Limits...
EEG resolutions in detecting and decoding finger movements from spectral analysis.

Xiao R, Ding L - Front Neurosci (2015)

Bottom Line: These new features reveal some similar, but more different spatial and spectral patterns as compared with classic mu/beta rhythms.Decoding results further indicate that these new features (91%) can detect finger movements much better than classic mu/beta rhythms (75.6%).The capability in decoding fingers (and hand gestures in the future) from EEG will contribute significantly to the development of non-invasive BCI and neuroprosthesis with intuitive and flexible controls.

View Article: PubMed Central - PubMed

Affiliation: School of Electrical and Computer Engineering, University of Oklahoma Norman, OK, USA.

ABSTRACT
Mu/beta rhythms are well-studied brain activities that originate from sensorimotor cortices. These rhythms reveal spectral changes in alpha and beta bands induced by movements of different body parts, e.g., hands and limbs, in electroencephalography (EEG) signals. However, less can be revealed in them about movements of different fine body parts that activate adjacent brain regions, such as individual fingers from one hand. Several studies have reported spatial and temporal couplings of rhythmic activities at different frequency bands, suggesting the existence of well-defined spectral structures across multiple frequency bands. In the present study, spectral principal component analysis (PCA) was applied on EEG data, obtained from a finger movement task, to identify cross-frequency spectral structures. Features from identified spectral structures were examined in their spatial patterns, cross-condition pattern changes, detection capability of finger movements from resting, and decoding performance of individual finger movements in comparison to classic mu/beta rhythms. These new features reveal some similar, but more different spatial and spectral patterns as compared with classic mu/beta rhythms. Decoding results further indicate that these new features (91%) can detect finger movements much better than classic mu/beta rhythms (75.6%). More importantly, these new features reveal discriminative information about movements of different fingers (fine body-part movements), which is not available in classic mu/beta rhythms. The capability in decoding fingers (and hand gestures in the future) from EEG will contribute significantly to the development of non-invasive BCI and neuroprosthesis with intuitive and flexible controls.

No MeSH data available.


Related in: MedlinePlus

Confusion matrices of finger movement decoding using combined spectral features. Row and column labels are same as in Figure 6.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4555046&req=5

Figure 7: Confusion matrices of finger movement decoding using combined spectral features. Row and column labels are same as in Figure 6.

Mentions: In Figures 6, 7, confusion matrices of five fingers movements from individual or combined spectral features are illustrated. The rows of these matrices stand for predicted condition labels, while the columns represent actual condition labels. For features from individual PCs, similar performances are achieved in all individual PCs and actually moved fingers were dominantly and correctly identified in the confusion matrices (diagonal elements with larger values than off-diagonal elements). Furthermore, the misclassifications are spread almost evenly in four fingers other than the actual one (off-diagonal elements with similar low values). Considering different fingers, thumb and little seem usually better classified than other fingers. For features from alpha and beta bands, only thumb is classified with relatively high accuracies, while the decoding accuracies of other fingers are close to the guessing level (i.e., 20%). Moreover, other four fingers are all confused to thumb, which might be the reason for thumb having high decoding accuracy. Spectral features from PCs show obvious better performance than features from mu/beta PSDs (best mean decoding accuracy in each category: 33.1 vs. 23.4%).


EEG resolutions in detecting and decoding finger movements from spectral analysis.

Xiao R, Ding L - Front Neurosci (2015)

Confusion matrices of finger movement decoding using combined spectral features. Row and column labels are same as in Figure 6.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4555046&req=5

Figure 7: Confusion matrices of finger movement decoding using combined spectral features. Row and column labels are same as in Figure 6.
Mentions: In Figures 6, 7, confusion matrices of five fingers movements from individual or combined spectral features are illustrated. The rows of these matrices stand for predicted condition labels, while the columns represent actual condition labels. For features from individual PCs, similar performances are achieved in all individual PCs and actually moved fingers were dominantly and correctly identified in the confusion matrices (diagonal elements with larger values than off-diagonal elements). Furthermore, the misclassifications are spread almost evenly in four fingers other than the actual one (off-diagonal elements with similar low values). Considering different fingers, thumb and little seem usually better classified than other fingers. For features from alpha and beta bands, only thumb is classified with relatively high accuracies, while the decoding accuracies of other fingers are close to the guessing level (i.e., 20%). Moreover, other four fingers are all confused to thumb, which might be the reason for thumb having high decoding accuracy. Spectral features from PCs show obvious better performance than features from mu/beta PSDs (best mean decoding accuracy in each category: 33.1 vs. 23.4%).

Bottom Line: These new features reveal some similar, but more different spatial and spectral patterns as compared with classic mu/beta rhythms.Decoding results further indicate that these new features (91%) can detect finger movements much better than classic mu/beta rhythms (75.6%).The capability in decoding fingers (and hand gestures in the future) from EEG will contribute significantly to the development of non-invasive BCI and neuroprosthesis with intuitive and flexible controls.

View Article: PubMed Central - PubMed

Affiliation: School of Electrical and Computer Engineering, University of Oklahoma Norman, OK, USA.

ABSTRACT
Mu/beta rhythms are well-studied brain activities that originate from sensorimotor cortices. These rhythms reveal spectral changes in alpha and beta bands induced by movements of different body parts, e.g., hands and limbs, in electroencephalography (EEG) signals. However, less can be revealed in them about movements of different fine body parts that activate adjacent brain regions, such as individual fingers from one hand. Several studies have reported spatial and temporal couplings of rhythmic activities at different frequency bands, suggesting the existence of well-defined spectral structures across multiple frequency bands. In the present study, spectral principal component analysis (PCA) was applied on EEG data, obtained from a finger movement task, to identify cross-frequency spectral structures. Features from identified spectral structures were examined in their spatial patterns, cross-condition pattern changes, detection capability of finger movements from resting, and decoding performance of individual finger movements in comparison to classic mu/beta rhythms. These new features reveal some similar, but more different spatial and spectral patterns as compared with classic mu/beta rhythms. Decoding results further indicate that these new features (91%) can detect finger movements much better than classic mu/beta rhythms (75.6%). More importantly, these new features reveal discriminative information about movements of different fingers (fine body-part movements), which is not available in classic mu/beta rhythms. The capability in decoding fingers (and hand gestures in the future) from EEG will contribute significantly to the development of non-invasive BCI and neuroprosthesis with intuitive and flexible controls.

No MeSH data available.


Related in: MedlinePlus