Date of Award
Doctor of Philosophy (PhD)
Modeling Simul & Visual Engineering
Frederic D. McKenzie
This study investigates the thesis that given cerebral response samples of an individual's left, right, both, and imagined finger tapping, continuous wave (CW) functional Near Infrared (fNIR), unregistered with fMRI, can differentiate between any two of the four categories.
Fifty subjects were outfitted with a single source/detector attached to a single, square pad, affixed to their heads using devices such as elastic bands and caps for light shielding. Slides depicting arrows pointing left, right, both directions, or made of dashed lines were presented to each subject, with a slide of text interspersed between each. Subjects tapped with their left finger, right finger, both left and right finger, or imagined tapping, depending on the type of arrow. Text was presented in between each tapping slide and was read with no tapping. Each slide was presented for twenty seconds and each type of tapping occurred three times in an eight minute, 20 second period.
Classification was performed using Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and decision tree algorithms. Results indicated that left finger tapping can be distinguished from right, both, and imagined right finger-tapping with error rates ranging from 24.92% to 29.51% (SVM), 40.05% to 42.69% (LDA), and 23.34% to 28.85% (decision tree). The decision tree algorithm produced results, on an individual trial basis, with greater than 95% confidence that the results were not due to chance.
These results were obtained with no screening out due to individual characteristics such as hair thickness. The generalizations included the use of a large sample of subjects for which the selection criteria only included statutory minimum and maximum ages.
This study also produced validation of a method of mitigating hair effect. Raising the sensor was shown to still produce valid results that could not be attributed to chance at a confidence level of 95%.
The results are directly applicable to brain-computer interfaces in a number of areas. These relate to validating the ability to classify data collected by a device with a single source/detector, from non-prescreened individuals, with real-time algorithms in a normal environment.
Stoudenmire, Eugene A..
"Functional Near Infrared Detection of Real and Imagined Finger Taps Using Support Vector Machine, Linear Discriminant Analysis, and Decision Tree Classification Methods"
(2013). Doctor of Philosophy (PhD), dissertation, Modeling Simul & Visual Engineering, Old Dominion University, DOI: 10.25776/d9da-0551