Errors associated with bin boundaries in observation-based posture assessment methods

Abstract

The trunk posture misclassification errors made by novice and experienced operators were quantified as a function of the angular distance from posture bin boundaries, similar to those used in observation-based posture assessment tools such as 3DMatch. The effect that these misclassification errors had on cumulative and peak low back loads was also determined in three simulated lifting scenarios. Ninety subjects in 3 experience groups were randomly presented with images of known trunk angle via a monitor. Subjects were instructed to make quick and accurate bin selections using standardized pictures included below the images on the monitor. Mean % bin misclassification errors were approximately 32% and 22% for the flexion/extension and lateral bend views, respectively. More bin classification errors were made the closer a viewed image was to a posture bin boundary, regardless of expertise level, and the number of errors made decreased as operator experience increased. Approximately 99% of bin selections were made either in the correct bin or in the bins immediately adjacent to the correct bin in both views. Misclassification errors made in the 3 simulated lifting scenarios induced errors in peak and cumulative loads in 66% of the cases assessed, with an average absolute difference of 13.5% across all load variables. Future work is aimed at determining the effect of training and bin size on the error misclassification rate for all body segments and views.