Patient tailored virtual rehabilitation

Abstract

Virtual rehabilitation should be adaptable to the patient need and progress. To do so, patient in-game performace and ability are monitored to maintain an adequate level of challenge. A novel adaptation strategy is proposed by which patient control and speed are dynamically interrogated to adjust the game difficulty. The strategy is based on a Markov decision process seeding a therapist-guided reinforcement learning algorithm. The optimal learning scheme for the algorithm is established (α = 0.5). Convergence to an optimal therapeutic plan is demonstrated for patients with non-deterministic behaviour. The proposed adaptation algorithm can enhance existing virtual reality-based motor rehabilitation platforms by tailoring the games response to the patient changing needs.