Biofeedback Systems for Gait Rehabilitation of Individuals with Lower-Limb Amputation: A Systematic Review

Individuals with lower-limb amputation often have gait deficits and diminished mobility function. Biofeedback systems have the potential to improve gait rehabilitation outcomes. Research on biofeedback has steadily increased in recent decades, representing the growing interest toward this topic. This systematic review highlights the methodological designs, main technical and clinical challenges, and evidence relating to the effectiveness of biofeedback systems for gait rehabilitation. This review provides insights for developing an effective, robust, and user-friendly wearable biofeedback system. The literature search was conducted on six databases and 31 full-text articles were included in this review. Most studies found biofeedback to be effective in improving gait. Biofeedback was most commonly concurrently provided and related to limb loading and symmetry ratios for stance or step time. Visual feedback was the most used modality, followed by auditory and haptic. Biofeedback must not be obtrusive and ideally provide a level of enjoyment to the user. Biofeedback appears to be most effective during the early stages of rehabilitation but presents some usability challenges when applied to the elderly. More research is needed on younger populations and higher amputation levels, understanding retention as well as the relationship between training intensity and performance.


Introduction
Lower-limb amputation (LLA) is associated with major rehabilitation challenges and lifelong mobility limitations. Limb loss not only hinders aspects of motor control, but it also reduces the sensory feedback information and proprioception that are associated with the peripheral nervous system [1][2][3][4]. As a result, individuals with LLA often walk slower and expend more energy than non-amputees [5]. They also exhibit atypical gait and loading patterns [5][6][7] that may be associated with long-term secondary health issues including chronic back pain and joint problems [8]. Moreover, poor balance and gait function in individuals with LLA can lead to the fear of falling and a greater incidence of falls [9][10][11][12], with more than half of ambulating adults with LLA falling at least once per year [11]. The consequences of these falls include injury and hospitalization [12,13], heightened fear of falling leading to prosthesis disuse [14,15], and the subsequent social withdrawal reducing their ability to recover from the trauma, both physically and psychologically [14,16].
Improving balance and gait is an important part of the rehabilitation process. Gait retraining, which is typically provided by a physiotherapist or prosthetist, includes the observation of gait

Inclusion and Exclusion Criteria
The article inclusion/exclusion criteria were divided into three main sections: (i) study population, (ii) biofeedback application, and (iii) publication type (Table 2).

Screening and Data Extraction
After the duplicates were removed, two independent reviewers screened the titles and abstracts of retrieved studies for relevance using the predefined eligibility criteria ( Table 2) (R.E. & A.M.). The remaining studies received full-text assessments. Articles with titles and abstracts that did not provide enough information for the article screening process were fully reviewed. The data were extracted based on the study aims. Accordingly, the following aspects for data extraction were considered: (i) year of publication; (ii) BFB objective and application; (iii) characteristics of the sample population; (iv) BFB design (BFB modality, BFB device, feedback strategy, sensors/transducers); (v) testing conditions (clinical/laboratory settings or field-based studies and treadmill or overground walking); (vi) outcome measures (targeted gait parameters, physical, physiological, or any other parameters, including questionnaires); (vii) experimental protocol (information related to subject's testing, such as number of sessions, duration, frequency, number of trials, follow-up interventions); and, (viii) key findings that were related to the efficacy and effectiveness of current BFB systems as gait rehabilitation tools for individuals with LLA. A third reviewer (J.A.) resolved the ambiguities or disagreements in the independent reviews of the articles (reviewers R.E. & A.M.). Additionally, the references of all included articles were scanned to identify other relevant studies that were missed in the original search.

Risk of Bias (Quality) Assessment
Most of the inclusions were peer-reviewed journal articles thus maintaining the quality of this systematic review and reducing the risk of publication bias. Two independent reviewers assessed all articles that met the inclusion criteria (R.E. & A.M.). In addition, a quality assessment was performed using a customized data extraction formula ( Table 3). The approach was based on previous Sensors 2020, 20, 1628 5 of 26 standardized methods [69][70][71] and reviews, and allows for data extraction that is relevant to the topic of interest [72,73]. For instance, Peters et al. [72] assessed 20 reviewed articles by using 19 appraisal questions as quality indicators. The appraisal questions were designed to collect information regarding the main research aims. Similarly, Ku et al. [73] utilized 14 appraisal questions to evaluate 23 articles that were related to balance control of individuals with LLA during quiet standing. The evaluation process in the current systematic review was adapted from these previously established appraisal criteria (Table 3) [69][70][71][72][73]. Accordingly, the score of each article provided a standardized measure for assessing the quality of research among the articles. Reviewers R.E. & A.M. independently applied the ratings and reviewer J.A. resolved disagreement.

Search Results
The initial search yielded 2456 studies (i.e., Medline 440, PubMed 426, Embase 550, IEEE 221, Web of Science 281 and Scopus 538). After the duplicates were removed, the title and abstract of 1419 articles were screened for potential relevance. Seventy-two (n=72) full-text articles were assessed for eligibility. Following the application of the eligibility criteria, thirty-one (n = 31) full-text studies were included in this systematic review. The flow diagram summarizes the overall review process ( Figure 1). The most common reasons for the exclusion of articles during full-text assessment included: (1) BFB systems not being tested on individuals with LLA, (2) mainly used for gait event detection, and (3) used to assess user's sensory perception (i.e., reaction time and subject's accuracy in response to stimulation). assessed for eligibility. Following the application of the eligibility criteria, thirty-one (n=31) full-text studies were included in this systematic review. The flow diagram summarizes the overall review process ( Figure 1). The most common reasons for the exclusion of articles during full-text assessment included: 1) BFB systems not being tested on individuals with LLA, 2) mainly used for gait event detection, and 3) used to assess user's sensory perception (i.e., reaction time and subject's accuracy in response to stimulation).  Table 4 depicts the results of the criteria applied to assess the quality of the reviewed articles. Most articles were high quality and included complete information about the research objectives, study design, participants characteristics, BFB modality and application, BFB components, primary gait outcome measures, key findings supported by results, and conclusions. However, some articles presented limited information about experimental protocol and study limitations. Whereas, other studies presented limited or null information about statistical analyses and key findings supported by other literature. The results showed that 16 out of the 31 studies satisfied at least 85% of the criteria. Nine studies ranged from 70% to 85%, and six studies scored less than 70%.

Quality of Reviewed Articles
It should be noted that Lee et al. published five articles with similar methodology, data sets, and outcomes between the years of 2007-2013 [56,[74][75][76][77].

Key Data Extracted from Reviewed Articles
BFB systems applied as a gait rehabilitation tool for prosthetic users has gained popularity over recent years, with most articles published from 2007 to 2019 (n = 21, 68%). The participants with transtibial amputation were most studied (n = 17, 55%) and the included studies had a median sample size of seven participants. Only two studies [34,66] compared BFB performance versus a control group of healthy subjects. Most of the studies included middle-aged (aged 30-59 years) and elderly (above 59 years) prosthetic users. Study participants had prosthetic experience ranging from one month to 53 years.
Only six studies presented detailed characteristics about the prosthetic components (i.e., prosthetic joint, foot, and socket) utilized to evaluate BFB [34,59,[78][79][80][81]. During BFB testing, most of the participants wore their prescribed prosthesis (i.e., passive mechanical or microprocessor-  Table 4 depicts the results of the criteria applied to assess the quality of the reviewed articles. Most articles were high quality and included complete information about the research objectives, study design, participants characteristics, BFB modality and application, BFB components, primary gait outcome measures, key findings supported by results, and conclusions. However, some articles presented limited information about experimental protocol and study limitations. Whereas, other studies presented limited or null information about statistical analyses and key findings supported by other literature. The results showed that 16 out of the 31 studies satisfied at least 85% of the criteria. Nine studies ranged from 70% to 85%, and six studies scored less than 70%.

Quality of Reviewed Articles
It should be noted that Lee et al. published five articles with similar methodology, data sets, and outcomes between the years of 2007-2013 [56,[74][75][76][77].

Key Data Extracted from Reviewed Articles
BFB systems applied as a gait rehabilitation tool for prosthetic users has gained popularity over recent years, with most articles published from 2007 to 2019 (n = 21, 68%). The participants with transtibial amputation were most studied (n = 17, 55%) and the included studies had a median sample size of seven participants. Only two studies [34,66] compared BFB performance versus a control group of healthy subjects. Most of the studies included middle-aged (aged 30-59 years) and elderly (above 59 years) prosthetic users. Study participants had prosthetic experience ranging from one month to 53 years.
Only six studies presented detailed characteristics about the prosthetic components (i.e., prosthetic joint, foot, and socket) utilized to evaluate BFB [34,59,[78][79][80][81]. During BFB testing, most of the participants wore their prescribed prosthesis (i.e., passive mechanical or microprocessor-controlled knee prostheses). In terms of BFB effectiveness and prosthetic components, the results showed that BFB systems were capable of improving the gait performance of individuals with LLA, regardless of the type of prosthetic components (i.e., passive mechanical knee or microprocessor-controlled knee or powered knee prostheses) [35,59,[79][80][81][82]. FSRs (force sensitive resistors) sensors that were attached to the plantar surface of the prosthetic foot were the most frequently used transducer for measuring the targeted gait parameters. The most commonly targeted gait parameters were related to limb loading, ground reaction forces, and symmetry ratios for stance or step time. Visual feedback was the most used modality, followed by auditory and haptic. Haptic feedback has been most frequently used in recent studies. For instance, 10 out of 15 studies published from 2012 to 2019 utilized some type of haptic feedback (i.e., vibrotactile, electrotactile, electrocutaneous, or intraneural stimulation) when compared to two out of 16 studies during years 1975 to 2011.
Most studies assessed the performance of the BFB systems under laboratory conditions either walking on a treadmill or over ground. Most studies (above 50%) also performed only one gait training session in which BFB was delivered to the participants. Most studies compared subject's gait performance with and without wearing the BFB system, walking at self-selected speed. Most of the studies did not report any follow-up sessions with the BFB system to test for retention. In addition, few studies evaluated changes on metabolic consumption [34,59], physical fatigue [59], and cognitive load or mental effort [58,59]. Most studies presented positive gait outcomes that were related to one or more physical and physiological parameters. However, six (n = 6) studies [34,36,57,[83][84][85] reported mixed results, showing gait improvements for some participants and not others after BFB. None of the studies reported negative effects BFB on gait. Only two studies [84,85] reported non-persistent lasting effects and/or periods of retention after training. Table 5 details the key information that was retrieved from each article.

Discussion
The primary aim of this systematic review was to consolidate published evidence that was related to the development and testing of BFB systems as a gait retraining tool for individuals with LLA. The quality of the identified studies (n = 31) was generally high, particularly in more recent years (i.e., since 1990).

Sample Size
The average sample size across studies was 13 ± 3 non-amputee subjects, 7 ± 2 individuals with transtibial amputation, and 3 ± 1 individuals with transfemoral amputation. A few studies indicated that statistical tests could not be accurately performed and the findings cannot be extrapolated to larger populations due to their low sample size [56,58,96]. It was indicated that, although a small sample limits generalization of findings, a case study could provide pilot data and allow for exploratory research across a diverse population [97,98].

User Demographicss
The majority of lower limb amputees are over 50 years of age and most of the amputations are due to complications that are associated with vascular diseases [99,100]. The age of participants with LLA in this review ranged from young adults (as young as 19 years) to the elderly (aged 60 years and above). None of the studies focused on children and youth; a population that might benefit most due to a lifetime of prosthetic use and savviness for technology. There exist limitations that are associated with the transferability of findings between young and older populations. One reason for this is the differences in ambulation ability and the capacity to regain mobility function [100]. Elderly patients may require longer practice time, yet they typically suffer from lower physical endurance [17]. Further, with an older population come challenges with BFB usability. A recent study showed that, in contrast to healthy young adults, elderly healthy subjects were unable to utilize BFB information to reduce trunk sway while walking distracted [101]. Walking can be considered to be an unconscious (i.e., low cognitive) activity for healthy subjects [102]. However, for prosthetic users, walking often increases cognitive load and energy expenditure [103,104]. Prosthetic users usually depend on additional information (i.e., visual, auditory, and somatosensory) to ambulate safety [10].
Phantom limb pain affects many individuals with lower-limb amputation. It is manifested as sensations or pain from a body part that no longer exists [105]. Techniques, such as mirror therapy and BFB, have demonstrated the potential to reduce phantom limb pain in prosthetic users [59,86,105]. An important benefit of BFB systems when compared to mirror therapy approaches is that, in addition to reducing phantom pain, BFB systems can potentially improve the overall gait performance and prosthetic function of individuals with LLA [59,86]. Accordingly, participant characteristics (i.e., age, etiology, level of amputation, and prosthetic experience) should be considered in the development an effective and user-friendly BFB system, as these may dictate the most suitable BFB modalities (i.e., visual, auditory, or haptic feedback) and BFB strategies (i.e., control algorithms utilized to convey sensory feedback to BFB users).

Level of Amputation
Transfemoral amputees typically demonstrate more severe gait deviations than transtibial amputees [106]. There is also increased energy cost, loss of mobility function and decrease in walking efficiency with higher levels of amputation [97]. However, transfemoral amputees are generally underrepresented in the BFB research. Additionally, the targeted gait parameters appear to be dependent on the specific amputee demographic. For instance, gait symmetry ratios, stance times, hip and knee flexion/extension angles, and trunk sway are more relevant for transfemoral than those for transtibial amputees [6,107]. One study found it difficult to recruit transtibial amputees with asymmetric gait and recommended recruiting transfemoral amputees [36]. On the other hand, the gait of transfemoral amputees and their ability to make gait adjustments is highly dependent on the function of the prosthesis, and particularly the prosthetic knee joint [56]. In one study, for example, because of the inability of the artificial knee-joint to actively adjust, the TF amputee could only vary gait speed with the healthy limb [108]. Hence, improvements in gait may require gait retraining, but also adjustments to the prosthetic setup, particularly with higher-level amputations. While this might become less of an issue with self-adjusting microprocessor knee joints, it will be imperative that the BFB and prosthetic control systems are designed to work symbiotically. It was also suggested that future research should determine how different amputee populations can benefit from different BFB modalities and different BFB strategies [28,56,88].

Prosthetic Experience and Time Since Amputation
The best practices in amputee rehabilitation encourage physiotherapy, prosthetic fitting, and training to be provided as soon as possible post amputation. One study involving established prosthetic users (>2 years post amputation) suggested that research should focus on how effective BFB is in the early stages of rehabilitation [56]. Others suggest that the best results occur with novice users, but that experiments should be done on expert prosthesis users to examine effectiveness [28]. To this point, one study found that newer, less established amputees were better able to adapt their gait patterns [36]. While early rehabilitation using BFB has been suggested for other populations, such as stroke, this approach presents certain challenges in amputee research and rehabilitation [109]. Issues, such as prosthetic fit and residuum healing can complicate experimentation on new amputees [97]. Overloading in the post-operation stage might result in tissue breakdown [90] and premature rehabilitation might affect the incision and cause healing issues [110]. BFB systems have been developed to improve the healing process in the early postoperative period by warning amputees applying excessive pressure on the residual limb [90]. Several studies had confounded results, since the patients were being provided conventional physiotherapy during the time of the BFB experiments; this is a potential issue when testing with recent amputees receiving standard care [81,86,97]. Similarly, to limit the confounding effects, one study withheld prosthetic alignment changes as the participants began to exhibit better gait patterns. However, this goes against standard care and might have contributed to poorer results in the study [81]. Based on these findings, BFB training should be provided as soon as the residuum is healed and volume stabilized, a satisfactory and stable prosthetic setup, alignment and fit have been achieved, and conventional physiotherapy treatments have concluded to exclusively assess the effectiveness of the BFB systems.

BFB Intervention (Experimental Protocol)
None of the studies tested BFB under a randomized control trial (RCT). RCTs minimise the risk of confounding factors that might alter the results. For this reason, RCT studies are the golden standard for validating the effectiveness of an intervention. Over one-quarter of the studies reported using a single gait training session. It is important to allow for the user to have adequate training with the BFB system to enhance learning [64]. Training is important, as making errors drives motor learning [39]. Studies often do not report instructions given or training methods and research should be done to determine the best method of conducting the training sessions [111]. The effects of training intensity were mentioned in a few of the included papers and conclusions were mixed. One study concluded that outcomes are better with more intensive training [56], while others suggest gradually setting attainable targets [88]. Moreover, as in previous reviews, most studies did not report any follow-up sessions with the BFB system to test for retention [22,64]. Finally, the literature is unclear as to how BFB should be integrated into conventional physiotherapy. For example, systems, such as the one described by Redd et al., can be used with little specialized training and without the supervision of a physiotherapist/prosthetists [88]. It is likely that BFB might be most effective in combination with existing physiotherapy and gait training practices [112].

Treadmill vs. Overground Walking
Most studies used a treadmill during the experimental procedure. Nagano et al. has shown that temporal gait parameters, such as double stance time and swing time differ when walking on a treadmill compare than those walking overground [101]. Another study showed that walking on a treadmill reduces dorsiflexor and knee extensor moments [102] and increases hip extensor moments in the sagittal plane [102]. This suggests that BFB strategies utilized to alter gait symmetry might need to be modified, depending on treadmill or overground walking. Further, it has been shown that difficulty might arise when translating locomotor skills from treadmill training to overground walking [103].

BFB Parameter Measurements
Only some of the studies validated the BFB system's accuracy in sensing the targeted gait parameter(s). For example, Isakov et al. validated their pressure sensing insoles with a force plate [66] and Yang et al. used a previously validated BFB system consisting of a force plate and motion analysis system [36,113]. The inaccuracy of a goniometer for knee angle measurements was a limitation in one study [87]. Another study noted a source of error in the algorithm to detect heel-strike and toe-off [88]. The accurate and reliable measurement of gait deviations is crucial to the success of wearable BFB systems, as erroneous feedback and false positives calling for corrections in gait can confuse and frustrate BFB users. Few studies mentioned time delays in their BFB system [30,87,114]. For instance, Crea et al. [30] and Liu et al. [114] reported delays less than or equal to 200ms due to wireless communication with the sensors embedded in the shoe-insoles. Consequently, delays were produced in the detection of the feedback stimulus. Accordingly, effective BFB systems must have low latency, especially when sensing and providing real-time feedback during dynamic activities, such as walking.

BFB Modality
Visual feedback was the most common method provided to the participants, but there is some debate on the most appropriate and effective feedback modality. In one study, visual feedback was deemed the most intuitive modality that was based on user preference and usability [88]; however, the authors suggested that more work should be done to improve the usability/ease of use of vibrotactile and auditory feedback [88]. One study found that amputees and physiotherapists valued auditory over visual feedback and the participants adapted more easily to auditory feedback [56]. These preferences appear to be related to BFB design, testing, and safety aspects. BFB users might prefer or find more useful the BFB modality that provides more intuitive and relevant feedback information according to the gait parameter and task performed. For example, visual feedback is typically confined to specific locations (e.g., treadmill walking while watching a display under laboratory conditions). Safety issues relating to falling can arise during activities of daily living, for example, as prosthetic users walk and negotiate obstacles, such as curbs and stairs while watching a display or a smartphone. For this reason, auditory and haptic (e.g., vibrotactile) feedback may be more suitable for field or community-based systems [64]. A previous mapping review came to similar conclusions. It noted that, while visual feedback was most commonly used and studied, it might not be the most effective feedback modality for practical use [64]. Visual feedback is more appropriate for the perception of spatial information [115], while auditory and haptic feedback are better suited for the perception of temporal [115] and spatiotemporal information [39,115], respectively, according to the literature in motor learning. Accordingly, BFB systems need to appropriately align feedback modalities and strategies with measured gait parameters. For instance, in terms of vibrotactile feedback, diverse feedback strategies (i.e., a combination of vibration patterns-vibration levels and activation sequences-including different locations and number of vibrating motors) have been utilized to improve gait performance [28,36,40,58,88,116]. However, a systematic comparison of the current implemented feedback strategies is missing to explore which strategy might produce greater positive gait outcomes on individuals with LLA.
Of the reviewed studies, six (n = 6) provided multimodal feedback [58,[74][75][76]88]. Some researchers have suggested that multimodal feedback reduces cognitive load and can enhance motor learning [39,58,83]. For instance, Crea et al. evaluated the cognitive load of a visual-vibrotactile BFB system by adding a secondary cognitive task (i.e., serial subtraction) during walking with vibrotactile feedback [58]. The results showed that gait symmetry was improved without significant increases on cognitive load in the presence of feedback walking [58]. In another study, Pagel et al. showed that, when cognitive load increases, the imbalance between intact and prosthetic limb becomes more pronounced [84]. Cognitive impairment appears to be more common in individuals with LLA than in the general population-this is linked to difficulties with regaining mobility and independence after amputation [117]. Thus, if the feedback is too cognitively taxing, it might be counterproductive and more difficult for amputees to process the information, potentially even resulting in worsened gait and mobility performance [112]. For instance, Fernie et al. [85] designed an auditory BFB system to maintain knee extension through the stance phase; however, the BFB system was found to alter knee flexion instead. Chow et al. [90] originally designed an auditory BFB to encourage participants to increase the loading of the residual limb, but in fact the audio BFB prevented adequate loading of the residual limb.

Feedback Strategies
Feedback strategies mainly utilized baseline (no feedback) conditions to obtain an initial value of the specific gait parameters. Most of the studies set a gait target threshold for participants prior to BFB testing. Few studies set these thresholds based on the feedback provided by a physiotherapist, who assessed the participant's walking ability to personalize the BFB treatment or provided verbal cues prior or during the early stages of BFB gait retraining [17,34,36,81,85,94]. Feedback can be provided in real-time (concurrent feedback) or after the trial has finished (terminal feedback) [112]. All the studies applied concurrent feedback, and none provided terminal feedback. A recent review [112] showed that concurrent feedback produces the best short-term results, while terminal feedback produces the best long-term results [112]. Therefore, an effective training strategy could be to provide concurrent BFB and terminal feedback from the physiotherapist during training sessions.

Other BFB design and Application Considerations
When designing a BFB system, it is important that the BFB strategy is non-obtrusive and enjoyable for the user. Program adherence has been linked to program enjoyment [97]. This will ensure that the BFB systems will be used in the long-term. Motivation for walking is a predictive factor for successful rehabilitation of amputees [118]. If a wearable daily use system is not the goal for the researcher, virtual reality (VR) systems may be a good option to motivate and encourage program adherence, since users might perceive to have more interaction with the system. For balance, the Wii-Fit has been shown to improve the balance and gait in older adults [97], children and adolescents with amputation [119], as well as individuals with Alzheimer's and Parkinson's disease [120]. Although the Wii-Fit is no longer commercially available, other VR options have been used for rehabilitation purposes, including the CAREN system and C-Mill (both from Motekforce Link, Amsterdam). The two included studies used VR environments, specifically the CAREN system [34,81]. Virtual reality might be the best option for BFB because program adherence is important [97]. Alternately, if BFB systems are envisioned as wearable systems that are built into prostheses that provide as needed feedback during activities of daily living, they must do so unobtrusively and seamlessly. However, to date, only two studies tested their systems in the field [59,86]. Moreover, retention and fading, which are important considerations for the continuous use of feedback systems, are not well understood and require further attention [64].

Limitations of the Systematic Review
It was not possible to conduct a meta-analysis on the data due to the wide range of target gait parameters, outcome measures, and methods used. Further, six databases as well as the references from included studies were used to search for articles. Articles that were not written in English were not included.

Conclusions
Although most individuals with amputation are older, there is a lack of research on technology-based feedback for a younger or paediatric population. Further, the older population might have difficulty with the usability and response times of BFB systems. Different amputation levels may benefit from different feedback strategies and/or target parameters; therefore, it is important to investigate the effect on gait of different feedback strategies to ensure that the target gait parameter and the sensory information are appropriate for the target population. BFB training should be provided as soon as possible in the rehabilitation stage, but the training should not start until the stabilization of the early stages of the rehabilitation process. Auditory and vibrotactile feedback are more wearable and different population ages may respond to feedback differently, and it is important to align feedback modality and feedback strategies appropriately with the measured gait parameter(s). The relationship between training intensity and performance is unknown and future work should be conducted to investigate a possible correlation.
In terms of developing an effective, robust, and user-friendly wearable biofeedback system to improve the gait of individuals with LLA, the following aspects should be considered: (1) target gait parameters should be clinically relevant for the targeted population. For instance, gait symmetry ratios, stance times, hip and knee flexion/extension angles, and trunk sway are more relevant for transfemoral than those for transtibial amputees; (2) BFB modalities (i.e., visual, auditory, haptic, and multimodal feedback) should take into consideration usage conditions (i.e., laboratory, clinical, or home-care settings), including user's age, level of amputation, and prosthetic experience; (3) feedback information (i.e., BFB strategies) should be easy to perceive, discriminate, and utilize by BFB users, allowing for them to transfer feedback information with low cognitive demand; (4) wearable sensors, such as pressure sensors, load cells, electrogoniometers, etc. should be fully integrated into BFB systems to improve wearability. In addition, accelerometers, gyroscopes, or inertial measurement units (IMU sensors) are encouraged to be used for gait event detection, which might improve accuracy and portability of the BFB systems; and, (5) program adherence and program enjoyment should be sought to ensure the long-term use of BFB systems. Effective BFB systems might be achieved by designing a goal-oriented experimental intervention and by considering the previously mentioned points regarding BFB design.  Acknowledgments: The authors would like to thank Pui-Ying Wong for her support as an experienced librarian during the search strategy process. And, to Firdous Hadj-Moussa for her support in reviewing and editing the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.