• EEG signals
• closed-loop neuroprosthetics
• event-related potentials
• visual stimuli decoding

Electroencephalography (EEG) is a non-invasive method for the recording of brain signals. Cortical signals, such as those obtained through EEG, allow for a comparison between natural and prosthetic senses and may thus become very useful instruments to close the loop in neuroprosthetics for visual restoration. However, the low spatial resolution of EEG makes the decoding of low-level aspects of a visual stimulus highly challenging. While the decoding of images in semantic categories is possible, no study has ever attempted to decode structural low-level properties, such as the location or size of a visual stimulus.

In this thesis, we propose a novel experiment to determine whether EEG signals can be used to decode the size, position, and orientation of visual stimuli. We perform an exploratory and statistical analysis of event related potentials (ERPs) to investigate how different subjects respond to the same stimulus and how the response varies when diverse stimuli are presented. The classification of EEG signals is carried out with support vector machines (SVMs). We perform single-trial decoding and introduce a multi-trial procedure, implemented to improve performances when using EEG signals in a translational setting. We conclude by performing a multi-subject classification to study how the model performs on unseen individuals.