Ultrasound Based Wrist Intent Recognition Method for Robotic-Assisted Stroke Rehabilitation

Rehabilitation is the most effective procedure for the stroke patients to regain their physical skills and improve activities of daily living. Recovering upper limb function after stroke requires intensive rehabilitation under the guidance of physical therapists, a costly and protracted process. Rehabilitation protocols that can be performed using robotic systems remotely at home with minimal assistance would decrease the cost of rehabilitation while reducing recovery time. Exoskeleton based robotic-assisted rehabilitation devices that can deliver high-intensity high-frequency training have been recently introduced. Such systems can be used independently without supervision of physical therapist, utilizes actuators and kinematic sensors to improve voluntary wrist movement of the stroke survivor while interacting with a goal-oriented interface. Although it has been clinically shown to improve functional abilities, motivation, and commitment to the rehabilitation programs, it requires users to have some degree of voluntary movement on their upper limb. This limits severely impaired stroke survivors who have very limited or even no motion on their limbs to benefit from these robotic-assisted systems. In this study, we present an ultrasound imaging-based method which can augment rehabilitation assistance capabilities of robotic systems by providing continuous motion intent recognition of the wrist. We implemented a recurrent neural network-based model which classifies extension, flexion and no movement wrist sequences with 87.10% accuracy.