Agreement between a 3D camera system and an inertial measurement unit for assessing the range of motion, head repositioning accuracy and quality of movement during neck and head movements

Abstract Purpose Altered head range of motion (RoM) and head repositioning accuracy (HRA) are commonly reported in neck pain. However, the quality of motion (QoM) is currently not easy to assess clinically. This study investigated the agreement of head rotation recordings using a 3D camera system compared to a commercially available inertial measurement unit (MOTI). Materials and methods Thirty participants, mean age 26.5 years old (SD 4.4), partook in this study. Participants wore a Headband with MOTI and markers for 3D motion capture analysis during head rotations. The two systems recorded active head RoM in rotation, HRA, and QoM. Agreement of RoM, HRA and QoM data was compared between the two systems using Interclass correlation coefficients (ICC; 2.1) and Bland-Altman plots. Results Good to excellent agreement between the two systems was seen for RoM (ICC: 0.998), HRA (0.75–0.88) and QoM (ICC: 0.911–0.913). The Bland-Altman plots revealed a systemic offset where the MOTI device measured higher values for RoM (mean bias: −0.56 ± 0.65°), HRA (mean bias: 0.48 ± 0.76°) and QoM (mean bias: −16.9 ± 51.6 A.U.). Conclusion The present study found that the MOTI device can accurately measure RoM, HRA and QoM during head rotation. MOTI may be preferred over a 3D camera system for clinical use.


Introduction
Neck pain is one of the most frequent complaints in the general population [1].In fact, in less than 30 years, the number of people seeking rehabilitation for neck pain has increased by a staggering 79% [2].Altered cervical range of motion (RoM) [3] and sensory-motor control, such as altered head repositioning accuracy (HRA) [4,5] is commonly reported in neck pain populations.
Reduced RoM has been linked to self-perceived pain and disability in neck pain populations [6,7] and may reflect a protective behaviour to avoid further pain [8,9].For altered HRA, the underlying mechanism is thought to be altered afferent proprioceptive input from cervical structures, which seems to be most notable during rotation movements [7,9].As reduced RoM and HRA are common findings in clinical populations, these are considered important when assessing neck pain patients in clinical practice and monitoring changes over time during rehabilitation [10][11][12].Quality of movement (QoM) is quantified through jerk, the first time derivative of acceleration, and the closer the jerk is to zero, the smoother the motion.Increased jerk during movements and thereby reduced smoothness are found in people with clinical neck pain of both insidious and traumatic onset compared to a healthy population [13].Nevertheless, QoM remains challenging to assess clinically as this relies on advanced kinematic recordings which are currently not widely used in a clinical setting.
For research purposes, expensive high-tech equipment such as 3D camera systems are often used to assess RoM and HRA in laboratory settings [14][15][16].Such systems give a precise estimate of human movement.Yet, these camera systems may not be appropriate in clinical settings for obvious reasons such as costs, requirements of specific technical skills, and the time needed to use such a setup [17,18].With this in mind, clinicians are recommended to focus on simpler measurements for clinical use [19,20].Traditionally, cervical RoM [21,22] and HRA [10,20] have been assessed using analogue equipment in clinical settings.However, in recent years, several studies have shown promising results using smartphone applications to assess neck movements in clinical settings [23].Unlike analogue recordings of head movements, digital recordings using a smartphone allow for estimating variations in movement, such as QoM, which is otherwise very difficult to estimate [17].One downside to using a smartphone is the size which may make it difficult to attach it to the head when recording movements.In addition, the added weight of a smartphone along with the fixture for attaching it to the head could potentially influence both QoM and RoM.This might be especially important when assessing clinical populations with neck pain where motor control of head movements may already be impaired.Hence, a smaller and lighter device may be preferable for clinical use [17].Here, an inertial measurement unit may be optimal for recording RoM, HRA and QoM with the least amount of interference of normal cervical kinematics due to its size and weight.The potential benefit of being able to accurately assess RoM, HRA and QoM on a routine basis in clinical practice is considerable as this may not only help inform clinicians on these parameters in their clients, but it may also help guide rehabilitation and track progress over time.However, before implementing new technologies into clinical practice, it is important to critically evaluate if these can accurately provide the needed data if they are to replace existing equipment for assessing RoM.
This study set out to compare the agreement between a small and lightweight inertial measurement unit connected to a smartphone using Bluetooth, to a 3D camera system for measuring RoM, HRA and QoM during head rotations.It was hypothesised that the inertial measurement unit would produce comparable results for RoM, HRA and QoM during head and neck movements compared to those recorded with a 3D camera system.

Participants
A convenience sample of thirty healthy, pain-free participants were included in this randomised cohort study.Participants were recruited among staff and students during the fall semester of 2021 at Aalborg University and provided informed consent before enrolling in the study.Data were collected in a laboratory setting at Aalborg University between September and October 2021.On average, participants were 26.5 (SD 4.4) years old, weighed 75.0 (SD 15.7) kg, and were 175.8 (SD 9.8) cm tall.Inclusion criteria were normal pain-free RoM of the cervical spine which was ensured by a short physical examination of active cervical range of motion performed by a registered physiotherapist.In addition, participants had to be able to read and comprehend Danish or English.Participants were excluded if they had experienced any neck pain during the past three months or had a previous history of neck surgery.Additionally, participants were excluded if they took any analgesic medicine or suffered from any neurological or musculoskeletal condition that could influence normal cervical RoM.The study was conducted in line with the rules and regulations of the regional ethical committee, which state that studies focussing on calibration and validation of measurement devices do not require ethical approval [24].In addition, the study is reported in line with the Guidelines for Reporting Reliability and Agreement Studies (GRRAS) [25].

Experimental setup
Participants sat comfortably in a chair with a backrest and were fixated with a chest strap.They wore a blindfold and hearing protection to reduce external disturbance (Figure 1).The participants' feet were placed flat on the floor with knees and elbows flexed to 90 � and their hands resting on their thighs.The participants wore a headband mounted with MOTI and active markers for 3D recordings.A Bosch Quigo laser level (Robert Bosch Power Tools GmbH, Stuttgart, Germany) with a horizontal and a vertical line was placed behind the participant with the 'þ' aimed at a white piece of tape on the headband.Participants were asked to find their neutral position, after which the 'þ' from the laser level was marked with a pen on the white tape.Participants were then asked to turn their head at a self-selected pace to the end of range into either right or left rotation before returning to their self-perceived neutral position.Recordings on both systems were initiated simultaneously by the same person prior to asking the participants to initiate the head movement and were stopped simultaneously after they had returned to their self-perceived neutral position.Participants were required to perform three repetitions in one direction before repeating this in the other direction.The average for each side was used for further analysis.The direction order (right or left) was randomised and balanced, so half the participants started with moving towards the right side first before moving to the left and vice versa.The randomisation order was concealed in individual envelopes and was first revealed when a participant chose an envelope at the beginning of a session.Following each repetition, the assessor moved the participants' head back to their self-selected neutral position, guided by the laser level.The participants' head was adjusted irrespectively if this was needed or not to ensure that the trials felt similar and no feedback on performance was provided [15].Data was stored on a computer until the data analysis was conducted.Both the participant and the assessor were blinded to the results during data collection.

Movement variables
MOTI is a commercially available inertial measurement unit.MOTI is controlled through a smartphone application (MOTI-Research, version RC1, Aalborg, Denmark) which was installed on a Huawei P Smart 2019 (Huawei Technologies Co., Ltd., Shenzhen, China) smartphone and was used to assess RoM, QoM and HRA during head rotations.RoM was defined as the angle between the start and end position of the head movement.QoM was defined as the time integral of the squared jerk (see signal processing section).HRA was defined as the difference between the start and end position i.e. when the participants had returned to their self-perceived neutral position (details on how RoM, QoM and HRA were calculated are outlined in the section on signal processing).
To assess the accuracy of MOTI recordings of RoM, QoM and HRA during head movements, these were simultaneously recorded with the Optotrack Certus 3D-camera-based system (NDI, Ontario, Canada) using two active markers.The Optotrak Certus has been appraised for its ability to accurately track movement [18,26] with a 3D accuracy of 0.1 mm and a resolution of 0.01 mm [27].This is supported by studies demonstrating the camera's high precision [28] and excellent reliability [28,29].In order to limit movement artefacts between the camera markers and MOTI, all units were fixated on the same plastic cluster and mounted on a headband (70.7 g), worn by the participants during head movements along with the hearing protection (364 g).The position of the active makers was sampled at 100 Hz.Participants were seated at an oblique angle to the camera system to ensure that the markers were visible throughout the entire movement (Figure 1).The position of the chair was changed depending on the direction of movement (left and right rotation).Markings were made on the floor to ensure a similar chair placement for each session.

Signal processing
The angle from head rotation was directly extracted from the MOTI device.For the 3D camera system, the arctangent was calculated between two 3D vectors defined by the markers.The first vector was the natural neck position, determined as the average of the initial 15 frames before movement onset.The second vector was calculated during the head rotation and represented the neck position over time.From here, all signal processing was performed in the same way for both systems in MATLAB (v.R2022a, The MathWorks, Inc., Natick, MA).Data for the angular position were filtered with a lowpass Butterworth filter (zero lag, 1.5 Hz, 4th order) [17].Angular velocity, acceleration, and the jerk over time were calculated by sequential derivatives of the angle.All data were automatically trimmed between the start and end of the movement.RoM for head rotation was calculated as the angular position range between the start and end of the movement.
QoM during the head rotation was calculated as the time integral of the squared jerk (equation 1) [30][31][32].Because jerk varies with movement duration and distance, jerk (J) must be normalised.The squared jerk was normalised by multiplying the integrated squared angular jerk by time ðTÞ lifted to the power of five divided by the range of the movement (h) lifted to the power of two.Which makes the jerk a unit-free measure.
Normalized Jerk ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi HRA following head rotation was calculated as the difference in head rotation angle between the first frame of the movement and the last.An example of position, velocity, acceleration, and jerk for both systems is presented in Figure 2.

Statistics
A sample size calculation for ICC statistics [33] was conducted using an online calculator [34].This assumed a minimum acceptable ICC of 0.6.The expected ICC was set to 0.9 based on a conservative estimate from a previous study comparing recordings of RoM in rotation from a smartphone to 3D camera recordings [17].The alpha was set to 0.01, beta to 90%, giving the needed sample size of 26 participants.As this is the first study on head movements using the inertial measurement unit, MOTI (MOTI, Aalborg, Denmark), a conservative number of 30 participants were recruited for this study.Data were assessed for normality using the Shapiro-Wilk test.The agreement of RoM, HRA and QoM recordings between the 3D camera system and MOTI was investigated using Intraclass Correlations (ICC; 2.1).Results were interpreted as poor (ICC < 0.5), moderate (ICC 0.5-0.75),good (0.75-0.9) and excellent (ICC � 0.9) [35].Measurement agreement between the two devices was visually inspected with Bland-Altman plots for RoM, HRA and QoM.To explore the correlation between repetitions for RoM, HRA and QoM, Pearson's (r) correlation coefficients were calculated for each side (left, right) and categorised using Cohen's criteria [36,37] as small (0.1), moderate (0.3) or large (�0.50) effect sizes.
Statistical analyses were performed in SPSS v27 (SPSS Inc., Chicago, IL) with an Alpha level set to 0.01.Data are presented as mean (standard deviation) unless stated otherwise.

Results
The mean age of the participants (13 females and 17 males) was 27.2 (range 24-34) years.They were, on average 182.6 (SD 5.1) cm tall and weighed 84.9 (SD 12.4) kg.
The camera recorded a mean rotation of 77.93 � 10.2) on the right side and 77.09 � (SD 9.9) on the left (Figure 3).Additionally, the mean error during the HRA test for rotation was 2.40 � (SD 0.91) on the right side and 2.71 � (SD 1.02) on the left (Figure 3).Good to excellent agreement was observed between MOTI and the camera system for RoM, HRA and QoM during head rotations, with ICCs ranging between 0.75 and 0.99, Table 1.MOTI measured between 0.42 � and 0.66 � higher RoM and HRA between 0.40 � and 0.56 � higher during head rotations compared to the camera system, Table 1, and Figure 3. MOTI also measured a higher QoM of 13-21.9A.U. during head rotations compared to the camera system.Bland-Altman's plots supported this overestimation tendency and showed that MOTI systematically measured greater RoM, HRA, and QoM values during head rotations than the camera system (Figure 4).
Correlation between each of the three trials for each side were found to be large for RoM and QoM during head rotations while it was medium to large for HRA (see supplementary tables).

Discussion
The present study is the first to investigate the agreement of MOTI compared to a 3D marker-based camera system.The findings indicate good to excellent agreement between the measurement systems for RoM, HRA and QoM during head rotations, thereby showing comparable results on all investigated parameters.

Assessing head movements
Examining cervical RoM, including rotation, is an important part of a physical assessment and provides valuable information to the clinician [3,11].This study found a bilateral mean RoM of approximately 77.5 � of rotation and a mean HRA of approximately 2.6 � (Figure 3) irrespectively of the recording device.The current values are in line with what would be expected within a healthy population with regard to RoM in rotation [3,38] and HRA [4,5].
While approximations of RoM can be made visually, a goniometer will improve the outcome [21].The CROM goniometer has shown a high degree of correlation compared to a 3D camera system for rotation of the head (r ¼ 0.89-0.94)with a mean bias of 1.7-2.2degrees depending on the side of rotation [39], similar to the current findings.However, MOTI has several advantages compared to analogue goniometers.MOTI allows for assessing not only RoM but also HRA and QoM.While this can be assessed using a smartphone, the added weight of the phone and equipment needed to ensure the phone is stable and yet possible to operate during recordings may influence the results [17].Had the hearing protection not been used in the current study, the setup would have been only 70.7 g.Furthermore, if MOTI had been used on its own, attached with a double-adhesive tape, the weight would be only 22 g as compared to a smartphone weighing easily þ150 g excluding the setup to mount it on the head increasing both weight and impracticality [17,40].With this in mind, MOTI may be a better option for assessing RoM, HRA and QoM than a smartphone, not only based on the current performance compared to the previous study [17] but also based on the potential influence of the weight of the equipment.Taken together, MOTI provides comparable results to that of a 3D marker-based camera system for assessing RoM, HRA and QoM during neck rotations.These findings are encouraging with regards to implementing this technology in clinical practice.This applies especially to the ability to accurately  assess QoM in populations such as people with neck pain, which is something that is currently not commonly available in a clinical setting.

Limitations
A limitation of the MOTI system is that it relies on a single measurement unit and therefore has difficulty calculating angles over a single joint if both body segments that the joint connects move relative to each other.Hence, it is important to ensure that only one part of the body (the head in this case) is moving relative to the rest of the body.
The current study addressed the potential influence of body movement on neck rotation by adding a chest strap to minimise body movement during recordings.Such limitations are no different from those clinicians' face when employing handheld goniometers in clinical practice.Furthermore, MOTI cannot assess the movement of individual cervical vertebrae but only that of the head relative to the starting position.However, this would apply to all types of goniometers except for using video-fluoroscopy, which allows tracking movements of specific vertebrae in 2D [41].
Another possible limitation in the current study is that only movements in the transverse plane (rotation) were investigated.Although movement planes should not impact data recordings from either the 3D camera system or MOTI, this cannot be confirmed based on the current study.Future studies should therefore investigate movements in the sagittal and coronal planes to cover all movements normally included in a clinical assessment of the cervical spine.In addition, an evaluation of how MOTI compares to a clinical cervical assessment is warranted.
As it was not possible to use either the 3D camera setup or MOTI as a trigger for the other system, the recordings had to be started manually.However, as these were initiated simultaneously by the same person, any potential difference in timing is not believed to have played any difference for the investigated parameters.Additionally, the participants were instructed to initiate the head movement after the recordings had started.The recordings were first stopped when they returned to their self-perceived neutral position.Furthermore, if the timing of recordings impacted the data, any potential discrepancy between systems would likely have resulted in poorer ICC values than seen here.
The present results showed that when compared to the 3D camera system, MOTI tended to overestimate the output systematically.However, the difference of 0.4-0.8� is modest in the context of measuring RoM.A likely explanation for this overestimation could be a drift in the sensor, a known issue with inertial measurement units [42,43], and this error would also apply to smartphones as both devices employ the same sensor technology.With this in mind, using inertial measurement units may be less feasible for long recordings as the drift will increase over time.For optimal performance, measurements should be recorded one repetition at a time, similar to using analogue goniometry.

Conclusion
This study found that MOTI, an inertial measurement unit, could accurately assess RoM, HRA and QoM during head rotations when compared to a 3D camera system.The results are promising and suggest that RoM, HRA, and QoM can be assessed in a clinical setting with comparable results to what was previously only possible in a research lab equipped with a 3D camera system.

Figure 1 .
Figure 1.Study setup.(A) Schematic overview of the setup optimised for assessment of a left rotation.(B) The participant is seated in a comfortable position on a chair, fixed with a chest strap, wearing a blindfold, hearing protection, a headband mounted with 3D markers and MOTI.

Figure 2 .
Figure 2. Example of raw data.Example of raw data from 3D camera system (red dotted line) and MOTI (blue line) during a movement cycle from the beginning and to the end of right rotation.Data for angle (A), velocity (B), acceleration (C) and jerk (QoM) is depicted as a % of the movement cycle.

Figure 3 .
Figure 3. Mean values for recorded data.Showing mean values for range of motion (RoM), head repositioning accuracy (HRA), jerkiness/quality of motion (QoM) for the 3D camera system (Gray dots) and MOTI (Black dots) in left and right rotation.

Figure 4 .
Figure 4. Bland-Altman plots.Bland-Altman plots showing mean bias (solid line) along with upper and lower limits of agreement (dotted lines) for range of motion (RoM), head repositioning accuracy (HRA) and normalised jerk/quality of movement (QoM).