The current study aims to examine the level of accuracy of the Oculus Rift during cervical spine movement, as well as to investigate the amount of error contributed by different error sources. Overall, the results indicate that using the Oculus Rift to measure cervical spine mobility during the task performed in this study could provide an approximate estimate of the RoM of lateral bending, axial rotation and flexion/extension. However, the measurement error of unilateral RoM can be greater than the error of the full RoM (Table 1). This is mainly related to the drifting error of the inertial sensor as well as the movement of the trunk during the task. While such error sources exist, in the case of one side RoM being over/ underestimated and the other side RoM being under/overestimated, and then the entire RoM could be less affected. The average trace of ISðtÞRGoðtÞ over time showed a prominent drift for approximately 61 in the first 3 s at the beginning of a trial. The drift over the rest of the time until the end of a trial was approximately 21. As the Rift was designed to deliver a VR environment to gamers, it is important that the content displayed in the Rift and the Rift both share the same orientation in the global coordinate system of the physical environment. In order to achieve that, the Rift artificially drifts the orientation of the inertial sensor to eliminate the accelerometer-based tilting angle in the few seconds after the initialization of the Rift (LaValle, 2013). However, this feature may positively or negatively impact the measured cervical spine kinematics. When the trunk and the head are well aligned along the vertical y-axis, this artificial drift can be helpful in correcting the error due to the initial misalignment between the Rift and the head, and the orientation of the inertial sensor at time t should be calculated relative to the time that the artificial drift has ceased. If the trunk and the head cannot be vertically aligned, such as when a user lies on an incline, this feature could introduce more errors to cervical spine kinematics. It should be noted that
the advantage of the Rift is to provide an affordable immersive VR environment for the gaming industry, and the provided customized
algorithm to track the goggles movement is not necessarily optimized for monitoring cervical spine mobility. If a VR environment is not
included in a study, one can use multiple inertial sensors and functional calibrations to derive a better accuracy for cervical spine
measurement, as proposed by Duc et al. (2014). There was minimal trunk movement involved in this study (Fig. 2a). It should be noted, however, that the task performed in the VR environment did not require the participants to move their head to an extreme posture since it was designed based on 80% of the mean neck RoM. During the full RoM movement, the trunk is more likely to move and the movement of the trunk can contribute more errors to the Rift-based cervical spine kinematics. In addition, each task was only performed for 1 min. For a longer duration of data collection, the cumulative trunk movement could be greater and
result in more error. Therefore, the trunk orientation needs to be well fixed during the data collection to improve the accuracy of the Riftbased cervical spine kinematics. The trace of TðtrÞRGoðtrÞ and GoðtÞRHðtÞ indicates that there were some misalignments between the Rift and the trunk at the beginning of the task, and between the Rift and the head. Further analysis indicates that between the Rift and the trunk, the Rift deviated from the trunk coordinate system by 1.71 (3.11) in lateral bending, 1.01 (6.51) in axial rotation, and 0.01 (8.21) in flexion/extension. This suggests that the misalignment between the Rift and the trunk at the beginning of the task varied between individuals but was unbiased over all the participants. Between the Rift and the head, the Rift
deviated from the head coordinate system by 0.81 (2.21) in lateral bending, 0.81 (2.41) in axial rotation, and 5.91 (5.21) in flexion/
extension. Such results suggest that, compared with the head coordinate system, the Rift flexed down for most of the participants.
This is possibly because the Rift sometimes was slightly flexed down when mounting it on the head in order to have a more secure contact with the floor of the orbits. There are some limitations that need to be addressed. First, the task performed in the current study only involved moderate head movement. Given that the inertial sensor error is influenced by the magnitude of angular velocity (Guo and Zhong, 2013), the accuracy level of using the Rift to measure cervical spine mobility can be altered when the head movement is more drastic. Second, due to the moderate head movement, it was assumed that the head and the Rift moved together,
and the marker clusters placed on the Rift was used to track the movements of the head and the Rift. With a more drastic head movement, relative movement between the head and the Rift can be expected, which would introduce additional error to the measurement of cervical spine mobility. Third, the current study used the head movement relative to the trunk to represent cervical spine mobility. The orientation of each cervical spine vertebra, however, remains unclear. In conclusion, when using the Oculus Rift to measure cervical spine kinematics, it is recommended that the head and trunk be vertically aligned, that the front cover of the Rift is along a vertical plane at the time that the Rift is initialized, and the trunk remains stationary during the task. These steps will be helpful for minimizing errors in Rift-based cervical spine kinematics measurement.