This work is presented to trying to solve for an open issue that afflicts actual Brain computer interface loops, I.e the impossibility to correct an erroneous command delivered by the decoder. It is known, in fact, that when an erroneous event is presented to a person, the brain elicits a specific and well known signal, called Error-related Potential (ErrP), composed by a complex of positive and negative deflections (N200, P300, N400). So, the aim of the thesis is to provide a method able to detect and decode Error potentials in EEG signals in order to use them as trigger to correct an erroneous decision taken by a BCI model during a Cybathlon race. From the next Cybathlon games edition will be, in fact, possible to implement this type of correction. This type of solution is, however, extendible to other applications beyond the Cybathlon game. To trying to achieve the desired goal, an offline protocol is adopted, in which the subjects control the game through a joystick, instead of through the BCI. This approach was thought in order to have a better control over the total amount of wrong commands and over the nature of commands delivered by subjects. Data are analyzed both in time and frequency domain and then used to train a SVM classifier. Finally the accuracy, sensitivity and specificity of the model are taken into account as results to be discussed. In conclusion, this solution can make the BCI usage more easy and accurate, with respect to the subject’s real motor intentions.
This work is presented to trying to solve for an open issue that afflicts actual Brain computer interface loops, I.e the impossibility to correct an erroneous command delivered by the decoder. It is known, in fact, that when an erroneous event is presented to a person, the brain elicits a specific and well known signal, called Error-related Potential (ErrP), composed by a complex of positive and negative deflections (N200, P300, N400). So, the aim of the thesis is to provide a method able to detect and decode Error potentials in EEG signals in order to use them as trigger to correct an erroneous decision taken by a BCI model during a Cybathlon race. From the next Cybathlon games edition will be, in fact, possible to implement this type of correction. This type of solution is, however, extendible to other applications beyond the Cybathlon game. To trying to achieve the desired goal, an offline protocol is adopted, in which the subjects control the game through a joystick, instead of through the BCI. This approach was thought in order to have a better control over the total amount of wrong commands and over the nature of commands delivered by subjects. Data are analyzed both in time and frequency domain and then used to train a SVM classifier. Finally the accuracy, sensitivity and specificity of the model are taken into account as results to be discussed. In conclusion, this solution can make the BCI usage more easy and accurate, with respect to the subject’s real motor intentions.
Decoding error potentials during the Cybathlon BCI Race game
VIVIAN, MARCO
2021/2022
Abstract
This work is presented to trying to solve for an open issue that afflicts actual Brain computer interface loops, I.e the impossibility to correct an erroneous command delivered by the decoder. It is known, in fact, that when an erroneous event is presented to a person, the brain elicits a specific and well known signal, called Error-related Potential (ErrP), composed by a complex of positive and negative deflections (N200, P300, N400). So, the aim of the thesis is to provide a method able to detect and decode Error potentials in EEG signals in order to use them as trigger to correct an erroneous decision taken by a BCI model during a Cybathlon race. From the next Cybathlon games edition will be, in fact, possible to implement this type of correction. This type of solution is, however, extendible to other applications beyond the Cybathlon game. To trying to achieve the desired goal, an offline protocol is adopted, in which the subjects control the game through a joystick, instead of through the BCI. This approach was thought in order to have a better control over the total amount of wrong commands and over the nature of commands delivered by subjects. Data are analyzed both in time and frequency domain and then used to train a SVM classifier. Finally the accuracy, sensitivity and specificity of the model are taken into account as results to be discussed. In conclusion, this solution can make the BCI usage more easy and accurate, with respect to the subject’s real motor intentions.File | Dimensione | Formato | |
---|---|---|---|
Vivian_Marco.pdf
accesso riservato
Dimensione
38.1 MB
Formato
Adobe PDF
|
38.1 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/36799