**3.7 Confounding factors**

306 Neuroimaging – Cognitive and Clinical Neuroscience

comes from the devices as well as from physiological parameters not a priori linked to the stimulation (eg. Exercise) and are thus undesirable (Nolte et al., 1998). This kind of noise is considered high frequency with regards to the frequencies of interest (Cui et al., 2010a). Low-pass filters are used to remove heart rate, blood pressure variations, breath, swallowing etc. Usually, the cut-off frequency ranges between 0.1 and 1Hz. Detrending is performed using a high-pass filter when NIRS signals slowly drift throughout the experimental session. High-pass filters usually range between 0.01 and 0.05Hz. However experimenters must care as the frequencies of interest could be part of this range. Finally, experimenters have several tools to choose from to remove movement artefacts. If possible set a marker during the experimental session when the subject moved his head is a good start. Retrospectively, the eye of the physiologist is the first tool which can be used. However, its somehow objective behaviour and its inability to treat large amounts of data make its main limits. Abrupt changes in the signals can be detected and corrected by algorithms (Lloyd-Fox et al., 2010; Wilcox et al., 2008). However, the thresholds must be defined carefully in order to preserve the changes that supposedly belong to the awaited

Since NIRS is a relatively new technique for brain investigations, there is no standardised method to analyse data. Up to date, the only invariant is that different experimental designs

In block-designed studies, experimenters are used to analysing time series by averaging multiple trials of the same condition. Mean variations and mean time courses are then obtained for each condition. The critical points of such techniques are the determination of the relevant windows of the time series and the baseline which it is compared to. Once determined, student t-test and analyses of variance are the most often used statistical

More complex, three main freeware packages are downloadable and provide analysis methods derived from the BOLD signal of fMRI: HomER (Huppert et al., 2009), fOSA (Koh et al., 2007) and NIRS-SPM (Ye et al., 2009). The general linear model (GLM) and the statistical parametric mapping (SPM) offer the possibility to create three dimensional pictures of the brain, where activated/inhibited cortex areas are colour encoded (Friston et al., 1999; Plichta et al., 2007; Schroeter et al., 2004; Zarahn et al., 1997). In most studies, the NIRS records are performed off the MRI scan. Then, the input of the three dimensional coordinates of the optodes/channels is crucial for the reconstitution of the pictures. In the case of a co-record of NIRS and fMRI techniques, the coordinates of the NIRS optodes can be precisely assigned; else, skull measurements and probe placement are made either by reference to the 10-20 EEG system or by kinematic acquisition using such devices as

Doubtlessly, the toughest part of the NIRS based studies, is to draw physiological and cognitive conclusions from the data. Multi-channel setups cover wide cortical zones and result in several time series and three dimensional coloured images in which probability to give statistically significant results is high. The question experimenters inevitably face is "What do those results mean?". A typical NIRS channel includes a great number of capillary

hemodynamic response (Gervain et al., 2011).

require different analysis techniques.

**3.5 Data analysis** 

methods.

optotrack or fastrack.

**3.6 Dos and don'ts** 

At this stage of the article, the most impeding factors have been brought to discussion. However, some factors, not directly linked to the NIRS concepts nor to brain characteristics must be debated. Before entering the tissue of interest, light travels through the skin and the fat layers (as well as the hair and skull layers in case of brain investigations, Fig. 3). The skin colour (and hair colour) has been shown to influence light absorption (Pringle et al., 1999). Intuitively, human eyes perceive various skin colours because skin absorbs and reflects light depending on its properties. The same (or the opposite) happens in the near infrared portion of the spectrum. Light skins are believed to absorb light more than dark skins, while Asian originated skins are the less absorbent. NIRS gain or laser power must then be modulated to fit with the skin properties of a given subject; which can be performed automatically by the NIRS hardware before starting the data acquisition.

Skin blood flow is one of the main confounding factors as the haemoglobin molecules present in the capillary beds located in the skin are the first (and last) exposed to NIRS light (Tew et al., 2010). In exercising subjects, blood flow is increasing in proportion to the intensity of exercise, for well-known thermoregulation reasons. However, skin is not believed to consume more oxygen at high intensity as compared with low intensity exercises. This means that skin blood flow overcomes by far the local metabolic demands; which necessarily biases the NIRS measurements.

The fat and bone layers are probably easier to take into account as they can be integrated in the automatic gain setup which occurs in most modern NIRS devices, before data acquisition.

Finally, gender has been shown to influence NIRS responses to various stimuli, notably motor, cognitive tasks and emotions (Marumo et al., 2009; Yang et al., 2009).
