An artificial neural network predicts knee loading using 3d marker trajectories of anatomical landmarks

Knee osteoarthritis (KOA) is a painful and debilitating disease that is accelerated by excessive loading in the joint. Personalized gait modifications that reduce the first peak of the knee adduction moment (P1KAM), can improve joint pain [3], but require visits to a laboratory equipped with expensive motion capture cameras and force plates. In this study, we used 3D marker trajectories of anatomical landmarks measured during walking with KAM-altering gait modifications to predict the P1KAM. We analyzed 125,415 steps from 98 subjects walking on an instrumented treadmill and computed a true P1KAM using inverse dynamics. An artificial neural network (ANN) model predicted the P1KAM with a test r2 value of 0.80. Given recent advancements in 3D joint position tracking using 2D video, our results suggest a method to evaluate the effect of personalized gait modifications on P1KAM in any clinical setting, without the use of force plates.