Interpolated functional manifold for functional near-infrared spectroscopy analysis at group level

Shender-María Ávila-Sansores; Gustavo Rodríguez-Gómez; Ilias Tachtsidis; Felipe Orihuela-Espina

doi:10.1117/1.NPh.7.4.045009

Interpolated functional manifold for functional near-infrared spectroscopy analysis at group level

Neurophotonics. 2020 Oct;7(4):045009. doi: 10.1117/1.NPh.7.4.045009. Epub 2020 Nov 27.

Authors

Shender-María Ávila-Sansores¹, Gustavo Rodríguez-Gómez¹, Ilias Tachtsidis², Felipe Orihuela-Espina¹

Affiliations

¹ Instituto Nacional de Astrofísica, Óptica y Electrónica, Santa María Tonatzintla, Puebla, Mexico.
² University College London, London, United Kingdom.

Abstract

Significance: Solutions for group-level analysis of connectivity from fNIRS observations exist, but groupwise explorative analysis with classical solutions is often cumbersome. Manifold-based solutions excel at data exploration, but there are infinite surfaces crossing the observations cloud of points. Aim: We aim to provide a systematic choice of surface for a manifold-based analysis of connectivity at group level with small surface interpolation error. Approach: This research introduces interpolated functional manifold (IFM). IFM builds a manifold from reconstructed changes in concentrations of oxygenated $Δ c {HbO}_{2}$ and reduced $Δ c HbR$ hemoglobin species by means of radial basis functions (RBF). We evaluate the root mean square error (RMSE) associated to four families of RBF. We validated our model against psychophysiological interactions (PPI) analysis using the Jaccard index (JI). We demonstrate the usability in an experimental dataset of surgical neuroergonomics. Results: Lowest interpolation RMSE was $1.26 e - 4 \pm 1.32 e - 8$ for $Δ c {HbO}_{2}$ [A.U.] and $4.30 e - 7 \pm 2.50 e - 13$ [A.U.] for $Δ c HbR$ . Agreement with classical group analysis was $JI = 0.89 \pm 0.01$ for $Δ c {HbO}_{2}$ . Agreement with PPI analysis was $JI = 0.83 \pm 0.07$ for $Δ c {HbO}_{2}$ and $JI = 0.77 \pm 0.06$ for $Δ c HbR$ . IFM successfully decoded group differences [ANOVA: $Δ {cHbO}_{2}$ : $F (2,117) = 3.07$ ; $p < 0.05$ ; $Δ c HbR$ : $F (2,117) = 3.35$ ; $p < 0.05$ ]. Conclusions: IFM provides a pragmatic solution to the problem of choosing the manifold associated to a cloud of points, facilitating the use of manifold-based solutions for the group analysis of fNIRS datasets.

Keywords: connectivity analysis; functional connectivity; functional near-infrared spectroscopy; manifold; topology.