**Gait Quality Assessment in Survivors from Severe Traumatic Brain Injury: An Instrumented Approach Based on Inertial Sensors**

**Valeria Belluscio 1,2, Elena Bergamini 1, Marco Tramontano 1,2, Amaranta Orejel Bustos 1, Giulia Allevi 2, Rita Formisano 2, Giuseppe Vannozzi 1,\* and Maria Gabriella Buzzi <sup>2</sup>**


Received: 6 October 2019; Accepted: 28 November 2019; Published: 3 December 2019

**Abstract:** Despite existing evidence that gait disorders are a common consequence of severe traumatic brain injury (sTBI), the literature describing gait instability in sTBI survivors is scant. Thus, the present study aims at quantifying gait patterns in sTBI through wearable inertial sensors and investigating the association of sensor-based gait quality indices with the scores of commonly administered clinical scales. Twenty healthy adults (control group, CG) and 20 people who suffered from a sTBI were recruited. The Berg balance scale, community balance and mobility scale, and dynamic gait index (DGI) were administered to sTBI participants, who were further divided into two subgroups, severe and very severe, according to their score in the DGI. Participants performed the 10 m walk, the Figure-of-8 walk, and the Fukuda stepping tests, while wearing five inertial sensors. Significant differences were found among the three groups, discriminating not only between CG and sTBI, but also for walking ability levels. Several indices displayed a significant correlation with clinical scales scores, especially in the 10 m walking and Figure-of-8 walk tests. Results show that the use of wearable sensors allows the obtainment of quantitative information about a patient's gait disorders and discrimination between different levels of walking abilities, supporting the rehabilitative staff in designing tailored therapeutic interventions.

**Keywords:** wearables; inertial sensors; traumatic brain injury; dynamic balance; gait disorders; gait patterns; head injury; gait symmetry; gait smoothness; acceleration

#### **1. Introduction**

Head injuries are considered a major health problem as they are associated with high mortality and disability in young adults (<45 years of age) [1,2]. Nearly 70% of all brain-injury cases are males [3], and most events are caused by falls (28%), followed by motor vehicle accidents (20%) and blows (19%) [4]. Traumatic brain injuries (TBI) impress a significant burden on the health care system, due to the need for therapy to address physical, communicative, and psychological problems [5]. Costs are usually more elevated when the traumatic brain injury is considered severe [5]; that is with an initial Glasgow coma scale score (GCS) of 8 or less [6]. Neuropsychological and cognitive impairments, such as anxiety and depression, selective/sustained attention, language, and executive function deficits have been well documented in the literature [7–12]. Less attention has been placed on motor impairments, in striking contrast with available data on other neurological populations, such as

stroke and Parkinson's disease ones [13–18]. The available studies mainly focused on impaired balance and altered coordination. Specifically, Rinne and colleagues [19] described that well-recovered men with TBI had impaired balance and agility compared to healthy controls. A recent review performed by Williams and colleagues [20] evidenced that people with TBI walked more slowly than healthy controls, primarily due to reduced step length. A few authors emphasized the impact of post-traumatic parkinsonism or post-traumatic cerebellar syndrome [21,22], two conditions that interfere with walking and balance performances in persons surviving from TBI. Additionally, balance abnormalities have also been reported in terms of increased postural sway during quiet standing or functional tasks, with altered sensory inputs [23–25]. Additionally, gait analysis has been used in few studies: Chou and colleagues [26] showed that people who suffered from TBI usually present a gait pattern with a significantly slower speed and a shorter stride length, confirming previous results [27]. Basford and colleagues [28] reported that gait analysis, balance, and vestibular testing could document subtle biomechanical changes among participants with TBI, suggesting the appropriateness of gait and balance testing in this population, even when motor disorders are not clinically evident.

Taken together, evidence exists for persistent motor deficits after TBI. However, these studies have focused on mild (GCS > 13) and moderate (GCS between 9 and 13) TBI, and to the authors' knowledge, no quantitative information is available about motor ability in people who have incurred a severe TBI (sTBI). An objective characterization of their level of motor impairment could be an important step in the rehabilitation process of this population, helping in obtaining not only physical improvements, but also increasing the independence in daily life and the overall quality of life. This characterization, in order to be helpful and informative, should be ecological and as non-intrusive as possible.

In this framework, attention is growing on miniaturized and wearable instruments that quantify movement patterns in a non invasive way: inertial measurement units (IMUs), embedding accelerometers and gyroscopes, have been widely used in the last two decades since they present many advantages compared to the traditional gait analysis approach based on stereophotogrammetry and force platforms. From the data measured by these units, spatiotemporal gait parameters [29] and stability-related parameters [13,30] can be extracted, allowing fall risk to be assessed [31], and allowing one to differentiate gait patterns between healthy and pathological populations [13,32–34]. However, in the sTBI population, an instrumented approach with IMUs has never been proposed and no information is available about their capability to discriminate among different levels of walking ability, as defined by currently administered clinical scales, such as the dynamic gait index scale [35]. An integrated approach based on the "gold standard" clinical evaluation method which relies on clinical scales and the proposed sensor-based assessment would overcome the limitations of a subjective evaluation, depending on the operator's specific training, helping in revealing changes hardly detectable using clinical scales. In addition, this integration would allow to assess patients in ecological contexts, where they perform tasks more similarly to those of real life, providing objective motor ability characterization.

Given these premises, the aims of the present study were twofold: (i) to quantify gait patterns in sTBI population using a set of wearable inertial sensors; (ii) to investigate the association of the estimated gait quality indices with the level of walking ability and the scores of commonly administered clinical scales. Specifically, spatiotemporal parameters and gait quality indices (dynamic stability, symmetry, and smoothness) were investigated considering clinical performance tests commonly used in the routine assessment [36,37].

The hypothesis is that the instrumental approach could be a valid support to the traditional clinical evaluation in order to obtain quantitative and objective information about sTBI patients' motor impairments, discriminating between different levels of walking abilities, and helping clinicians with defining and evaluating the efficacy of personalized rehabilitation treatments, as previously reported in the literature [38]. Furthermore, the correlation analysis could help with simplifying and facilitating routine evaluation in terms of time-consuming administration of clinical scales, possibly allowing a reduction in the number of scales used, maintaining those necessary to characterize the investigated population/motor task.

#### **2. Materials and Methods**

The research was performed at the Santa Lucia Foundation and it was approved by the Local Independent Ethics Committee of Fondazione Santa Lucia IRCCS (Rome, Italy) (protocol number: CE/PROG.700).

#### *2.1. Participants*

Twenty healthy subjects (control group, CG) (age: 33.9 ± 9.5 years), 15 males and 5 females, and 20 people who suffered from a sTBI (age: 33.4 ± 10.5 years), 15 males and 5 females, were involved in the study. This sample size complied with the minimum number of participants recommended by a power analysis purposely performed (α = 0.05; power (1-β) = 0.95, effect size d: 0.7) for non parametric comparisons [39]. Exclusion criteria for CG were the presence of any orthopedic, neurological, or other co-morbidities which could have influenced the motor performance. Inclusion criteria for sTBI were: (i) age between 15 and 65 years; (ii) Glasgow coma scale (GCS) score ≤ 8 (used to objectively describe the severity of impaired consciousness at the time of injury) [6]; (iii) level of cognitive functioning (LCF) ≥ 7 [40]; (iv) presence of disturbances in static and dynamic balance; (v) ability to understand verbal commands. Almost all the patients selected suffered from a sTBI as a consequence of a traffic accident (19 out 20 participants), whereas one person suffered from a sTBI due to a fall.

#### *2.2. Procedures*

#### 2.2.1. Clinical Assessment

The following clinical scales were administered by an expert physiotherapist to all sTBI participants, to assess static and dynamic balance, ambulation skills, and mobility deficits:


To codify for different levels of walking ability, sTBI patients were further divided into two sub-groups, according to their score in the dynamic gait index clinical scale: persons with a score >19 were considered severe (10 people, sTBI-1), while those with a score ≤ 19 were considered very severe (10 people, sTBI-2), according to [35]. The demographic characteristics of each subgroup are reported in Table 1.



#### 2.2.2. Motor Assessment

Each participant was asked to perform three different motor tasks in a randomized order: the 10 m walk Test (10mWT), the figure-of-8 walk test (F8WT), and the Fukuda stepping test (FST). All tests were carried out in a fully dedicated quiet area at the Santa Lucia Foundation, where the surface was accurately kept flat, and participants were asked to stay barefoot and to stand upright for at least 5 s at the beginning and at the end of each trial. Tasks were carefully explained and demonstrated by an instructor before the testing. The instructor also gave the patients start and stop commands and stayed close to participants to prevent dizziness and/or falls. A detailed description of the motor tasks is reported below.

#### 10 m Walk Test (10mWT)

The 10mWT is a widely used and recommended test for measuring gait speed in different populations [44]. The experimental protocol of the assessment was selected according to previous studies [13,45]: it consists of walking on a straight 14 m long walkway for three repetitions at the participant's preferred walking pace, with the middle 10 m marked on the floor and considered as steady-state walking for further analysis. The time taken to walk the middle 10 m was measured using a stopwatch and walking speed was calculated by dividing the distance covered (i.e., 10 m) by the time taken.

#### Figure-of-8 Walk Test (F8WT)

The F8WT requires a person to walk a figure-of-8 shape, as illustrated in Figure 1, marked on the floor with tape, with each circle diameter of 1.66 m (5.44 ft) [46]. Participants were instructed: (*i*) to stand still with feet side-by-side in the start position facing the "8"; (*ii*) to begin walking at their preferred pace when ready; (*iii*) to stop when returning to the start position, placing feet side-by-side again. The test was performed three times for each F8WT direction (clockwise and counterclockwise), alternating the two directions, and the entire trial was considered for further investigations.

**Figure 1.** Figure-of-8 shape used for the figure-of-8 walk test (F8WT). Clockwise and counterclockwise directions are indicated with grey and black arrows, respectively.

#### Fukuda Stepping Test (FST)

The FST is a test used for the diagnosis of vertigo-associated disease [47] and an instrumented version of this test has been recently proposed in the literature [48] and was adopted in this work. Participants were instructed to stand upright blindfolded with both arms frontally outstretched, creating a 90◦ angle between the arms and the body. Then, they were asked to step on the spot for one minute and to remain still in the final position. Lateral and forward displacements, as well as the amount and side of rotation, were marked on the floor by a piece of tape and subsequently reported as clinical FST parameters. For what concerns the sensor-based parameters, the first and last three strides were discarded in order to evaluate only steady-state stepping.

#### *2.3. Equipment*

While performing the three above mentioned motor tasks, each participant was equipped with five synchronized inertial measurement units (IMUs) (128Hz, Opal, APDM, Portland, Oregon, USA): one located on the occipital cranium bone close to the lambdoid suture of the head (H), one on the center of the sternum (S), and one at L4/L5 level, slightly above the pelvis (P), and were used to assess the upper-body stability. The other two IMUs were located on both shanks, slightly above the lateral malleoli, and were used for step and stride segmentation. Each IMU was securely fixed to the participant's body with Velcro straps, except for the head IMU, which was inserted in a tailored pocket of a swim cap worn by each subject.

#### *2.4. Data Processing*

All data processing was performed using the Matlab software (The MathWorks Inc., Natick, MA, USA). Each unit embedded three-axial accelerometers and gyroscopes (±6 g with g <sup>=</sup> 9.81 m·s<sup>−</sup>2, and ±1500 ◦/s of full-range scale, respectively) and provided the quantities with respect to a unit-embedded system of reference. To guarantee a repeatable reference system for the three IMUs located on the upper body, each unit was aligned with the corresponding anatomical axes (antero-posterior: AP, medio-lateral: ML, and cranio-caudal: CC) following the procedure proposed by [49]. The following spatiotemporal parameters were obtained, through a peak detection algorithm, on the ML angular velocity signals measured by the two IMUs on the shanks: average stride duration (SD = time to complete the test/total number of strides) and average stride frequency (SF = total number of strides/time to complete the test). The following gait quality indices were estimated:


$$\begin{aligned} \text{ACPS}\_{\circ} &= \left(1 - \frac{\text{RMS}\_{j}S}{\text{RMS}\_{j}P}\right), \\ \text{ACPH}\_{\circ} &= \left(1 - \frac{\text{RMS}\_{j}H}{\text{RMS}\_{j}P}\right), \\ \text{ACSH}\_{\circ} &= \left(1 - \frac{\text{RMS}\_{j}H}{\text{RMS}\_{j}S}\right). \end{aligned}$$

Each coefficient represents the variation of the acceleration from lower to upper-body levels. A positive coefficient indicates an attenuation of the accelerations, while a negative coefficient indicates an amplification of the accelerations from the lower to the upper body level.

• Improved harmonic ratio (iHR), as proposed by [51], was calculated for each acceleration component (*j*) measured at the pelvis level. This index is based on a spectral analysis of the acceleration signals and is a measure of hemilateral symmetry when stepping (0% = total asymmetry; 100% = total symmetry). It was calculated as follows:

$$\text{iHR}\_{\text{l}} = \frac{\sum \text{Power of intrinsic harmonics}}{\sum \text{Power of intrinsic harmonics} + \sum \text{Power of extrinsic harmonics}} \cdot 100.1$$

• SPectral ARC length (SPARC), as proposed by [52], calculated for each acceleration component (*j*) measured at the pelvis level. The calculation of SPARC was performed as follows:

$$-\int\_{\mathbf{0}}^{\bar{\omega}\_{\xi}} [(\frac{1}{\bar{\omega}\_{\xi}})^2 + (\frac{d\mathcal{A}(\bar{\omega})}{d\bar{\omega}})^2]^{\frac{1}{2}} \, d\bar{\omega} ; \mathcal{A}(\bar{\omega}) = \frac{\mathcal{A}(\bar{\omega})}{\mathcal{A}(\mathbf{0})}$$

$$\bar{\omega}\_{\xi} = \operatorname{mtn} \{ \bar{\omega}\_{\xi}^{\max} \operatorname{mtn} \{ \bar{\omega} \vee \mathcal{A}(r) < \dot{\mathcal{A}}, \forall r > \bar{\omega} \}$$

where *A*(ω˜ ) is the Fourier magnitude spectrum of the acceleration signal a(t) and *A*(ω˜ ) is the normalized magnitude spectrum.

#### *2.5. Statistical Analysis*

Descriptive and inferential statistical analyses were performed using IBM SPSS Statistics software (v23, IBM Corp., Armonk, NY, USA), and the alpha level of significance was set at 0.05. The normal distribution of each parameter was verified using the Shapiro–Wilk test. As most of the parameters were not normally distributed, the following non-parametric tests were performed:


#### **3. Results**

#### *3.1. Clinical Scale Score Results*

The scores of the administered clinical scales for sTBI-1 and sTBI-2 are reported in Table 2. Results show that sTBI-2 group (defined as very severe TBI according to DGI scores; see methods) presented worse, statistically significant scores in the three clinical scales compared to sTBI-1.


**Table 2.** Clinical scales results for sTBI. Mean ± standard deviation values are displayed. Statistically significant differences are indicated with \*.

#### *3.2. Spatio-Temporal Parameters and Clinical FST Parameters*

Results of temporal (stride frequency and stride duration) and clinical FST parameters (lateral and forward displacements; amount and side of rotation) for the three groups are reported in Table 3. Statistically significant differences were present for all three motor tasks when comparing CG with sTBI-2 and sTBI-1 with sTBI-2. In addition, statistically significant differences between CG and sTBI-1 were found in the spatio-temporal parameters of the FST. Concerning clinical FST parameters, no statistical differences are displayed in terms of lateral and forward displacements, or amount and side of rotation among the three groups. Walking speeds (mean ± standard deviation) obtained during the 10mWT were: 1.48 ± 0.20, 1.10 ± 0.23, and 0.53 ± 0.20, for the CG, sTBI-1, and sTBI-2 groups, respectively. Significant differences were found between CG and both sTBI-1 and sTBI-2, as well as between sTBI-1 and sTBI-2.

**Table 3.** Temporal and FST parameters. \* indicates statistically significant differences between CG and sTBI-2 (*p* < 0.001); § indicates statistically significant differences between sTBI-1 and sTBI-2 (*p* < 0.05); # indicates statistically significant differences between CG and sTBI-1 (*p* < 0.001). Clinical FST parameters: the values of the antero-posterior (AP) and medio-lateral (ML) displacements, the amount of rotation and the side of rotation of the three groups of subjects (CG, sTBI-1, sTBI-2) in the three tasks are reported (mean ± standard deviation).


#### *3.3. Root Mean Square, Attenuation Coe*ffi*cients, Improved Harmonic Ratio, and SPARC*

Significant differences were found for the three motor tasks (10mWT, F8WT, and FST) when comparing both sTBI against CG and sTBI-1 against sTBI-2. Results regarding the 10mWT, the F8WT, and the FST are reported in Figure 2a–c, respectively.

**Figure 2.** *Cont*.

**Figure 2.** Normalized root mean square (nRMS) values, attenuation coefficients (AC), improved harmonic ratio (iHR), and SPectral ARC length (SPARC) for the sTBI sub-groups and for CG in 10mWT (**a**), F8WT (**b**), and FST (**c**). Medians and interquartile ranges are reported. AP, antero-posterior; ML, medio-lateral; CC, cranio-caudal; P, pelvis; S, sternum; H, head. The horizontal lines indicate statistically significant between-groups differences. (**a**) 10 m walk test. (**b**) Figure-of-8 walk test. (**c**) Fukuda stepping test.

#### *3.4. Association of the Gait Quality Indices with the Clinical Scale Scores*

Correlation analysis (Table 4) shows that several indices displayed a significant correlation with the clinical scales scores in the three motor tasks, especially in the 10mWT and F8WT.


**Table 4.** Spearman's correlation coefficients (*p*) between each estimated parameter and each clinical scale. Statistical significance is indicated by asterisks (\* *p* < 0.05; \*\* *p* < 0.001). Abbreviations: BBS, Berg balance scale; DGI, dynamic gait index; CB&M, community balance and mobility scale, RMS, root

#### **4. Discussion**

The aims of this study were to quantify gait quality of a sTBI population with different levels of walking ability using a set of wearable inertial sensors and to investigate the association of the estimated gait quality indices with the scores of commonly administered clinical scales. Results show that the instrumented approach allows (*i*) obtainment of quantitative and objective information about patient's motor impairments; (*ii*) discrimination between different levels of walking abilities; (*iii*) exploration of the relationship between the estimated gait quality indices and the clinical scale scores. As expected, clinical scale scores displayed a consistent increasing trend from low to high walking ability levels, showing statistically significant differences between severe (sTBI-1) and very severe (sTBI-2) TBI participants (Table 2). A similar trend was observed when considering the spatio-temporal parameters: for what concerns walking speed, statistically significant differences were found between the control group (CG) and sTBI-2, and between sTBI-1 and sTBI-2 (Table 3). These results are consistent with the existing literature about healthy people [13,53] and TBI participants [26], and confirm the relevance of walking speed as an informative and concise parameter to discriminate between different level of walking ability.

In addition, the values of stride frequency and stride duration obtained in this study are consistent with previously reported results. In particular, in persons with sTBI, a reduced stride frequency, along with an increased stride duration, may be related to post-traumatic Parkinsonism [22,54,55]. Furthermore, as suggested in [56], it can be speculated that people with sTBI increase their stride duration in order to compensate for gait instability and counteract the fear of falling. This significant gait impairment was still observed despite the provision of optimal medication therapy, confirming the very close relationship between altered gait and postural instability in this population [57,58].

Interesting results come from the estimated gait quality indices: almost all parameters in the three motor tasks were able to discriminate between CG and both sTBI groups, especially the sTBI-2, as expected. The actual added value of the proposed approach, however, lies in its ability to detect possible differences between sTBI-1 and sTBI-2, facilitating discriminating between different levels of walking ability. In this respect, in the 10mWT, the two sTBI sub-groups presented differences in gait stability and symmetry. Specifically, considering gait stability, sTBI-2 showed higher nRMS compared to sTBI-1. High nRMS values have been associated with a higher amount of acceleration, and hence, decreased stability [13,29,30,33,34,50]. Both sTBI subgroups, and especially sTBI-2, displayed a decreased stability at the three upper body levels, particularly in the ML direction. This is consistent with previous studies dealing with other neurological populations [13,15,59]. In addition, the attenuation coefficient from pelvis to head in the ML direction discriminates between the two sTBI sub-groups, highlighting that the sTBI-2 sub-group exhibits a limited bottom-up attenuation of upper body accelerations. This result is related to a lack of ability to stabilize the head, impairing the consequent planning of adaptive motor strategies. Concerning gait symmetry, the iHR in the CC component discriminated between sTBI-1 and sTBI-2, showing a reduced gait symmetry, particularly evident in the sTBI-2. Reduced symmetry has been widely associated with an increased fall risk [60,61], thus indicating this parameter as a biomarker for the identification of patients at high risk of falling.

In the F8WT, the nRMS and iHR discriminated between sTBI-1 and sTBI-2, as reported for the 10mWT. In addition, differences were also pointed out considering the smoothness: in fact, the SPARC discriminated sTBI-1 and sTBI-2 well, probably because of greater upper body rigidity in sTBI-2 than sTBI-1 observed in this more difficult task. It is worth mentioning that, being characterized by a curved trajectory, the execution of the F8WT involves the activation of different cortical areas than those required in the planning of straight point-to-point movements. In fact, it is well known that the trajectory planning during curved-path conditions requires additional preparation time [62,63]. The results of the present study show indeed that the F8WT seems to be the walking test that better discriminates among different walking ability levels. This suggests that testing dynamic balance abilities during curved trajectories could be useful for assessing gait in conditions more relevant to cognitive-motor dual tasks, and thus, closer to daily living activities [64,65].

For what concerns the FST, results about clinical FST parameters confirm the previous literature [48,66]: no differences were found in terms of AP–ML displacements, nor side and degree of rotation among groups. The presence of rotation in either direction in the CG shows that turning while stepping on the spot also occurs in healthy people, confirming that clinical FST parameters are not able to distinguish between healthy and pathological subjects, confirming the doubts about its clinical use also for this population. Conversely, when considering gait quality indices obtained from the instrumented FST, the discrimination capability of the test greatly increased: in fact, significant differences were found, not only between CG and both sTBI sub-groups, but also between the sTBI sub-groups. Specifically, for what concerns stability, the nRMS did not discriminate between different levels of walking abilities in pathological subjects, as observed in the 10mWT and the F8WT. On the other hand, attenuation coefficients in the AP and CC directions distinguished between sTBI sub-groups, with sTBI-2 showing less ability in attenuating upper body accelerations from lower to higher levels. Additionally, a reduced symmetry was displayed by sTBI-2 with respect to sTBI-1, with the AP component of the iHR displaying a significant difference between the two sub-groups. The AP direction seems to be the most critical in very severe TBI and this result is in agreement with the existing literature about stroke patients [48], indicating the AP component as the most informative when comparing patients with different walking abilities. In addition, the absence of the visual input during the FST plays an important role on the sensory reweighting, which has been acknowledged as critical in the TBI population [67]. Therefore, these results confirm that the instrumented approach in this test provides valuable information about patients' motor strategies and useful data to tailor rehabilitation protocols [48].

When considering the second aim of the study, several correlations were found between clinical scales and gait quality indices, especially in the 10mWT and the F8WT. nRMS and attenuation coefficients, parameters related to dynamic stability, correlated well with all clinical scales, while worse correlations were present when considering the iHR and the SPARC in both the 10mWT and the F8WT, with no correlations at all for these two parameters in the FST. It should be acknowledged that the

proposed clinical scales do not consider tests in which the visual input is removed: this could be one of the reasons why only few correlations have been found when considering the FST. These results highlight the lack of specificity that some clinical scales exhibit [68], while confirming their ability to determine whether or not a patient has a motor impairments. Therefore, the integration of traditional scales and technology-based protocols could assist with improving current clinical routines and with designing rehabilitation treatments, helping to bringing more sensitive, specific, and responsive motor tasks to clinical practice.

Despite the promising results, this study presents some limitations: the main limitation is the heterogeneity of the sample, mainly due to the severities and the locations of the brain injuries. Increasing the sample would likely lead to reduce the heterogeneity of the sample. Furthermore, the relationship between gait characteristics and specific neurological deficits, such as post-traumatic parkinsonism or cerebellar syndromes, and the presence of possible cognitive and behavioral sequelae of sTBI, were not investigated. Although such analyses were beyond the scope of the present study, they could be considered in further studies, in order to obtain more detailed information and better discriminate among people suffering from sTBI.

#### **5. Conclusions**

People who suffer a sTBI often complain of balance and gait impairments, but despite the evidence that neuromotor deficits are a common consequence of a sTBI, the existing literature does not adequately describe balance strategies adopted by sTBI survivors. This lack of information depends on various factors: the heterogeneity and severity of the brain damage, the patient's age, and the presence of pre-morbid/co-morbid conditions are the most significant. Furthermore, subtle cognitive functioning deficits, such as executive functions, which are detectable even in persons with good recovery after sTBI [7,63], may interfere with dynamic performances.

The main contribution of the present work is represented by the analysis of gait stability, symmetry, and smoothness indices which objectively describe gait quality in patients with sTBI. Specifically, the lack of ability of both severe and very severe TBI patients to stabilize their head by attenuating body accelerations may have a big impact. In fact, the vestibular system is located at head level; therefore, a high head acceleration could be critical for the planning of adaptive motor strategies.

The data reported herein suggest the appropriateness of an integrated assessment using both clinical scales and wearable sensors to objectively evaluate gait and balance impairments during different dynamic tasks. This integrated approach may be useful to assessing the measures of changes during rehabilitation training aimed at improving patients' gait quality and limiting the risk of falling, supporting rehabilitative staff with designing effective and tailored interventions.

**Author Contributions:** Conceptualization, R.F., G.V., M.G.B., and M.T.; methodology, V.B., E.B., G.V., and G.A.; software, E.B.; formal analysis, V.B. and E.B.; investigation, V.B. and A.O.B.; resources, G.V., M.T., M.G.B., and R.F.; data curation, V.B. and E.B.; writing—original draft preparation, V.B.; writing—review and editing, all authors; visualization, V.B.; supervision, G.V., M.T., and M.G.B.; project administration, G.V. and R.F.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

### **Wearable Electronics Assess the E**ff**ectiveness of Transcranial Direct Current Stimulation on Balance and Gait in Parkinson's Disease Patients**

#### **Mariachiara Ricci 1, Giulia Di Lazzaro 2, Antonio Pisani 2, Simona Scalise 2, Mohammad Alwardat 2, Chiara Salimei 2, Franco Giannini <sup>1</sup> and Giovanni Saggio 1,\***


Received: 25 October 2019; Accepted: 8 December 2019; Published: 11 December 2019

**Abstract:** Currently, clinical evaluation represents the primary outcome measure in Parkinson's disease (PD). However, clinical evaluation may underscore some subtle motor impairments, hidden from the visual inspection of examiners. Technology-based objective measures are more frequently utilized to assess motor performance and objectively measure motor dysfunction. Gait and balance impairments, frequent complications in later disease stages, are poorly responsive to classic dopamine-replacement therapy. Although recent findings suggest that transcranial direct current stimulation (tDCS) can have a role in improving motor skills, there is scarce evidence for this, especially considering the difficulty to objectively assess motor function. Therefore, we used wearable electronics to measure motor abilities, and further evaluated the gait and balance features of 10 PD patients, before and (three days and one month) after the tDCS. To assess patients' abilities, we adopted six motor tasks, obtaining 72 meaningful motor features. According to the obtained results, wearable electronics demonstrated to be a valuable tool to measure the treatment response. Meanwhile the improvements from tDCS on gait and balance abilities of PD patients demonstrated to be generally partial and selective.

**Keywords:** balance; gait; Parkinson's disease; transcranial direct current stimulation; wearable electronics; IMUs

#### **1. Introduction**

Wearable electronics are gaining increasing attention and importance as a valid tool for healthcare practitioners in medical treatment [1–3] and patient monitoring [4–6]. In particular, wearable sensors have been applied for assessing the motor performance of patients with neurodegenerative disorders, as it is for Parkinson's disease, in both home and clinical environments [7–12].

Parkinson's disease (PD) can be characterized by motor deficiencies, such as bradykinesia and a combination of rest tremor, rigidity, as well as gait and balance impairment [13]. In routine clinical care, the evaluation of those deficiencies is mainly based on severity-rating standardized scales, such as the Movement Disorder Society Unified Parkinson's disease rating scale (MDS-UPDRS) [14], based on patients' reports and clinicians' vision-based evaluations, and clinical investigators determine the effectiveness of a therapy of a drug by using the MDS-UPDRS score [15]. Inconveniently, patient reports can be affected by mood and unfamiliarity with forms, and clinicians' evaluations can be

biased by personal beliefs, experiences, and a priori expectations, resulting in inter- and intra-rater score variability [15,16]. Furthermore, the MDS-UPDRS is quantified according to a discrete scale (0–4, unity step) only, and the human eyes of clinicians hardly detect subtle motor changes during the monitoring of patients. These limitations compel investigators to employ more rigorous, and thus costly, clinical trial designs, with a random assignment of patients, thus blinding investigators to treatment assignment.

The aforementioned limitations can be in some way reduced or overcome through the use of wearable inertial sensors (hereafter wearables), which provide measures of human postures and kinematics, paving the way for objective assessment in clinical trials [17]. In fact, wearables can gather motion parameters in a continuous (analog) or high-step density (digital) scale, and avoid intra- and inter-rater variability, thereby reducing the sample size and simplifying the assessment of the patients, objectively quantifying a possible beneficial effect of a therapeutic intervention. For this reason, even if wearables are still poorly used (only 2.7% of ongoing clinical trials [15]), there is growing attention given to this technological tool, and some pharmaceutical companies are working to develop their own devices [18–20].

Our work approaches the utilization of wearables in the particular case of objectively demonstrating the therapeutic beneficial effects, if any, of transcranial direct current stimulation (tDCS) treatment on the motor impairments of patients affected by Parkinson's disease.

The proven appeal of tDCS is evident as it is a non-invasive, inexpensive, painless brain stimulation technique with many clinical and research applications, ranging from the treatment of depression to neurorehabilitation [21,22]. It consists of applying a direct positive (anodal) or negative (cathodal) 1–2 mA current to the scalp. This stimulation supports the depolarization or hyperpolarization of neurons, thus leading them closer to, or farther away from firing, acting on synaptic transmission or synaptic plasticity [21,23]. Further, tDCS has been used alternatively to (or sometimes concurrently with) dopaminergic drug therapy, because the latter can lose its efficacy during the natural course of the disease, in particular regarding its benefit on postural and gait disorders. Gait is now considered a higher level of cognitive function that involves the integration of attention, planning, memory and other motor, perceptual and cognitive processes. In fact, walking and balance constitute a combination of automatic movement processes, afferent information processing, and intentional adjustments that require a delicate balance between various interacting neuronal systems. In PD, to compensate the loss of motor task, cognitive resources as attention and executive function performed by the dorsolateral pre-frontal cortex (DLPFC) plays a critical role in the relief of gait disorder [24]. In addition, previous studies have shown that anodal tDCS stimulation to either the motor area (M1) or dorsolateral prefrontal cortex (DLPFC) had a significant impact on the motor, non-motor, and balance functional outcomes in PD patients. In fact, brain activation patterns in M1 and DLPFC are extremely involved in successful locomotion performance in patients with PD [21,25–27]. Further, the effectiveness of tDCS for alleviating gait and postural instability seems promising [28–31], however, evidence of its benefit remains unclear and controversial [23,32] because different tDCS protocols and target areas of scalp have been considered, leading to conflicting evidence on MDS-UPDRS scores [23,28].

Our work aims to objectively quantify the motor performance improvements, if any, due to tDCS treatment in a population of patients with PD and gait disturbances. To this aim, we used wearables to measure specific motor tasks, and analyzed the related results by means of the standardized response mean (SRM) index, comparing them with those obtained by the clinical evaluation.

#### **2. Materials and Methods**

#### *2.1. Subjects*

Ten PD patients (Table 1) with postural and gait disturbances were recruited at Tor Vergata University Hospital, Rome, Italy. Idiopathic PD was diagnosed according to the MDS clinical diagnostic criteria for PD [13], and patients were enrolled at Hoehn & Yahr disease stages between 1.5 and 4, and with MDS-UPDRS III scores related to a gait higher than 1. Exclusion criteria were age (younger than 30 or older than 85), dementia (mini mental status evaluation, MMSE, score < 24 [33]), therapy changes in the last three months, orthopedic comorbidities, other neurological disorders, and therapy with drugs possibly interfering with motor function (e.g., antipsychotics).


**Table 1.** Patients' information.

This study was conducted in agreement with the ethical principles of the Helsinki declaration. Informed consent was obtained from each participant and ethical approval was obtained by the local committee (RS 190/18). Patients consented to participate and did not change the therapy during the study, from T0 to T2 (Figure 2), in order to minimize any alteration of motor performance due to dopaminergic therapy variations.

#### *2.2. Motor Tests*

We requested each participant to perform six motor tasks which, according to clinical standards, are relevant for a comprehensive evaluation of balance and gait. Tasks included stance feet together (SFT), tandem stance (TS), the pull test (PT), timed up and go test (TUG), stop and go test (S&G), and narrow walking test (NW). In particular, SFT and TS are useful to test balance; PT corresponds to the item 3.12 of MDS-UPDRS III to test postural response; TUG, S&G and NW are used to assess mobility and gait. Wearables were placed by means of Velcro strips on segments of the body, according to the particular test, as schematized in Figure 1. The descriptions of the tests and corresponding placements of the wearable sensors are specified in the following.

#### 2.2.1. Stance Feet Together (SFT) and Tandem Stance (TS)

In SFT and TS tests, the patient has to stand and maintain the posture for 30 s. More particularly, in the SFT with feet side-by-side and close together, in TS with feet in tandem position (i.e., one ahead, aligned and close to the other). The wearables were placed on the posterior trunk at the level of T5 and on the external parts of the calf segments of both legs.

#### 2.2.2. Pull Test (PT)

The subject, comfortably standing upright with shoulders to the examiner, is rapidly and vigorously pushed backward on his/her shoulders so as to be forced to make one, or more, steps backwards, recovering his/her balance. The sensors were placed as for SFT and TS.

#### 2.2.3. Timed Up and Go (TUG)

The subject starts seated on a straight-backed chair with arms across the chest, then gets up, walks straight 6 m, turns around, walks straight back and, turning on his/her-self, sits down returning to the initial condition. The sensors were placed on the patient's pelvis at the level of L5, posterior trunk at the level of T5, on the external parts of thighs and calf segments of both lower limbs, arms, and forearms.

#### 2.2.4. Stop and Go (S&G)

The subject walks for six meters in a straight line, turns around, walks six meters back while the examiner tells him/her to stop and go for 6 times. The sensors were placed on the patient's pelvis at L5 level, posterior trunk at T5 level, on the external parts of thighs and calf segments of both lower limbs. The time, when the examiner tells the patient to stop was recorded.

#### 2.2.5. Narrow Walking (NW)

The subject walks 6 m straight, but passing through a 70 cm narrow door in the middle of the path. The sensors were placed on the patient's pelvis at L5 level, posterior trunk at T5 level, on the external parts of thighs, and calf segments of both lower limbs. The time, the time when the patient passes through the door was recorded.

**Figure 1.** Sensors, labeled from S1 to S10, as located on the body of the patients. Different motor tests resulted with a different number of used sensors.

#### *2.3. tDCS Stimulation*

Direct current (DC) was delivered to stimulate the left dorsolateral-prefrontal cortex (DLPFC) by means of a tDCS low-intensity stimulator (BrainStim, EMS Srl, Bologna, Italy). Two saline-soaked electrodes (35 cm2) were placed on F4 (according to the 10–20 international EEG nomenclature) and on the right forearm, respectively. The stimulation was of 2mA DC (0.057 mA/cm<sup>2</sup> in density) delivered for 20 min (30 s step-up ramp, 30 s step-down ramp), repeated ten times, obtaining one session/day, for five consecutive days. Such a stimulation session was followed by two non-stimulation days, and again by another five days of long stimulation (Figure 2). During each tDCS application, patients were at rest without any concurrent motor tasks.

**Figure 2.** Flow diagram showing the study design and stimulation protocol.

#### *2.4. Wearable Electronics*

Different technologies can furnish data in terms of gait and balance performances. We can refer, for instance, to pressure sensors embedded into the floor and electro-goniometers, etc., with the optical-based systems considered as the gold standard because of their high accuracy. However, optical-based systems have some important drawbacks, such as the necessities of a free line of sight, time-consuming calibration procedures, necessity of skilled personnel and, above all, a very high cost. Wearable electronics have none of those drawbacks, and have been demonstrated to perform with the appropriate accuracy for our purposes [34,35].

Wearable electronics constitute a network of validated inertial measurement units (IMUs) termed Movit (by Captiks Srl, Rome Italy) [7,34,35], each housing a 3-axis accelerometer (±8 g) and a 3-axis gyroscope (±2000◦/s), synchronized to a personal computer receiver, with a 50 Hz data transfer rate. A proprietary application, termed Motion Studio, processes and stores data.

The number of used IMUs and the position of patients' bodies (by means of elastic bands) varied according to the particular motor tasks performed. Measured data consist of accelerations, angular velocities, and joint angles, computed from the related quaternions via Euler decomposition. In turn, the quaternions are generated using a Kalman filter on data coming from the accelerometers and the gyroscopes, sampled at 200 Hz. By means of a patented calibration procedure, the spatial orientations of the dressed IMUs are represented on a computer screen as a human avatar, which replicates patient movements, with his/her joint angles gathered with a forward kinematic procedure in a parent-child hierarchy.

#### *2.5. Features*

For each task, we obtained several features, as reported in Table 2 and described in the following paragraphs.

#### 2.5.1. Stance Feet Together (SFT) and Tandem Stance (TS)

Eleven features from the sensor located on the trunk were taken into consideration: range of accelerations, angular velocities and angles of the trunk in the medial-lateral (ML), anterior-posterior (AP) and vertical (V) directions; Jerk and Sway Area. In particular, Jerk, gathered from the accelerometers, represents the time derivative of acceleration [36], and is used as an empirical measure of the smoothness of the movements [37,38]. The Sway Area is the area of the ellipse that encompasses 95% of the values of medial lateral and anterior posterior accelerations around their mean values.

#### 2.5.2. Pull Test (PT)

The PT test is useful to evaluate the postural responses to an unexpected external perturbation. We extracted the 11 features as for the SFT, plus the number of steps following the pushing as resulted from data gathered by the sensors placed on the ankles.

#### 2.5.3. Time Up and Go (TUG)

TUG is one of the most widely used clinical tests and allows for the assessment of several aspects of gait. Parkinsonian gait is characterized by a slowed speed, decreased arm swing, shuffling steps, and difficulty to turn [39]. TUG is composed by four phases: the sit-to-stand phase (patient gets up from the sitting position with arms across the chest), the walking phase (patient walks for 6 m forth and back), the turning phase (the patient turns 180◦), and the turn-to-sit phase (the patient turns and sit back on the chair). Each phase is segmented considering data gathered by the IMU on the trunk. We detected the sit-to-stand and turn-to-sit phases considering the interval between the two local minimum values before and after a local maximum of the accelerometer data, in the AP direction, corresponding to the flexion/extension movement of trunk. The turning phase is identified using thresholds on the trunk angle in the vertical direction (the turning component looks as a positive or negative ramp, depending on the direction of the turn). Further details on the segmentation of TUG test are reported in [7].

From these segmentations, 24 features were computer, as described in Table 2, including:


#### 2.5.4. Stop and Go (S&G) & Narrow Walking (NW)

Parkinsonian gait problems are often triggered by some circumstances such as spaces with a narrow passage (e.g., a door), unexpected visual or auditory stimuli, stressful situations, cognitive load anxiety and difficulty in starting and stopping [39]. The results are a decreasing step length and step time, decreasing velocity, and increasing variability of step length and time [40,41]. The S&G and NW tests are used to provide evidence for these symptoms. We computed seven features for each task.

For the S&G test, we computed the duration of steps, stance and swing, as well as the angular velocity of the leg of the first steps at the beginning of gait, thus, after each stop signal of the examiner and the variability of the temporal step variables (CV Step, CV Stance, CV Swing).

For the NW test, we computed the same features but extracted them during the 3 s when the patient was passing through the door.


#### **Table 2.** Extracted Features from each motor test.

#### *2.6. Clinical and Wearables-Based Evaluations*

Motor test performances of each of the ten PD patients just before the stimulation protocol (T0 time), just soon after the protocol (T1 time), and 1 month after (T2 time) were evaluated in order to quantify the effect of the tDCS and its persistence, if any.

The evaluations were performed both as standard clinical ones and by the analysis of data gathered through the wearable electronics.

All patients were evaluated by a movement disorder specialist, with general neurological examination, clinical tests, and questionnaires. Clinical tests consisted in the administration of MDS unified Parkinson's disease rating scale (MDS-UPDRS) and the Berg balance scale (BBS) [43], a clinical five-point ordinal scale that assess balance. Each patient was also evaluated with the freezing of gait questionnaire (FOG-Q) [44], a 6-item questionnaire used to assess gait disturbance severity in patients with PD, and the Hoehn and Yahr scale (H&Y) [45], a commonly used system for describing the progress of symptoms.

To evaluate the responsiveness of a treatment, we considered two aspects. First, we assessed the ability of wearable features to detect change over a particular time frame. Then, we evaluated the relationship between a change in the feature values and the external measure (e.g., the clinical score).

The standardized response mean (SRM) [46] was used to assess the responsiveness to the tDCS therapy. A reason for choosing SRM is because, differently from the paired t-test, it has no dependence on sample size [47]. The SRM expresses the ratio of TT:SDC, where TT is the mean change between T1 and T0 and between T2 and T1, and SDC the standard deviation of the change. Empirically, an SRM value of 0.20 represents a small, 0.50 a moderate, and 0.80 a large responsiveness, respectively.

We used Spearman's rank correlation coefficient to investigate the relation between the clinical scores and the features. Stance feet together (SFT) and tandem stance (TS) tasks were used to evaluate the static balance, assessed by the clinicians using the BBS scale. Features extracted from SFT and TS are compared with the BBS score. PT features were correlated to the corresponding UPDRS III item 3.12 score (PT is part of UPDRS III tasks). Features extracted from gait related tasks (TUG; ST and NW) were correlated with the UPDRS III gait item score (3.10). The significance level was set at 0.05.

#### **3. Results**

Table 3 shows the mean, standard deviation values, and SRM of the clinical evaluation results. Tables 4–9 report the motor features of SFT, TS, PT, TUG, S&G and NW tests, and correlation analysis between the features and the corresponding clinical evaluation.


**Table 3.** Clinical evaluation.

**Table 4.** Stance feet together (SFT): feature values at T0, T1, T2; values of SRM comparing times; correlation with BBS score.


\* *p* value < 0.05.

**Table 5.** Tandem stance (TS): features values at T0, T1, T2; values of SRM comparing times; correlation with BBS score.


\* *p* value < 0.05.


**Table 6.** Pull test (PT): feature values at T0, T1, T2; values of SRM comparing times; correlation with UPDRS item 3.12 (PT) score.

\* *p*-value < 0.05.

**Table 7.** TUG: feature values at T0, T1, T2; values of SRM comparing times; correlation with UPDRS item 3.10 (Gait) score.


\* *p*-value < 0.05.


**Table 8.** Stop and go (S&G): feature values at T0, T1, T2; values of SRM comparing times; correlation with UPDRS item 3.10 (gait) score.

\* *p*-value < 0.05.

**Table 9.** Narrow walking (NW): feature values at T0, T1, T2; values of SRM comparing times; correlation with UPDRS item 3.10 (gait) score.


\* *p*-value < 0.05.

#### *3.1. Clinical Evaluation*

MDS-UPDRS sections two and three, BBS, and FOG-Q (Table 3) demonstrated moderate responsiveness to tDCS at the end of the treatment. The effect appears stable after one month with some improvement in BBS and MDS-UPDRS Section 2 score.

#### *3.2. Stance Feet Together (SFT) and Tandem Stance (TS)*

Jerk demonstrated a decrement, but only in a small percentage, in SFT (Table 4) and TS (Table 5) in both T1 and T2. During TS, Sway Area, range of the accelerations and angular velocities in the three directions decreased in T1 with a responsiveness around 0.4. The effect is stable at T2 compared to T1 with low improvements in some features.

The BBS score correlates significantly with almost all the features extracted from SFT and TS such as Jerk, Sway area (only TS, r = −0.37) and range of the accelerations and angular velocities. So, features highly reflect the clinical evaluation in this case.

#### *3.3. Pull Test (PT)*

During the PT, the obtained results (Table 6) showed an unchanged number of steps after tDCS treatment, a small increment of Jerk, and a small reduction of Sway Area at the end of the treatment and one month after.

Regarding the clinical evaluation, only few features (Range Acc V, r = −0.45; Range Gyr ML, r = −0.47) correlated with the UPDRS PT sub score.

#### *3.4. Time Up and Go (TUG)*

It was found that tDCS showed a moderate effect on the duration of sit-to-stand and walking phase in T1 and T2, as compared to the baseline (Table 7). A lower duration of the Turning phase is present only at T2. In correlation with a lower duration of the walking phase, our results show a reduction of the number of steps and stance duration. No changes were found in features related to the upper limbs. Conversely, the velocity of the lower extremities meaningfully increased. Finally, patients increased the velocity to turn and sit at T1 and T2, with comparison to the baseline values.

The UPDRS gait item score correlates significantly with several features extracted from TUG. Significant correlations regard the features representing the duration of the TUG phases (namely tug time, walk time and turning time). So, patients that take time to complete TUG have higher score on gait item. Weak correlation was for the temporal gait characteristics with the exception of number of steps and CV step. Gait item correlates significantly with features related to lower limb movements (Flex Leg, Average Vel Thigh, and Average Vel Leg) and the turning phase (Turning Vel, Steps Turning).

#### *3.5. Stop and Go (S&G) & Narrow Walking (NW)*

Both S&G (Table 8) and NW (Table 9) tests show a shorter duration of the step and swing phase and decreased variability of step duration in both T1 and T2 with respect to the baseline. The velocity remained unchanged in S&G but increased in NW. Large responsiveness is found in NW related to step duration, swing duration, velocity, and all the temporal step variability features.

One feature from S&G (step duration, r = −0.42) and two features from NW (Step Velocity, r = −0.44; CV Swing, r = 0.35) are significantly related to the UPDRS gait item.

#### **4. Discussion**

The response to dopaminergic drug replacement therapy in PD may lose its effectiveness during the course of the disease. Postural and gait disturbances, in particular, are symptoms that are difficult to treat with currently available pharmacological therapies.

Recent studies suggest a potential positive impact of tDCS on gait and balance in PD patients, symptoms of the late stage of PD, poorly responding to the classic dopaminergic treatment.

Our work focused on objectively quantifying the effect of tDCS on gait and postural stability from measured data gathered by wearable electronics used during motor tests of Parkinson's disease patients.

Within this context, the obtained results demonstrate the impact of wearable electronics with respect to standard clinical evaluation, allowing for interesting insights on the range of change on motor performance following the therapy. In fact, wearable electronics can evidence key elements of postural instability or gait abnormalities, both for evaluating the progression in PD and even to identify the disease at early stages [7,48–50]. Accordingly, in this study, specific motor tests were considered to assess the effects of tDCS therapy on balance and gait disturbances, taking into account the effects on measured motor features, soon after the delivery and one month later.

For balance assessment, three different motor tests were adopted to evaluate the equilibrium in three different conditions: SFT for static balance, TS to assess the balance when a low perturbation is introduced, and PT to assess postural responses to an unexpected perturbation. According to the kinematic assessment, Jerk is the only feature that presents a significant variation in SFT, TS and PT, suggesting that it is a highly sensitive measure of balance. This confirms the finding reported in previous studies, wherein Jerk was suggested as a valid biomarker of PD [7,49].

For gait assessment, the TUG test was useful to evaluate the slower speed, decreased arm swing, shuffling steps and difficulty to turn. Further S&G and NW tests were useful to evaluate step time, velocity, and variability of steps, due to the difficulty to start/stop and pass through a narrow door.

Our results show a reduction of step and stance duration and an increment of lower limb velocity during TUG, S&G and NW tests. These achievements confirm the findings reported in other works, which evidenced some improvement of hypokinetic gait in PD after tDCS treatment [29,30,51]. The effect is more evident in NW test, where we observed a large responsiveness to tDCS. The reason why PD patients tend to decrease step time and velocity when approaching a narrowed space is not completely understood [39], however tDCS in some way improves this aspect. We evidenced an improvement of gait in turning and standing tasks during TUG test too, when patients increased the velocity to turn and sit after the stimulation protocol. In particular, changes in turning are one of the early motor deficiencies in PD, as previously reported [50]. The wearable impact in analyzing this complex motor task is relevant. In fact, clinical evaluation alone demonstrated an amelioration in gait and pull test items but was not able to disclose which features of these two motor functions improved. Being able to thoroughly phenotype patients' motor performances is crucial to understanding the effect of a therapeutic intervention and to allow for speculation with respect to its dynamics.

In order to provide clinical validity for our approach, we investigated the relation between the clinical scores, given by the examiners, and the measured features. Clinical vs. wearables outcomes demonstrated general significant results (Tables 4–9). In particular, a higher correlation was found between features extracted from static balance tasks (SFT and TS) and BBS scores and between TUG features and UPDRS gait item scores.

Not all of the features presented a perfect correlation with clinical rating, and this is also expected since these measures should be more sensitive than clinical scales, mostly due to the fact that clinical examination is based on a rating scale with only a few steps, while wearables produce a density scale with a high number of steps [52]. For example, in the TUG test, the duration of the performance is a significant parameter for both the classical clinical exam and "technology-based assessment". Conversely, the average velocity of lower limbs was significantly and accurately measured only by the wearable sensors. The same consideration applies for the other features extracted from the balance and gait tests. These results are in accordance with a recent work [7], evidencing that several features extracted by sensors were able to detect subtle abnormalities in early stage PD patients where the corresponding clinical score, obtained by visual examination, was considered normal for the majority of subjects.

It could be argued that a better sensitivity can be clinically irrelevant, detecting differences too small to have a real impact on a patient's life and functioning. Alternatively, it allows investigators to better phenotype motion alterations and their changes after a therapy, and to objectively measure the benefit from a standard intervention, in view of its customization and relevant optimization.

We are aware of some limitations of the present study. First, tDCS was adopted for patients under other medical treatments that had already been adjusted for the optimal dose. We did not use a test-retest design, thus we cannot exclude variability due to participants' physical or mental conditions, or to drug response fluctuations. To minimize the effects of the aforementioned limitations, we performed the study at the same time of the day for every patient, and no modification to the therapy was allowed in the three months preceding the study and during its course. The study cannot exclude a placebo effect. Moreover, we performed the experiment on a small sample size. Indeed, further studies, on larger cohorts, are mandatory in order to confirm our findings.

#### **5. Conclusions**

Our study aimed to demonstrate the advantages of outcomes from technology-based measures in clinical trials. These advantages are particularly important for revealing the effectiveness of tDCS protocols in late stage PD patients. This is because the benefit of tDCS remains unclear and controversial, thus the outcomes from electronic wearables can help the clinical rating of the tDCS effectiveness. In particular, our results provide evidence of the wearable electronic impact, as a complementary tool to the standard clinical evaluation.

The adoption of wearables furnished a number of motor features, some of them with a good correlation with standard clinical assessment, others adding information not evident to human eyes.

Nonetheless, even if wearables can provide motor features for an insight of each patient's motor performances, they remain rarely adopted in clinical trials. We believe that relevant reasons for this can be ascribed to the lack of an integrated platform that can be easily used by nurses and clinicians, and a lack of regulatory approval and appropriate cost–benefit ratios [15,52]. However, the idea to develop and integrate technologies into the assessment of therapy effectiveness has become so evident that several academic centers and companies have started to bring them to the market.

**Author Contributions:** conceptualization, writing—review and editing, M.R., G.D.L., A.P., F.G. and G.S.; methodology and investigation, M.R., G.D.L., A.P., S.S., M.A., C.S. and G.S.; software, M.R.; formal analysis and data curation, M.R. and G.D.L.; validation, M.R., G.D.L., A.P., F.G. and G.S.; writing—original draft preparation, M.R., and G.D.L.; supervision, A.P., F.G. and G.S.

**Funding:** This research received no external funding.

**Acknowledgments:** We acknowledge the support given by Luca Pietrosanti in measurements.

**Conflicts of Interest:** F.G. and G.S. owe 6% each of Captiks srl. The other authors declare no conflicts of interests. Except from the authors, no others had role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

**Ethical Statements:** All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of Fondazione PTV Policlinico Tor Vergata.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Cueing Paradigms to Improve Gait and Posture in Parkinson's Disease: A Narrative Review**

#### **Niveditha Muthukrishnan 1, James J. Abbas 1, Holly A. Shill <sup>2</sup> and Narayanan Krishnamurthi 1,3,\***


Received: 11 November 2019; Accepted: 9 December 2019; Published: 11 December 2019 -

**Abstract:** Progressive gait dysfunction is one of the primary motor symptoms in people with Parkinson's disease (PD). It is generally expressed as reduced step length and gait speed and as increased variability in step time and step length. People with PD also exhibit stooped posture which disrupts gait and impedes social interaction. The gait and posture impairments are usually resistant to the pharmacological treatment, worsen as the disease progresses, increase the likelihood of falls, and result in higher rates of hospitalization and mortality. These impairments may be caused by perceptual deficiencies (poor spatial awareness and loss of temporal rhythmicity) due to the disruptions in processing intrinsic information related to movement initiation and execution which can result in misperceptions of the actual effort required to perform a desired movement and maintain a stable posture. Consequently, people with PD often depend on external cues during execution of motor tasks. Numerous studies involving open-loop cues have shown improvements in gait and freezing of gait (FoG) in people with PD. However, the benefits of cueing may be limited, since cues are provided in a consistent/rhythmic manner irrespective of how well a person follows them. This limitation can be addressed by providing feedback in real-time to the user about performance (closed-loop cueing) which may help to improve movement patterns. Some studies that used closed-loop cueing observed improvements in gait and posture in PD, but the treadmill-based setup in a laboratory would not be accessible outside of a research setting, and the skills learned may not readily and completely transfer to overground locomotion in the community. Technologies suitable for cueing outside of laboratory environments could facilitate movement practice during daily activities at home or in the community and could strongly reinforce movement patterns and improve clinical outcomes. This narrative review presents an overview of cueing paradigms that have been utilized to improve gait and posture in people with PD and recommends development of closed-loop wearable systems that can be used at home or in the community to improve gait and posture in PD.

**Keywords:** Parkinson's disease; cueing; gait; posture; rehabilitation; wearable sensors

#### **1. Introduction**

Parkinson's disease (PD), which is the second most common progressive neurodegenerative disease, results in motor and non-motor dysfunctions caused by the degeneration of dopamine-producing cells of the substantia nigra and other brain regions [1,2]. Clinical motor symptoms include bradykinesia, tremor, rigidity, freezing of gait, and instability of posture and gait [3–6]. Some of the common manifestations of PD that affect gait and posture are stooped posture and shuffling of gait, increases in gait asymmetry and double support time, reductions in step length and walking speed, impairments in postural responses to perturbations, and increases in variability of step/stride time as well as step/stride length [7]. Considerable efforts are being taken to improve options for treating mobility deficits in persons with PD because of the associated risk of falls and loss of independence.

Pharmacological and deep brain stimulation (DBS) surgical treatments have been demonstrated to be partially effective in managing some of the manifestations of gait impairments and postural instability. As the primary pharmacological treatment in PD, the dopamine replacement therapy (i.e., levodopa) improves stride length, gait speed, and double support time variability, whereas it does not have any significant benefits on cadence and other temporal characteristics of gait [8]. The effects of levodopa on postural sway is controversial [9,10]. Regarding inadequate postural responses (compensatory stepping) leading to falls in PD, levodopa seems to offer no benefit [11,12]. Thus, the effects of levodopa on gait and posture in PD is inconsistent.

Concerning the effects of the DBS, stimulation of subthalamic nucleus (STN-DBS) consistently improved stride length but no effects on stride time and its variability were found. Stimulation of globus pallidum internus (GPi-DBS) significantly improved gait velocity but without any significant improvements in stride length. Also, many people with PD reported postoperative worsening of gait and increased risk of falls [13]. In the case of the stimulation of pedunculopontine nucleus (PPN-DBS) at 15–70 Hz, improvements in postural instability and gait disorder, including freezing of gait and falls, have been noticed. However, the improvements varied depending on the duration of follow-up and types of outcome measures obtained [14]. Low-frequency STN-DBS and GPi-DBS (below 100 Hz) have shown encouraging beneficial effects on axial symptoms in PD; however, higher levels of evidence with randomized and blinded studies are needed to confirm the benefits [15]. Also, the overall benefits of low-frequency STN-DBS decrease with long-term use [16].

#### **2. Pathophysiology of Motor Dysfunction in PD**

The loss of dopaminergic neurons in the substantia nigra pars compacta within the basal ganglia leads to classical parkinsonian motor symptoms. The basal ganglia play significant roles in the production and control of automatic and well-learned motor movements. First, the basal ganglia generate internal cues or trigger to facilitate the initiation of movement sequences without attention. Second, they contribute to the cortical "motor set", i.e., they aid in the preparation and maintenance of motor schemes in a state of action readiness thereby enabling appropriate motor function execution. The widely accepted model of basal ganglia consists of two circuits, the direct and indirect pathways, which originate from striatal neurons and project to various output structures. The direct pathway is postulated to promote movement by direct inhibitory projections to the globus pallidus internus/substantia nigra reticulata (GPi/SNr), whereas the indirect pathway is hypothesized to inhibit movement projecting to the GPi/SNr through globus pallidus externus (GPe) and subthalamic nucleus (STN). In PD, striatal dopaminergic depletion results in the reduced inhibitory direct pathway and increased indirect pathway output onto the GPi/SNr and, subsequently, increased GPi/SNr inhibition to the output structures. This consequently leads to deficiencies in the execution of a movement [6,17–19] (Figure 1). This deficiency in execution results in hypokinesia, a central feature in PD, or lack of movement together with muscular rigidity.

Evidence indicates that the basal ganglia are also important for sensorimotor integration. Striatal cells are robustly activated when a sensory event functions as a cue for a movement. In addition, the caudate nucleus and substantia nigra contain a large proportion of cells that are multisensory; such cells could be used to integrate sensory inputs and form a multimodal representation of the environment in the basal ganglia. Disruption of basal ganglia processes enhances the response of pallidal neurons to passive limb movement, suggesting an impaired gain mechanism because of dopamine depletion [2,4,20]. A common consequence of striatal dopamine loss is attenuation of

the transfer of critical information to the basal ganglia which leads to a decrease in the ability to detect relevant internal sensory and or movement cues [1,21,22]. Such a disruption of information flow to the basal ganglia may worsen impaired movement selection and sequencing in striatum with dopamine loss thereby resulting in gait impairments [23]. The pattern of deficits in people with PD is consistent with a disruption of this integration mechanism. Persons with PD may become increasingly dependent on external stimuli to initiate and shape motor output and may be unable to effectively execute movements because of the lack of critical proprioceptive information [24–27].

**Figure 1.** (**A**) Sensory-motor areas for movement execution in the basal ganglia and the impaired motor pathways in Parkinson's disease (PD) with the prevalence of the indirect pathway over the direct pathway and the affected SN's input to the circuit. SN—Substantia nigra, GPi—globus pallidus internus, GPe—globus pallidus externus, Put—putamen, Th—thalamus, CN—caudate nucleus, STN—sub-thalamic nucleus. This results in increased neuronal firing activity in the output nuclei of the basal ganglia that leads to excessive inhibition of thalamo-cortical and brainstem motor systems which, in turn, interferes with movement onset and execution [28,29]. (**B**) Representation of brain areas activated during external cueing reported from findings of image analysis studies conducted on people with PD during cueing experiments [17,30–32].

The presentation of cues in PD is hypothesized to compensate for the pathology by increasing cortical activation which diminishes pathological activity (10–30 Hz) in the basal ganglia [33], mainly by suppressing the subthalamic nucleus through direct pathways [34]. In the case of visual cues, the unaffected visual-motor pathways are believed to play a major role in facilitating movements bypassing the basal ganglia [35].

#### **3. Methodology**

In this review, the current state of scientific knowledge associated with cueing to improve gait and posture in PD is presented. The search for research articles involving cueing/feedback to improve gait/posture in PD used combinations of the following keywords: Parkinson's disease, cueing/cues/cue, real-time feedback, gait, and posture. From the set of 304 articles returned by the search, only studies that used quantitative gait and posture outcome measures (e.g., step length, stride length, walking speed, cadence, and posture) were included. The set of studies were then categorized by type of feedback implemented (visual, auditory, somatosensory), by wearability/non-wearability of the cueing device/mechanism, and by study duration (single-session or long-term training). References cited in the selected publications were also examined for other relevant studies to be considered. Studies were excluded if they were not directed for people with PD, did not measure spatiotemporal parameters of gait and/or posture, or used non-cue-based gait and posture rehabilitation strategies.

#### **4. Cueing for Rehabilitation in PD**

Given the limited ability of pharmacological and surgical treatments to address gait and postural impairments in PD, various forms of external cueing (visual, auditory, or somatosensory) are being investigated for inclusion in neuromotor rehabilitation programs. Cueing can be defined as a mechanism of applying a spatial or a temporal stimulus to facilitate initiating or maintaining motor activity [32]. Numerous studies have shown that external cueing can improve the amplitude and timing of the intended movement by increasing body position/movement awareness, making it a suitable modality for gait and posture rehabilitation [25,26,36–40]. In addition, cueing has also been increasingly used in helping with the initiation of a movement [41].

Cueing studies could be classified as open-loop cueing or closed-loop cueing based on how the cue is presented. In open-loop cueing, the user is presented a series of cues in a periodic or preset manner that is independent of the user's performance. Metronome beats and a set of lines on the floor separated by a preset distance are examples of open-loop temporal and spatial cues, respectively. Open-loop studies have most widely utilized auditory or visual forms of cues to improve gait in people with PD. While auditory cues have most often been delivered as rhythmic auditory stimulation (RAS) or metronome beats in accordance to the user's preferred gait speed or cadence [36,42–46], other types of cues such as highly rhythmic music or verbal instructions have also been investigated in some of the studies [43,44,47]. Most forms of visual cueing present lines or markers on the floor as targets for foot placement. Markers such as stripes/tapes on the floor, projections from laser pointers, and lights mounted on the user or embedded on a walking stick or walker [48–52] have been utilized. Visual cues were spaced at distances based on the subject's average step/stride length measured at baseline trials. A few studies have investigated the use of somatosensory cues [53–55] using vibrating wrist-worn devices and a combination of audio/visual and or/somatosensory cues for rehabilitation [40,56,57].

Studies of open-loop cueing used as a therapeutic modality have demonstrated short-term and long-term gait improvements [41–43,48,58–61]. Short-term studies investigated immediate effects with and without different cue interventions [62–64]. Laboratory-based long-term training studies compared walking with cues to without cues [50]. One long-term auditory cueing study investigated differences between ecological-based footstep cues (sound recorded while walking on gravel) to artificially synthesized RAS [43] and compared walking with auditory cues to walking with visual cues [40]. In the studies that presented cues as training, cues were provided progressively [65] or in combination with physical therapy improved step time variability [59], posture and bradykinesia [66], stride length, gait speed, and cadence [42].

In contrast with open-loop cues, closed-loop cueing provides feedback on the user's performance in real-time which can facilitate modifying one's performance to achieve the desired movements. Real-time feedback of step length [67–70] and uprightness of posture [69] have been investigated for targeting PD-specific gait and posture deficits. However, these studies used treadmill-based cueing systems and, therefore, are not suitable for overground locomotion during free-living conditions.

Many studies have been performed using virtual reality (VR) which provides visual stimuli that can help in motor and cognitive training [60,62,71–76]. These studies have used augmented visual/auditory- or somatosensory-based feedback for training, but a meta-analysis [77] indicated that there is only limited evidence of improvements in gait and balance due to the use of VR compared to an active intervention without the VR component. Importantly, most of these VR systems require a very sophisticated and expensive setup and may not be suitable for use at home.

#### **5. Benefits of Open-Loop Cueing on Gait in PD**

Evaluation of the acute/immediate effects of cues demonstrated that gait variables, such as cadence [48,52,56,59,78,79], speed [48,52,57,59,80], and step length [42,44,45,48,63,80], increased during walking with rhythmic auditory stimulation (RAS) when compared to walking without cues. In some instances, the improvements in step length were reported to be a consequence of using a cadence that was higher than the baseline. In addition to improving stride length and temporal measures, RAS also reduced stride-time variability [81] and helped persons without freezing of gait (FoG) more than those with FoG [79]. It was suggested that RAS might provide an external rhythm that can compensate for the defective internal rhythm of the basal ganglia in PD [45,50,80].

Use of visual cues, on the other hand, consistently improved step/stride length [49–51,70,71,73] with or without increasing walking speed or cadence. Plausible explanations for these acute effects could be that visual cues may help fill in for the motor set deficiency by providing visual-spatial data [17,82] and help in focusing attention on gait [57,73]. However, in studies that involved visual cueing during treadmill walking, it is not clear whether the gait benefits were due to the visual cueing or to the external pacemaker effect of the treadmill. Also, treadmill walking at speeds greater than the comfortable speed may demand more attention to the task of walking itself, which may result in worsening gait automaticity (ability to perform upper and lower limbs movements automatically during gait with little attention) which is already reduced in PD compared to age-matched controls [52,83]. An investigation of a visual cueing strategy that used a subject-mounted light device to present step length cues at a preset distance in front of the user reported improvements in stride length and gait speed [49]. Cueing studies that combine auditory, visual, or somatosensory cues [40,56] also reported improvements in cadence, gait speed, and stride length. Moreover, studies that focused on attention strategy by asking people with PD to think about taking larger strides were found to be effective in normalizing gait deficits observed in PD [47,57].

Notably, studies that have investigated the impact of long-term training demonstrated that RAS was effective in improving both temporal and spatial gait measures, such as walking speed, cadence, and step/stride length, regardless of the type of sound stimulation (ecological, synthetic auditory cue) that was provided. A follow-up evaluation conducted after three months revealed that the effects of the training were still largely maintained. When RAS was used for one-week training in PD people with FoG [58], walking speed was increased, but no change in freezing episodes was noted, whereas, in another study that used RAS for a three-week training, stride length, walking speed, and cadence were significantly increased [42]. Effects of long-term gait training with and without visual cues showed increases in step length and gait speed [50]. An open-loop cueing study that demonstrated improvements in both temporal gait parameters and stride length attributed temporal improvements with the use of auditory cues and improved stride length to the visual cues [40]. Results from a similar study [66] showed improvements in postural stability and bradykinesia (as measured using Unified Parkinson's Disease Rating Scale (UPDRS)-Part III items) that were retained six weeks after the training period was completed.

#### **6. Benefits of Closed-Loop Cueing on Gait in PD**

Closed-loop cueing provides feedback based on the user's movements in real-time so that the user can be aware of their performance and modulate it to achieve the desired/target performance. Studies that investigated closed-loop cueing are fewer in number and are more recent as compared to open-loop strategies. Both single-session and long-term training studies using closed-loop cueing

were conducted using auditory, visual, somatosensory, and combined cueing strategies to evaluate their effects on gait and posture. Studies that used closed-loop feedback systems have demonstrated a higher degree of gait and posture improvement as well as residual carry-over effects in comparison with open-loop, feed-forward systems [39,81]. This could be because performance-based cues have been shown to help the user understand the delivered cue.

A single-session closed-loop study provided visual feedback based on the patient's own motion using eye-glasses and observed acute improvements in walking speed and stride length [84]. Two studies used treadmill walking with closed-loop visual cueing to demonstrate that people with PD could successfully follow the cues and improve the targeted gait parameters; one involved projection of target step length and uprightness cues (only one type of feedback was used at a given time) on the monitor in front of the treadmill [69] and the other projected visual cues (transverse lines) on the treadmill belt [70]. A few studies developed smartphone applications and utilized data from inertial measurement units to measure surrogates of current gait performance, which were obtained by calculating an average of the parameter over several steps, and provided feedback when the gait parameter was not in the target zone [85,86]. The feedback was provided to the user only when the gait pattern was insufficient and was referred to as "on-demand" feedback [86]. Of the closed-loop cueing studies listed in Table 1, two of them examined the immediate/acute effects of auditory cues in a single session using a wearable sensor system. The Ambulosono sensor system and the StepPlus system [87,88] were developed to provide auditory feedback to inform users when their current spatiotemporal gait parameters are out of a specified target range. Both systems [87,88] were designed for use by people with PD but have not yet been tested in people with PD. Preliminary results on a control population (a group of individuals without PD) showed improvements in stride length, stride length CoV, and cadence. The Armsense device, a portable device to measure arm-swing and provide tactile feedback, was tested in a single-session study on individuals with PD and demonstrated improvements in spatiotemporal gait parameters [89].

With mounting evidence suggesting greater gait and posture improvements as a result of closed-loop cueing training, a pilot study [69] was extended to assess the performance of cues on improving gait and posture in PD in a six-week training study [90,91].

Other long-term training studies using closed-loop visual and auditory cueing evaluated the effects of closed-loop cueing on a variety of gait parameters: gait speed [67,68,70,86,92,93], cadence [70,86], stride length [67,68,70,86], fall incidences [94], and other gait and dynamic balance measures [74,84,95] at follow-up and post-training. Only two of the long-term training studies used a wearable sensor-based, closed-loop system [86,95].

Some closed-loop training studies used augmented reality devices and game-based motion therapy for combinational cueing [72,75,77,84,96,97]. Results from these studies suggested that the closed-loop sensory feedback with or without long-term training was an effective non-pharmacologic intervention for gait and balance improvement in PD. The abovementioned studies involving virtual reality and game-based visual cueing have provided feedback to the user using monitors placed at the eye-level which may help people with PD to be upright at least while following the feedback.

The regular practice of being upright during the training and any sustained benefits may reduce the issue of stoopness experienced by people with PD. To date, only a few closed-loop studies [70,74,86,94,95] included a randomized control trial (RCT) research design to confirm that the gait and posture improvements observed are mainly due to the presentation of cues.


**1.** Summary of key findings of closed-loop cueing strategies that were included in this review.

**Table** 



reach test (FRT), Randomized

 controlled trial (RCT). The table includes only experimental

 studies and does not list the review articles mentioned in the text.

#### *Sensors* **2019**, *19*, 5468

#### **7. Discussion**

#### *7.1. Di*ff*erent Cueing Types May Engage Di*ff*erent Mechanisms*

Findings from the literature indicate that both types of cueing (i.e., auditory and visual) result in improved gait and posture in individuals with PD. The hypothesized neural mechanism for external cueing, suggested by Morris et al. [92], bypasses the hypoactive basal ganglia-supplementary motor cortex (SMA) circuit by slightly altering the way the neural circuits control movement in individuals with PD [31,35,99]. In general, sensory cues are known to enable the dorsolateral pre-motor control system [30,32,63] which bypasses the SMA that is deficient in PD. Specifically, it has been suggested that auditory cues help in improving the temporal parameters, such as cadence and gait speed, and that external cues help because they are able to bypass the internal rhythm deficit associated with PD. Visual cues, on the other hand, are believed to enable the visual–cerebellar motor circuit that influences the spatial aspects of gait, such as step/stride length [71,82,92,100].

#### *7.2. E*ff*ect of Disease Stage on Cueing Strategy*

The effect of cueing in PD rehabilitation may depend on the stage of the disease and the type of dominant symptoms being experienced. The studies included in this review focused predominantly on individuals classified as Hoehn and Yahr stages II, III, and IV. For people in the early stage of disease severity, external cues can compensate for small deviations from their normal gait pattern thereby maintaining optimal gait quality and preventing deconditioning through training. Severely affected individuals with PD rely on external cues to compensate for deficits in the automatic control mechanisms (i.e., the ability to automatically generate normal stride length in a timely manner) thus improving gait and reducing the incidence of falls and freezing of gait [30,32,37,55].

#### *7.3. Open-loop Cueing: Challenges and Limitations*

The primary challenge in cueing, whether open-loop or closed-loop, is to present the cue in a manner that is informative, but does not have detrimental side effects on gait or balance. Despite numerous studies that demonstrated the benefits of cues on gait in PD, most of them did not investigate cues effects on balance control. Also, studies that specifically used cues to improve balance control in PD are very limited, and they were focused on improving posture during quiet stance, sit-to-stand, and dynamic balance maneuvers [101–103]. It is possible that the presentation of visual cues in the form of markers on the floor or on the treadmill belt [38,70,81] may further degrade posture and stability because it requires people with PD, who may already experience stooped posture, to look down.

Similarly, the use of auditory cues provided via earphones may reduce the awareness of environmental sounds which may make it unsuitable for use outside of a laboratory environment. This could be particularly problematic if sounds are provided continuously, i.e., with every step. Another major limitation associated with open-loop systems is that the user is required to detect any mismatch between the cue and their performance and decide how to respond in a manner that will get them entrained (in sync) with the cues. For auditory cues, the user might have to make a quick or a long-duration step in order to get in phase with the cues; for visual cues, the user might have to make a short or a long step in order to get in phase. Finally, although the literature on open-loop cueing in PD includes several studies that observed considerable improvements in spatiotemporal parameters, future studies along these lines could help to move the field forward by documenting how well users are able to follow the cues and by utilizing an RCT research design. Documentation of performance in following the cues could provide insight into the limitations of the cue presentation technique and could help to document the progression of learning throughout an intervention; the use of an RCT design would provide more reliable and actionable evidence for a decision to use a technique in the clinic.

#### *7.4. Closed-loop Cueing: Challenges and Limitations*

As with open-loop cueing, closed-loop strategies must also present information in a manner that is informative and does not have detrimental side-effects on gait. In addition, closed-loop paradigms must also measure/calculate the feedback parameter in real-time and, if it is to be useful outside of the laboratory, the entire system should be wearable and affordable. Setups that use motion capture systems in the laboratory or clinic are expensive and require travel and staff time. Treadmill-based systems pose limitations because some people with PD do not feel comfortable walking on a treadmill, whether that be at home or a facility with supervision. For these reasons, a low-cost wearable system that could readily be used on a daily basis during overground walking might be more widely accepted. However, there are technical challenges in measuring gait parameters from wearable sensors in real-time and conveying feedback in a manner that is safe and easy-to-use. Our group and others are working to develop low-cost, wearable systems for real-time feedback in home or community environments [87,104,105]. Once the technical development challenges are overcome, these systems will be evaluated for accuracy and safety and then clinical efficacy will have to be assessed in an RCT. These types of technologies have potential for widespread use, but they would require regulatory approval before commercialization and marketing.

#### **8. Conclusions**

Based on the review of the literature presented here, it is clear that cueing can be an effective component of locomotor therapy for people with PD who experience gait deficits. Rhythmic auditory cueing has been the most widely used technique, but it is most effective only in influencing the temporal parameters of gait. Visual cueing techniques have been used to increase spatial parameters, such as step/stride length, and to reduce step/stride length variability and asymmetry. Such improvements could have a high clinical impact, as they are important factors in gait and posture rehabilitation for people with PD. However, the usefulness of visual cueing techniques has been limited by challenges in presenting cues in a manner that is practical outside the laboratory and in a manner that encourages upright walking. To overcome the limitations of currently available techniques, several groups are developing unobtrusive wearable systems for closed-loop cueing to provide feedback of performance on a step-by-step or on-demand basis. These systems seek to improve locomotion during activities of daily living by providing feedback of gait and posture parameters that are often deficient in PD and by providing it in a way that can be readily used on a regular basis in the home or the community. Recent engineering developments have produced technology that is suitable for applications that require wearable sensors. Current challenges are to develop algorithms to interpret information from the sensors in real-time and to present it to the user in a manner that is intuitive, non-distracting, and actionable. Such advances that lead to technology for cueing that is effective, affordable, and wearable may enable adoption of these techniques by individuals with PD for use on a regular basis at home and in the community.

**Author Contributions:** N.K., J.J.A., and N.M. conceptualized the article, wrote the text and performed critical revision of the manuscript; H.A.S. provided support for interpretation of important clinical content and for review and editing of the entire manuscript; Funding was obtained by N.K., J.J.A., and H.A.S.

**Funding:** The preparation of the manuscript was supported by funds from the National Institutes of Health (1R21NR017484).

**Conflicts of Interest:** The authors have no conflict of interest/financial disclosures to report.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Review*
