Using Fiberless, Wearable fNIRS to Monitor Brain Activity in Real-world Cognitive Tasks

Functional Near Infrared Spectroscopy (fNIRS) is a neuroimaging technique that uses near-infrared light to monitor brain activity. Based on neurovascular coupling, fNIRS is able to measure the haemoglobin concentration changes secondary to neuronal activity. Compared to other neuroimaging techniques, fNIRS represents a good compromise in terms of spatial and temporal resolution. Moreover, it is portable, lightweight, less sensitive to motion artifacts and does not impose significant physical restraints. It is therefore appropriate to monitor a wide range of cognitive tasks (e.g., auditory, gait analysis, social interaction) and different age populations (e.g., new-borns, adults, elderly people). The recent development of fiberless fNIRS devices has opened the way to new applications in neuroscience research. This represents a unique opportunity to study functional activity during real-world tests, which can be more sensitive and accurate in assessing cognitive function and dysfunction than lab-based tests. This study explored the use of fiberless fNIRS to monitor brain activity during a real-world prospective memory task. This protocol is performed outside the lab and brain haemoglobin concentration changes are continuously measured over the prefrontal cortex while the subject walks around in order to accomplish several different tasks.


Introduction
Abnormality of function within prefrontal cortex, and especially the most anterior subpart (rostral prefrontal cortex, or BA10) is common in a range of developmental, psychiatric and neurological conditions. It causes marked disturbances in problem-solving, memory, and attentional abilities in everyday life that are very disabling 1,2 . However, these kinds of problems are difficult to diagnose in the lab or clinic. This is because the mental processes that BA 10 supports are involved in dealing with novel, open-ended situations, where behaviour is self-initiated 3 . Such situations are hard to recreate successfully in the lab, since the formal, artificial and tightly constrained situation the participant typically faces in the lab can change their behaviour and the way that they approach the task. This can significantly reduce the validity of the measurement for either clinical or research purposes, with a strong risk of under-diagnosis 4 . One of the cognitive abilities supported by the frontal lobes where this is most apparent is prospective memory (i.e., the ability to remember to carry out a future action), where it has long been known that there can be significant disagreement between measurements taken in everyday life and the lab 5 . These methodological issues could be largely circumvented if researchers and clinicians investigating prefrontal cortex function, including prospective memory, could do so by taking their measurements in "real-world" situations.
While neuroimaging techniques represent a powerful tool to investigate brain function in a non-invasive and objective way, most of these techniques impose physical constraints on the subject, and are thus not appropriate for use in everyday life settings (e.g., functional magnetic resonance (fMRI), magnetoencephalography (MEG), positron emission tomography (PET)). Given the need to bring functional imaging instruments outside the lab and given recent technological improvements, portable and wearable electroencephalography (EEG) and functional near infrared spectroscopy (fNIRS) systems have been developed [6][7][8][9][10][11] . One of the major advantages of fNIRS over EEG is its higher spatial resolution. Moreover, it is less sensitive to motion artifacts, blinking and eye movements 12 . Wearable fNIRS is thus better suited for use in dailylife contexts, as it imposes fewer physical constraints than EEG and allows free movement in a more natural environment. fNIRS non-invasively irradiates the head with near-infrared light (650-900 nm). As the biological tissue is relatively transparent in that wavelength range, the light can reach the brain and get absorbed by haemoglobin. fNIRS thus measures the concentration changes of both oxyhemoglobin (HbO 2 ) and deoxyhemoglobin (HHb) giving information of oxygenation and haemodynamic changes associated with brain activity. More specifically, brain functional activation is defined as a concurrent increase in HbO 2 and a decrease in HHb . The new generation of miniaturized and fiberless fNIRS systems offers the possibility to explore brain activity in realistic situations on freely moving participants and without significant physical restraints. Realistic situations are particularly valuable when exploring human executive functions and fiberless fNIRS systems may provide a unique insight into human brain functions. The first fiberless systems were equipped only with a small number of channels (e.g., single channel 15 and 2 channels 16 ) limiting the investigation to small areas. More recently, multichannel wireless and wearable fNIRS devices have been developed 6,7,[17][18][19][20] giving the possibility to monitor larger portions of the head on freely moving participants.
In this study, a new multichannel wearable and fiberless fNIRS system was used to monitor and to map prefrontal cortex activity during a realworld prospective memory (PM) task. The fNIRS system is primarily composed of a flexible probe unit (headset) that covers both the dorsolateral and the rostral prefrontal cortex (Figure 1), which is connected to a processing unit (portable box) that is worn on the participant's waist ( Figure  1D). The headset is made up of 6 surface emitting laser diodes with two wavelength (705 nm and 830 nm) and 6 silicon photodiodes. The absence of optical fibers reduces the weight and the bulk of the probe, being more comfortable and robust against motion artifacts. The optodes are arranged in an alternating geometry ( Figure 1A) with an inter-optode separation of 3 cm, creating 16 source-detector combinations (e.g., 16 measurement channels) 6 . In order to shield the headset from the surrounding light, a shading cap is provided ( Figure 1D).
The aim of this study was to investigate prefrontal cortex function, during a prospective memory task in the real-world. During prospective memory tasks, participants are asked to remember to respond to an infrequent cue (e.g., a familiar face or a parking meter) while performing another demanding task known as an "ongoing task". In two different blocks of the task, social prospective memory cues (a person) are contrasted to non-social prospective memory cues (a parking meter). This contrast was chosen because it represents a major distinction made between different forms of cue in event-based prospective memory tasks and so the experimental paradigm can be kept close to a "real-life" situation 21 . While BA 10 is known to be sensitive to the processing of social versus non-social information in some situations (e.g., Gilbert et al., 2007 22 ), recent evidence suggests that haemodynamic changes in BA 10 related to prospective memory tasks are relatively insensitive to cue differences (see Burgess  The goal of this study is to evaluate the feasibility of using the fNIRS system to monitor prefrontal cortex hemodynamic and oxygenation changes induced by a real-world cognitive task. Here we report a single case study (one healthy adult participant, 24 years old) on the use of the fNIRS device during a prospective memory task, conducted outside in a typical London street location and mimicking the demands of everyday life. In particular, whether haemodynamic changes in response to social and non-social PM cues can be recorded is investigated.
3. Save the digitized coordinates and use the Spatial Analysis Tool (http://brain-lab.jp/wp/?page_id=52) of the open-source Platform for Optical Topography Analysis Tools (POTATo) software (see the Table of Materials for further information) to register fNIRS data onto a Montreal Neurological Institute (MNI) brain template. NOTE: The implemented algorithm for the probabilistic registration converts the digitized locations in the real world coordinate system into the MNI coordinate system and then projects and localizes them onto the MNI brain surface ( Figure 1E) 26,27 . 1. Open POTATo through the Matlab command P3. 2. Select "Spatial Analysis" from the menu on the main window of the POTATo Graphical User Interface (GUI) and click the "Spatial Analysis" button. 3. Load the digitized coordinates by clicking the "Empty 10-20" button on the Spatial Analysis Data Viewer window. 4. Click the "Empty MNI" button. 5. Select the 10/20 reference points on the MNI estimation window and start the spatial registration. 4. Check for the correct location of the fNIRS channels on the template brain surface ( Figure 1E): check if channel 8 and channel 9 overlap the inter-hemispheric fissure 28 . If correct, save the channel configuration file for further analyses; otherwise replace the digitizing band re-aligning channels 8 and 9 to the Fpz-Fz line and overlapping channel 9 to Fpz. Then repeat the digitizing procedure. 5. Place the fNIRS headset aligning channels 8 and 9 to the Fpz-Fz line and overlapping channel 9 to Fpz, in agreement with the digitizing headband, and remove the headband (Figure 1B-C). Make sure that the probe is well attached to the participant's head. 6. Place the shading cap with the head camera mounted on it over the fNIRS headset. 7. Explain the experimental rules to the participant. Include device-related precautions (e.g., 'Take as little time as possible without rushing or leaving the experimenter behind (NO running)") as well as task specific rules (e.g., 'Do not go outside the Queen Square area into neighbouring streets or areas"). 8. Have the participant successfully memorize all the rules and go outside to start the experiment.

fNIRS Signals Quality Assessment
1. Use the fNIRS system in wireless mode first to visually inspect signals quality on the fNIRS laptop: 1. Press the "Power" button on the portable box and turn on the fNIRS in the wireless mode. Open the fNIRS acquisition software on the fNIRS laptop and establish the connection with the portable box. 2. Press the "Probe Adjustment" button to optimize the detectors gain on the base of the detected light. 3. Check the probe adjustment results on the software "Probe Adjustment" window and check if each detector receives enough light from the sources by checking if all the channels are classified as "Normal". If channels are marked as "Stray" or "Under", re-place the shading cap and maximize the optodes coupling with the forehead. If channels are marked as "Over", set the power of the laser source to "low". NOTE: As the lateral channels cover the dorsolateral prefrontal cortex, in some cases it may be necessary to move the hair off the forehead to maximize the received light. 4. Press the "Ready" button and then "Start" to acquire data for a minute and check if heartbeat (haemoglobin oscillations of ~1 Hz) is visible on concentration signals, which ensures a good signal quality.
2. Turn off the portable box in the wireless mode pressing the "Power" button on it. Press the "Power" button in conjunction with the "Mode" button on the portable box to turn on the fNIRS in the stand-alone mode. NOTE: The stand-alone mode ensures that the participant can move freely around the experimental area and avoids the necessity to be close to the fNIRS laptop to maintain the wireless connection.

Data Acquisition
1. Turn on the head camera and the experimenters' cameras and start filming. Press the "Probe Adjustment" button on the fNIRS portable box to optimize the detectors gain and then press the "Play/Stop" button to start the fNIRS acquisition (sampling frequency=5 Hz). 2. Add a marker to the fNIRS data manually by using the "Mark" button on the fNIRS portable box in conjunction with an audio trigger (e.g., a beep). The audio trigger must be clearly recorded on all video cameras. Then start the experiment. NOTE: This allows a robust time synchronisation between the different video cameras and the fNIRS recording. , but in addition, have him respond to one of the experimenters who acts as a confederate who moves around into pre-specified positions within the experimental area. Have the participant go over to them and give them a "fist bump" greeting. 5. Use an additional Ongoing condition (Contaminated Ongoing) after the PM conditions (e.g., participants has to count the number of unobstructed stairways within the testing area). 6. Repeat the two Rest conditions described above in opposite order (Rest 2 and then Rest 1).

Experimental Protocol
NOTE: This allows evaluation of walk-related systemic changes at the end of the experiment.

Data Analysis
1. Open the fNIRS software and export data from the portable box flash card into the fNIRS laptop. NOTE: The fNIRS system processing unit uses the modified Beer-Lambert law and calculates the relative changes in HbO 2 and HHb from an arbitrary zero baseline at the beginning of the measurement period. Concentration values are hence expressed in molar concentrations (mmol/l) multiplied by the path length (mm) 6 as they are not corrected for the optical path length.
2. Save concentrations data and import them into Matlab through an in-house pre-processing software.
3. Pre-process the signals following these steps ( Figure 3B 3. Motion Artifact Correction: 1. For each channel, identify and remove motion artifacts through a wavelet-based method 31 . Improve signals quality by applying the Correlation Based Signal Improvement (CBSI) method 32 .

Complex wavelet transform:
1. Use a Morlet mother wavelet, scaled and translated over time, to compute the wavelet transform of each channel through the wavelet toolbox (Matlab function: wt) provided by Grinsted et al. 33 (http://noc.ac.uk/using-science/crosswavelet-waveletcoherence). NOTE: From the wavelet spectrum, it is possible to evaluate the spectral content of signals in a time-frequency space.  Figure 3 presents an example of HbO 2 and HHb un-processed signals (channel 8) recorded during the life-based PM experiment in this case study ( Figure 3A) and the corresponding signals ( Figure 3C) after being pre-processed ( Figure 3B). Figure 4 shows the wavelet power spectrum of channel 8 HbO 2 and HHb signals in which the rectangle indicates the frequency range preserved with the band-pass filter.

Representative Results
Considering the fact that the participant was walking outside throughout the experiment and moved his head to perform the task, the fNIRS system was robust against motion artifacts and sunlight. In fact, HbO 2 increments and HHb decrements can be found in correspondence to nonsocial ( Figure 3D) and social ( Figure 3E) prospective memory events. These trends typically denote functional brain activity 13,35 . In fact, when a brain area is activated, the neurons' metabolic demand for oxygen increases with consequent increases in regional cerebral blood flow. As most of the oxygen is delivered to cells through haemoglobin, increments in HbO 2 and decrease in HHb concentrations are observed during functional brain activity 9 . Regions within the prefrontal cortex that exhibit these trends can be assessed by the spatial distribution of HbO 2 and HHb concentration values mapped over the forehead (Figure 5, Video 1, Video 2). An example of how brain responses to a social PM event are distributed across all the channels is shown in Figure 5. Figure 5A and Figure 5B report respectively the spatial distribution over the forehead of

Discussion
The aim of this study was to evaluate the potential use of wearable and fiberless fNIRS to monitor brain haemodynamic and oxygenation changes related to brain neuronal activity during real-world situations. A wearable and fiberless multichannel fNIRS system was used to measure brain activity over the prefrontal cortex during a prospective memory task performed outside the lab. The case study reported here explored whether brain changes in HbO 2 and HHb on a freely moving participant in response to social and non-social PM cues in an experiment outside the lab can be monitored continuously and robustly.
The use of fNIRS on freely moving participants in life-based experiments represents a challenging situation. In fact, head movements can cause probe displacements with consequent motion artifacts that corrupt the optical identification of brain activity 36 . Moreover, optical sensors are sensitive to stray light (e.g., sunlight when experiments are performed outside), creating additional noise in fNIRS signals. The reported case study provides a preliminary demonstration of the feasibility of the fNIRS system in such real life applications. The absence of optical fibers in such devices prevents optical coupling between the scalp and the optodes resulting in more robust measurements against motion artifacts. In addition, the shading cap ensures a good shielding from the stray light which avoids detectors saturation and low Signal-to-Noise Ratio (SNR). Moreover, increases in HbO 2 and decrease in HHb concentrations were found in correspondence of social and non-social PM hits (Figure 3D-E) 11,37 further supporting its feasibility. In order to assess if the haemodynamic trends observed in Figure 3D-E are statistically significant and to locate activated regions within the prefrontal cortex ( Figure 5, Video 1, Video 2, Figure 6, Figure 7), group-level analyses are required. In order to make inference and to identify functionally specialized prefrontal cortex regions 38,39 , future works will present group data and statistical analyses based on Statistical Parametric Mapping (SPM) using a General Linear Model (GLM) approach.
Even though results have to be considered preliminary, it has been demonstrated that fiberless fNIRS can be effectively brought outside the traditional lab settings and used for real time monitoring of brain activity. This opens up new directions for neurological and neuroscience research. There are at least two obvious areas for application in this respect. The first relates to ecological validity. Cognitive neuroscience researchers investigate patterns of brain activity while people are performing cognitive tasks (using e.g., blood oxygen level dependent signal change as a proxy in functional MRI) in order to try to discover how the brain supports our mental abilities. In some cases, it is possible to create experimental situations in the scanner that match very closely the situation in everyday life where the process of interest is used. Consider, for example, reading. Reading words on a display while in a MRI scanner likely makes such similar demands to reading words in a book when at home that it is almost taken for granted that the results gleaned in the scanner can help explain how the brain implements reading in everyday life. However, for many forms of human behaviour and cognition, this assumption is more precarious. For instance, the cognitive processes that a participant uses when a social situation is presented in an MRI scanner (where the participant is immobile, on their own, and in a very unfamiliar and tightly controlled environment) may well be different in important regards to those engaged when the participant is socialising in real life 40 . This is particularly important in social neuroscience where the investigation of the neuronal correlates of inter-personal dynamics (termed hyperscanning, for review see Babiloni and Astolfi, 2014 41 ) requires a more naturalistic environment. NIRS-based hyperscanning 42,43 may thus represent a new tool to simultaneously monitor brain activity from two or more people in realistic situations. Indeed, there are some mental abilities that cannot be studied well in the highly artificial and physically constrained environment of a MRI, PET or MEG scanner. Those involving ambulation or large amounts of bodily movement as well as those involving social interactions are obvious candidates. For this reason, being able to study the brain activity of participants in naturalistic situations is highly desirable for researchers.
However, although wearable fNIRS systems show potential for real-world observations, there are other limitations that have to be addressed when using fNIRS during natural walking. Since the infrared light travels through the scalp, it is sensitive to processes that happen both at the cerebral and extra-cerebral compartments of the head. Previous studies demonstrated that a certain amount of the signals measured through fNIRS arises from systemic changes 34,39,44 that are not directly related to brain activity (see Scholkmann et al. 9 for a review). As intra and extracerebral hemodynamic are affected by systemic changes both task-evoked and spontaneous (e.g., heart rate, blood pressure, respiration, skin blood flow), physiological changes related to the walking activity should be considered. They originate from the autonomic nervous system (ANS) activity, which regulates heart rate, respiration, blood pressure and vessels diameter through its efferent fibers. More precisely, the sympathetic division of the ANS is hyper-activated during exercise leading to heart rate, blood pressure and respiration increments 45 . For example, previous studies have demonstrated that respiration induces changes in partial pressure of carbon dioxide in the arterial blood (PaCO 2 ) which in turn influence cerebral blood flow and cerebral blood volume 46,47 . In addition, Figure 3A shows an example of periodic HHb increases and HbO 2 decreases that occur within walking periods that can be confused with brain deactivation. In order to make consistent comparisons between conditions (e.g., assess if significant changes in concentration occur respect to a baseline period), all the experimental phases should be measured under the same physical activity state. For this reason, a walked rest phase (Rest 2) was included in our life-based protocol. A proper interpretation of fNIRS data requires also a good SNR. This is usually achieved with conventional block and event-related designs where stimulations are repeated several times. Trial repetitions and structured designs are not always possible in life-based experiments. For this reason, additional sensors and appropriate analysis techniques to account for systemic changes 48 and motion artifacts are necessary to improve the SNR and to correctly interpret brain signals. We plan to investigate the impact of such walk-related systemic changes through the use of portable devices to monitor breathing rate, heart rate and walking pace. Moreover, the problem of events recovery needs to be addressed, too. In cognitive neuroscience experiments, brain activity is investigated in relation to stimuli or environments encountered by participants', and their behaviour in response to, or anticipation of them. Experimenters therefore need to (a) know what is currently available to the participant in their environment, and (b) have a moment-by-moment record of the participant's behaviour. In a typical lab situation these factors can be readily controlled since the experimenter can constrain what participants encounter, and the form and number of behaviours that the participant can evince. However, this is not the case in "real-world" environments outside the lab, where many events and experiences that the research participant will have are beyond the strict control of the experimenter 49 . Accordingly, in "real-world" type tasks of the kind studied here, video records are used for analysis (e.g., Shallice and Burgess, 1991 3 ). This allows to recover both sustained (e.g., block-level) and transient (e.g., event-related) processes that support different aspects of performance (for review see Gonen-Yaacovi and Burgess, 2012 21 ). The events to be recovered from the video recordings will depend on the theoretical question being addressed in the experiment. In the reported case study, event onsets were recovered from the videos filmed by the 3 cameras. This procedure of determining the onset and termination of particular cues and behavioural responses is laborious and requires skill when carried out on life-based data. A central issue is that with "real life" type experiments there is usually not the same degree of a priori knowledge of events as with the lab-based ones, and participants usually have more scope in the way they can respond. Moreover, as participants are free to move in a natural and uncontrolled environment, they are faced with a variety of rapidly-changing stimuli and it is difficult to recover the haemodynamic response to the real event of interest. For example, in the case study, the haemodynamic trends observed for HbO 2 and HHb (Figure 3D-E) are not phase-locked to the video-recovered onset like the typical eventrelated haemodynamic response 38 . HbO 2 and HHb start respectively to rise and decrease 20 sec before the stimulus onset and reach a peak after it. Further analyses are thus needed to establish whether PM cues events are happening actually when the participant sees the target, when he approaches towards it or when he reaches it. Given the potential of fiberless fNIRS technologies for real life clinical applications, future work will address the video-coding problem by developing new algorithms to identify event onsets in a more objective way, as well as exploring the possibility of doing it directly from fNIRS data.

Disclosures
The authors declare that they have no competing financial interests.