A computational algorithm for classifying step and spin turns using pelvic center of mass trajectory and foot position

Transient changes in direction during ambulation are typically performed using a step (outside) or spin (inside) turning strategy, often identified through subjective and time-consuming visual rating. Here, we present a computational, marker-based classification method utilizing pelvic center of mass (pCOM) trajectory and time-distance parameters to quantitatively identify turning strategy. Relative to visual evaluation by three independent raters, sensitivity, specificity, and overall accuracy of the pCOM-based classification method were evaluated for 90-degree turns performed by 3 separate populations (5 uninjured controls, 5 persons with transtibial amputation, and 5 persons with transfemoral amputation); each completed turns using two distinct cueing paradigms (i.e., laser-guided “freeform” and verbally-guided “forced” turns). Secondarily, we compared the pCOM-based turn classification method to adapted versions of two existing computational turn classifiers which utilize trunk and shank angular velocities (AV). Among 366 (of 486 total) turns with unanimous intra- and inter-rater agreement, the pCOM-based classification algorithm was 94.5% accurate, with 96.6% sensitivity (accuracy of spin turn classification), and 93.5% specificity (accuracy of step turn classification). The pCOM-based algorithm (vs. both AV-based methods) was more accurate (94.5% vs. 81.1-80.6%; P < 0.001) overall, as well as specifically in freeform (92.9 vs. 80.4-76.8%; P < 0.003) and forced (96.0 vs. 83.8-81.8%; P < 0.001) cueing, and among individuals with (92.4 vs. 80.2-78.8%; P < 0.001) and without (99.1 vs. 86.2-80.8%; P < 0.001) amputation. The pCOM-based algorithm provides an efficient and objective method to accurately classify 90-degree turning strategies using optical motion capture in a laboratory setting, and may be extended to various cueing paradigms and/or populations with altered gait.