Gait disabilities, mainly caused by strokes, compromise daily independence, quality of life, professional and social inclusion, and increase the risk of falling in adults ^[1][2][1,2]. Stroke survivors may regain their quality of life through neuroplasticity phenom, elicited by biofeedback systems (BSs) ^[3][4][5][3,4,5].

In the context of this resviearchw, a BS is a robotic device that measures gait-related unconscious parameters through sensors and feedback in real-time this information to users through visual, auditory, or haptic cues, using appropriate actuators [6]. Therefore, patients are aware of their abnormal behaviour, and they are intensively and repetitively encouraged to self-control it (fostering neuroplasticity) towards recovery.

A review on BSs designed for post-stroke gait rehabilitation is essential to find evidence on the BSs’ rehabilitation effects and to identify the challenges and the limitations to be tackled in future research. The most recent reviews on this topic ^{[7][8][9][10]}[7,8,9,10] revealed the promising impact of biofeedback on motor recovery. Linda van Gelder et al. [7] suggested future directions on biofeedback research to improve gait function, although they had considered a large variety of participant groups (healthy, runners, stroke/hemiplegia, Parkinson’s, incomplete spinal cord injuries, cerebral palsy, multiple sclerosis, amputees, diabetics, and knee injuries). Similarly, Thomas Bowman et al. [9] analyzed the BSs’ effects in heterogeneous participants, including patients with Parkinson’s disease, stroke, and mild cognitive impairment. It is important to note that the unique post-stroke sensorimotor deficits suggest that a review on the clinical effects of BSs’ merits should be pathology-specific rather than crossing different pathologies. However, the reviews ^[7][9][7,9] do not focus on post-stroke participants, who are the target users for several BSs developed for gait training ^{[7][8][10][11]}[7,8,10,11]. On the other hand, Rosalyn Stanton et al. [8] investigated the efficacy of biofeedback to improve performance in lower limb activities (namely, sitting, sit to stand, standing, and walking) after stroke compared to conventional therapy. This resviearchw was not specific for gait rehabilitation. Since gait recovery, due to its complexity, requires balance and coordinated activation of muscles, a review focusing on gait rehabilitation is valuable [12]. Jacob Spencer et al. [10] reviewed current evidence and future research directions related to post-stroke gait biofeedback, but they excluded studies involving biofeedback in adjunction with robotic assistive gait training (also not found in ^[8][9][8,9]). However, biofeedback in adjunction with assistive gait training can encourage patients’ active participation, preventing motor dependence on assistive devices [13].

Therefore, there is a need to conduct a scoping review and appoint future research directions on the design of BSs exclusively related to post-stroke gait rehabilitation, including the adjunctive use with assistive devices such as exoskeletons. Moreover, physiotherapist-oriented sensory cues provided according to biofeedback parameters should be innovatively studied once the physiotherapist’s involvement can assure effective use of the robotic devices and foster the patient’s motivation [14].

There was no sensor technology found that was common to most studies, as appointed in [9]. In of spite that, sensors that measure kinetic and spatiotemporal parameters were usually employed in the reviewed studies. Pressure sensors and force platforms evaluated kinetic biofeedback parameters, and force and load sensors additionally measured spatiotemporal biofeedback parameters. IMUs and motion capture systems determined kinematic and spatiotemporal biofeedback parameters. Cameras assessed the body’s movement. EMG and EEG systems measured muscular activation from the tibialis anterior and quadriceps femoris muscles and alpha-band EEG signals, respectively.

Even though most of the studies implied only one sensor on the BS, Byl et al. and Mottaz et al. ^[15][16][31,37] combined pressure sensors and MRI with IMUs and EEG, respectively. As reviewed in Byl et al. and Mottaz et al. ^[15][16][31,37], future research should study the effectiveness of combining multiple biomechanical and physiological sensors to personalize biofeedback for post-stroke users given their variable motor deficits. Further, the use of multimodal sensors may enable a more holistic BS-based gait training and assessment, attending to intra- and inter-subject motor variability ^[7][10][7,10].

Regarding sensors’ wearability, wearable sensors were used in most studies. Wearability allows unique assessments of body motion during ambulatory training in a non-fixed facility ^[10][17][10,47]. Thus, allowing the users to practice in multiple spaces, encouraging training dosing to increase in everyday scenarios and, consequently, accelerating recovery ^[7][18][7,48]. However, EMG and EEG sensors required a time-consuming preparation in opposition to IMUs, pressure, force, and load sensors that were fast positioned on the feet. This finding guides future research to select wearable sensors with fast positioning.

Concerning biofeedback mode, most studies reported visual biofeedback mode, as concluded in [9], using screens from monitors, televisions, tablets, or projectors. Auditory and haptic biofeedback modes were usually applied using speakers integrated into computers, televisions, or tablets, and vibrators, respectively. Even most studies have implied one mode, visual and auditory cues were combined once multimodal biofeedback can reduce the user’s cognitive load compared to a single-mode [7]. Therefore, there is space to explore auditory and haptic modes, as retained in ^[7][10][7,10], so that post-stroke users with multiple sensorial deficits can take advantage of multimodal biofeedback rehabilitation. In this manner, the physiotherapist has the necessary resources to personalize the training according to the patient’s imminent sensorial deficits.

Regarding actuators wearability, only haptic biofeedback was provided using wearable actuators in most studies, allowing ambulatory practice on daily-like scenarios as overground walking. Thus, future research should study the impact of using wearable actuators as earphones and augmented reality glasses for auditory and visual modes, respectively, benchmarking the results with non-wearable solutions. Haptic actuators were placed on both upper or lower limbs, fostering the conclusion that haptic feedback on the body, either at or away from the desired body segment to be changed, can improve motor performance ^[18][48].

Most biofeedback control strategies compared sensor data with a threshold or reference obtained from the user’s limbs, a baseline trial, maximum voluntary contraction, body weight, or data from healthy subjects, having a need to benchmark these methods. Regarding periodicity, most BSs for gait training have controlled the sensory cues at each gait cycle. Studies exploring EMG- and EEG-based biofeedback, performed during sitting, did not mention the control’s periodicity or attended a fixed-time control, respectively. Future studies should clearly state the control’s details (type, quality, and periodicity), as concluded in [7], fostering the research’s reproducibility. Less than half of the studies updated the threshold or reference and the control’s periodicity during training agreeing with the time for achieving the threshold or reference, intended as the user’s imminent disability level. In this sense, the rehabilitation was personalized according to the user’s imminent disability level, avoiding frustration and dependence on the sensory cues, respectively ^[16][19][37,49]. However, there is space to continue the investigation of biofeedback control strategies’ personalization according to the user’s imminent disability level, reporting significant evidence of this adaptation.

Positive reinforcement was employed in most biofeedback control strategies, as referred in the systematic review of Bowman et al. [9], drawing the user’s attention to a condition that should be learned and repeated. Even most studies with multiple biofeedback modes applied positive reinforcement on both. In this manner, there is space to explore if combining both positive and negaztive reinforcements, taking into account the cognitive effort, accelerates the relearning process by aware the patients of motor conditions that must be repeated and avoided, respectively.

Visual cues were usually active during training, which can lead to visual reliance and high cognitive effort ^[20][21][50,51]. They provided detailed biofeedback through graphs and scenarios modulated on shape, color, and size according to the control strategy. Auditory and haptic cues were usually inactive, being enabled according to the control strategy at a fixed intensity once intensity adaptation may not be perceived by post-stroke users due to their sensory deficits ^[22][52].

Non-wearable assistive devices, mostly treadmills, were used in adjunction with BSs. Treadmills encourage an intensive practice of walking at a controlled and stable gait speed [11]. Robotic assistive devices such as exoskeletons, robotic arms, FES systems, robotic platforms, or cable-driven robotic systems were also applied. They were usually controlled to provide 100% guidance assistance. These robotic systems intensively and repetitively assisted patients on motor tasks while BSs fostered active participation during training ^[23][22]. The future directions should continue exploring the contribution of BSs, as an alternative and complementary medicine approach, behind different closed-loop controlled wearable assistive devices.

Physiotherapists guide and instruct the patients according to their specialized knowledge towards recovery ^[24][53]. Additionally, physiotherapists’ intervention during training allows the safe and effective use of BSs and robotic assistive devices [6]. Although it was not stated in most studies, BSs could provide objective real-time information about the patients’ motor behaviour to the physiotherapists during the training, complementing visual inspection of the patients, as in ^{[15][23][25][26]}[22,25,31,42]. Moreover, BSs could fulfil the lack of physical contact between the physiotherapists and patients during training with robotic assistive devices ^[27][54]. Future research should explore the design of physiotherapist-oriented sensory cues according to the physiotherapist’s needs to enrich their contribution to rehabilitation.

Clinical studies related to the effects of BSs on post-stroke recovery had been carried out with a median sample size of 17 post-stroke participants, serving as a reference for future research. Post-stroke participants were selected and characterized mostly by age, gender, time post-stroke, hemiplegic side, stroke aetiology, and body mass, avoiding influencing the research evidence. Future clinical studies may include quantitative characterization of participants’ disability level before the intervention, increasing the reliability of their conclusions.

Literature includes randomized balanced controlled and uncontrolled studies as high-quality and proof of concept research designs, respectively, addressing a median training dose of three times a week for 4 weeks at 30 min/session. Familiarization was typically not stated. However, future research should appoint the existence and duration of this procedure once it can influence the research reproducibility. Moreover, further randomized controlled studies need to be conducted to find clinical evidence regarding the efficacy of biofeedback on post-stroke motor recovery; and potentiate benchmarking of biofeedback technologies and the standardization of their technical and clinical specifications ^[7][9][10][7,9,10].

Biofeedback training usually involved walking at a self-selected speed, but neurofeedback and biofeedback complemented with EMG-FES were trained during sitting and virtual reality games during standing. Evaluation time points occurred at pre-training and post-training in most studies, evaluating the effects of BSs on post-stroke recovery considering sensor-based and clinical outcomes. Retention procedures were also performed to evaluate motor learning immediately (in most studies) and follow-up post-training. Follow-up evaluations usually occurred 1 month post-training to assess the long-term benefits of the intervention.

In comparison with kinematic and physiological outcomes, spatiotemporal and kinetic ones were the most evaluated sensor-based outcomes. It was expected that most studies would apply kinetic or spatiotemporal biofeedback parameters. Step length and walking speed were highlighted between spatiotemporal outcomes. Clinical outcomes assessed motor effects in most of the studies using the 10 m walk test and TUG. Less than half of the studies also measured the post-stroke stage, mental and sensory effects through clinical outcomes. BSs exert direct action on cognitive and sensory functions once the patients are encouraged to self-control their motor behavior according to the coding scheme of the sensory inputs [13]. Therefore, future studies should fulfil this gap by conducting a user-specific holistic assessment, also including kinematic and physiological sensor-based outcomes, post-stroke stage, mental and sensory clinical effects.

Most studies achieved statistically significant improvements on at least one spatiotemporal, kinetic, kinematic, and physiological sensor-based outcomes, post-stroke stage, motor and mental clinical effects between at least two evaluation time points. Although most studies implied kinetic or spatiotemporal biofeedback parameters, they obtained promising results concerning kinematic and physiological effects and clinical-based outcomes. Moreover, most controlled studies exhibited higher statistically significant improvements on the experimental group than the control group concerning spatiotemporal, kinetic, and physiological sensor-based outcomes, motor and mental clinical effects. In this manner, these positive trends indicate the promise of the efficacy of biofeedback on rehabilitation, as concluded in ^{[7][8][9][10]}[7,8,9,10]. Future research should evaluate clinical sensory effects using statistical tests to power the research conclusions.