Muscle intent-based continuous passive motion machine in a gaming context using a lightweight CNN

This paper presents a novel approach to control and actuate a Continuous Passive Motion (CPM) machine by integrating a deep learning-based control strategy using convolutional neural networks in a gaming context for providing post-surgical therapy and knee rehabilitation. Electromyography and inertial measurement unit sensors are interfaced with the patient's thigh muscles to record the patient's intent signals and classify them as three states: forward, backward, and rest. Comparison studies have been performed to prove the novelty of the proposed lightweight convolutional neural network architecture over other architectures and machine learning methodologies for real-time implementation. Additionally, gaming software has been interfaced, making the recovery process motivating to deal with the psychological aspects of rehabilitation. A low-cost, ecofriendly alpha prototyped CPM machine is prototyped for implementing the algorithms. Experiments are performed on three healthy subjects to establish the feasibility of this home rehabilitation device under professional guidance. Thus, this study aims to improve home-based knee rehabilitation effectiveness, offering complete recovery to the patients, delivering intensive and motivational rehabilitation.