TY - GEN
T1 - Optimizing steady-state visual evoked potential classifiers for high performance and low computational costs in brain-computer interfacing
AU - Kæseler, Rasmus Leck
AU - Struijk, Lotte N. S. Andreasen
AU - Jochumsen, Mads Rovsing
PY - 2021/12
Y1 - 2021/12
N2 - While assistive robotic devices can improve the quality of life for individuals with tetraplegia, it is difficult to provide a high-performing interface that can be fully utilized, with little to no motor functionality. While a brain-computer interface (BCI) can be used with little to no motor functionality, it typically has a low performance. Steady-state visually evoked potentials (SSVEP) provide some of the best performing signals for a BCI, but are rarely investigated for online asynchronous control where not only accuracy is important, but also the computational costs. This study investigates and compares three classifiers: the well-known and high-performing task-related component analysis (TRCA), the computational efficient Spatiotemporal beamformer (STBF) build on the stimulus-locked inter-trace correlation (SLIC) algorithm and our proposed novel algorithm which combines the two: the SLIC-TRCA. Results show the SLIC-TRCA achieving higher accuracies (95.00±5.36% with a 1s classification window) compared to the TRCA (88.25±14.58%) and similar compared to the STBF (91.00±11.02%) while having a much lower computational cost (519% faster than the TRCA and 144% faster than the STBF). We, therefore, believe this algorithm has an exciting potential as it will allow a high classification accuracy without requiring a high-performing CPU.
AB - While assistive robotic devices can improve the quality of life for individuals with tetraplegia, it is difficult to provide a high-performing interface that can be fully utilized, with little to no motor functionality. While a brain-computer interface (BCI) can be used with little to no motor functionality, it typically has a low performance. Steady-state visually evoked potentials (SSVEP) provide some of the best performing signals for a BCI, but are rarely investigated for online asynchronous control where not only accuracy is important, but also the computational costs. This study investigates and compares three classifiers: the well-known and high-performing task-related component analysis (TRCA), the computational efficient Spatiotemporal beamformer (STBF) build on the stimulus-locked inter-trace correlation (SLIC) algorithm and our proposed novel algorithm which combines the two: the SLIC-TRCA. Results show the SLIC-TRCA achieving higher accuracies (95.00±5.36% with a 1s classification window) compared to the TRCA (88.25±14.58%) and similar compared to the STBF (91.00±11.02%) while having a much lower computational cost (519% faster than the TRCA and 144% faster than the STBF). We, therefore, believe this algorithm has an exciting potential as it will allow a high classification accuracy without requiring a high-performing CPU.
U2 - 10.1109/BIBE52308.2021.9635303
DO - 10.1109/BIBE52308.2021.9635303
M3 - Article in proceeding
SN - 978-1-6654-4262-6
T3 - International Conference on Bioinformatics and Bioengineering
BT - IEEE 21st International Conference on Bioinformatics and Bioengineering (BIBE)
PB - IEEE
T2 - 21st IEEE International Conference on BioInformatics and BioEngineering, BIBE 2021
Y2 - 25 October 2021 through 27 October 2021
ER -