Using the Built-in iPhone Body Tracking System for Neurological
Tests: The Example of Assessing Arm Weakness in Stroke Patients:
A Preliminary Evaluation of Accuracy and Performance
Vittorio Lippi
1,2 a
, Isabelle Daniela Walz
2,3 b
, Tobias Heimbach
2
, Simone Meier
2
, Jochen Brich
2c
,
Christian Haverkamp
1d
and Christoph Maurer
2e
1
Institute of Digitalization in Medicine, Faculty of Medicine and Medical Center, University of Freiburg,
Freiburg im Breisgau, Germany
2
Clinic of Neurology and Neurophysiology, Medical Centre, University of Freiburg, Faculty of Medicine,
University of Freiburg, Breisacher Straße 64, 79106, Freiburg im Breisgau, Germany
3
Department of Sport and Sport Science, University of Freiburg, Freiburg, Germany
{simone.meier.neurol, jochen.brich, christian.haverkamp, christoph.maurer}@uniklinik-freiburg.de
Keywords: iPhone Arkit, Neurology, Diagnostic Tests, Body Tracking.
Abstract: Timely treatment of stroke is critical to minimize brain damage. Therefore, efforts are being made to educate
the public on detecting stroke symptoms, e.g., face, arms, and speech test (FAST). In this position paper, we
propose to perform the arm weakness test using the integrated video tracking from an iPhone—some general
tests to assess the tracking quality and discuss potential critical points. The test has been performed on 4 stroke
patients. The result is compared with the report of the clinician. Although presenting some limitations, the
system proved to be able to detect arm weakness as a symptom of stroke. We envisage that introducing a
portable body tracking system in such clinical tests will provide advantages in terms of objectivity,
repeatability, and the possibility to record and compare groups of patients.
1 INTRODUCTION
"Time is brain" – the later the treatment for a large
vessel ischemic stroke, the more brain neurons are
lost, and each hour costs around 3.6 years of normal
aging (Saver, 2006). The magnitude of the insult
plays a pivotal role in determining the course of
action for rescue operations. It determines whether
the patient is taken to a nearby hospital, typically for
smaller infarctions, or a comprehensive stroke center
(CSC), usually for larger infarctions (Václavík et al.,
2018). Early thrombolytic therapy leads to a
significantly better functional outcome in patients
(Ospel et al., 2020). It is, therefore, crucial to assess
the extent of the infarction as early as possible,
but preferably when the emergency call is made.
a
https://orcid.org/0000-0001-5520-8974
b
https://orcid.org/0000-0002-8033-1429
c
https://orcid.org/0000-0001-6325-1892
d
https://orcid.org/0000-0001-8165-4783
e
https://orcid.org/0000-0001-9050-279X
Figure 1: The "Robot" from the custom tracking application
shows the kinematic as the system tracks it.
Lippi, V., Walz, I., Heimbach, T., Meier, S., Brich, J., Haverkamp, C. and Maurer, C.
Using the Built-in iPhone Body Tracking System for Neurological Tests: The Example of Assessing Arm Weakness in Stroke Patients: A Preliminary Evaluation of Accuracy and Performance.
DOI: 10.5220/0012208600003543
In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023) - Volume 2, pages 181-188
ISBN: 978-989-758-670-5; ISSN: 2184-2809
Copyright © 2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
181
Figure 2: Healthy subject simulating a positive result for the
test.
Efforts are therefore being made to educate patients,
family members, and the general public (Saver et al.,
2010). In rural areas, particularly when faced with a
significant infarction, it becomes difficult to hit the
within-3-hour window for an early thrombolytic
therapy (Morris et al., 2000). Thus, efforts are
underway to develop reliable methods for detecting
substantial blockages, if possible, even from
laypersons. In prehospital settings, the Face, Arms,
Speech, Time (FAST) test is the most frequently
employed scale to identify signs of stroke (Aroor et
al., 2017; Budinčević et al., 2022). Especially the
Arm Test in the FAST seems to be good for detecting
a large vessel occlusion (LVO). Severe hemiparesis
but also mono paresis are considered the most
recognizable symptoms of an LVO stroke (Nakajima
et al., 2004). The test is easy and straightforward to
perform, does not require specific equipment, and can
be performed everywhere, even on patients who
cannot leave the bed. The test is qualitative by nature
(as a result is "positive" or "negative") but is based on
the judgment of the examiner. The introduction of an
easy and quantitative measure (i.e., hand tracking) by
only recording the arm test may provide advantages
in terms of A faster detection from laypersons and
easy documentation for emergency service.
Nowadays, sensors in technologies have improved
and are built into wearable devices like smartphones.
Since 2020, the iPhone has been equipped with a
LiDAR scanner capable of assessing three-
dimensional scanning (Bhandarkar et al., 2021). So
far, evaluation of the LiDAR sensor in mobile devices
is still ongoing. The first results in scanning
landscapes and little objects lead to the results that
only larger objects can be measured (Teppati Losè et
al., 2022). For objects >10 cm, an absolute accuracy
of 1 cm is estimated. Therefore, we want to introduce
a first easy and quantitative measure (i.e., hand
tracking) by recording the arm test. This may provide
advantages in terms of faster detection from
laypersons and easy documentation for emergency
services for people with stroke.
2 MATERIALS AND METHODS
2.1 The Smartphone Application
In order to perform the presented tests, an application
has been developed with a unity engine, exploiting C
code to interface with the ARKit library. The
application can work on smartphones (iPhone) and
tablets (iPad, Apple Inc., Cupertino, California).
Specifically, the smartphone used in patient tests was
the iPhone 12 Pro (2020, 128 GB, A14 Bionic chip,
Model A2341), and the iPad (2021, 512 GB, 5
th
Generation, Apple M1 chip, Model A2378) was used
for the preliminary tests. Both devices were equipped
with LiDAR and True Depth capability. The
preliminary test used the Captury system (§2.4) as a
reference.
2.2 Tracking Library Overview
The Arkit Library for iPhone and iPad is designed for
augmented reality. With this purpose, it provides the
capturing of body motion in 3D: tracking a subject in
the physical environment and visualizing their motion
by applying the same body movements to a virtual
character. The library relies on the camera and a
LIDAR system, using lasers as a light source and the
Time-of-Flight technique to measure distances
(Gillihan, 2023). A skeleton (see the "robot" in Fig.
1) is fit on the tracked body, providing joint angles
and link positions as output.
2.3 A Preliminary Tapping Test with
the Captury System
In order to visualize the precision of the iPhone/Arkit
system, a task with a defined hand movement has
been used. Specifically, participants were told to
touch two platforms with one hand as quickly as
possible for 20 seconds. They were sitting on a chair
while doing this arm-movement test. The platforms
were positioned on the floor, aligned with the
participant's feet. The platforms were 75 cm tall, and
there was a 26 cm distance between them. This test
was proposed by (Walz et al., accepted) and is based
on the water-pouring task in the Fahn-Tolosa-Marin
Clinical Rating Scale for Tremor (Fahn et al., 1988).
ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics
182
Figure 3: Tracking of the right hand during the tapping test. The plot shows one trial with the subject facing the camera. As
the “skeletons” used in the iPhone and the Captury systems differ, the right hand tracked by the former is compared with the
right wrist tracked by the latter. The components x,y, and z are expressed with respect to the camera and represent the left/right
position (x), the vertical position (y), and the distance from the camera. The tracking is worse in the z-axis, as it is evident
that there is a ripple effect. In this case (frontal view), the error is not much relevant in identifying the task.
We used the markerless system Captury (The
Captury GmbH, Saarbrücken, Germany). to provide
ground truth test movements (repetitive repeated arm-
movements, as shown in Fig. 3). The Captury uses a
method called visual hull and background subtraction
to identify the shape of the subject. Then, it fits a
skeleton on such a shape using an automatic scaling
process. The camera system captures the movements
at a rate of 50 Hz, giving us precise data on the
positions of different body parts like the wrist, elbow,
shoulder, hip, knee, and ankle and joint angles. In
order to align the data between the two systems, the
cross-correlation between the respective tracking of
the moving hand is used:
∗





(1)
Specifically, the delay between the two systems is
the that maximizes eq (1), where and are the
profiles tracked with the two systems. The three-
position components are used; hence, the product
between the two functions is a scalar product
.
All the computations were performed in Matlab
(R2019b; MathWorks, Natick, Ma).
2.4 Participants
We recruited a total of 4 stroke patients for our study.
All participants demonstrated their comprehension of
the study procedures and provided written consent in
accordance with the Declaration of Helsinki (World
Medical Association, Declaration of Helsinki: Ethical
principles for medical research involving human
subjects, 2013).
2.5 The Arm Weakness Test
Before the main study, we conducted a pilot test,
leading to minor changes in our test instructions.
Specifically, patients were instructed to extend both
arms in front of them, forming a V-shape to allow for
better visibility of both arms. They were asked to fully
extend their elbows and wrists while keeping their
palms open and facing upward. Once patients closed
their eyes, a 10-second recording was initiated to
capture their movements. A clinical expert evaluated
the patients' performance while simultaneously
recording the motion data. The arm weakness was
assessed via item 5 of the NIHSS rating scale (0 - no
drift. 1 - drift, 2 - some effort against gravity, 4 - no
movement). All recordings were made from a frontal
perspective, with the patients seated upright.
Using the Built-in iPhone Body Tracking System for Neurological Tests: The Example of Assessing Arm Weakness in Stroke Patients: A
Preliminary Evaluation of Accuracy and Performance
183
Figure 4: Tracking of the right hand during the tapping test. The plot shows one trial with the iPhone at 45° with respect to
the subject. As the “skeletons” used in the iPhone and the Captury systems differ, the right hand tracked by the former is
compared with the right wrist tracked by the latter. The components x,y, and z are expressed with respect to the iPhone camera
and represent the left/right position (x), the vertical position (y), and the distance from the camera (z). The output of the
Captury system has been rotated by 45° around the y-axis accordingly. The overall distortion is smaller than the one in Fig.
3, especially on the x-axis, as reported in Table 1.
3 RESULTS
3.1 Preliminary Test
The trials were performed with the iPhone in front of
the subject and with an approximately 45° angle with
respect to the subject. As the "skeletons" used in the
iPhone and the Captury systems differ, the right hand
tracked by the former is compared with the right wrist
tracked by the latter. Fig. 3 shows the movement of
the hand with the phone in front of the subject. In this
configuration, the tracking of the right hand during
the tapping task is better in the frontal plane than in
the tracking of the depth (z-axis). The plot in Fig. 4
depicts a single trial where the iPhone is positioned at
a 45° angle from the subject. In Fig 5, the same
tracking with a 45° angle is shown for the left hand
(not moving). The capture system demonstrates the
absence of hand movement. However, the left hand,
partially occluded, exhibits a rippling movement
attributed to an artifact caused by the skeleton
"shaking." Notably, the frequency of this artifact
movement matches that of the intended movement
performed by the subject. Overall, the iPhone
tracking system identifies the movement of a body
Table 1: Tracking error expressed as the standard deviation
of the difference between the tracking performed by the
iPhone and the Captury. The difference is computed over
30 seconds of tracking of the tapping test. The right and the
left wrist are tracked. The right hand is performing the task;
the right hand is kept in a relaxed position resting on the
thigh.
Frontal 45° de
g
rees
Right* Lef
t
Right* Lef
t
X 0.0610 0.0499 0.0438 0.0571
Y 0.0640 0.0871 0.0292 0.0532
Z 0.1282 0.0711 0.0226 0.0510
total 0.1558 0.1230 0.0573 0.0932
segment and its timing (i.e., the frequency is
consistent with the one observed with the Captury)
but has some error in movement amplitude in
agreement with what was observed in early
experiments (Reimer et al., 2021). Table 1 reports the
difference between the Captury and the iPhone
tracking in the first 30 seconds (after that, the subject
stopped performing the movement, as is visible in
Fig. 2,3, and 4) in terms of mean squared error. It is
interesting to notice how, for the right hand, the 45°
view produced a smaller tracking error, especially on
the z-axis.
ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics
184
Figure 5: Tracking of the left hand during the tapping test. The plot shows one trial with the iPhone at 45° from the subject.
The capture system shows that the hand is not moving. The partially occluded left hand is affected by a rippling movement
that is an artifact due to the skeleton's "shacking." The frequency of the artifact movement is one of the voluntary movements
performed with the right hand.
Figure 6: Three healthy subjects simulating a positive test. Hand trajectory tracking in the frontal plane from the iPhone.
3.2 The Arm Weakness Test
A preliminary check with healthy subjects simulating
a test with positive results (Fig. 2) produced the
trajectories shown in Fig. 6. It is evident how the
range of motion of the hand that is dropping allows
for positive identification of the movement and
lateralization of the problem. The test with the four
patients produced the outcome shown in Fig. 7 and
described in Table 2. For the two subjects, S2 and S3,
the iPhone tracking shows an evident drop (they were
rated 1 by the doctor). The most severe case is
represented by S4, rated 3. It was recorded in a lying
position because of the impossibility of the patient to
sit-stand upright. This produced less stable tracking
and more "shaky" than the other cases. The excursion
on the y-axis shows a large drift, although the patient
returned to the original position. The tracking of hand
pronation is not always reliable, as shown in Fig. 8.
In some cases, e.g., subject 3 (S3), the pronation is
clearly visible; in some recordings, it is ambiguous,
like for subject 4 (S4). This is because, in general, the
tracking of hand orientation is not always stable.
Table 3 shows the amplitude of the vertical drop for
the hands recorded in the tests (also the ones
simulated by the healthy subject). It is evident that the
hand affected by weakness is moving at a larger
distance compared to the stable one. This allows for
identifying the positive cases with a simple threshold,
e.g., in the presented cases, a threshold of 30 cm
would identify all the positive cases addressed by the
doctor.
Using the Built-in iPhone Body Tracking System for Neurological Tests: The Example of Assessing Arm Weakness in Stroke Patients: A
Preliminary Evaluation of Accuracy and Performance
185
Figure 7: Test on stroke patients. In red, the right hand. In green, the left. The large vertical excursion (hand drop) is evident
in the cases classified as positive by the doctor (i.e., S2, S3, S4).
Table 2: Report from the physician with the rating and some comments. Rating based on the NIHSS item 5 – Motor arm, 0 -
no drift. 1 - drift, 2 - some effort against gravity, 4 - no movement).
Patient
ID
Rating Comments
S1 0 Pronation on the left, slightly bent elbow on the right, but then corrected it.
S2 1 Pronation and minimal lateral descent of the right arm <10cm
S3 1 Not sinking all the way to the bottom
S4 3
4 DISCUSSION CONCLUSIONS
AND FUTURE WORK
The results show that the iPhone body tracking system
is suitable for identifying hand drops in patients and
capturing movement features such as movement
frequency for tasks like tapping in Fig 3,4, and 5.
Although equipped with a LIDAR that allows for
direct measurement of depth, the iPhone can be prone
to significant errors on the z-axis when the pose is
ambiguous (Fig.3). This agrees with previous analysis
reporting that, while the precision of the tracking is
limited the system can provide useful features to be
applied to patient examination (Reimer et al., 2021).
ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics
186
Figure 8: Tracking hand pronation. The figure shows the hand's orientation that is dropping during the test. On the left, the
pronation is visible, accounting for a total rotation of about 20° around the z-axis. A similar outcome is visible on the right
with the healthy subject H3. Subject S4 exhibits a very noisy trajectory from which it is impossible to assess the pronation
properly.
Since the tasks examined in this paper require no
feedback from the user, no considerations have been
made about the delay that is not visible in the figures
as the Captury and the iPhone recordings were
aligned using eq. (1). For this purpose, it should be
considered that the patient could perceive the delay
from about 70 – 80 ms (Morice et al., 2008), and such
an amount of delay can degrade the performance in
the performed task (Lippi et al., 2010). This may be
relevant in some applications as the iPhone tracking
system is estimated to introduce around 120 ms of
latency (Unlu & Xiao, 2021).
One intrinsic limitation of the iPhone is that being
a single camera system, it is prone to the problem of
occlusion (see the artifact movements of the hidden
hand in Fig. 5). On the other hand, the experiments
with different recording angles showed good
flexibility in the possibilities to record the patient as
required for the performed test. While the position
tracking is good enough to identify the drop and be
recorded as a quantitative measure, but the rotation
was not always reliable in the examples. The
integration of inertial units could improve overall
precision, as shown in some applications, mainly
using additional IMUs that the iPhone can read (e.g.:
Table 3: Vertical hand excursion was recorded with the
healthy subjects and the simulated test.
Subjec
t
Lef
t
-hand
drop (m)
Righ
t
-hand
drop (m)
H1 0.0320 0.4934
H2 0.0704 0.4478
H3 0.0496 0.3314
S1 0.2528 0.3479
S2 0.2367 0.3982
S3 0.0417 0.2777
S4 0.7277 0.4672
in Kask & Kuusik, 2019; Monge & Postolache,
2018). To the best of our knowledge, there are no
examples of test where the iPhone is handed to the
patient for the test. The use of IMUs, on the other
hand, would come at the costs of complicating the
experimental test setup.
Overall, applying the iPhone as a diagnostic tool
for neurological patients seems promising. One open
point is storing the data from such tests in a way that
is useful to make statistics on the patients and refine
the model while preserving the subject's privacy. The
design of an aggregated storage system will be the
object of future work.
REFERENCES
Aroor, S., Singh, R., & Goldstein, L. B. (2017). BE-FAST
(Balance, Eyes, Face, Arm, Speech, Time): Reducing the
Proportion of Strokes Missed Using the FAST
Mnemonic. Stroke, 48(2). https://doi.org/10.1161/
STROKEAHA.116.015169
Bhandarkar, A. R., Bhandarkar, S., Jarrah, R. M., Rosenman,
D., & Bydon, M. (2021). Smartphone-Based Light
Detection and Ranging for Remote Patient Evaluation
and Monitoring. Cureus. https://doi.org/10.7759/
cureus.16886
Budinčević, H., Meštrović, A., & Demarin, V. (2022). Stroke
Scales as Assessment Tools in Emergency Settings: A
Narrative Review. In Medicina (Lithuania) (Vol. 58,
Issue 11). https://doi.org/10.3390/medicina58111541
Fahn, S., Tolosa, E., & Marin, C. (1988). Clinical Rating
Scale for Tremor. In Parkinson’s disease and movement
disorders.
Gillihan, R. (2023). Accuracy Comparisons of iPhone 12 Pro
LiDAR Outputs. University of Colorado Denver.
Kask, T., & Kuusik, A. (2019). Performance comparison of
smartphones and a wearable motion sensor for patient m-
assessment. Proceedings of the Biennial Baltic
Using the Built-in iPhone Body Tracking System for Neurological Tests: The Example of Assessing Arm Weakness in Stroke Patients: A
Preliminary Evaluation of Accuracy and Performance
187
Electronics Conference, BEC, 2018-October.
https://doi.org/10.1109/BEC.2018.8600973
Lippi, V., Avizzano, C. A., Mottet, D., & Ruffaldi, E. (2010).
Effect of delay on dynamic targets tracking performance
and behavior in virtual environment. Proceedings - IEEE
International Workshop on Robot and Human Interactive
Communication, 446–451. https://doi.org/10.1109/
ROMAN.2010.5598607
Monge, J., & Postolache, O. (2018). Augmented Reality and
Smart Sensors for Physical Rehabilitation. EPE 2018 -
Proceedings of the 2018 10th International Conference
and Expositions on Electrical And Power Engineering.
https://doi.org/10.1109/ICEPE.2018.8559935
Morice, A. H. P., Siegler, I. A., & Bardy, B. G. (2008).
Actionperception patterns in virtual ball bouncing:
combating system latency and tracking functional
validity. J Neurosci Methods, 1(255), 255–266.
Morris, D. L., Rosamond, W., Madden, K., Schultz, C., &
Hamilton, S. (2000). Prehospital and Emergency
Department Delays After Acute Stroke. Stroke, 31(11),
2585–2590. https://doi.org/10.1161/01.STR.31.11.2585
Nakajima, M., Kimura, K., Ogata, T., Takada, T., Uchino,
M., & Minematsu, K. (2004). Relationships between
Angiographic Findings and National Institutes of Health
Stroke Scale Score in Cases of Hyperacute Carotid
Ischemic Stroke. American Journal of Neuroradiology,
25(2).
Ospel, J. M., Menon, B. K., Demchuk, A. M., Almekhlafi,
M. A., Kashani, N., Mayank, A., Fainardi, E., Rubiera,
M., Khaw, A., Shankar, J. J., Dowlatshahi, D., Puig, J.,
Sohn, S. Il, Ahn, S. H., Poppe, A., Calleja, A., Hill, M.
D., & Goyal, M. (2020). Clinical course of acute
ischemic stroke due to medium vessel occlusion with and
without intravenous alteplase treatment. Stroke.
https://doi.org/10.1161/STROKEAHA.120.030227
Reimer, L. M., Weigel, S., Ehrenstorfer, F., Adikari, M.,
Birkle, W., & Jonas, S. (2021). Mobile Motion Tracking
for Disease Prevention and Rehabilitation using Apple
ARKit. Studies in Health Technology and Informatics,
279. https://doi.org/10.3233/SHTI210092
Saver, J. L. (2006). Time is brain - Quantified. In Stroke (Vol.
37, Issue 1). https://doi.org/10.1161/01.STR.000019695
7.55928.ab
Saver, J. L., Smith, E. E., Fonarow, G. C., Reeves, M. J.,
Zhao, X., Olson, D. M., & Schwamm, L. H. (2010). The
“Golden Hour” and Acute Brain Ischemia. Stroke, 41(7).
https://doi.org/10.1161/strokeaha.110.583815
Teppati Losè, L., Spreafico, A., Chiabrando, F., & Giulio
Tonolo, F. (2022). Apple LiDAR Sensor for 3D
Surveying: Tests and Results in the Cultural Heritage
Domain. Remote Sensing, 14(17). https://doi.org/
10.3390/rs14174157
Unlu, A. E., & Xiao, R. (2021). Pair: Phone as an augmented
immersive reality controller. Proceedings of the ACM
Symposium on Virtual Reality Software and Technology,
VRST. https://doi.org/10.1145/3489849.3489878
Václavík, D., Bar, M., Kleč
ka, L., Holeš, D., Čábal, M., &
Mikulík, R. (2018). Prehospital stroke scale (FAST
PLUS Test) predicts patients with intracranial large
vessel occlusion. Brain and Behavior, 8(9).
https://doi.org/10.1002/brb3.1087
Walz, I. D., Waibel, S., Kuhner, A., Gollhofer, A., &
Christoph, M. (n.d.). Age-related changes in mobility
assessments correlate with repetitive goal-directed arm-
movement performance. BMC Geriatrics.
World Medical Association Declaration of Helsinki: Ethical
principles for medical research involving human
subjects. (2013). In JAMA (Vol. 310, Issue 20, pp. 2191–
2194). https://doi.org/10.1001/jama.2013.281053
ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics
188