Automatic and Personalized Adaptation of Therapy Parameters for Unsupervised Robot-Assisted Rehabilitation: a Pilot Evaluation

Growing evidence shows that increasing the dose of upper limb therapy after stroke might improve functional outcomes and unsupervised robot-assisted therapy may be a solution to achieve such an increase without adding workload on therapists. However, most of existing robotic devices still need frequent supervision by trained personnel and are currently not designed or ready for unsupervised use. One reason for this is that most rehabilitation devices are not capable of delivering and adapting personalized therapy without external intervention. Here we present a set of clinically-inspired algorithms that automatically adapt therapy parameters in a personalized way and guide the course of robot-assisted therapy sessions. We implemented these algorithms on a robotic device for hand rehabilitation and tested them in a pilot study with 5 subacute stroke subjects over 10 robot-assisted therapy sessions, some of which unsupervised. Results show that our algorithms could adapt the therapy difficulty throughout the whole study without requiring external intervention, maintaining performance around a predefined 70% target value (mean performance for all the subjects over all the sessions: 64.5%). Moreover, the algorithms could guide patients through the therapy sessions, minimizing the number of actions that subjects had to learn and perform. These results open the door to the use of robotic devices in an unsupervised setting to increase therapy dose after stroke.