Non-invasive Brain Computer Interface (BCI) technology can be incredibly important for those who are no longer able to communicate to due loss of muscle control. However, in comparison to other methods of non-muscular communication, such as an eye tracker, the BCI method of communication can be slow. Therefore it is important to implement techniques that can improve both accuracy and speed of BCI performance. Currently, the P300 Speller allows communication at a rate up to eight selections per minute. Given this relatively slow rate of communication highly accurate classification is of great importance. One method of improving accuracy in BCI performance has been the presentation of familiar faces rather than traditional letter flashes or character flashes. Previous studies have shown that the use of faces elicits both an N400 and N170 component in conjunction with the P300 component, resulting in greater speed and accuracy of letter selection. Functional MRI studies have shown that images of familiar locations produce unique brain responses located in distinct brain regions (i.e. parahippocampal place area). These two types of stimuli (images of faces and images of familiar locations) were incorporated into simultaneous two-stimulus presentation paradigm with the intention of developing two distinct classifiers to discriminate between the cognitive responses produced by the spatially disparate areas. By developing stimulus specific classifiers, the BCI system could eliminate half of the characters in the matrix as potential selections, thus reducing the amount of error in performance. The present study aims to provide proof of concept. Ten able-bodied participants completed one experimental session consisting of two calibration phases. Each participant completed two calibration phases: one for face stimuli and one for house stimuli. In each calibration phase, participants spelled three-six letter words using an 8x9 matrix of alphanumeric characters and symbols. During each calibration phase, only one of the two images were presented to the participant (i.e., faces only or houses only). Participants then completed a copy spelling portion that consisted of spelling six words containing six characters each. During this time, participants received feedback in regard to the accuracy of the BCI systems character selections. Participants were instructed to continue spelling the word without correcting errors made by the BCI system. Offline analyses were conducted to examine BCI accuracy, selections per minute, and bitrate for stimulus specific classifiers. Results indicated no significant differences in accuracy; however, results showed a significant interaction of classifier (i.e., face or house classifier) by image type (i.e., face or house) for selections per minute and bitrate. Specifically, the house classifier produced higher selections per minute and bit rate when applied to house data than the house classifier applied to face data, and vice versa for the face classifier. These results indicate that stimulus specific classifiers may be able to eliminate half of the characters located in the matrix as potential character selections, thereby increasing overall BCI performance.
Identifer | oai:union.ndltd.org:ETSU/oai:dc.etsu.edu:etsu-works-2-1898 |
Date | 01 April 2017 |
Creators | Gardner, Aaron, Kellicut-Jones, Marissa R., Kazmark, Ashley, Sellers, Eric W. |
Publisher | Digital Commons @ East Tennessee State University |
Source Sets | East Tennessee State University |
Detected Language | English |
Type | text |
Source | ETSU Faculty Works |
Page generated in 0.0013 seconds