Development of machine learning models for wearable sensors fusion to improve home-based rehabilitation after stroke

One of the repercussions of stroke is the paralysis of the upper limb, which affects the majority of patients and limits their quality of life. Recent research shows that stroke patients at any stage, even chronic, benefit from high-quality and high-dose rehabilitation training. In stroke patients with upper-limb paresis, the main issue is that the kinematic activity is small and noisy, while the muscular activity is still present. The electromyographic activity appears even if there is no movement because the contractions are not large enough to evoke a voluntary movement, especially if there is spasticity that provokes involuntary muscle contractions. At Ramoslab (http://www.ramoslab.org ) in Tübingen a novel system of wearables was developed, which is able to record parameters of patients' movements using multiple sensing modalities and locations. The system is mobile and wireless and can thus be used in a home setting enabling such training. The project is called ReHome. During the internship at Ramoslab, in this thesis work, a model of the movements of the upper limb of stroke patients is going to be developed, exploiting machine learning techniques. The data involved are collected from the kinematic sensors (inertial measurement units, IMU), recorded at multiple locations, and electromyographic (EMG) readings, capturing residual muscle activity from these patients. The main goal of this thesis is to create a model that can predict the IMU derivatives (angles, speeds, and accelerations) from the EMG data, in 12 healthy subjects and in one stroke patient. This model will be later used in an uncontrolled environment by stroke patients to perform home-based rehabilitation. The model will help to provide feedback to the patients, guiding the therapists to customize their rehabilitation and increase the independence of the patient at home.