Investigating the possibilities of using singing imagery to enhance EEG-based active BCIs

Mohammadpour, Hadi (2023) Investigating the possibilities of using singing imagery to enhance EEG-based active BCIs. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Active brain-computer interfaces are a novel communication/interaction pathway that relies on the performance of imagery tasks. The patterns of brain activity associated with these tasks are detected and decoded as commands for operating an external device. This study focuses on augmenting the practicality of such systems by investigating the mental task of singing imagery. Singing imagery is the simple act of imagining singing a song in your head. Despite its straightforward nature, the potential of singing imagery as an alternative task for active BCIs or for increasing their number of commands has yet to be thoroughly investigated. The research described in this thesis comprises two phases. In the first study, singing imagery is combined with the commonly used imagery tasks in BCI research (i.e., 4- and 5-class combinations consisting of the imagined movement of the left hand, right hand, feet, and tongue, as well as a \rest" state). Filter bank common spatial patterns algorithm and the random forest classifier are utilized to incorporate a singing imagery task in the 2-, 3-, 4-, and 5-class combinations. These analyses resulted in comparable classification accuracies to conventional motor imagery tasks. Hence, based on the survey results, singing imagery could be considered as a potentially more intuitive alternative mental task. Furthermore, singing imagery may also be a practical approach for increasing the number of commands to six, where accuracies as high as 60.7% were achieved. The second study investigated the potential of using \dual imagery" tasks (i.e., the simultaneous performance of two single tasks, in this case, singing imagery and one of the conventional motor imagery tasks) as additional BCI control tasks. Here, the 3- and 4-class analyses of the dual tasks and their constituent single tasks (alongside a \rest" state for the 4-class) were carried out to verify the possibility of differentiating them. Using an extended version of filter bank common spatial patterns and regularized linear discriminant analysis classifiers, average accuracies as high as 64.1% and 63% were achieved for the 3, and 4-class scenarios, respectively. Next, the dual imagery tasks were combined with conventional single motor imagery tasks to investigate increasing the number of commands to seven or eight. As a result, for the 7- and 8-class scenarios, accuracies as high as 55.4%, and 50.5%, which are well above the corresponding chance levels of 14.3% and 12.5%, were obtained. Increasing the number of commands a BCI can recognize is important as it can significantly impact the user's experience with the device. Specifically, a BCI with a more intuitive list of commands can help the user avoid a high mental workload. Moreover, a higher number of commands can be helpful by allowing users to communicate with a higher information transfer rate. Based on the results of this thesis research, singing imagery appears to be a potentially viable solution for improving active BCIs.

Actions (login required)

View Item