Scoliosis assessment from trunk rotation based on torsobarography using machine learning

Abstract: An Angle of Trunk Rotation (ATR) over 7° is a common indicator of scoliosis. Torsobarography as a novel surface topographic system for radiation-free posture analysis is capable to quantify the asymmetry of the trunk. This study investigates on 71 subjects the potential of torsobarography to differentiate between ATR < 7° and ATR ≥ 7° using machine learning algorithms. A Cohen's Kappa κ of 0.76 was achieved for an optimized classifier model. However, a generalizability verification revealed an average κ of 0.54. In conclusion, the rotation of the trunk can be characterized by machine learning algorithms based on torsobarographic images.