Assessment of the effect of data length on the reliability of resting-state fNIRS connectivity measures and graph metrics. (September 2019)

• Record Type: Journal Article
• Authors: Aarabi, A.; Huppert, T.J.
• Is Part Of: Biomedical signal processing and control. Volume 54(2019)
• Publication Date: 2019-09
• Subjects: Data length -- Reliability analysis -- Stability analysis -- Pearson's correlation -- Percentage bend correlation -- Mutual information -- Partial correlation -- Graph metrics -- Functional connectivity analysis -- Resting state fNIRS; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101612
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: Data length had a significant effect on the reliability of fNIRS graph metrics. Connectivity metrics except partial correlation stabilized with increasing data lengths. Partial correlation showed stability only with short data lengths. Majority of network metrics showed high reliability with increasing data lengths. Abstract: The reliability assessment of connectivity measures and graph metrics is crucial for characterizing topological properties of resting-state brain networks that are intrinsic to the functioning of the brain and not biased by variability across subjects and data lengths. In this study, we investigated the effect of data length on the reliability and stability of four functional connectivity measures, Pearson's Correlation (PC), percentage-Bend Correlation (BC), Mutual Information (MI) and Partial Correlation (PtC), and twelve graph theoretical metrics derived from resting state functional near-infrared spectroscopy (rsfNIRS) data using data lengths ranging from 0.5 to 4.5 min. We analyzed rsfNIRS data collected in two separate sessions from 13 healthy adult subjects using an optical probe covering the whole brain. Our results showed that PC and BC stabilized with data lengths longer than 1–2.5 min depending on concentration signals. The stabilization for MI occurred with medium to long-range data lengths (more than 2.5 min). PtC showed stability only for data lengths shorter than 2.5 min. The reliability of the majority of the PC, BC and MI-derived network metrics improved significantly by data lengths of at least 1.5 to 2.5 min, depending on functional connectivity (FC) measures and concentration signals. For the PC and BC and MI-based networks, degree, global efficiency, characteristic path length, clustering coefficient and transitivity, graph radius and diameter exhibited high reliability. For these networks, the betweenness, modularity and vulnerability metrics showed moderate to high reliability with increasing data length for oxyhemoglobin (HbO), deoxyhemoglobin (HbR) and/or total-hemoglobin (HbT) signals. The participation coefficient, however, showed no specific pattern of changes or improvement with increasing data length. The hierarchy measure also showed variable reliability trends with increasing data length. The PtC-derived network metrics exhibited moderate to high reliability only with short-range data lengths shorter than 2 min for HbO, HbR and/or HbT. Our results show that data length can significantly affect the results of the FC analysis as well as the topological properties of weighted functional brain networks. This suggests that caution should be taken when comparing results from studies on functional network organization when FC analysis is performed with different data lengths.