Premio per Tesi di Laurea sul tema della Disabilità o dei Disturbi Specifici dell'Apprendimento (DSA)

Neuromuscular disorders, such as stroke, often result in impaired motor control affecting the patient’s autonomy and lifestyle. Nowadays, Brain-Computer Interfaces (BCIs) represent a gold-standard in motor rehabilitation protocols based on brain plasticity. A BCI is a system that measures central nervous system activity and converts it into an action, providing a new artificial channel to convey messages and commands directly from the brain to the external world. BCIs can restore functions in patients with severe motor impairments, enhancing signals from a damaged cortical area and detecting user’s motor intentions with a very short delay according to the Hebbian principle. This allows to send an effective feedback to the patient, inducing brain plasticity and improving functional recovery. The design is mostly suited to counteract the maladaptive motor re-learning. Since stroke causes connectivity changes between the central and the peripheral nervous system, researchers developed a hybrid BCI (hBCI) based on the interplay between brain and muscular activities in order to enhance the functional motor recovery after stroke. For these reasons, both brain and hybrid features can be used to control a hBCI defining different paradigms for rehabilitation. In this thesis, three control features have been investigated: the Cortico-Muscular Coherence (CMC), the Sensorimotor Rhythms (SMR) and the Movement-Related Cortical Potential (MRCP). While the CMC is a hybrid feature that measures the synchronization between primary motor cortex and peripheral muscles involved in a motor task, both the SMR and the MRCP are brain features. SMR (or ERD) consists in a modulation of the EEG activity within certain frequency oscillations that occur during a motor task. On the other hand, MRCP is a time-domain potential generated during the preparation and planning of a motor task: MRCP peak negativity is associated with movement onset. The chosen and developed pipelines were able to extract CMC and SMR patterns and MRCP waveforms. The three control features were analyzed and characterized using three different spatial filtering conditions: no filtering, Common Average Reference (CAR) filter and Laplacian filter. The performances of four classification algorithms were compared to evaluate their ability in discriminating hand extension from hand grasping executed by 13 healthy subjects. The classification results allowed to establish the best spatial filter and the best combination classifier-number of features for each control feature. No filtering promoted the best trade-off between high performances and low computational cost (few electrodes) for CMC, the Laplacian filter with its peculiar spatial blurring attenuation resulted the best one for SMR, while the CAR filter was able to extract the best MRCP waveforms in terms of recognizability and minor variance around the EMG onset. Moreover, couple SVM-2 was chosen as the best combination classifier-number of features for all the features. The final comparison analysis between the three control features confirmed CMC-based BCI systems as the best ones in the discrimination between simple upper limb movements, such as hand extension and hand grasping. However, the SMR-based SVM-2 classifier with Laplacian filter showed a very good classification quality. Further analysis with a larger group of participants would be needed to increase the statistical power and make the results more consistent and accurate. Moreover, the pipelines followed for the three different control features might be extended to pathological subjects in order to allow a comparison with the healthy ones.