Low-Cost Tracking Systems Allow Fine Biomechanical Evaluation of Upper-Limb Daily-Life Gestures in Healthy People and Post-Stroke Patients

Since the release of the first Kinect in 2011, low-cost technologies for upper-limb evaluation has been employed frequently for rehabilitation purposes. However, a limited number of studies have assessed the potential of the Kinect V2 for motor evaluations. In this paper, a simple biomechanical protocol has been developed, in order to assess the performances of healthy people and patients, during daily-life reaching movements, with focus on some of the patients’ common compensatory strategies. The assessment considers shoulder range of motion, elbow range of motion, trunk compensatory strategies, and movement smoothness. Seventy-seven healthy people and twenty post-stroke patients participated to test the biomechanical assessment. The testing protocol included four different experimental conditions: (1) dominant limb and (2) non-dominant limb of 77 healthy people, and (3) the more impaired limb of 20 post-stroke hemiparetic patients, and (4) the less-impaired limb of 11 patients (subgroup of the original 20). Biomechanical performances of the four groups were compared. Results showed that the dominant and non-dominant limbs of healthy people had comparable performances (p > 0.05). On the contrary, condition (3) showed statistically significant differences between the healthy dominant/non-dominant limb and the less-affected limb in hemiparetic patients, for all parameters of assessment (p < 0.001). In some cases, the less-affected limb of the patients also showed statistical differences (p < 0.05), with respect to the healthy people. Such results suggest that Kinect V2 has the potential for being employed at home, laboratory or clinical environment, for the evaluation of patients’ motor performances.


Introduction
The evaluation of motor performances of neurological patients is a common practice in clinical environment [1]. In fact, evaluations are at the basis of a correct selection of therapies to be administered, and are needed for measuring their effect. In a clinical environment, the standard tools for the assessments are clinical scales. Clinical scales are surveys and questionnaires that associate a score to specific performances, related to motor and cognitive aspects [2,3]. Despite providing a wide variety of assessments, clinical scales are inter-and extra-operator dependent, have intrinsic low sensibility, and suffer of ceiling and floor effects [4,5].
A deeper and quantitative assessment can be achieved with motion analysis or wearable sensors, which are some of the main techniques used in clinics to assess the motor capabilities of neurological patients. The clinical status and the effects of therapies can be evaluated in terms of motor performances, mentioned, a reduced number of studies evaluating Kinect V2 performances on patients, in comparison to a golden standard marker-based systems method was found. In their work, Otte et al. [37] found an excellent agreement between the Kinect and Vicon gold standard, as well as retest reliability for a variety of kinematic parameters extracted from different motor tasks of clinical interest, on a wide sample of neurological patients [37]. In another study [38], a detailed analysis was performed to verify the suitability of Kinect V2 as a tool to evaluate rehabilitation of the upper-limb, coming to the encouraging conclusion that the device is suitable for rehabilitation applications. However, it should be underlined that several studies have indicated the adequacy of the sensor on healthy people tracking, aimed at motor evaluations, suggesting its application to pathological movement. In fact, comparing Kinect V2 and a marker-based system, in Otte et al. [37], it was found that, in summary, most clinical parameters showed a high absolute agreement and no systematic bias between systems, and that the parameters that showed moderate absolute agreement, mostly showed a high consistency in agreement, as well. Similar results were found in the Parkinson Disease Assessment [39], in gait analysis and evaluation [40], and for dynamic movements in rehabilitation scenarios [41]. A small cohort of neurological patients was clinically evaluated by the means of Kinect V2 [42]. A recent study [43] assessed Kinect V2 as a tool for evaluating spinal muscular atrophy patients, matched with healthy controls, concluding that the Kinect V2 sensor had the potential of being developed into a complementary output measure, as it provided reproducible, objective, and detailed information of body point motion. In addition, in the issue of Kinect V2 repeatability, with promising results that were to be further investigated, was addressed in a recent study [43]. Kinect V2 was used to record the kinematics of the upper-limb as trigger for functional electrical stimulation of a robotic set-up, aimed at assistance in home environment, confirming a high confidence on the system reliability [44]. Furthermore, a detailed study, not oriented to rehabilitation, underlined that Kinect V2 performances were evidently higher than the Kinect ones [45]. Embedded algorithms for joint tracking made it one of the most valuable, despite affordable, substitute of marker-based systems.
Last, there is a growing literature that works at merging clinical scales and low-cost tracking devices. In fact, the recent work of some authors, was aimed at automatizing some clinical, commonly used assessments, such as proximal arm non-use (PANU) [46,47], the Fugl Meyer Assessment [48], or the Reaching Performance Scale [49]. This approach seemed a natural and valuable use of the device, exploiting its ease-of-use, to standardize the already existing clinical exams.
However, only a reduced number of studies have assessed in fine details, the potential of the system in discriminating healthy people and pathology, and in comparing performances between healthy people and neurological people performances, especially in a context of biomechanical evaluations, rather than in a training platform test.
In the framework of a group of Research Projects in Northern Italy aiming at developing technologies for the improvement and evaluation of patients with motor disabilities, a simple but consistent upper-limb functionality evaluation module, coupled with low-cost technologies, could provide valuable improvements for both clinical and home therapies and monitoring. In the clinics, or in little rehabilitation centers and laboratories, it could support clinical scales evaluation, providing low-cost and timesaving motor assessments. At home, during the execution of unsupervised domestic training, or during daily-life activities, the low-cost technology, coupled with biomechanical evaluation modules could be helpful for setting training difficulty, when integrated into virtual applications, giving feedback to the patients for motivation and monitoring the quality of training and life at distance.
Supported by promising evidences found in the literature, especially when considering healthy people, in this paper Kinect V2 is presented as a tool for the evaluation of the motor performances of neurological patients, during the execution of the Reaching Movement (RM), further exploiting the data presented in previous studies [50]. Thus, this paper investigates and characterizes the presence of differences in performances of healthy subjects, during the execution of RM, compared to post-stroke hemiparetic patients, with a more or a less-impaired limb, with the aim of better understanding the possibility of building strong rehabilitation paradigms, based on low-cost tracking technologies.

Aim
The aim of the study was to investigate the biomechanical performances of healthy subjects and neurological hemiparetic patients in RM movements, with the Kinect V2 commercial, low-cost sensor, and test whether the sensor is able to discriminate differences between pathology and healthiness.

Settings
The experiment on healthy people was conducted at the Institute of Intelligent Industrial Systems and Technologies for Advanced Manufacturing (STIIMA) of the Consiglio Nazionale delle Ricerche (CNR), Lecco and Milano, Italy. Recruitment of participants took place at the Institute of Intelligent Industrial Systems and Technologies for Advanced Manufacturing (STIIMA) of the Consiglio Nazionale delle Ricerche (CNR), Milano, Italy. The experiment on the patients was conducted at the Villa Beretta Rehabilitation Hospital (Costa Masnaga, Italy).

Participants
For healthy people, criteria for eligibility were-being neurologically and orthopedically intact [50]. A cohort composed of 77 healthy subjects (46 males, 31 females, mean age 41.87 ± 19.34), unaware of the purpose of the study, was enrolled for the experiment, after giving an informed consent. The characteristics of the subjects are summarized in Table 1. Patients were in the late subacute or chronic phase of the disease (>5 months after the stroke); recruitment criteria included a correct understanding of the task to perform. Their characteristics are summarized in Tables 1 and 2. Written informed consent was obtained from each subject, before inclusion in the study. The study was reviewed and approved by the CNR Ethics: Commissione per l'etica della Ricerca e la Bioetica del CNR, and the local Ethics Committee at A. Manzoni Hospital, Lecco, and was conducted in compliance with the Declaration of Helsinki. The protocol approval number was 0044338/2018. For the patients, the criteria of eligibility were not imposed on the impaired limb functionality, while capability of understanding the task to be performed was requested.

Experimental Set-up
During the trials, the experimental set-up ( Figure 1) was composed of:

Experimental Set-up
During the trials, the experimental set-up ( Figure 1) was composed of: • A Microsoft Kinect V2 sensor version 2.0, mounted on an easel and placed at about a 2.0 m distance from the torso of the subject. The 2.0 m distance was chosen since it is halfway from 0.8 and 3.2 m, which were indicated by Microsoft as the range of distances to use for exploiting the tracking functions of the device. The Kinect V2 was placed about 10 cm under shoulder height, and precisely in front of the subject, about halfway between the limbs. A tolerance of some centimeters had to be accepted, due to the patients' different anthropometry and sitting postures. Considering that the use of Kinect V2 was meant for future use, even in environments that are non-supervised by medical personnel (such as the patient's home), where even less-controlled conditions are found, this approximations were considered as reasonable.
• In house software C# for on-line feedback and data logging;   The enrolled subjects performed the RM (portrayed in Figure 2), according to the protocol described in [50]. The RM was chosen as it is a paradigmatic gesture for upper-limb rehabilitation, and is fundamental for autonomy in daily life activities, because it simultaneously: (1) involves a multi-joint coordination, (2) involves the capability of elevating the arm against gravity, (3) allows to reach for desired objects, (4) allows interaction with the environment. The coordinated capability of performing RM, allows a wide exploration of the workspace of the upper-limb and a purposeful interaction with the environment. Such choice of motor task also stresses the capability of moving against gravity, which is one of the capabilities that might strongly affect the quality of life of post-stroke patients. The enrolled subjects performed the RM (portrayed in Figure 2), according to the protocol described in [50]. The RM was chosen as it is a paradigmatic gesture for upper-limb rehabilitation, and is fundamental for autonomy in daily life activities, because it simultaneously: (1) involves a multi-joint coordination, (2) involves the capability of elevating the arm against gravity, (3) allows to reach for desired objects, (4) allows interaction with the environment. The coordinated capability of performing RM, allows a wide exploration of the workspace of the upper-limb and a purposeful interaction with the environment. Such choice of motor task also stresses the capability of moving against gravity, which is one of the capabilities that might strongly affect the quality of life of poststroke patients. Subjects were sitting comfortably on a chair, with their back straight. They were asked not to move their torso, during the experiment, and to perform the motor tasks, only by limb motions. The considered motor gestures are shown in Figure 3. A target for the RM was set at shoulder height, indicating the point toward which subjects had to point. Subjects were requested to perform the repetitions of the gestures at a natural, comfortable, self-selected speed, with no pauses between one repetition and the following. The starting position was with the elbow flexed, at about 90°, with a pronated hand leaning on the thigh. Each subject performed two acquisitions-RM performed with their dominant and non-dominant limbs. Twelve repetitions of the motor task were performed per acquisition.
The Microsoft Kinect V2 sensor was used to record the movement execution. In total, 77 healthy subjects and 20 hemiparetic patients executed two acquisitions, for a total of 154 records on healthy people and 40 records on patients.

Patient Clinical Assessment: The Reaching Performance Scale
A physicist evaluated each patient with the Reaching Performance Scale [49,51], which is a recently developed assessment used for considering the main motor capabilities involved in the reaching gesture.

Data Sources and Measurements
Tracking data acquired with the Kinect V2 were recorded and logged with an in-house C# software developed for visual feedback and data acquisition. Offline analysis was performed by the means of an in-house developed Matlab software.
3D joints tracking data of shoulder, elbow, wrist, and time labels were recorded and logged for off-line analysis. In order to eliminate noise, the tracking data were low-pass Butterworth filtered, third-order, with a cut-off frequency of 6 Hz. An algorithm for automatic phase detection was implemented. It was needed to separate the forward phase of the movements from the backwards one. All outcome measures were performed in the forward phase. Subjects were sitting comfortably on a chair, with their back straight. They were asked not to move their torso, during the experiment, and to perform the motor tasks, only by limb motions. The considered motor gestures are shown in Figure 3. A target for the RM was set at shoulder height, indicating the point toward which subjects had to point. Subjects were requested to perform the repetitions of the gestures at a natural, comfortable, self-selected speed, with no pauses between one repetition and the following. The starting position was with the elbow flexed, at about 90 • , with a pronated hand leaning on the thigh. Each subject performed two acquisitions-RM performed with their dominant and non-dominant limbs. Twelve repetitions of the motor task were performed per acquisition. such reaching movements [49]; the more velocity peaks are found, the more jerky is the movement. Thus, the number of velocity peaks is a measure of the movement smoothness.
The protocol was designed to be a simplified, synthetic version, made of a selection of the parameters computed in the clinical environments, during robotic therapies [51].
The present study aims at analyzing the influence of limb dominancy, gender and age, on the biomechanics of daily-life gestures, when measured with a Kinect V2 sensor.

Study Design
In this work, four main experimental cases were considered: (1) Healthy Dominant Limb test; (2) Healthy non-Dominant Limb test; (3) Patients with Impaired Limb test; (4) Patients with Less-Impaired Limb test. Each subset was tested in comparison to the others, on each variable of the biomechanical assessment. Means and standard deviations were computed along with p-values of the comparisons, for a total of six multi-group comparisons, one for each of the The Microsoft Kinect V2 sensor was used to record the movement execution. In total, 77 healthy subjects and 20 hemiparetic patients executed two acquisitions, for a total of 154 records on healthy people and 40 records on patients.

Patient Clinical Assessment: The Reaching Performance Scale
A physicist evaluated each patient with the Reaching Performance Scale [49,51], which is a recently developed assessment used for considering the main motor capabilities involved in the reaching gesture.

Data Sources and Measurements
Tracking data acquired with the Kinect V2 were recorded and logged with an in-house C# software developed for visual feedback and data acquisition. Offline analysis was performed by the means of an in-house developed Matlab software.
3D joints tracking data of shoulder, elbow, wrist, and time labels were recorded and logged for off-line analysis. In order to eliminate noise, the tracking data were low-pass Butterworth filtered, third-order, with a cut-off frequency of 6 Hz. An algorithm for automatic phase detection was implemented. It was needed to separate the forward phase of the movements from the backwards one. All outcome measures were performed in the forward phase.

Outcome Measures
The recorded data were used to compute the following outcome measures (according to the revision of a previously presented protocol [50]  Referring to Figure 3, Shoulder Elevation angle is defined as: Shoulder Rotation along Vertical Axis angle is defined as: Elbow Extension angle is defined as: For the conventions adopted, see Figure 3. The above listed parameters accounted for the movement kinematics (range of motion) at the end of the forward phase of the movement. The parameters listed above account for the conventional motor strategies used by patients, to compensate some of their motor impairments. The trunk compensus is a normalized parameter that takes into account the amount of movement, produced by displacing the trunk, rather than by moving the hand towards the target. Defining the trunk displacement from the origin position as OT, and the total displacement of the hand (from begin to end position) as OH, TC is defined as follows:

Compensatory Strategies
If the trunk is not displaced, OT

•
Trunk Compensus (TC, []) • Scapular Elevation (ScE, m) The parameters listed above account for the conventional motor strategies used by patients, to compensate some of their motor impairments. The trunk compensus is a normalized parameter that takes into account the amount of movement, produced by displacing the trunk, rather than by moving the hand towards the target. Defining the trunk displacement from the origin position as OT, and the total displacement of the hand (from begin to end position) as OH, TC is defined as follows: If the trunk is not displaced, OT  0, while, for a "virtual movement" done only with the trunk compensus, OT  1. In neurological patients, TC might be increased with respect to healthy people, since trunk compensus strategies are used.
Scapular elevation is measured in meters (m), and addresses how much the shoulder is translated, vertically, to achieve the target. In neurological patients, scapular elevation tends to be increased with respect to healthy people.

Motor Control and Motion Quality
Average number of velocity peaks (VP), used as a measure of smoothness. The above listed parameter accounted for the movement's quality of execution, representing smoothness. It is well-known that bell-shaped velocity profiles (single peak) are usually employed in 0, while, for a "virtual movement" done only with the trunk compensus, OT . Compensatory Strategies

•
Trunk Compensus (TC, []) • Scapular Elevation (ScE, m) The parameters listed above account for the conventional motor strategies used by patients, to pensate some of their motor impairments. The trunk compensus is a normalized parameter that s into account the amount of movement, produced by displacing the trunk, rather than by ing the hand towards the target. Defining the trunk displacement from the origin position as OT, the total displacement of the hand (from begin to end position) as OH, TC is defined as follows: If the trunk is not displaced, OT  0, while, for a "virtual movement" done only with the trunk pensus, OT  1. In neurological patients, TC might be increased with respect to healthy people, e trunk compensus strategies are used. Scapular elevation is measured in meters (m), and addresses how much the shoulder is slated, vertically, to achieve the target. In neurological patients, scapular elevation tends to be eased with respect to healthy people.

. Motor Control and Motion Quality
Average number of velocity peaks (VP), used as a measure of smoothness. The above listed parameter accounted for the movement's quality of execution, representing othness. It is well-known that bell-shaped velocity profiles (single peak) are usually employed in 1. In neurological patients, TC might be increased with respect to healthy people, since trunk compensus strategies are used.
Scapular elevation is measured in meters (m), and addresses how much the shoulder is translated, vertically, to achieve the target. In neurological patients, scapular elevation tends to be increased with respect to healthy people.

Motor Control and Motion Quality
Average number of velocity peaks (VP), used as a measure of smoothness. The above listed parameter accounted for the movement's quality of execution, representing smoothness. It is well-known that bell-shaped velocity profiles (single peak) are usually employed in such reaching movements [49]; the more velocity peaks are found, the more jerky is the movement. Thus, the number of velocity peaks is a measure of the movement smoothness.
The protocol was designed to be a simplified, synthetic version, made of a selection of the parameters computed in the clinical environments, during robotic therapies [51].
The present study aims at analyzing the influence of limb dominancy, gender and age, on the biomechanics of daily-life gestures, when measured with a Kinect V2 sensor.

Study Design
In this work, four main experimental cases were considered: (1) Healthy Dominant Limb test; (2) Healthy non-Dominant Limb test; (3) Patients with Impaired Limb test; (4) Patients with Less-Impaired Limb test. Each subset was tested in comparison to the others, on each variable of the biomechanical assessment. Means and standard deviations were computed along with p-values of the comparisons, for a total of six multi-group comparisons, one for each of the biomechanical parameters.

Statistics
In the main investigation, for each dependent variable belonging to the evaluation protocol and to a specific data subset to be tested, the normality of the distribution was assessed using the Kolmogorov-Smirnov normality test. Normality and statistical tests were performed with Matlab 2018a. For testing the biomechanical differences on each of the six measures used in the assessment, the one-way ANOVA test was used. For each of the measures, four groups were tested-healthy dominant limb, healthy non-dominant limb, patients with impaired limb, and patients with less-impaired limb. In case biomechanical differences between groups were found, Post-hoc tests (Matlab multcompare) were conducted to understand which group(s) differed from the others. The alpha-error significance level was set to 0.05, for all tests.

Shoulder Elevation Angle
Results related to Shoulder Elevation Angle are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −6 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients More-Affected Limb group was different from one of the other groups.

Shoulder Rotation Angle (Along Vertical Axis)
Results related to Shoulder Rotation Angle are reported. Statistical differences were found between the four groups, according to the ANOVA test (p = 0.0004). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients group was different from the one of the other groups.

Shoulder Rotation Angle (Along Vertical Axis)
Results related to Shoulder Rotation Angle are reported. Statistical differences were found between the four groups, according to the ANOVA test (p = 0.0004). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients group was different from the one of the other groups.

Elbow Extension Angle
Results related to Elbow Extension Angle are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −10 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients More-Affected Limb group was different from one of the other groups.

Elbow Extension Angle
Results related to Elbow Extension Angle are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −10 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients More-Affected Limb group was different from one of the other groups.

Scapular Elevation.
Results related to Scapular Elevation are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −10 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients More-Affected Limb group was different from one of the other groups.

Scapular Elevation
Results related to Scapular Elevation are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −10 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients More-Affected Limb group was different from one of the other groups.

Trunk Compensus.
Results related to Trunk Compensus are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −8 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients Less Affected Limb group was different from the one of the other groups. The Patients with the Less-Affected Limb group was also different from the others, probably due to outliers.

Trunk Compensus
Results related to Trunk Compensus are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −8 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients Less Affected Limb group was different from the one of the other groups. The Patients with the Less-Affected Limb group was also different from the others, probably due to outliers.

Average Number of Velocity Peaks
Results related to the Average Number of Velocity Peaks are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −8 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients with More-Affected Limb group was different from the one of the other groups, as well as the Patients with Less-Affected Limb group, which was also different from the others.

Average Number of Velocity Peaks
Results related to the Average Number of Velocity Peaks are reported. Statistical differences were found between the four groups, according to the ANOVA test (p < 10 −8 ). Post-hoc tests (Matlab multcompare) revealed that the mean of the Patients with More-Affected Limb group was different from the one of the other groups, as well as the Patients with Less-Affected Limb group, which was also different from the others.

On the Biomechanical Results
As expected by previous findings in the literature [50], no observable differences in motor performances were found when considering the healthy dominant and non-dominant limbs in a simple gesture, such as the reaching movement, according to the Kinect V2. It is well-known that dominant and non-dominant limbs have a tendency to specialize in dynamical and static motor tasks respectively [52], and the dominant arm can achieve more varied and flexible control over movement trajectories, while accuracy and precision are comparable [53]. However, such differences would be particularly expected in relation to highly demanding motor tasks (hard to complete or very fast, engaging, not known a-priori) or in fine control (related for example to hand or finger dexterity [54]), "in favor" of the dominant limb, rather than in the daily-life, well-known gestures, such as RM. Considering the quite high number of involved subjects, this result can be considered to be a solid benchmark condition for the assessment of the performances of impaired people. At the same time, this assessment provides further evidence that, at least for the considered simple gesture, the biomechanical performances of the two limbs were comparable. For patients, the assessments The whiskers comprise all data points of distribution except the outliers, which are plotted separately using the '+' mark.

On the Biomechanical Results
As expected by previous findings in the literature [50], no observable differences in motor performances were found when considering the healthy dominant and non-dominant limbs in a simple gesture, such as the reaching movement, according to the Kinect V2. It is well-known that dominant and non-dominant limbs have a tendency to specialize in dynamical and static motor tasks respectively [52], and the dominant arm can achieve more varied and flexible control over movement trajectories, while accuracy and precision are comparable [53]. However, such differences would be particularly expected in relation to highly demanding motor tasks (hard to complete or very fast, engaging, not known a-priori) or in fine control (related for example to hand or finger dexterity [54]), "in favor" of the dominant limb, rather than in the daily-life, well-known gestures, such as RM. Considering the quite high number of involved subjects, this result can be considered to be a solid benchmark condition for the assessment of the performances of impaired people. At the same time, this assessment provides further evidence that, at least for the considered simple gesture, the biomechanical performances of the two limbs were comparable. For patients, the assessments provided very different results. In fact, the biomechanical assessment presented in this paper showed, as expected, that in general, the motor performances of a heterogeneous group of patients detracted from the performances of healthy people and of the less-affected limb. This is a relevant result because the "capability of discrimination" of the performances was required to guarantee that Kinect might be a useful device for evaluations in neurorehabilitation. Very interestingly, even if the number of the available subjects was reduced, the patients' less-impaired limb provided a "halfway performance capability", which often performed better than the average of the more-affected limb, even if in a worse way, with respect to the healthy benchmark. These results also need to be confirmed on a larger cohort of people. This finding suggests that low-cost tools, such as Kinect might discriminate different level of impairments, as already suggested by previous research [50], and at least identify peculiar features of a group of people or patients (i.e., on average, patients with less-impaired limb show less velocity peaks than the more-affected limb, but both have a more than healthy benchmark).

Kinect in Real Applications
The use of Kinect for rehabilitation has already been tested in the literature, and exploited in several commercial and research applications. In fact, such an approach might substitute, or better integrate, some of the clinical evaluations that are usually administered to patients. A quite detailed overview of the patient's status can be portrayed and summarized in short reports, highlighting the level of disability with respect to the population of healthy people. Furthermore, this potential can be exploited for the automatization of clinical scales with a quantitative method that does not depend on operator variability, as has already been proposed in previous studies that have tried to perform quantified evaluations of the Fugl-Meyer Assessment, the Reaching Performance Scale, and the Paretic Arm Non-Use assessments [46][47][48][49]. Furthermore, Kinect or equivalent markerless systems could be introduced in clinics and small-scale laboratories, as well as in the home environment, which would increase the possibility of executing tests for the evaluation of motor functionality. The Kinect assessment could also be used for a preliminary, fast evaluation of patients, to direct him to a compatible rehabilitation process, chosen according to his remaining residual functionality.

Limitations
This study has several limitations. First, only the paradigmatic gesture has been considered. The validity of the methodology should be tested on a wider group of motor gestures. Furthermore, while a quite high number of healthy people were involved (77 participants), the cohort of patients was not as large (20 patients), and data from both limbs were available only for a subgroup of patients (11 patients). Thus, while the statistical significance of some results was reported, the statistical power, especially on the sample of patients, was limited. However, some trends and preliminary statistics could be presented. The proposed protocol gave only a summarized overview of the patients' capabilities, but it still has a large margin for being enriched and improved with further assessments. Last, the authors consider it to be very important to test the presented approach in a real-life scenario, involving clinics, laboratories, and patients' home.

Conclusions
In this paper, Kinect V2 was used to test a simple biomechanical protocol on the dominant and non-dominant limbs of healthy people, and on the less and more impaired limbs of post-stroke patients. Preliminary results suggest that healthy dominant and non-dominant limbs show comparable performances, as well as patients' less affected limb. On the contrary, Kinect V2 and clinical scales detect poorer performances for the more affected limb. Further developments of the concept presented in this paper will include refinement of the biomechanical protocol, and testing in scenarios including patients' home and clinics.
Author Contributions: A.S. designed the experiment, analyzed the data, and wrote the paper. F.M. recruited the patients, gave the clinical directives, and revised the paper. L.M.T. revised the paper for intellectual contribution and provided the funding.
Funding: This paper was funded by the RIPRENDO@home, a Regional research project funded under the framework agreement between Regione Lombardia and the National Council of Research (D.G.R. n. 3728-11 July 2012). It was also supported by the Future Home for Future Communities, a Regional research project funded under the framework agreement between Regione Lombardia and the National Council of Research (D.G.R. n. 3866-17 July 2015), and by the Project Empatia.