Effect of blood oxygen-level-dependent signal data trimming on functional connectivity metrics

In the big data era, with a lack of comparable functional neuroimaging data, researchers try to combine heterogeneous data of different lengths, trimming those to the same number of timepoints (NTPs). However, the effects of trimming blood-oxygen-level dependent (BOLD) signal data on functional connectivity (FC) are still poorly understood.

Resting-state functional magnetic resonance imaging data from thirty healthy subjects were pre-processed for five different NTPs, from which FC matrices were computed. These BOLD signal correlation matrices were binarized for several thresholds, excluding weak correlations. Graph metrics were computed to study FC differences between different NTPs. The study included node degree analysis for each brain region and assessment of small-worldness coefficients ($\sigma$ and $\omega$), whereas in small-world networks, characteristic values are $\sigma$ > 1 and $\omega$ $\approx$ 0, indicating a balance between high clustering coefficients and short characteristic path lengths.

A tendency of decreasing global network degrees for higher NTPs was observed, translating the loss of stronger correlations with longer BOLD signals. Trimming such data affects brain regions differently, probably due to brain network dynamics. Regarding small-worldness, we observed that $\sigma$ was greater than 1 for all the different NTPs, showing an increasing trend for higher NTPs (median value: ${\sigma }_{\mathrm{BRAIN}}=$ 3.05). In addition, $\omega$ consistently remained greater than 0 for all NTPs, gradually approaching 0 as the NTPs increased (median value ${\omega }_{\mathrm{BRAIN}}=$ 0.20). As such, the results suggest a tendency for an increase of small-world properties for increasing NTPs. Nonetheless, the overall properties of brain networks almost remain constant. In conclusion, trimming BOLD signal data leads to small differences in FC.