File Download
There are no files associated with this item.
Supplementary
-
Citations:
- Appears in Collections:
Conference Paper: Neurofeedback improves SSVEP BCI performance on subjects with both 'high' and 'low' performance
Title | Neurofeedback improves SSVEP BCI performance on subjects with both 'high' and 'low' performance |
---|---|
Authors | |
Issue Date | 2018 |
Publisher | Brain-Computer Interface. |
Citation | Seventh International BCI Meeting: BCIs: Not Getting Lost in Translation, Pacific Grove, CA, May 21-25, 2018. How to Cite? |
Abstract | Introduction: Our previous work has demonstrated that the SSVEP BCI users with low performance (i.e.
classification accuracy <80%) can be improved by alpha down-regulation neurofeedback training (NFT)
[1]. However, it is unclear whether the NFT benefits on a larger population with both 'high' and 'low'
initial performances. Furthermore, whether the improvement of BCI performance after NFT is
predictable using initial BCI performances is an important question for subject selection in the NFT.
Thus, this study aims to answer the above two questions. Material, Methods and Results: In total 12
subjects (age: 28 ± 6 years, 4 females) including 8 subjects with high classification accuracy completed
the 10 NFT sessions in two consecutive days. The experiment protocol and settings remain consistent
with previous work [1]. Results show that IAB is reduced during NFT and the protocol is effective to
improve the BCI performance for subjects with both 'high' and 'low' initial performances in SSVEP trials
in analysis time length starting form 1s to 3.5s. Using 3s analysis time length of SSVEP flashing as an
example (see Fig. 1), a paired t-test revealed a significant improvement on both the SSVEP signal SNR
(t(11) = 4.168, p = 0.002) and the BCI classification accuracy (t(11) = 3.310, p = 0.007) for all subjects.
More specifically, the 'high' performance group showed an average increase of 11.9% in the SSVEP
signal SNR and an average increase of 15.11% in the BCI classification accuracy. For the 'low'
performance group, the signal SNR has increased 16% and the BCI classification accuracy has increased
45.29%. Furthermore, Spearman correlation test showed that the percentage change of accuracy after
NFT was significantly correlated with initial BCI accuracy (r=-0.796, p=0.002) in 3s analysis, and such
correlation was also found in the analysis data with 2s, 2.5s and 3.5s. Discussion: The results indicate
that NFT can benefit wide range users with both 'low' and 'high' BCI performances. However, the
improvement of BCI accuracy in 'high' group is much smaller than the 'low' group, which may be
explained by ceiling effect and one outlier (accuracy increased from 48% to 94%) in total 4 subjects with
'low' initial performance. Significance: This study verifies that alpha down-regulation neurofeedback
training can improve the SSVEP BCI performances for subjects with both high and low initial BCI
performances, and the improvement of BCI accuracy is predictable using initial BCI accuracy. |
Persistent Identifier | http://hdl.handle.net/10722/260788 |
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Tang, Q | - |
dc.contributor.author | Nan, W | - |
dc.contributor.author | Wan, F | - |
dc.contributor.author | Hu, Y | - |
dc.date.accessioned | 2018-09-14T08:47:25Z | - |
dc.date.available | 2018-09-14T08:47:25Z | - |
dc.date.issued | 2018 | - |
dc.identifier.citation | Seventh International BCI Meeting: BCIs: Not Getting Lost in Translation, Pacific Grove, CA, May 21-25, 2018. | - |
dc.identifier.uri | http://hdl.handle.net/10722/260788 | - |
dc.description.abstract | Introduction: Our previous work has demonstrated that the SSVEP BCI users with low performance (i.e. classification accuracy <80%) can be improved by alpha down-regulation neurofeedback training (NFT) [1]. However, it is unclear whether the NFT benefits on a larger population with both 'high' and 'low' initial performances. Furthermore, whether the improvement of BCI performance after NFT is predictable using initial BCI performances is an important question for subject selection in the NFT. Thus, this study aims to answer the above two questions. Material, Methods and Results: In total 12 subjects (age: 28 ± 6 years, 4 females) including 8 subjects with high classification accuracy completed the 10 NFT sessions in two consecutive days. The experiment protocol and settings remain consistent with previous work [1]. Results show that IAB is reduced during NFT and the protocol is effective to improve the BCI performance for subjects with both 'high' and 'low' initial performances in SSVEP trials in analysis time length starting form 1s to 3.5s. Using 3s analysis time length of SSVEP flashing as an example (see Fig. 1), a paired t-test revealed a significant improvement on both the SSVEP signal SNR (t(11) = 4.168, p = 0.002) and the BCI classification accuracy (t(11) = 3.310, p = 0.007) for all subjects. More specifically, the 'high' performance group showed an average increase of 11.9% in the SSVEP signal SNR and an average increase of 15.11% in the BCI classification accuracy. For the 'low' performance group, the signal SNR has increased 16% and the BCI classification accuracy has increased 45.29%. Furthermore, Spearman correlation test showed that the percentage change of accuracy after NFT was significantly correlated with initial BCI accuracy (r=-0.796, p=0.002) in 3s analysis, and such correlation was also found in the analysis data with 2s, 2.5s and 3.5s. Discussion: The results indicate that NFT can benefit wide range users with both 'low' and 'high' BCI performances. However, the improvement of BCI accuracy in 'high' group is much smaller than the 'low' group, which may be explained by ceiling effect and one outlier (accuracy increased from 48% to 94%) in total 4 subjects with 'low' initial performance. Significance: This study verifies that alpha down-regulation neurofeedback training can improve the SSVEP BCI performances for subjects with both high and low initial BCI performances, and the improvement of BCI accuracy is predictable using initial BCI accuracy. | - |
dc.language | eng | - |
dc.publisher | Brain-Computer Interface. | - |
dc.relation.ispartof | International BCI Meeting | - |
dc.title | Neurofeedback improves SSVEP BCI performance on subjects with both 'high' and 'low' performance | - |
dc.type | Conference_Paper | - |
dc.identifier.email | Hu, Y: yhud@hku.hk | - |
dc.identifier.authority | Hu, Y=rp00432 | - |
dc.identifier.hkuros | 290218 | - |
dc.publisher.place | Pacific Grove, CA | - |