3D free reaching movement prediction of upper-limb based on deep neural networks

Quantitative assessment of motor disorder is one of the main challenges in the field of stroke rehabilitation. This paper proposes a simplified kinematic model for human upper limb(UL) using seven main joints of both the dominant and non-dominant side. With this model, a deep neural network (DNN) is used to predict the 3D free reaching movement of UL of a healthy participant. The experimental results show that the prediction trajectories can achieve high similarities with trajectories of real movements, indicating the promising accuracy in 3D movement estimation of UL achieved by the DNN. With the capability of identifying specific reaching movements in realtime, the trajectories predicted by this data-driven model can be utilized to inform the rehabilitation assessment and training in the future studies as a personalized therapy approach.