By Felipe Orihuela-Espina
The recent special issue on “Algorithms for Functional Near-Infrared Spectroscopy, Cerebral Oximetry and Near-Infrared Imaging” in the journal Algorithms was announced. It covers new developments regarding algorithms, methods and ideas related to signal processing and data analysis for fNIRS, cerebral oximetry and NIRI measurements.
The first paper presented the NIRS Brain AnalyzIR Toolbox from Ted Huppert’s group (see Figure 1). The manuscript introduces the reader to an all-purpose software package covering image reconstruction, processing, registration and analysis. It also provides an option to generate synthetic and semi-synthetic data. The underlying algorithms are founded on rigorous statistical models, which are clearly described in the paper. As one would expect of such a comprehensive toolbox, it permits first and second level analysis, voxel and ROI based analysis, graph theory-based analysis of connectivity. It also permits delineation of pipelines for workflow. Finally, the paper outlines a convenient way to compare the effectiveness of different pipelines. Among the almost overwhelming number of features included in the described tool two ‘jewels’ merit highlighting: multimodal imaging integration and hyperscanning analysis. This toolbox represents the most comprehensive analysis software for NIRS data thus far and will likely dominate different tools for analysis of data from application studies in the coming years.
Figure 1. Example of data plotting in the NIRS AnalizIR toolbox. Figure reproduced with permission from [Santosa et al (2018) Algorithms, 11:73].
The second paper, a collaboration between Ilias Tachtsidis’ group at UCL and our group in Mexico, examines the in-common information of the connectivity networks from [HbO2] and [HHb] – a topic that has not been adequately explored despite obvious implications for interpretation of fNIRS data. This exploratory work was done with the use of the novel differential symmetry index (DSI) (illustrated in Figure 2) to quantify the extent of symmetry between [Hb]-derived connectivity networks. The analysis is based on a naturalistic dataset collected outside of the lab where systemic changes induced by walking can potentially affect any expected symmetry among the [Hb] responses.
Figure 2. The differential symmetry index in action. Discrimination of connectivity network symmetry perceived from the Hb species. Figure reproduced with permission from [MonteroHernandez et al (2018) Algorithms, 11:70].
Finally, the third paper from Jeffrey Dunn’s group in Calgary provides rare yet valuable evidence of how our processing pipelines have dramatic effects on our signal quality (Figure 3) and may substantially affect interpretation of observations. Topics covered include assessment of cardiac pulsation and other systemic physiology (i.e., low frequency component filtering), as well as optode movement detection and correction. The de-facto standard general linear model (GLM) is used as a base to establish the implications of the statistical analysis. The work is presented in terms of qualitative comparisons of several popular approaches to addressing factors affecting the NIRS signal. These are tested on a novel dataset acquired using a traditional finger tapping task. The insights provided by this paper about sources of variability in data analysis and interpretation are invaluable.
Figure 3. Oscillations in an fNIRS signal. Processing pipelines have different impact on the quality of the surviving information. Figure reproduced with permission from [Hocke et al (2018) Algorithms, 11:67].
To learn more: