Exploring the Potential of a Robot-Assisted Frailty Assessment System for Elderly Care

THE PAPER SUBMITTED CAN BE READ AT THE FOLLOWING LINK: LINK TO THE PAPER

Aniol Civit ^a, Antonio Andriella ^b, Maite Antonio ^c, Casimiro Javierre ^d, Concepción Boqué ^c and Guillem Alenyà ^a.

^a Institut de Robòtica i Informàtica Industrial, CSIC-UPC, C/ Llorens i Artigas 4-6, 08028 Barcelona

^b Artificial Intelligence Research Institute (IIIA-CSIC), Campus de la UAB, 08193 Bellaterra, Barcelona

^c Oncohematogeriatrics Unit, Institut Català d’Oncologia, IDIBELL Hospitalet, 08908 Barcelona

^d Unitat de Fisiologia. Departament de Ciències Fisiològiques II, Facultat de Medicina (Campus de Bellvitge), Ctra. Feixa Llarga, s/n. 08907 L’Hospitalet de Llobregat, Barcelona

Abstract: Frailty assessment plays a pivotal role in providing older adults care. Currently, the process is time-consuming and involves exercises that are measured using stopwatches. To help reduce the burden on healthcare personnel, this paper introduces a set of methods to be used in robots to autonomously perform frailty assessments. The system employs camera-derived skeletal data to measure standard total time and additional frailty-related metrics. The study presents results from a cohort of 22 elderly participants, in which the assessments were conducted by both medical professionals and our system. The results demonstrate the precision of our approach, achieving a performance close to that of a human. Moreover, we extract supplementary metrics requested by the doctors to enhance the diagnosis. This research represents a significant stride towards the advancement of social robotics in healthcare, offering potential improvements in patient care and clinical decision-making.

Short Introduction: In this work, it is defined and validated a set of methods to automate the commonly used Short Physical Performance Battery (SPPB) and the Timed Up and Go (TUG) tests. The algorithms provide the times and additional and frailty-related metrics that cannot be seen with the human eye. To do so, a ZED2i camera with its SDK is used to extract and track the skeleton of the patient during the test execution. The comparison is done with OptiTrack as a Gold Standard.

Proposed tests: Examples and distribution

The different individual tests consist on:

(SPPB) Standing Balance: The patient has to maintain balance for ten seconds with the feet in three predetermined positions: together, tandem, and semi-tandem.
(SPPB) Gait Speed: The patient has to walk six meters in a straight line. The first meter corresponds to acceleration and the last one to deceleration. Time is taken on the constant speed phase corresponding to the 4 central meters.
(SPPB) 5 Times Sit-Stand: It involves standing up and sitting down from a chair five times while measuring the total time taken.
(TUG) Timed Up and Go: It consists of standing up from a chair, walking three meters towards an object, surrounding it, walking back, and sitting in the same chair.

Examples of the skeleton tracked for each of the tests are shown below.

Gait Speed Test

TUG Test

5 Times Sit-Stand Test

Standing Balance Test - Feet together

Standing Balance Test - Feet semi-tandem

Standing Balance Test - Feet tandem

A distribution example for the tests is shown in Figure 1.

Figure 1. Space distribution and representation for the execution of the SPPB and TUG tests.

Tracking System Validation

To conduct this validation, 7 volunteers performed each of the tests three times wearing the OptiTrack suit with 41 markers. The skeleton 3D points were recorded using both ZED SDK and Mediapipe algorithms. The volunteers were asked to vary their behavior in each run of the tests to have more variance between the samples. The following results represent the metrics of the profitable runs where metrics could be acquired.

Mediapipe Results

We tried to obtain the skeleton from an actual state of the art algorithm that is used in the literature. It is the case of the Mediapipe, which showed bad tracking results at a certain distance (>5 meters). We show a picture of the output of the Mediapipe algorithm in a recording of the TUG test during the experiments with the OptiTrack.

Figure 2. Sequence of the Mediapipe output plot during the TUG test, in the left image, the user skeleton is not correctly assigned. In the middle image, the skeleton is adjusted properly. In the right image, as the user moves away from the camera, the skeleton is partially lost.

ZED SDK Results

Here, the obtained metrics from the tests, captured with the ZED2i camera using its SDK, are validated using the OptiTrack system as a ground truth. The following table contains the means of each metric, both from the OptiTrack and our system. Also the Mean Absolute Percentage Error (MAPE) is shown.

Test	Metric	Optitrack (M)	Our System (M)	MAPE (%)
Gait Speed	Time (s)	4.71	4.76	4.73
Gait Speed	Left Stride Length (m)	1.25	1.16	8.41
Gait Speed	Left Stride Velocity (m/s)	0.99	0.96	5.12
Gait Speed	Left Step Length (m)	0.67	0.52	22.26
Gait Speed	Left Step Velocity (m/s)	1.07	0.93	15.49
Gait Speed	Left Step Width (m)	0.22	0.13	38.79
Gait Speed	Right Stride Length (m)	1.24	1.13	9.53
Gait Speed	Right Stride Velocity (m/s)	0.99	0.97	5.34
Gait Speed	Right Step Length (m)	0.66	0.67	12.26
Gait Speed	Right Step Velocity (m/s)	1.05	1.09	11.32
Gait Speed	Right Step Width (m)	0.21	0.14	34.48
Gait Speed	Range Balance (m)	0.05	0.08	68.74
Gait Speed	Mean Balance (m)	0.03	0.04	66.02
Gait Speed	Number Steps	7.11	7.68	11.11
TUG	Time (s)	9.29	9.36	2.03
TUG	Left Stride Length (m)	1.11	1	12.41
TUG	Left Stride Velocity (m/s)	0.86	0.81	9.5
TUG	Left Step Length (m)	0.59	0.56	13.09
TUG	Left Step Velocity (m/s)	0.92	0.91	9.73
TUG	Left Step Width (m)	0.19	0.14	33.28
TUG	Right Stride Length (m)	1.06	1.02	9.27
TUG	Right Stride Velocity (m/s)	0.82	0.79	6.52
TUG	Right Step Length (m)	0.59	0.52	13.54
TUG	Right Step Velocity (m/s)	0.94	0.8	15.41
TUG	Right Step Width (m)	0.24	0.17	31.61
TUG	Range Balance (m)	0.57	0.48	17.47
TUG	Mean Balance (m)	0.42	0.33	20.55
TUG	Number Steps	14	13.78	9.52
Standing Balance	Time Together (s)	10	10	0
Standing Balance	Time Semi-Tandem (s)	10	10	0
Standing Balance	Time Tandem (s)	10	10	0
Standing Balance	Box Together (m³)	0.0021	0.0019	14.98
Standing Balance	Box Semi-Tandem (m³)	0.0001	0.0002	69.51
Standing Balance	Box Tandem (m³)	0.0011	0.0008	33.87
5 Times Sit-Stand	Time (s)	21.83	22.04	1.2
5 Times Sit-Stand	Fatigue (s)	1.55	1.29	24.91
5 Times Sit-Stand	Number Sequences	5	5	0

Validation Results

The individual samples for each of the metrics are plotted below. Every plot contains the values of the same metric for all the different runs. The plots compare the value acquired with the ZED SDK (y-axis) and the OptiTrack (x-axis). The linear regression is also plotted in each graphic (orange line), with the identity (green line) which can be used as a reference to check the proper fit of the data. The plots also show the precision and the repeatability of the data, when the samples are closer to the identity line, they are considered more precise.