Figure 5. Schematic for time-series analysis workflow.

From the cohort of n = 33 RRMS patients, we performed the time-series analysis on the subset of n = 24 patients who showed contractions in ventricle volume greater than the ± 6% range of normal variation. Ventricle volumes were measured at 13 monthly time points. At the same time points, an additional 8 MRI parameters and 4 clinical parameters were measured. This allowed each of these measures to be considered as a time series. Using the cross-correlation function, the cross-correlation coefficients between 2 time series can be calculated; significant coefficients indicate that events of one series precede (negative time lag) or follow (positive time lag) the events of another series. In the current study, we limited the consideration of significant cross-correlation coefficients to ± 2 time lags (i.e., ± 2 months). From n = 24 patients, n = 12 variables, and n = 5 time lags (including the 0 time lag), a total of 1440 coefficients was calculated. From the coefficients with nominal P < 0.05, the FDR correction for multiple comparisons was applied, to yield the corrected significant cross-correlation coefficients, which were then displayed in the 3D plots.