Estimating scapular kinematics from individualized regression models: Effects of humeral input parameters

Individualized linear regression (ILR) models have been used to estimate subject-specific scapular kinematics from humeral kinematics. This study examines how the number of training poses and the kinematic representation of the humeral predictor variables affects the estimation of scapular kinematics from an ILR model. Twenty-two healthy individuals (12 females / 10 males) participated in this study. ILR models were developed from 11 or 6 training poses, and with the humeral predictor variables expressed as helical angles, Euler XZY angles or Euler ZXY angles. Six ILR models were defined for each participant: Helical 11, Helical 6, Euler XZY 11, Euler XZY 6, Euler ZXY 11, and Euler ZXY 6. The ability of each ILR model to predict scapular kinematics during 5 novel test poses and 3 dynamic movements was examined. Comparisons were made against palpation and an acromial marker cluster (AMC). Root mean square errors (RMSEs) of estimated scapular joint angles indicated that the ILR models developed from 11 training poses were more accurate than their 6 pose counterparts. The Helical 11 model was the only ILR model whose RMSEs were < 10° across all test poses and axes. Waveform comparisons between the ILR models and the AMC during the dynamic movements highlighted plane-specific instability in the Euler ILR models. Expressing humeral predictor variables as helical angles may lead to more robust and consistent ILR models, as they will not be affected by rotation sequence, accumulation of error across axes or gimbal lock.