**1. Introduction**

Among chronic neurodegenerative diseases, Parkinson's disease (PD) is recognized as the second most common disorder after Alzheimer's disease. It causes an important negative impact on the quality of life characterized by a progressive impairment in motor functions [1].

Neurologists employ clinical assessment scales, such as the Part III of the Unified Parkinson's Disease Rating Scale (UPDRS) [2], as a common basis to assess the motor impairment severity and its progression over time. During the patient assessment, particular features of the movements (e.g., amplitude, speed, rhythm, hesitations) or of the posture (e.g., trunk flexion, one-side leaning and posture recover capabilities) are subjectively evaluated by neurologists on a discrete scale of five classes of increasing severity, with reliability limitations due to intra and inter-rater variability [3]. Aiming to

improve the clinical management and the quality of life of individuals with PD, more objective and automated approaches to disease assessment, also suitable for at home use, have been proposed.

The majority of these approaches employ wearable technologies [4,5], specifically in lower limbs UPDRS tasks assessment [6]; fewer of them are based on optical tracking, smartphones and other technologies [7,8]. Wearable and optical based technologies exhibit complementary aspects: the first ones are more ubiquitous but they are also more invasive and require more management efforts; the second ones are suited for spot assessment in a localized environment, but they are non-invasive. In general, for the assessment, both approaches make use of the correlation existing between the severity of the impairment, as assessed by UPDRS, and the static and/or kinematic parameters characterizing the pose and the movements [9,10].

Recently many low-cost, optical, body and hand tracking systems [11–14] have been employed successfully in the health care context. Among these, the Microsoft Kinect® v1 device has been used to monitor people with PD [15], in rehabilitation [16,17], in body sway and balance [18,19], in gait assessment [20] and gait anomalies detection [21], in identifying different subjects by kinematic signature [22], in hand tracking [23] and to prevent falls [24]. The Microsoft Kinect v2 is more robust and accurate as compared to Microsoft Kinect v1 [25], and it has been recognized a viable tool for tracking human movement in clinical applications [26], standing balance and postural stability [27], gait [28], body sway [29] and clinical motor functions [30]. In the specific context of neuro-degenerative diseases, it has been used for time up and go test [31], in assessing different types of PD patients [32], to classify gait disorders [33] and in neurological rehabilitation [34].

The work presented here is part of a more extensive project aimed to bring UPDRS compliant automated assessments at patients' home. In a former paper [35], we presented our work on the upper limb UPDRS tasks. Instead, in this case, the assessment is based on a Microsoft Kinect v2 device and it is focused on the analysis of posture and lower limb tasks, as specified by UPDRS [2]. This approach guarantees both to compare results with the standard clinical assessment scales, accepted and used by neurologists, and to define explicitly posture and movements to be performed. Furthermore, we select those UPDRS motor items which are suitable to be self-managed by patients at home, considering that some motor tasks are not feasible in any home environments. For example, according to task specifications [2], Gait task (UPDRS task 3.10) requires a safe straight walking path of 4–10 m, which is not usually available at home, as well as the Postural Stability task (UPDRS task 3.12) that cannot be self-administered being based on a retropulsion test.

Specifically, we examine the following UPDRS tasks (Section 3, items 3.8, 3.9 and 3.13): Leg Agility (LA), Arising from chair (AC) and Posture (Po). Concerning Postural Stability UPDRS task (PSretrop, item 3.12), the standard retropulsion test used for the assessment is not a good predictor of fall risk. Furthermore, the related step count parameter is a too rough estimator of the postural instability [36]. Nevertheless, postural stability assessment is important to prevent falls and injury risk in PD, especially in advanced stages [37]. Postural stability deficits in PD subjects can be highlighted by concurrent cognitive tasks or by secondary motor tasks during steady standing stance tasks [38]. Several studies have found that, during the quite stance, the continuous movement of the center of mass (CoM), named "postural sway" or "body sway", contributes to balance control [39–41]. Alterations of body sway can reveal balance dysfunctions in PD, long before their clinical assessment [42] and they can be used to differentiate between motor subtypes of PD [43]. Recently, low cost RGB-Depth devices as Microsoft Kinect v2, have been used to assess objectively balance dysfunctions [44–46]. Given the importance of postural stability in PD progression and risk of fall prediction, we analyze the postural stability by CoM movements (PSCOM) during the posture task. Furthermore, we investigate also the potential correlation between this method and a standard clinical measure of postural stability PSPIGD based on the Postural Instability and Gait Difficulty (PIGD) subscale score of UPDRS, defined as the sum of the scores assigned to the AC, Gait, PSretrop and Po tasks of UPDRS [47]. To our knowledge, this is the first time, in the context of PD, that a set of UPDRS tasks (namely, LA, AC, Po) and the Postural

Stability (PSCOM) during quite stance are automatically assessed by the use of low-cost optical body tracking devices.
