Design of SmartWalk for Estimating Implication of Pathway With Turn and Task Condition on Postural and Gait Indices: Relevance to Fear of Fall

Fear of Fall (FoF) is often associated with postural and gait abnormalities leading to decreased mobility in individuals with Parkinson’s Disease (PD). The variability in knee flexion (postural index) during heel-strike and toe-off events while walking can be related to one’s FoF. Depending on the progression of the disease, gait abnormality can be manifested as start/turn/stop hesitation, etc. adversely affecting one’s cadence along with an inability to transfer weight from one leg to the other. Also, task demands can have implications on one’s gait and posture. Given that individuals with PD often suffer from FoF and their dynamic balance is affected by task conditions and pathways, in- depth investigation is warranted to understand the implications of task condition and pathways on one’s gait and posture. This necessitates use of portable, wearable device that can capture one’s gait-related indices and knee flexion in free-living conditions. Here, we have designed a portable, wearable and cost-effective device (SmartWalk) comprising of instrumented Shoes integrated with knee flexion recorder units. Results of our study with age-matched groups of healthy individuals (GrpH) and those with PD (GrpPD) showed the potential of SmartWalk to estimate the implication of task condition, pathways (with and without turn) and pathway segments (straight and turn) on one’s knee flexion and gait with relevance to FoF. The knee flexion and gait-related indices were found to strongly corroborate with clinical measure related to FoF, particularly for GrpPD, serving as pre-clinical inputs for clinicians.

O NE'S Fear of Fall (FoF) refers to the lack of selfconfidence in performing activities of daily living without falling [1]. The FoF is often associated with gait and postural abnormalities leading to decreased mobility in older adults and commonly in patients with Parkinson's Disease (PD) [2]. Varying postural abnormality while walking [3] accompanied with reduced automaticity in postural control [4] are common and debilitating symptoms of PD. The postural abnormality can be manifested as dropped head, stooped posture, flexed knee [3] that can be quantified by measuring the relevant joint angles. The variability in the knee joint angle (i.e., knee flexion defining one's posture during walking) is a key component of analysis of healthy and abnormal gait [5]. The variability in knee flexion during heel-strike (important for bipedal stability [6]) and toe-off (critical for maintaining balance [6]) while walking (besides other factors) can be related to one's FoF. Again, depending on the progression of the disease, decreased mobility (coupled with gait abnormality) can be manifested as minimal forward movement (linked with hesitation in taking a step) [7] adversely affecting one's cadence [8] along with an inability to transfer weight from one leg to the other leg (in preparation of taking a step) leading to increased Double Limb Support Time [9] are particularly true for those with PD.
One's gait (e.g., cadence, step time, double limb support time [6]) and posture (e.g., knee flexion) during walking [5] can be affected by pathways having turns or varying task demands. For walking on pathways, turning on a curve along a pathway not only threatens stability, but also requires precise balance of each limb [10] thereby affecting one's gaitrelated indices and posture. Such ability is often compromised in elderly [11] and those with PD [10]. In fact, based on the disease progression, individuals with PD may report gait initiation failure (i.e., start hesitation), accompanied with postural transition inducing FoF [7]. Also, these individuals often approach turns in ways that are different from their healthy counterparts [12] and may exhibit turn hesitation (while walking on a turn along a pathway) [13]. Further, they may demonstrate carryover effect (post a turn) [14] and This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ stop hesitation while approaching the destination of a pathway [13]. In addition, task demands imposed by task conditions, e.g., Single and Dual task conditions [15] during walking can have implications on one's gait and posture. In fact, research has shown that different task conditions, e.g., walking without talking (i.e., Single task condition), walking while counting backwards (i.e., Dual task) or holding a tray while walking and counting backwards (i.e., Multiple Task) [16] can have varying implications on one's gait and posture. Specifically, Dual and Multiple task conditions (often experienced in daily life) challenges one's gait, posture and executive function thereby inducing FoF since these tasks require one to divide attention across various components of the task [15] which are often compromised in individuals with PD.
Given that elderly individuals often experience falls [1] and their dynamic balance is affected by task conditions and pathways [11], [15] with the effect (on gait and posture) being pronounced among those with PD [15], in-depth investigation is warranted to understand such implications on gait and posture (with relevance to FoF). This would need use of portable and wearable devices that can capture one's gaitrelated indices and knee flexion (postural index) in free-living conditions while walking on pathways under varying task conditions. Research has shown use of portable and wearable shoes having force sensors and also use of accelerometers, gyro sensors, goniometers to quantify one's gait.

II. RELATED WORKS
Various investigators have used shoes embedded with multiple Force Sensitive Resistors (FSRs) located at different positions under the feet region to sense the pressure profile below one's foot [17], [18] and quantify gait of individuals. For example, in one of the studies [19], insoles embedded with force sensors positioned under the toe, metatarsal, and heel regions were used to identify the gait disturbance of twelve individuals with PD while walking along a pathway that needed them to start and stop walk, navigate turns and walk-through doorways of different widths. In another study, researchers used wearable sensor network comprising of four FSRs with two FSRs placed underneath the heel pad (one medially and the other laterally) and other two placed under the first and fifth metatarsal heads in Gait Shoe [20] to quantify gait of sixteen participants (10 healthy, 6 individuals with PD) while walking overground at self-selected speeds. Though use of multiple FSRs offer improved measurements, yet, it increases the cost and hardware complexity making it infeasible for translational research. Again, too few FSRs can fail to accommodate cases of foot inversion/eversion [21], often seen in individuals with gait disorders. Thus, it is important to select cardinal points below one's foot while accommodating issues of foot inversion/eversion to sense gait events, e.g., heel-strike, toe-off for estimating one's gaitrelated indices. Alternatively, accelerometers, gyro sensors, goniometers are portable, wearable [22] and have been widely used in developing wearable systems for PD gait analysis by measuring gait-related indices along with joint angles. For example, in one of the studies [23], researchers have proposed instrumented Timed Up and Go Test (iTUG) for which they have used a combination of accelerometers and gyroscopes along with seven inertial sensors attached to one's forearm, shank, thigh and sternum. Results of their study with twelve individuals with PD and twelve age-matched controls performing the TUG test thrice showed a significant difference in the gait (quantified in terms of cadence, turning duration, time to perform turn-to-sit and sit-to-stand tasks) of the two participant groups. Again, in another study, two inertial sensors were fastened to the anterior shank (proximal to the ankle joint) and the anterior aspect of the mid-thigh of each leg of the participants (six individuals with PD and seven healthy controls) [24] who walked at self-selected speed on a pressure mat. In another study with twenty-four individuals with PD, twenty-four age-matched healthy controls and twenty-four young healthy controls, inertial sensors placed on the Medial Gastrocnemius and Tibialis Anterior muscles of one's legs were used to measure variation in gait while the participants walked on a pathway having turn and obstacles. Results showed that variance in their gait could significantly discriminate the individuals with PD from the healthy cohort [25]. Though the findings using inertial sensors are promising, yet, one needs to be cautious to address the drift-related issues [22] limiting their use in translational applications.
Motivated by the need, we have come up with a portable, wearable and cost-effective device (SmartWalk henceforth) that comprises of instrumented shoes (for measuring gait-related indices while accommodating issues of foot inversion/eversion) integrated with knee flexion recorder units. Our objectives were to (i) design a SmartWalk that can measure one's gait-related indices and knee flexion while walking and (ii) carryout a study with two age-matched groups of healthy individuals and those with PD to understand the potential of SmartWalk to estimate the implication of (a) task condition, (b) pathways (with and without turn) and (c) specific segments (e.g., straight and turn) of a pathway on one's knee flexion and gait that can have relevance to FoF. In addition, we wanted to understand whether the knee flexion and gaitrelated indices corroborated with clinical measure related to FoF, particularly for the individuals with PD.
The rest of the paper is organized as follows: Section III presents the system design; Section IV presents the methodology used for the study; Section V provides our findings obtained during the study and Section VI summarizes the research findings and discusses the limitations of the current research as well as the direction of future research.

A. Knee Flexion Recorder
A pair of Knee Flexion Recorder units (Knee Flex unit henceforth) ( Fig. 1 (b)) each consisting of a 4.5" bend sensor (from Spectra Symbol) were used for recording one's knee flexion (angle) of both legs. The flexion (angle) refers to the angle between the line connecting the greater trochanter and knee, and the line connecting knee and lateral malleolus [3]. The flexion (angle) was calibrated using a stepper motor-hinge setup [26] for varying bend positions of the sensor. Given that the knee flexion during one's gait is <90 • [27], we mounted the bend sensor in the knee-cap while choosing the bend location to allow us a measurement range of 0 • to ∼100 • with the sensor output changing linearly (R 2 = ∼ 0.99) with bending angle [26] and with an average error of ±0.13 • . The analog signal (0-5V) from the bend sensor along with time stamping was acquired by Microcontroller-based Central module (described below).

B. Instrumented Shoes
A pair of Shoes having Instrumented insoles impregnated with Force Sensitive Resistors (FSR henceforth) was used for recording one's gait events (e.g., heel-strike, toe-off.). We used 0-445 N FSR (FlexiForce A201 from Tekscan) with active diameter of 9.53 mm. The FSRs were placed at the toe, lateral and medial heel locations of each Shoe ( Fig. 1(c)) to accommodate any possible foot inversion/eversion. The Instrumented Shoes were calibrated with VICON (from Vicon Motion Systems Ltd.). The calibration results of a pilot study with five healthy participants (Mean (SD)=27(±3.67) years) revealed good agreement between gait-related indices recorded by the Instrumented Shoes and the VICON setup with average %Absolute error being 0.71% (on an average ∼10.44 ms) for Stride time and 0.8% (on an average ∼6 ms) for Step time, respectively [28]. The analog signal (0-5V) from each FSR along with time stamp was acquired by the Microcontrollerbased Central module.

C. Ultrasonic Sensor Unit
The Ultrasonic Sensor Unit (HCSR04 [29]) was used to transmit synchronizing markers to the Microcontroller-based Central Module to keep track of the segments of a pathway (Section IV.B) traversed by an individual.

D. Microcontroller-Based Central Module
A Central Module comprising of a Microcontroller (ATMEGA 2560) and Data Storage unit (64 GB SD card from SanDisk Ultra) was mounted on a waist belt. The Microcontroller was used to acquire analog (0-5 V) data from the Knee Flex units, the FSRs (of Instrumented Shoes; followed by 10-bit Analog-to-Digital conversion) and the Ultrasonic Sensor Unit at ∼200 samples/second. The data was acquired along with time-stamping for synchronization and extraction of gait events from the Instrumented Shoes. This data (along with shoe ID (namely, 'left' and 'right')) was routed to the Data Storage unit for subsequent offline analysis of one's (i) postural index in terms of knee flexion and (ii) gaitrelated indices namely, Cadence and double-limb support time computed from the information on heel-strike and toe-off events (Figs. 2 (a) and (b)).

1) Extraction of Heel-Strike Event:
We extracted one's heelstrike event (important contributor to bipedal stability [6]) by identifying the earliest valid peak (magnitude ≥ preselected threshold based on a pilot study with 10 healthy (Mean (SD)=58(±6.72) years) individuals and 10 individuals with gait disorders (Mean (SD)=62(±4.27) years)) from any of the FSRs placed at the lateral and medial heel locations of left (L) and right (R) legs and labeled these events as 'L+' and 'R+' (Fig. 2 (b)), respectively. While deciding the threshold, we started by considering that ∼60% of one's weight is distributed over the heel region and that in turn being distributed as ∼32% (of the weight) on the medial heel region and the remaining ∼28 % (of the weight) on the lateral heel region [30] while these can vary with gait abnormality. For this, a pilot study was conducted in which participants (weight ranging from 40 kg-120 kg) were asked to walk overground while wearing the Instrumented Shoes when their heel-strike events were recorded (in terms of digital equivalent on a scale of 0-1023). The minimum value of the peak from any of the FSRs (0-445 N connected in a resistive voltage divider network) placed at the lateral and medial heel locations was found to be 50 units during heel-strike events and 0 during heel-off events.
2) Extraction of Toe-Off Event: We also extracted one's toeoff event (important while maintaining balance [6]) by finding the valley point of the toe FSR data (represented by ' * ') immediately following the valid peak representing the heelstrike event ( Fig. 2 (b)). We represented these events as 'L-' and 'R-' for left and right legs, respectively ( Fig. 2 (a)).
3) Computation of Cadence: One's cadence is the number of steps taken per minute [6]. Here, we computed one's Cadence from heel-strike events. Subsequently, we computed the Normalized Cadence on a scale of 0 to 1 (Eq. (1)) for presentation of our findings.
Here, i represents the i th individual, Min and Max represent the minimum and maximum values, respectively of the Cadence (while considering all the participants).

4) Computation of Double Limb Support Time:
A gait cycle includes various phases, out of which some have both the feet (of an individual) being in contact with the base of support (i.e., double limb support) [6]. This can be defined in terms of initial double limb support time (IDLST henceforth) and terminal double limb support time (TDLST henceforth). In our case, the IDLST, for example, was computed as the duration between the heel-strike event of right leg and the toe-off event of left leg whereas, the TDLST was the duration between the toe-off event of right leg and the heel-strike event of the left leg, while considering one's right leg (Fig. 2). Subsequently, we added the IDLST and TDLST to compute the total double limb support time (DLST henceforth; Eq. (2)) followed by computing the Normalized DLST on a scale of 0 to 1 (Eq. (3)) for presentation of our findings.
Here, i represents the i th individual, Min and Max represent the minimum and maximum values, respectively of the DLST (considering all the participants).

5) Computation of Coefficient of Variation (%CV):
We computed the Coefficient of Variation (%CV (Eq. (4)) of the knee flexion, since this can help in capturing the fluctuation in knee flexion that in turn can be used as a proxy measure of instability associated with fear of fall [31].

6) Computation of Correlation Coefficient:
We computed the correlation coefficient to measure the strength of the relationship (higher values indicate stronger relation) between two variables [32] using Eq. (5).
Here, x and y represent the average values of variables of interest, i.e., x and y, respectively.

A. Participants
Fourteen healthy elderly (Grp H henceforth; (H1-H14; Age>50 years)) and fourteen age and gender-matched individuals with PD (Grp PD henceforth; (P1-P14; Age>50 years (Table I)) volunteered. The individuals with PD were recruited through physiotherapist's referral from nearby hospital where they were undergoing treatment. All the individuals with PD were on medication and reported to experience mild symptoms of freezing of gait. The inclusion/exclusion criteria for healthy participants were (i) age between 18 and 90 years, (ii) can understand experimenter's instructions and (iii) have no neurological, musculoskeletal or vestibular impairment. The individuals with PD were checked for their ability to perform the 10m walk-test [6] while walking over-ground without any external support, e.g., orthosis, canes, and other aids. Further, Falls Efficacy Scale (FES) [33] score, Unified Parkinson Disease Rating Scale motor part (UPDRS III) and H&Y stage [34] (indicating severity of the symptom) were collected by the accompanying therapist in the OFF state.

B. Experimental Setup
The experimental setup comprised of (i) SmartWalk consisting of a waist-belt (housing the Microcontroller-based Central Module), Knee Flex units (in a pair of knee caps adjustable with Velcro belts (Fig. 1 (d)), Instrumented Shoes and Ultrasonic Sensor Unit and (ii) 10m (d say) pathway segmented into Seg A (0m ≤ d ≤ 0.5m), Seg B (3m < d ≤ 4m), Seg C (4m < d ≤ 6m), Seg D (6m < d ≤ 7m) and Seg E (9.5m < d ≤ 10m) identified by a pathway delineator (having markers to demarcate each pathway segment) along with 'START' and 'STOP' lines (Fig 1 (e)). Specifically, Seg A , Seg B , Seg C , Seg D and Seg E were chosen to understand whether the SmartWalk can sense (i) one's gait initiation failure (start hesitation) often reported by patients suffering from PD when initiating their walk [7], (ii) how the participants approach turns that might be differentiated between individuals with PD and their healthy counterparts [12], (iii) one's turn hesitation that is often reported by patients with PD [13], (iv) any carryover effect after leaving the turn segment [14] and (v) one's stop hesitation that is often reported by the patients with PD [13] while approaching the destination of a pathway, respectively. Please note that the range of values of d used for the different pathway segments was chosen as an initial approximation. The 10 m pathway was of 2 types based on the pathway turn angle, e.g., 0 • (Path 0 henceforth) and 180 • (Path 180 henceforth) with each pathway having straight segments (e.g., Seg A , Seg B , Seg D and Seg E ) and a turn segment (i.e., Seg C ), except Path 0 (that had only the straight segments).

C. Procedure
Our study had three walking task conditions, namely Single Task (S T henceforth), Dual Task (D T henceforth) and Multiple Task (M T henceforth) conditions. In each of S T , D T and M T , one was instructed to walk with self-selected speed on the pathways (Path 0 and Path 180 ). Also, in S T , one was instructed not to speak while walking. The D T was similar to the S T , except that one was asked to walk while counting backwards [15]. Again, in M T , one was asked to hold a tray (with a bottle of water placed on it) and count backwards while walking [16]. In addition, the participant was instructed not to match his/her steps with backward counting. The order of presentation of S T , D T and M T was randomized across participants to eliminate ordering effects [35]. Our study required ∼30 min of commitment from each participant. On entering the study room, the participant was offered a chair to sit and relax. Then the experimenter showed the experimental setup to the participant and demonstrated what he/she was expected to do in the study. One was also told that he/she was free to ask for a break at any time or discontinue from the study if uncomfortable. The participant signed the consent form after expressing that he/she understood the task and was willing to participate in the study. For Grp PD , the accompanying therapist collected the UPDRS III and H&Y scores. The experimenter helped the participant to wear the waist-belt, a pair of knee caps (adjustable with Velcro belts) housing the Knee Flex units and the Instrumented Shoes with Ultrasonic Sensor Unit. For the Grp PD , data was collected in the OFF-state. Also, before starting walk in each of S T , D T and M T , the participant was asked to stand straight with both legs touching the 'START' line (Fig 1 (e)) for baseline recording of the knee flexion (Angle Baseline henceforth). This was followed by one walking on the pathway.

D. Unsupervised k-Means Clustering
While quantifying one's gait and posture using SmartWalk, we wanted to understand whether these indices can classify the Grp PD and Grp H using unsupervised k-Means clustering. The k-Means clustering [36] is one of the most widely used unsupervised techniques that provides 'k' number of clusters for n-data points based on the mean of the centroids of the data points. It partitions all the data points into 'k' nonoverlapping clusters so that each data point belongs to only one cluster. The non-overlapping clusters are formed in such a way that the sum of the Euclidean distances between data points and the centroid of a cluster is minimum. Here, we had applied the k-Means clustering algorithm with k = 2. We chose the measure for Grp H as the positive class and that for Grp PD as the negative class and calculated the classification parameters, such as accuracy, specificity and sensitivity [36].

E. Statistical Analysis
Post our study, we analyzed the statistical significance of the postural index (in terms of knee flexion) and gaitrelated indices computed for each of S T , D T and M T for (a) Path 0 and Path 180 and (b) Seg A , Seg B , Seg C , Seg D and Seg E of the pathways for both Grp H and Grp PD . We applied non-parametric dependent sample Wilcoxon Signed Rank test [37] and independent sample Mann Whitney test [37] for investigating the statistical significance of postural index and gait-related indices during data analysis.

V. RESULTS AND DISCUSSION
While the participants (belonging to Grp H and Grp PD ) wore the SmartWalk and walked overground on Path 0 and Path 180 under different task conditions (S T , D T and M T ), we measured their knee flexion (postural index) and gaitrelated indices (namely, Cadence and Double Limb Support Time). Subsequently, we analyzed these indices for statistical significance using non-parametric test since the postural index and gait-related indices were not normally distributed (as evident from Shapiro-Wilk test [37]). Also, we computed Pearson's Correlation [32] to understand the clinical relevance of these indices with regard to FoF.

A. Potential of SmartWalk to Estimate the Implication of Task Condition and Pathway With Varying Turn on Knee Flexion of Grp H and Grp PD With Relevance to Fear of Fall
We wanted to understand the potential of SmartWalk to investigate the implication of pathway without and with turns (Path 0 and Path 180 ) and task condition on the %CV in the knee flexion during heel-strike and toe-off events with relevance to FoF for Grp H and Grp PD .
Our results showed that irrespective of the pathway, the %CV of the knee flexion during heel-strike ( Fig. 3(a)) and toe-off ( Fig. 3(b)) events for both Grp H and Grp PD increased with increasing complexity of task condition. Such an observation might infer that increase in complexity of walking task challenges one's postural performance leading to increased FoF [15]. However, while comparing the increase in %CV of knee flexion during heel-strike and toe-off events between Grp H and Grp PD across task conditions, Grp H showed smaller change in %CV than that of Grp PD for both Path 0 and Path 180 (TABLE II). For Grp PD , the increase in %CV was statistical across all task conditions (for both heel-strike and toe-off events and irrespective of the pathways). In contrast, for Grp H , the increase in %CV was found to be statistical only while comparing S T and M T (irrespective of the pathways and only for toe-off event). Stronger effect on knee flexion of Grp PD than that of Grp H due to increase in task complexity can be possibly attributed to the fact that D T and M T are more attention-demanding than S T inducing increased postural instability [15], adversely affecting the balance of Grp PD thereby making them more prone to falls [38].
Again, irrespective of the heel-strike ( Fig. 3(a)) and toeoff ( Fig. 3(b)) events, while considering the contribution of pathway (with and without turn) towards one's FoF, it was found that Grp H had higher %CV of the knee flexion while walking on Path 180 than that on Path 0 (none being statistically different). Again, Grp PD showed higher %CV of knee flexion while walking on Path 180 than that on Path 0 (all being statistically different). In fact, the Grp PD demonstrated statistically higher %CV of knee flexion than that of Grp H irrespective of the heel-strike and toe-off events. Such an observation on the effect of pathway with turn on %CV of knee flexion might infer that pathway with turn necessitates one to change the direction of straight ahead walk making the task of walking more difficult [15], adversely affecting one's balance [15] leading to increase in FoF.

B. Potential of SmartWalk to Estimate the Implication of Task Condition and Segments of Pathway on Knee Flexion of Grp H and Grp PD With Relevance to Fear of Fall
Given that walking on pathways having turns (with turn being one of the pathway segments (Section IV.B)) is often associated with variations in knee flexion, we wanted to understand the implication of specific segments of the pathway (Path 0 and Path 180 ) and task condition on the %CV of knee Our results (Figs. 4(a) and (b)) showed that irrespective of the pathway, the change in the postural index from that at baseline was statistically higher (p-value<0.05) for Grp PD than that of Grp H across all task conditions and for each of Seg A , Seg B , Seg C , Seg D and Seg E of the pathway. This might infer that the Grp PD showed lesser postural control across all the pathway segments compared to that of Grp H [4]. In addition, for Grp H , irrespective of the pathway (Path 0 and Path 180 ), the change in the postural index from that at baseline remained nearly the same (while considering each task condition) for Seg A , Seg B , Seg C , Seg D and Seg E . In contrast, for Grp PD , the change in the postural index from that at baseline for Path 180 was maximum for the turn segment (Seg C ; while considering each task condition) among all the pathway segments, reflecting maximum effect on the posture of the Grp PD (who are often reported to be adversely affected by pathway turn [10]) was due to the turn segment. However, for Path 0 , the change in the postural index from that at baseline corresponding to Seg A , Seg B , Seg C , Seg D and Seg E was nearly the same (while considering each task condition).
Again, we found that for Path 0 (Fig. 4(b)), all the pathway segments (except Seg E ) caused statistical variation (p-value<0.05) in the change in the postural index from that at baseline only between S T and M T for Grp PD unlike that for Grp H (for whom none were statistically significant). However, for Path 180 (Fig. 4(b)), we found that change in the postural index of Grp PD from that at baseline was statistically different (p-value<0.05) across all the task conditions (unlike that for Path 0 ) for each of the pathway segments (except Seg E in which there was statistical variation in the postural index only between S T and M T ). Thus, for Path 180 , both the D T and M T (triggering divided attention [15]) challenged the postural performance of the Grp PD with the variations being more prominent for the turn segment of the pathway. In short, the SmartWalk can be used to investigate one's start hesitation, effect on posture (in terms of knee flexion) while approaching turns, turn hesitation, carryover effect on posture after leaving the turn segment and destination hesitation, respectively with relevance to one's FoF.

C. Potential of SmartWalk to Estimate the Implication of Task Condition and Segments of Pathway on Gait-Related Indices of Grp H and Grp PD With Relevance to Fear of Fall
Given that individuals with PD suffer from postural and gait-related milestones [2], [3], [4], we were interested to investigate their gait-related indices while using SmartWalk, added to exploring their postural index under varying task conditions and pathway segments. In addition, we wanted to understand whether the SmartWalk can be used to sense the start hesitation, effect while approaching turns, turn hesitation, carryover effect after leaving the turn segment and destination hesitation, respectively in terms of one's gait-related indices, such as Cadence and Double Limb Support Time (DLST).
1) Implication on Normalized Cadence: Based on the progression of the disease, patients with PD might show shuffling gait with minimal forward movement (that is linked with hesitation in taking the next step) [7] adversely affecting their cadence along with an inability to control the cadence [8] making them prone to fall [8]. From Figs. 5 (a) and (b), we can say that irrespective of the pathways (with and without turn), the normalized cadence (Section III. D.3) was statistically higher (p-value<0.05) for Grp PD than that of Grp H (for whom the normalized cadence remained nearly the same) across all the task conditions and this was true for Seg A , Seg B , Seg C , Seg D and Seg E of the pathways (Path 0 and Path 180 ). Again, for Grp PD , all the segments of Path 0 had minimal effect on their cadence and the turn segment (Seg C ; while considering each task condition) for Path 180 had maximum effect among all the pathway segments. High cadence of Grp PD during the turn segment might be because of lesser control on gait [10] triggered by the turn segment, leading to their increased FoF. In fact, the Path 180 had a stronger effect on the cadence of the Grp PD under D T and M T than that under S T , mostly attributed to the turn segment of Path 180 since turning (while walking on a pathway) is a less automatic task than straight ahead walking [15], particularly true for patients with PD.

2) Implication on Normalized Double Limb Support Time:
Individuals with PD often demonstrate inability to transfer their weight from one leg to the other in preparation of taking a step [9] resulting in increased Double Limb Support Time (DLST) [9] that might be associated with falls [9]. From Figs. 6 (a) and 6 (b), we see that irrespective of Path 0 and Path 180 , the normalized DLST was statistically higher (p-value<0.05) for Grp PD than that of Grp H across all task conditions and for Seg A , Seg B , Seg C , Seg D and Seg E of the pathways. Higher DLST might be related with one spending more time while keeping both feet in contact with the base of support [9], a phenomenon often associated with a feeling of 'feet being stuck to the base of support' that has relevance to one's FoF [9]. Again, for Grp PD , all the segments of Path 0 had minimal effect on their DLST and the turn segment (Seg C ; considering each task condition) for Path 180 had maximum effect (similar to that for Normalized cadence). Also, the Path 180 had stronger effect on the DLST of the Grp PD under D T and M T than that under S T , mostly attributed to the turn segment of Path 180 .

D. Postural and Gait-Related Indices (quantified by SmartWalk) -Corroboration With Clinical Measure
Falls Efficacy Scale (FES) is a clinical measure, often used to predict one's FoF [33]. This scale can measure one's confidence in performing a range of activities of daily living without falling. We wanted to understand whether the postural index (knee flexion) and the gait-related indices (cadence and DLST) measured by SmartWalk corroborated with FES scores, particularly for Grp PD . Pearson's Correlation [32; Eq. (5)] between their FES scores and postural index along with gaitrelated indices (irrespective of the pathway and task condition) were strongly correlated (value between 0.7-1.0 [32]; with the  Double Limb Support Time emerging as the most powerful index (closely followed by posture and the other gait-related index studied here)) (Fig. 7). Thus, the SmartWalk can be used to offer pre-clinical inputs in terms of one's postural index and gait-related indices that can have relevance to one's FoF.

E. Classification of Grp H and Grp PD
Having understood the importance of (i) pathway with turn (i.e., Path 180 ), (ii) Task Condition requiring divided attention (i.e., D T and M T ) and (iii) Double Limb Support Time (DLST) as the most powerful measure that has clinical relevance with regard to FoF (among the posture and the gait-related indices studied here), we wanted to investigate the potential of these aspects in classifying Grp H and Grp PD . Results of k-Means clustering (Section IV.D) applied for D T and M T for Path 180 while considering the DLST across all the  (Table III). Again, going deeper, within M T , the Seg B and Seg C outperformed the other segments in classifying the two participant groups emphasizing the importance of approaching turn and the turn segment of the pathway in segregating Grp H and Grp PD .
In summary, the results of our study with SmartWalk showed that the Double Limb Support Time was powerful in discriminating the Grp H and Grp PD while emerging as the most powerful index (closely followed by posture and the other gait-related index studied here) that corroborated strongly with clinical measure related to FoF, particularly for individuals with PD. Again, the Multiple Task Condition (M T ) seemed to be most powerful (among all the three task conditions) in affecting one's posture and gait-related indices. When it comes to pathways, Path 180 led to a stronger effect on one's posture and gait-related indices than the Path 0 . Going one more level deeper, the turn segment of the pathway, i.e., Seg C of Path 180 had the highest implication on one's posture and gait-related indices than the other segments of the pathways.

VI. CONCLUSION
In this study, a portable and wearable device (SmartWalk) setup was designed to measure one's knee flexion (postural index) and gait-related indices (namely, Cadence and Double Limb Support Time) that can have relevance to one's Fear of Fall (FoF). To understand the potential of SmartWalk to estimate the implication of task conditions and pathway (and its segments, with and without turn) on one's knee flexion and gait, we conducted a study with fourteen healthy volunteers (Grp H ) and fourteen age and gender-matched individuals with Parkinson's Disease (PD; Grp PD ) who were asked to walk overground on a 10 m pathway without turn (Path 0 ) and with turn (Path 180 ) under different task conditions with increasing complexity (e.g., Single, Dual and Multiple task conditions). Our results showed that irrespective of the pathway, the variability of the knee flexion during heel-strike and toe-off events increased with higher complexity of task condition for all the participants, with the effect being stronger for the Grp PD than Grp H . With regard to the implication of specific segments of the pathway and task condition on knee flexion and gait-related indices (with relevance to FoF), our results showed that irrespective of the pathway segment (i.e., straight and turn), the effect on knee flexion and gait was higher for Grp PD across all the task conditions with the implication being maximum for the turn segment of the pathway. In addition, our results showed that knee flexion and gait-related indices strongly corroborated with clinical measure related to FoF, particularly for the individuals with PD.
Though our results were promising, our study had certain limitations. One of the limitations was the restricted sample size. In future, we plan to include a larger participant pool. Further, we chose a large range of age in the inclusion/exclusion criteria (to ensure availability of participants), though the participants enrolled for this study had age > 50 years. In future, we plan to extend our study by enrolling participants of various age groups since the gait and posture automaticity varies with age [1]. Again, given the fact that FoF is also linked with various types of pathway terrains such as uphill/downhill [39], we plan to include pathways with various terrains in our future studies. Also, another limitation was the use of pathway with no intermediate obstacle. In future, we want to include such pathways in our study, since one's FoF is also affected by obstacles on a pathway [39]. Though we have analyzed gait from the perspective of Cadence and Double Limb Support Time and posture through knee flexion, yet there exists room for extended analysis of gait from various dimensions, such as Stance and Swing times that are often important with regard to disease severity and postural instability in individuals with PD [40]. Again, though the SmartWalk has the capability of online computation of the gait and posture-related indices, in our present study, we have carried out offline analysis subsequent to data collection since we did not aim to trigger any output in real-time based on the computed indices. In future, we plan to trigger cues, such as visual cues in response to the indices for which we need to utilize the online computation capability of SmartWalk. Triggering of external cues is needed because cueing is known for its de-freezing influence on individuals with Parkinson's Disease who often demonstrate gait disruption, a debilitating experience that is related to freezing of gait [19]. Notwithstanding the limitations, our study offers a wearable, portable and cost-effective solution for estimating the implication of pathways and task conditions on one's knee flexion (postural index) and gait-related indices that have relevance to one's Fear of Fall thereby offering pre-clinical inputs to clinicians working with individuals with Parkinson's Disease.