Abstract
Measurements of shoulder kinematics during activities of daily living (ADL) can be used to evaluate patient function before and after treatment and help define device testing conditions. However, due to the difficulties of making 3D motion measurements outside of laboratory conditions, there are few reports of measured shoulder 3D kinematics during ADL. The purpose of this study was to demonstrate the feasibility of using wearable inertial measurement units (IMUs) to track shoulder joint angles.
A nonrandom sample of 5 subjects with normal shoulders was selected based on occupation. The occupations were: dental hygienist, primary school teacher, mechanical project engineer, administrative assistant, and retail associate. Subjects wore two OPAL IMUs (APDM, Portland OR) as shown in Figure 1 on the sternum and on the upper arm for approximately 4 hours while at their workplace performing their normal work place activities and then up to 4 hours while off-work.
Orientation angles from IMUs have traditionally been estimated by integrating gyroscope data and calculating inclination angles relative to gravity with accelerometers. A significant problem is that inaccuracies inherent in the measurements can degrade accuracy. In this study, we used an Unscented Kalman Filter (UKF) with IMU output to track shoulder angles. The UKF mitigates the effect of random drift by incorporating domain knowledge about the shoulder normal range of motion, and the gyroscope and accelerometer characteristics into the state-space models. Initially, in the horizontal plane, without gravity measurements from the accelerometer to aid the gyroscope data, there were unacceptable errors in transverse rotation. To mitigate this error, additional constraints were applied to model gyroscope drift and a zero velocity update strategy was included. These additions decreased tracker errors in heading by 63%. The resulting accuracy with the modified tracker in all motion planes was about 2° (Figure 2).
Subjects commented that the IMUs were well tolerated and did not interfere with their ability to perform tasks in a normal manner. The overall averaged 95th percentile angles (Figure 3) were: flexion 128.8°, adduction 128.4°, and external rotation 69.5°. These peaks angles are similar to other investigator's reports using laboratory simulations of ADL tasks measured with optical and electromagnetic technologies, though this study's observations did show 17% greater extension and 40% greater adduction. Additionally, in these observations, occurrences of maximal internal rotation were rare compared to maximal external rotation and when maximum external rotation did occur, it was in combination with an average flexion angle of 103°. Finally, by performing a Fourier transform of the arm angles and using the 50th percentile frequency the number of arm cycles in a 10 year period was calculated at over 600,000 cycles.
Application of the UKF with the additional drift correction made substantial improvements in shoulder tracking performance and this feasibility data suggests that IMUs with the UKF are suitable for extended use outside of laboratory settings. The motion data collected provides a novel description of arm motion during ADLs including estimating the cycle count of the upper arm at more than 600,000 cycles over 10 years.
