, we compared results by conducting global normalization (PSGS) and not conducting global normalization (NO PSGS) based on a proportional scaling model. The orthogonality assumption of global normalization was violated, and significant changes between BOLD responses for NO PSGS and PSGS were shown in most subjects. Extensive deactivations of acupuncture in fMRI were the non-specifically pernicious consequences of global normalization. The central responses of acupuncture during De-qi are non-specifically activation-dominant at the somatosensory-related brain network, whose statistical power is specifically enhanced by PSGS. In conclusion, PSGS should be unjustified for acupuncture studies in fMRI. The differences including the global normalization or not may partly contribute to conflicting results and interpretations in previous fMRI acupuncture studies.

Figure 1

Experimental paradigm: scans of acupuncture stimulation (a) and tactile stimulation (b) lasted 8 min and consisted of three one-minute acupuncture/tactile stimulations.

Figure 2

Correlations between the global signal (GS), reference vector (RV), and voxel-level time series. (a) and (b) indicate the correlation between the GS-RV R-value and AVER-WB-RV R-value (NO PSGS) for the acupuncture and tactile runs, respectively. Dots in each panel represent the subjects. The larger dots in orange and blue indicate the significant values of positive and negative correlations, respectively (P < 0.05, corrected). (c) and (d) manifest the linear relationship between the GS-RV R-value and changes in the AVER-WB-RV R-value (PSGS-NO PSGS) for the acupuncture and tactile runs, respectively. Dots also represent the subjects. (e) shows the mean and standard deviation (SD) of the AVER-WB-RV R-value across subjects for NO PSGS with a dark grey bar and for PSGS with a light grey bar in the acupuncture and tactile runs, respectively. (f) shows the mean and SD of the number of activations and deactivations across subjects for NO PSGS and PSGS. The numbers of activations for NO PSGS and PSGS are shown in black and dark grey bars, respectively. The numbers of deactivations for NO PSGS and PSGS are shown in white and light grey bars, respectively. Each number is calculated at P < 0.00001, uncorrected.

Figure 3

Histogram and activation/deactivation of the group results for NO PSGS and PSGS. The left part of (a) and (b) indicates the histogram of the group results based on the random effects model (REM) in acupuncture and tactile runs respectively. The dark grey line shows the histogram for NO PSGS and the light grey line shows the one for PSGS. Three thresholds were marked on the histogram, P < 0.00001, uncorrected, P < 0.001, uncorrected, and P < 0.01, uncorrected. The number of activations and deactivations of the group results at the three thresholds for NO PSGS and PSGS is shown in the right part of (a) and (b).

Figure 4

Maps and histogram for the paired t-test between PSGS and NO PSGS. (a) and (b) indicate the paired t-test map for PSGS < NO PSGS in the acupuncture and tactile runs, respectively. Two thresholds were selected and overlapped on the map: P < 0.001, uncorrected (colored in blue), and P < 0.01, uncorrected (colored in green) with 5 contiguous voxels, respectively. (c) shows the histogram of the paired t-test between PSGS and NO PSGS. The red line represents the acupuncture run, and the blue line indicates the tactile run. P < 0.001, uncorrected, and P < 0.01, uncorrected, are marked on the histogram.

Figure 5

Changes in activations between NO PSGS and PSGS. The map is based on REM group results at P < 0.001, uncorrected with 5 contiguous voxels. (a) and (b) show the changes in activations in the acupuncture and tactile runs, respectively. For each panel, the “stronger” regions are colored in red, delegating the extent of the activations that are significant for both NO PSGS and PSGS, and show a larger absolute value of the t-value for NO PSGS. The “weaker” regions are colored in yellow, delegating the extent of the activations that are significant for both NO PSGS and PSGS, and show a smaller absolute value of the t-value for NO PSGS. The “disappeared” regions are colored in blue, delegating the extent of the activations that are significant for NO PSGS but nonsignificant for PSGS. The “arisen” regions are colored in green, delegating the extent of the activations that are nonsignificant for NO PSGS, but are significant for PSGS.

Figure 6

Changes in deactivations between NO PSGS and PSGS. The map is based on REM group results at P < 0.001, uncorrected with 5 contiguous voxels. (a) and (b) show the changes in activations in the acupuncture and tactile runs, respectively. For each panel, the meaning for each color is identical to that in Figure 5.

Figure 7

Map of the REM group results for the acupuncture run and two-sample t-test results between the acupuncture and tactile runs for NO PSGS. (a) indicates the REM group results evoked by acupuncture stimulation at P < 0.00001, uncorrected with 5 contiguous voxels. (b) shows the between-group results of “acupuncture-tactile” at P < 0.001, uncorrected with 5 contiguous voxels.