Prefrontal cortical activation measured by fNIRS during walking: effects of age, disease and secondary task

Background Cognitive processes are required during walking to appropriately respond to environmental and task demands. There are now many studies that have used functional Near-Infrared Spectroscopy (fNIRS) to record brain activation to investigate neural bases of cognitive contributions in gait. The aim of this systematic review was to summarize the published research regarding Prefrontal cortical (PFC) activation patterns during simple and complex walking tasks in young adults, older adults and clinical groups with balance disorders using fNIRS. Our secondary aim was to evaluate each included study based on methodological reporting criteria important for good data quality. Methods We conducted searches in June 2018 using four databases: Embase, PubMed, Scopus and PsycINFO. The strategy search used was: (((((near infrared spectroscopy) OR functional near infrared spectroscopy) OR nirs) OR fnirs) AND (((gait) OR walking) OR locomotion) AND (((((young) OR adult) OR older) OR elderly) NOT children)) AND (((Brain) OR cortex) OR cortical) for our search. The papers included met the specific review criteria: (i) used fNIRS to measure PFC activation patterns; (ii) included walking tasks (simple and complex) and; (iii) assessed young people, older people and/or clinical groups with balance disorders. Results Thirty five (describing 75 brain activation comparisons) of the 308 studies retrieved through our search met the inclusion criteria. Based on 6 methodological reporting considerations, 20 were of high quality, 10 were of medium quality and 5 were of low quality. Eleven/20 comparisons in young people, 23/37 comparisons in older people and 15/18 comparisons in clinical groups reported increased PFC activation with increased walking task complexity. The majority of comparisons that used verbal fluency, counting backwards or secondary motor tasks reported increases in PFC activation (83%, 64% and 58% of these studies, respectively). In contrast, no studies found secondary visual tasks increased PFC activation. Conclusion Increased PFC activation was most common in studies that involved walks comprising secondary verbal fluency and arithmetic tasks. Clinical groups generally showed increased PFC activation irrespective of type of secondary task performed during walking which suggests these groups require more attentional resources for safe walking. Systematic review registration number: PROSPERO 2017 - CRD42017059501.


INTRODUCTION
Walking relies heavily upon coordinated movement controlled by subcortical structures such as the basal ganglia (Takakusaki, Tomita & Yano, 2008). However, cognition is also important for locomotor tasks, particularly tasks that require attention and processing speed, such as multi-tasking and gait adaptability (Montero-Odasso et al., 2012;Caetano et al., 2017). Traditionally, the role of cognition has been assessed using dual-task paradigms (walking while performing a secondary cognitive task) which provide indications of the role of attention and executive function in the regulation of gait control (Montero-Odasso et al., 2012) and the negotiation of obstacles (Caetano et al., 2017;Caetano et al., 2018). Impaired cognitive processing has been associated with reduced gait speed and increased gait variability during complex gait (Killane et al., 2014;Hausdorff, 2005), however how higher level brain areas are activated during complex walking tasks is still unclear.
Functional near-infrared spectroscopy (fNIRS) is an optical neuroimaging technique for investigating cortical brain area activation while participants move freely. This technique is particularly useful for monitoring hemodynamic responses to brain activation (i.e., changes in oxygenated (oxyHB) and deoxygenated hemoglobin (deoxyHB)) in cortical regions before and after stimulation (i.e., resting followed by simple walking or simple walking followed by dual-task walking) (Leff et al., 2011).
Two overlapping theories have been posited for relative changes in cortical activity as measured with fNIRS. The first suggests reduced activity represents decreased use of a brain region and therefore increased efficiency (Lustig et al., 2009;Grady, 2012). The second suggests increased cortical activity is a compensatory mechanism and reflects over-recruitment and reduced efficiency (Cabeza et al., 2002;Reuter-Lorenz & Cappell, 2008;Grady, 2012).
Several reports of brain activation during walking using fNIRS have been published in the last decade. Activation of the Prefrontal Cortex (PFC) (easily accessible using fNIRS) has often been investigated during walking tasks (Leff et al., 2011). Brain motor areas investigated also include the Pre Motor Cortex (PMC), the Pre Supplementary Motor Area (preSMA), the Supplementary Motor Area (SMA) and the Sensory Motor Cortex (SMC) Koenraadt et al., 2014;Lu et al., 2015;Suzuki et al., 2004;Suzuki et al., 2008). Thus, there is now considerable literature that requires synthesizing and systematic review of the main findings related to the brain activation as assessed by fNIRS during walking tasks.
Further analysis and synthesis of published fNIRS studies are required to gain a better understanding of (i) brain activation changes during complex walking compared to simple walking or standing; (ii) brain activation patterns in healthy young people as this group provides the model of intact cognitive functioning; and (iii) brain functioning in diverse clinical groups with walking and neurological impairments. A methodological scale is also required to assist in the evaluation of the literature published to date.
Thus, we conducted a systematic review to summarize the published findings regarding brain activation patterns during simple and complex walking tasks in young adults, older adults and clinical groups with balance disorders, to gain an insight into neural processes required for ambulation. Our primary objectives were to determine whether (i) PFC activation patterns change when people perform gait tasks of increasing complexity requiring concomitant somatosensory, motor or cognitive tasks; (ii) PFC activation patterns during gait differ between young and older people and between patient groups and healthy controls. Our secondary aim was to evaluate each included study based on six methodological reporting criteria important for good data quality.

METHODOLOGY Search strategy
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statements defined by Moher et al. (2009) to identify and screen the articles included in this systematic review. We conducted searches in June 2018 using four databases: Embase, PubMed, Scopus and PsycINFO. A protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO) (registration number: PROSPERO 2017: CRD42017059501). We used the following Booleans terms: (((((near infrared spectroscopy) OR functional near infrared spectroscopy) OR nirs) OR fnirs) AND (((gait) OR walking) OR locomotion) AND (((((young) OR adult) OR older) OR elderly) NOT children)) AND (((Brain) OR cortex) OR cortical) for our search. We considered papers in English, Portuguese, Dutch and French.

Selection criteria
Study identification and screening were conducted independently by PP and MT or PP and JM with disagreements resolved by consultation and input from a third researcher (JM or SL). At stage 1 (identification), the researchers screened the manuscript titles and selected those that were consistent with the broad inclusion criteria. Studies were excluded if: (i) they were not in line with the review objectives; (ii) were conference abstracts with insufficient information for data extraction; (iii) were conducted in animals; (iv) were conducted in children/infants; (v) used fNIRS for other purposes (e.g., muscle studies); (vi) used fNIRS for standalone purposes (i.e., no walking assessment); (vii) used a device other than fNIRS (e.g., electroencephalogram) and/or; (viii) were published in a language other than English, Portuguese, Dutch or French.
At stage 2 (screening), the researchers screened the abstracts to identify papers that met the other specific review criteria: (i) used fNIRS to measure PFC cortical activation patterns; (ii) included walking tasks (simple and complex) other than stepping and; (iii) assessed young people, older people and/or clinical groups with balance disorders (defined as any peripheral or neurological condition that affects balance control). At stage 3 (eligibility), the full-text articles were assessed for eligibility. A manual search for additional relevant references from published reviews and articles was also conducted at this point. Articles were further excluded if: (i) they did not include a walking analysis; (ii) gait tasks involved walking speeds slower than 3 km/h; (iii) the participants performed tasks other than walking (i.e., stepping tasks); (iv) there was no baseline data comparison. Papers meeting all selection criteria were included at stage 4 (included papers) and relevant information was extracted from the papers by three authors (PP, MT and JM).
The primary outcome of the review was PFC activity change post stimulation. This was operationalized by changes in oxyHb, deoxyHb (gold standard measurement in brain magnetic resonance imaging (Obrig & Villringer, 2003;Lindauer et al., 2010), tissue oxygenation index (ratio of oxygenated to total tissue hemoglobin) and total hemoglobin level (sum of both oxyHb and deoxyHb). All hemodynamic changes reported in this review reflect statistically significant results reported by the authors of each study, i.e., p values <0.5.

Data extraction
From the included studies, relevant data were extracted and summarized for further analysis (Table 1). These included: (i) author and year; (ii) sample characteristics; (iii) study aims; (iv) gait assessment; (v) secondary task types; (vi) equipment details; (vii) fNIRS parameters used to describe the brain activation; (viii) control of motion artefacts and filtering; (ix) main findings; (x) study limitations; and (xi) conclusions.
A methodological reporting scale based on availability of information provided in the papers was devised. It comprised one point for the following 6 items: (i) equipment details described adequately, i.e., number of channels for the fNIRS, optode distances; (ii) movement artefacts/high frequency noise controlled for; (iii) use of either the 10-5, 10-10 or 10-20 electroencephalography electrode system to guide the optode placement; (iv) interference with external light controlled for; (v) heart rate changes and physiological noises controlled for; (vi) sample size in each group >10. Papers were classified as follows: low methodological reporting quality: total score < 2 points; medium methodological reporting quality: total score 3 and 4 points; high methodological reporting quality: total score ≥ 5 points.

RESULTS
A total of 308 study records were identified from the four databases; 103 unique studies with the removal of duplicates. The manual search of the references of these studies identified 12 further relevant studies. Of these 115 studies, 58 were deemed eligible for full-text assessment based on abstract review, with 35 meeting our final inclusion criteria (Fig. 1). The data extracted from these studies are summarized in Table 1. Nine studies involved young adults only (simple walking (1), fast walking (1), motor task (2), motor and cognitive tasks (2) and only cognitive tasks (3)); 4 studies involved young and older people (motor and cognitive tasks (1), only cognitive tasks (3)); 1 study involved young and older people and a clinical group with balance disorders (stroke survivors) performing simple walking, motor and cognitive tasks; 10 studies involved older people only (fast walking (2), motor and cognitive tasks (1), somatosensory and cognitive tasks (1), somatosensory and motor and cognitive tasks (1), only cognitive tasks (5)); 8 studies involved older people PFC. Tasks were conducted in a dimly illuminated room. Each channel was visually inspected and movement artefacts were corrected and data were reconstructed. Hemodynamic response function low pass filter and waveletminimum description length de-trending algorithm to remove possible global trends due to breathing, heartbeat, vasoconstriction or experimental influences.
No between-group difference in brain activation across all channels. No significant effect of verbal memory secondary task (vs. simple walking task) on brain activation in young and HOA.   Chaparro et al. (2017) 10 individuals with MS (mean 56 ± 5 years old, 8 women) and 12 HOA (mean 63 ± 4years, 9 women).
To examine the effects of partial body-weight support-30% body weight support on PFC activation while DT in HOA with and without MS.
One trial per condition (simple and complex walk in non-weight-bearing and partial bodyweight support conditions). OxyHb. PFC. Data were visually inspected and removed in case of saturation or dark noise. Low-pass filter with a cut-off frequency at 0.14 Hz to eliminate possible respiration and heart rate signals and unwanted high frequency noise. The additional frequency noise was identified and removed by an expert data analyst.
PFC oxyHb levels significantly higher in DT condition vs. single task irrespective of obstacle condition. Slow gait moderated the increase in PFC activation in obstacle conditions: compared to participants with normal gait speed, in those with slow gait, PFC activation levels were significantly increased in obstacle condition relative to simple walking.
No reporting of controlling for external light. The authors also reported that the absence of deoxyHb analysis is a limitation, as well as that other brain areas involved with obstacle negotiation were not analysed.
Individuals with mobility limitations (slow gait) utilized more cognitive resources when navigating around obstacles.
(continued on next page)   OxyHb. PFC. Low-pass filter with a finite impulse response filter at 0.14 Hz to eliminate possible respiration, heart rate signals, and unwanted high-frequency noise. Using a combined independent component analysis/principal component analysis, environmental and equipment noise and signal drifts were removed from the raw intensity measurements.
Significant age × condition interaction, whereby significant increase in activation levels in PFC bilaterally in WWT compared with NW. The increase is also significantly more pronounced in young compared with the old.  To eliminate possible respiration, heart rate signals, and unwanted highfrequency noise, raw intensity measurements at 730 and 850 nm were lowpass filtered with a finite impulse response filter with a cut-off frequency of 0.14 Hz. Noise (saturation or dark current conditions) was observed in 4% of the data that were subsequently excluded.
Higher oxyHb levels during walking while talking significantly higher compared with normal walk in normal. Central neural gait abnormalities was associated with significantly attenuated changes in oxyHb levels in walking while talking compared to single walking task.
Among participants without neurological gait abnormalities, higher oxyHb levels (versus lower oxyHb levels-median split) were related to better cognitive performance, but slower gait velocity. In contrast, higher oxyHb levels during walking while talking among older adults with peripheral neurological gait abnormalities were associated with worse cognitive performance, but faster gait velocity.
Unknown if lighting conditions were controlled for.
Neural confirmation of the posture first hypothesis emerged among older adults whose postural and locomotive abilities were compromised. Increased activation in the PFC during locomotion may have a compensatory function that is designed to support gait among older adults with peripheral neurological gait abnormalities.
Second order low-pass Butterworth filter with a cut-off frequency of 1.25 Hz was conducted to reduce high frequency noise. A second order high-pass Butterworth filter with a cut-off frequency of 0.01 Hz was used to reduce low frequency drift cause by fNIRS. After the correction for superficial interference, a second order low pass Butterworth filter with a cut-off frequency of 1 Hz was conducted. Continuous blood pressure monitoring. Short separation channels were used to remove hemodynamic changes in superficial tissue layers.
Task effect: Increased activation in the significant upper PFC channel (larger deoxyHb decrease) during the early-task (6-18.5s) for precision stepping compared to normal walking. No significant difference in motor cortices activation between the 2 walking conditions for either early or late phase. Phase effect in PFC: normal walking: increased activation in pre-task vs. early and latetask in normal walking (larger deoxyHb levels for early and late-task compared with pre-task) and versus late-task in precision stepping. Phase effect for motor cortices activation: in normal walking, larger pre-task oxyHb compared with early and late-task. The SMA, M1, and S1 revealed no significant differences between normal walking and precision stepping.
The authors pointed as limitation the small number of optodes used in this study.
The lack of M1/S1 activation during gait suggests that even in the current precision stepping task the control of ongoing gait depended mostly on subcortical automatisms, while motor cortices contributions did not differ between standing and walking. A prolonged activation of the PFC for precision stepping indicated that more attention was needed to perform precision stepping in comparison to normal walking.   OxyHb. PFC.
To eliminate physiologically irrelevant effects, a low-pass filter was applied with a finite impulse response filter, with a cut-off frequency at 0.14 Hz before processing the signals.
Walking alone demonstrated the lowest levels of oxyHb followed by walking + counting condition, followed by Walking + S7 condition, which was significantly different compared to the two other walking conditions. No significant differences in oxyHb levels were observed between usual walking and the standing condition or between standing with or without serial subtraction. During walking + S7, gait variability and the number of subtractions completed were inversely correlated with oxyHb levels.
The order of the tasks was not randomised. OxyHb.

PFC.
A band-pass filter with frequency of 0.01-0.14 Hz was used to reduce physiological noise such as heart beat and drift of the signal. To remove motion artefacts a wavelet filter was used, followed by correlation based signal improvement. Probes were shielded from ambient light using a black cloth.
HOA had significant increases in oxyHb levels during simple walking, relative to standing. Both groups showed significantly increased PFC activation in the DT and in the obstacle negotiation conditions compared with usual walking. Significant group effect whereby younger participants had lower oxyHb levels than HOA, in all conditions. Significant positive correlation between age and oxyHb levels, as well as between oxyHb levels and gait variability in HOA in the obstacle condition.
Age was a significant independent predictor of oxyHb levels in the usual walking. Age and gait speed were significant independent predictors of oxyHb levels in the obstacle condition but not the DT.
The authors pointed the limitations as: small sample size; few probes to analyse the brain activation in the PFC and also, that, other brain areas (motor cortex) should be investigated in future studies.
PFC activation during walking is dependent on age and task properties. HOA apparently rely more on cognitive resources even during usual walking.
To investigate the association between PFC activity and DT interference on physical and cognitive performance in stroke survivors.
Walking at a comfortable pace around a circle with a radius of 2.5 m for 60s. Baseline: 40s prior to calculation task and 20s after the task. The outcome measured was the trunk linear acceleration.
3 blocks (i) control period: the participants were instructed to repeat the number 1-10 in sequence; OxyHb and deoxyHb.
PFC. The devices were shielded from ambient light by covering the whole forehead with a black cloth. The movement artefact reduction algorithm was performed within each trial. The fNIRS signals were also linearly de-trended per trial and low-pass filtered at 0.1 Hz using a Butterworth filter to remove heart rate and other higher frequency physiological signals. Before starting any trial, participants stood for at least 1 min to minimize blood pressure fluctuations after standing up.
Bilateral PFC oxy-Hb concentrations were significantly increased during walking while serially subtracting and reciting digit spans when compared to rest. DeoxyHb concentrations did not differ between the walking tests and rest.
The authors pointed as limitations: the small sample; that they did not control blood pressure simultaneously with fNIRS that they did not control for superficial hemodynamics using short reference channels for example.
Using the new wireless fNIRS devices described in this paper, it is feasible to measure the PFC activity in PD during DT walking.
To determine the obesityspecific activation of the PFC using fNIRS during simple and complex ambulatory tasks in older adults.
Walking back and forth at a selfselected pace for 30s (4 trials). Baseline: 30s quiet seating. 10s quiet standing in between each trial and 2 min of seated rest between blocks. Gait outcome: gait speed. OxyHb and total Hb levels.

PFC.
Heart rate was continuously monitored using a heart rate monitor and was averaged across each trial. Participants instructed to avoid any sudden head movements during the tasks. Probes covered with a black headband to eliminate external lights.
Significant task * group interactions on oxyHb levels. Significantly increased oxyHb levels in DT and precision walking tasks compared to rest and usual walking, as well as in obese compared with non-obese individuals. Obesity was associated with three times greater oxyHb levels, particularly during the precision gait task, despite obese adults demonstrating similar gait speeds and performances on the complex gait tasks as non-obese adults.
The authors pointed: small sample size might limit the extension of these results; some conditions were not controlled (diabetes, higher blood pressure, etc.). Order of simple and DT trials counterbalanced but precision walking always last.
In order to maintain gait performance, obesity was associated with higher neural costs, and this was augmented during ambulatory tasks requiring greater precision control.
(continued on next page) Blood pressure, heart rate and arterial oxygen saturation were measured immediately before and after each task PFC activation was significantly greater when the participants ran at 9 km/h than when they walked at 3 km/h and at 5 km/h. Activations in the PFC, PMC and medial-SMC were significantly greater than that in the lateral SMC.
No mention of filtering or controlling for movement artefacts and ambient light in this study.
The PFC was significantly more activated during the periods before reaching a constant speed in the 9 km/h run compared with the 5 km/h walk and compared with the 3 km/h walk. PFC might be involved together with other structures in controlling locomotion to adapt to the increasing speed in the acceleration of phase of locomotion. Suzuki et al. (2008) 7 healthy young adults, 31.3 + 4.8 years old, 3 women.
To assess how a verbal instruction before walking would affect cortical activation and walking performance using fNIRS.
To evaluate the correlation between PFC activity and DT cost during smartphone use while walking in young and HOA.
Baseline: overground walking on a 2.5 m radius pathway at self-selected speed for 30s (5 trials). Gait outcomes: step time and acceleration magnitude. Rapid changes in oxyHb concentration more than 3 SD over the average for two consecutive samples were considered as movement artefacts. All blocks that were affected by these motion artefacts were excluded. A band pass filter of low pass 0.5 Hz was applied for the effects of Mayer waves and highfrequency fluctuations. A high-pass 0.01 Hz was used to account for baseline drift. Participants were instructed to keep their faces turned to the screen of the smartphone and to minimize head movements in all conditions.
No significant effect of age or site on oxyHb levels in the DT condition. In healthy young adults: significant positive associations between right PFC activation and DT cost on acceleration magnitude and negative association between left PFC activation and DT cost on error rate. In HOA: negative associations between middle PFC activation and DT cost on step time and between left PFC activation and DT on acceleration magnitude. DT costs on correct and mistake rates in the HOA group were significantly higher than in the young group.  and clinical groups with balance disorders (simple walking (1), motor and cognitive tasks (2), somatosensory and cognitive tasks (1), only cognitive tasks (4)); 3 studies involved solely clinical groups with balance disorders (simple walking (1) and only cognitive tasks (2)). These studies are summarized in Table 2.
Thirteen studies Chen et al., 2017;Clark et al., 2014a;Clark et al., 2014b;Hawkins et al., 2018;Lin & Lin, 2016;Lu et al., 2015;Maidan et al., 2016;Mirelman et al., 2014;Mirelman et al., 2017;Nieuwhof et al., 2016;Osofundiya et al., 2016; reported two or more comparisons, i.e., walking while crossing an obstacle and walking while performing a serial subtraction, both in comparison to a baseline condition such as simple walking. Many studies also contrasted brain activation in more than one area. In total, 75 brain activation comparisons are included in this review. Table 3 presents the individual and overall methodological reporting scores, and also describes how the scores were attributed for each paper. Twenty studies (57%) were classified as high quality, 10 (29%) as medium quality and five (14%) as low quality. Figure 2 shows the number of comparisons from the included studies showing an increase, decrease or no change in PFC activation when comparing either (a) walking to standing or (b) walking with an additional task to simple walking, according to group. In young people,    increased cortical activation in the SMA , while one found no change in the SMA as well as no change in brain activation in the preSMA and motor and sensorimotor areas (Primary Motor Cortex (M1) and Primary Sensorimotor Cortex (S1)) . One comparison also reported increased cortical activation in the PMC  while another found no change in activation in the PMC and the SMC (Suzuki et al., 2004). Of the 37 comparisons from 22 studies conducted in older adults, 23 comparisons reported significant increases in PFC activation with increasing locomotor task complexity Mirelman et al., 2017;Hawkins et al., 2018;Chaparro et al., 2017;Chen et al., 2017;Harada et al., 2009;Holtzer et al., 2011;Holtzer et al., 2015;Holtzer et al., 2016;Holtzer et al., 2017a;Holtzer et al., 2017b;Maidan et al., 2016;Verghese et al., 2017;Clark et al., 2014a;Clark et al., 2014b;Osofundiya et al., 2016;Hernandez et al., 2016); two comparisons reported a reduction in PFC activation Clark et al., 2014a) and twelve comparisons reported no change Eggenberger et al., 2016;Mori, Takeuchi & Izumi, 2018;Harada et al., 2009;Takeuchi et al., 2016;Clark et al., 2014b;Maidan et al., 2016;Caliandro et al., 2015;Al-Yahya et al., 2016).
There was no indication that the PFC activation was associated with methodological reporting scores; i.e., increased activation was reported in 23/36 (64%) comparisons in high quality studies, 14/18 (78%) comparisons in medium quality studies and 10/19 (53%) comparisons in low quality studies; (χ 2 = 2.56, df = 2, p = 0.279) ( Table 4). Regarding the effects of a secondary task during walking on PFC activation, 9/14 (64%) comparisons that used counting backwards reported increases, 20/24 (83%) that used verbal fluency reported increases, 11/19 (58%) that used complex motor tasks reported increases and 0/4 (0%) that used visual tasks reported increases (Table 5). Table 6 shows the effect of an additional motor or cognitive task relative to simple walking on gait outcomes in healthy young adults, healthy older adults and clinical groups with balance disorders. Reduced gait speed was reported in all studies investigating overground walking with the exception of one study that observed no changes in gait speed when older people walked while counting backwards by 3 or negotiated an obstacle course . No changes in gait speed were also observed in the two studies of treadmill walking where walking speed was controlled by the examiner Clark et al., 2014b). Shorter step/stride length was observed in all studies conducted on level surfaces except for one study that investigated counting backwards from a 3-digit number while walking on a treadmill  where both older people and stroke survivors exhibited increased stride length compared with simple walking. As expected, higher spatiotemporal variability was observed when people performed a secondary motor task that manipulated spatiotemporal characteristics, such as obstacle crossing and precision stepping Clark et al., 2014b;Mirelman et al., 2017), but also in one study where older people performed a verbal fluency task . No changes in gait variability were observed when young people walked while performing arithmetic tasks Meester et al., 2014), when young people walked carrying a tray , when somatosensory information was manipulated in older people (Clark et al., 2014a;Clark et al., 2014b) and when young and older people walked while negotiating obstacles or while counting backwards by 3 . One study, by Nieuwhof et al. (2016), showed decreased stride length variability in people with PD when performing the digit span task while walking. Table 7 presents group comparisons with respect to brain activation changes resulting from undertaking a complex walk between (i) healthy older and young adults and (ii) clinical groups with balance disorders and healthy peers. Of the five studies that contrasted PFC activation changes when conducting a dual task walk between young and older adults, only one study showed greater increases in PFC activation in older adults in all tasks performed . Another study , reported greater PFC activation in older people in simple walking in only the first half of an obstacle negotiation task and not when walking and performing a verbal fluency task. Two other studies reported no group differences Takeuchi et al., 2016) and one reported a relatively smaller increase in PFC activation in older people .
Thirteen comparisons from eight studies have contrasted PFC activation changes when conducting a dual task walk between clinical groups with balance disorders and healthy peers. Four of these reported a relatively larger increase in PFC activation in clinical groups with balance disorders. Two comparisons showed increased PFC activation when people with multiple sclerosis performed cognitive dual tasks Hernandez et al., 2016), while the other comparisons showed increased PFC activation when stroke survivors walked while performing a cognitive dual task  or in the first half of an obstacle negotiation task . Six comparisons reported no between-group PFC activation differences. In three of these comparisons, stroke survivors , PD  and obesity  performed motor tasks (people with obstacle negotiation and precision stepping group), while in three other comparisons, stroke survivors , people with neurological gait  and obesity group  performed cognitive tasks. Finally, the three reports of PFC decreases in clinical groups were observed when people with multiple sclerosis walked with partial body support  and when people with PD and stroke survivors performed a cognitive dual task Mori, Takeuchi & Izumi, 2018).

DISCUSSION
This systematic review summarizes the published findings regarding PFC cortical patterns of activation in healthy young adults, healthy older adults and clinical groups with balance disorders, to gain an insight into neural processes during simple and complex walking tasks. Approximately 60% of the study comparisons reported that healthy young and older adults exhibited higher PFC activation when performing a complex task while walking compared with a baseline simple walking task; this was also the case for more than 80% of study comparisons of clinical groups with balance disorders. Moreover, PFC activation appears to be related to the type of complex walk undertaken.

Brain activation in healthy young adults
Compared with simple walking, PFC activation increased when individuals performed: (i) fast walks (Suzuki et al., 2004); (ii) negotiated expected obstacles Maidan et al., 2018) and unexpected obstacles of different heights ; (iii) performed a secondary task while walking Mirelman et al., 2014); the last being the paradigm used in most studies that have assessed brain activation patterns in healthy young adults. In these studies, participants performed the following secondary tasks: subtracting numbers by 1, 3 or 7 Meester et al., 2014;Mirelman et al., 2014;Mirelman et al., 2017), walking while talking  and talking and carrying a bottle of water on a tray . However, increased hemodynamic responses in the PFC appear to be task-specific. Indeed, several studies of healthy young adults Koenraadt et al., 2014;Lin & Lin, 2016;Beurskens et al., 2014;Takeuchi et al., 2016) reported no change or even a decrement in PFC activation during either secondary motor task performance (e.g., precision stepping, crossing obstacles or walking on a narrow pathway) or cognitive task performance (e.g., visual checking, alphabet recall, memory task and manipulating a smartphone). Such findings suggest that increasing balance/ locomotor changes does not require additional PFC activation in healthy young adults, and might involve other cortical and subcortical areas involved in the control of locomotion. In addition, secondary tasks involving working memory (Lin & Lin, 2016;Beurskens et al., 2014) might also involve cortical areas in addition to the PFC that were not investigated in the published studies.

Brain activation in healthy individuals: age comparisons
Healthy older adults usually walk slower and have more difficulty performing dual tasks than healthy young adults, (Al-Yahya et al., 2011). However, of the five studies that investigated between age-group effects when performing walks with secondary tasks, only one study observed greater PFC activation in older people performing different secondary tasks (obstacle negotiation and counting backwards) . PFC activation patterns were not different between healthy young and older adults in three studies Takeuchi et al., 2016;Hawkins et al., 2018) and lower in older adults in one study . The limited number of studies which have explored the effects of aging as well as the nature of the secondary task used in these studies might account for the lack of an age effect.

Brain activation in clinical groups with balance disorders
In most clinical groups with balance disorders, including stroke survivors Hawkins et al., 2018), obese individuals , individuals with ataxia (Caliandro et al., 2015), multiple sclerosis Hernandez et al., 2016), peripheral neuropathy , and mild cognitive impairment (Doi et al., 2013), higher PFC activation has been reported regardless of the type of concomitant task performed during ambulation. This is also the case for comparisons made in studies in people with PD Thumm et al., 2018;Nieuwhof et al., 2016) with one exception; Maidan et al. (2016) found no change in PFC activation when comparing walking whilst performing a concomitant subtracting task with simple walking.

Theoretical considerations
Our findings of enhanced hemodynamic responses in the PFC apparent when older adults and individuals with balance disorders perform complex walking tasks align particularly with the notion that increased cortical activity reflects a compensatory mechanism (Cabeza et al., 2002;Reuter-Lorenz & Cappell, 2008;Grady, 2012). This might reflect the need to allocate more attentional resources to walking while performing secondary tasks, or the need to use a more direct locomotor pathway due to deficits in automaticity (e.g., as generally observed in individuals with PD) (Herold et al., 2017). The age-related differences are also consistent with the frontal lobe hypothesis of aging (West, 1996) and the cognitive reserve theory which supports that older adults increase brain activity by a larger degree to cope with elevated cognitive task difficulty (Stern, 2009). Moreover, these functional effects of aging mirror age-related structural changes proposed by the ''last-in-first-out'' hypothesis where late maturing brain regions decline first in later life (Raz & Kennedy, 2009;Tamnes et al., 2013;Bender, Volkle & Raz, 2015) and explain gait disturbances.
control for important aspects that could have affected the interpretation of the data, such as motion artefacts, external lighting and physiological noise (Lin & Lin, 2016), and one was conducted in people with PD . Obstacle negotiation and precision stepping tasks increased PFC in the three studies performed in clinical groups with balance disorders Maidan et al., 2016;Hawkins et al., 2018), but this may simply indicate any additional load may elicit such changes in such populations.
Finally, visual tasks such as visual checking and manipulating a smartphone did not increase PFC activation Takeuchi et al., 2016). For these task types, other brain areas such as the visual cortex might be more involved. However, further studies directly assessing visual cortex activation as well as other cortical regions are required to confirm this hypothesis.

Study limitations
Studies addressing gait with fNIRS are still at a relatively early stage, with best practice methodology evolving as experience with this technique is garnered. Studies may have not met particular quality criteria due to the pioneering nature of the studies using this new technology and/or omission of reporting of all methodological factors and most of the papers (89%) had one or more of the following methodological limitations: small sample sizes, no indication of removal or control of motion artefacts or physiological noise in data processing and sub-optimal number and positioning of optodes. Further, the secondary tasks used in many studies involved speaking (counting backwards and verbal fluency) requiring muscles adjacent to the PFC (Zimeo- Morais et al., 2018). Such muscle activity as well as different facial expressions (Balardin et al., 2017) may affect fNIRS signal quality.
To address the above, we recommend that in future studies, sample sizes be based on power analyses of expected effect sizes for spatiotemporal gait and hemodynamic measures to provide confidence in the study findings. Second, motion artefacts should be removed during the data processing or be controlled for; confounding physiological noise such as fluctuations in heart rate should either be monitored and reported, or controlled for using appropriate filtering. Third, baseline and test trials should be of sufficient duration to detect the slow changing hemodynamic signals as oxygenated blood starts flowing between 1 and 2s after stimuli onset and achieves its peak approximately 6s after stimulus onset . Finally, EEG standards (i.e., 10-5) and/or anatomical maps to define optode positions (i.e., Brodmann areas) for the brain regions of interest should be used. However, we acknowledge that given the limited number of fNIRS channels in current devices, this may prove to be an unavoidable limitation for some investigations Maidan et al., 2016).
We also acknowledge some limitations. First, there was considerable heterogeneity of study protocols in the present review. As such, variations in baseline conditions (e.g., sitting/standing/unspecified), walking speed (e.g., self-selected/controlled), duration and amount of trials, treadmill vs. overground walking, montage, inter-optode distance, etc. limited the clustering of studies and hindered the overall interpretation of the data. Finally, factors such as motor repertoire, physical activity, practice and skill levels, risk of falling and hemispheric asymmetry (Ekkekakis, 2009;Erickson et al., 2007;Jancke, Shah & Peters, 2000;Naito & Hirose, 2014) can affect cortical activity but were beyond the scope of this review. Complementary studies and reviews are required to elucidate the influence these factors have on cortical activity and associated balance control.

CONCLUSION
This systematic review revealed that the majority of studies found increased PFC activation with increased walking task complexity in young and older people and clinical groups with balance disorders. However, increased PFC activation was most common in studies that contained walks comprising secondary tasks of verbal fluency, arithmetic and alphabet reciting. The finding that clinical groups with balance disorders generally showed increased PFC activation irrespective of type of secondary task during walking suggests these groups require more attentional resources for safe walking.