Reliability and agreement of Azure Kinect and Kinect v2 depth sensors in the shoulder joint range of motion estimation

Abstract

Background: Depth sensor–based motion analysis systems are of interest to researchers with low cost, fast analysis capabilities, and portability; thus, their reliability is a matter of interest. Our study examined the agreement and reliability in estimating the basic shoulder movements of Azure Kinect, Microsoft’s state-of-the-art depth sensor, and its predecessor, Kinect v2, by comparing them with the gold standard marker-based motion analysis system. Methods: In our study, the shoulder joint ranges of motion of 20 healthy individuals were analyzed during dominant-side flexion, abduction, and rotation movements. The reliability and agreement between methods were evaluated using the intraclass correlation co- efficient (ICC) and the Bland-Altman method. Results: Compared to the gold standard method, the old- and new-generation Kinect showed similar performance in terms of reliability in the estimation of flexion (ICC ¼ 0.86 vs. 0.82) and abduction (ICC ¼ 0.78 vs. 0.79) movements, respectively. In contrast, the new- generation sensor showed higher reliability than its predecessor in internal (ICC ¼ 0.49 vs. 0.75) and external rotation (ICC ¼ 0.38 vs. 0.67) movement. Conclusion: Compared to its predecessor, Kinect Azure has higher reliability in analyzing movements in a lower range and variability, thanks to its state-of-the-art hardware. However, the sensor should also be tested on multiaxial movements, such as combing hair, drink- ing water, and reaching back, which are the tasks that simulate extremity movements in daily life.