The detection of mental states is fundamental for many applications as brain-computer interface (BCI). Currently, in the majority of works, features related to the activity of single brain regions is used, as power spectral density. Performances of BCI are non-optimal and a percentage of 20-30% of users is not able to use the interface. For these reasons, there is an increasing interest in developing new metrics and algorithms, focusing, in the majority of cases, in the classification block. On the contrary, the aim of the project is introducing a new feature for the control of brain computer interface. The proposed feature is the connectivity, that describes the functional interaction between different brain areas and it can be descriptive of complex tasks as motor imagery.
To the end of integrating connectivity metrics in BCI, the features extraction has to be online and it is necessary to develop a time-varying estimate of connectivity. For this reason, this project focuses in developing a method for the estimation of connectivity neglecting the hypothesis of stationarity of the brain signals, which are strongly non-stationary because of the rapid changing of user¿s brain activity to to control the BCI.
In this project, I intend to introduce a new feature to improve brain-computer interface. In fact, the performances are not optimal and a non-negligible percentage of users is not able to use it. For this reason, there is a growing interest in developing algorithms to increase results, which can possibly enlarge the use of these interfaces. My project is focusing in feature extraction block, introducing connectivity as feature to control BCI. In fact, connectivity describes the interactions between brain areas and this information can characterize brain reorganization, consequence of complex tasks as motor imagery ones. To this end, a time-varying connectivity estimation is necessary to provide an online control of the interface.
In this project, fields of signal processing and neuroscience are both necessary. The first is necessary to define the robust estimators, simulate EEG data and to realize the time-varying algorithm. On the other side, the latter is essential to interpret and analyze results in a neurophysiological perspective.
The fusion of the two fields and perspectives is achieved thanks to the collaborations of two laboratories, the Aramis team-Algorithms, models and methods for images and signals of the brain, (Brain and Spine Institute-Paris) and DIET department ( La Sapienza-Rome). Being a first year PhD student of a joint doctorate program between Edite de Paris at Sorbonne Université and PhD ICT at la Sapienza, I will foster the collaboration of the labs, with the goal of studying brain signals from a signal processing point of view, without losing the clinical perspective. The collaboration already gave results with an accepted publication for Eusipco 2018 [20].
[20] Cattai T., Colonnese S., Corsi M.-C., Bassett D.S., Scarano G., De Vico Fallani F., "Characterization of mental States through Node Connectivity between Brain Signals." European Signal Processing Conference (EUSIPCO), 2018 IEEE International Conference on. IEEE, 2017 (Accepted for publication)