Signal Processing in Functional Near-Infrared Spectroscopy (fNIRS): Methodological Differences Lead to Different Statistical Results

Even though research in the field of functional near-infrared spectroscopy (fNIRS) is being conducted for more than 20 years, a consensus on a signal processing methods is still lacking. A significant knowledge gap exists between established researchers and those newly entering the field. One major issue regularly observed in publications from new researchers in this field is the neglect of a possible signal contamination by hemodynamic changes unrelated to neurovascular coupling (i.e. scalp blood flow and systemic blood flow). This might be due to the fact that these researchers use signal processing methods with tools provided by the manufacturers of their devices without an advanced understanding of the performed steps. The aim of the present study was to investigate how different signal processing approaches (including and excluding approaches that partially correct for the possible signal contamination) affect the results of a typical functional neuroimaging study performed with fNIRS. In particular, we evaluated one standard signal processing method provided by a commercial company and compared it to three customized approaches and investigated the influence of the chosen method on the statistical outcome of a clinical data set (task-evoked motor cortex activity). No short-channels were used in the present study and therefore two types of multi-channel corrections based on multiple long-channels were applied. The choice of the signal processing method had a considerable influence on the outcome of the study. While methods that ignored the contamination of the fNIRS signals by task-evoked physiological noise yielded several significant hemodynamic responses over the whole head, the statistical significance of these findings disappeared when partially accounting for the contamination by using a multi-channel regression. We conclude that, when lacking the possibility of applying multi-distance measurements, the adoption of signal processing methods that correct for the physiological confounding effect, might yield more realistic results. Furthermore, we do not recommend to use standard signal processing methods as provided by the manufacturers without having an advanced understanding of every performed step.