Single trial detection of hand poses in human ECoG using CSP based feature extraction

Kapeller, Christoph; Schneider, Christoph; Kamada, Kazuhiro; Kunii, Naoto; Ortner, Rupert; Prückl, Robert; Guger, Christoph

doi:10.1109/embc.2014.6944648

Cited by 16 publications

(22 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Yanagisawa et al (2011) reported 68.3% in one subject for three different hand postures. For the same experiment, Kapeller et al (2014) published accuracies up to 95.9% averaged over two subjects, and Li et al (2017) scored on average 80.0% in three subjects. In a similar setup, Xie et al (2015) obtained correct classification rates of 92.7% over three hand gestures in four subjects on average.…”

Section: Introductionmentioning

confidence: 89%

“…Common spatial patterns (CSPs) are the de-facto standard for dimension reduction in EEG signal processing (Blankertz et al, 2008; Lotte et al, 2015) and are also popular in ECoG signal processing (Onaran et al, 2011; Kapeller et al, 2014; Xie et al, 2015; Jiang et al, 2017). This approach expects multivariate distributions of two classes and with covariances Σ and Σ , respectively.…”

Section: Methodsmentioning

confidence: 99%

“…Invasive BCIs that operate on high-gamma based features have gained considerable attention and are subject to intensive research. For example, this encompasses real-time passive functional mapping in the course of surgery planning (Ogawa et al, 2014), visual categorization tasks (Kapeller et al, 2018), or the development of BCI prototypes for prosthetic limb, hand, or finger control (Shenoy et al, 2007; Kubánek et al, 2009; Onaran et al, 2011; Yanagisawa et al, 2011; Pistohl et al, 2012; Chestek et al, 2013; Kapeller et al, 2014; Xie et al, 2015; Bleichner et al, 2016; Hotson et al, 2016; Branco et al, 2017; Jiang et al, 2017; Li et al, 2017; Pan et al, 2018).…”

Section: Introductionmentioning

confidence: 99%

“…A variety of classifiers for both offline and real-time BCIs exist. Besides linear programming machines (Shenoy et al, 2007), Bayesian approaches (Chestek et al, 2013), pattern matching (Bleichner et al, 2016; Branco et al, 2017; Kapeller et al, 2018), neural networks (Pan et al, 2018), and support vector machines (Onaran et al, 2011; Yanagisawa et al, 2011; Li et al, 2017), linear discriminant analysis (LDA) is widely used for both non-invasive and invasive BCI and all types of features (Bostanov, 2004; Scherer et al, 2004; Blankertz et al, 2008, 2011; Hoffmann et al, 2008; Prueckl and Guger, 2009; Onaran et al, 2011; Yanagisawa et al, 2011; Pistohl et al, 2012; Kapeller et al, 2014; Xu et al, 2014; Lotte et al, 2015; Xie et al, 2015; Hotson et al, 2016; Gruenwald et al, 2017a; Jiang et al, 2017; Li et al, 2017). LDA is robust, has low complexity due to linearity and performs well in line with more sophisticated methods (Garrett et al, 2003; Lee et al, 2005; Lotte et al, 2007).…”

Section: Introductionmentioning

confidence: 99%

“…If the dimension of the feature space is high, a spatial filter must be employed to reduce the number of features and to prevent the classifier from overfitting. The most straightforward approach is feature selection, either manual from a-priori data inspection or automatized via statistical algorithms (Kapeller et al, 2014; Xie et al, 2015; Bleichner et al, 2016; Hotson et al, 2016; Li et al, 2017; Pan et al, 2018). Another approach for feature reduction in invasive and non-invasive bandpower-based BCIs is common spatial patterns (CSPs), a linear projection scheme that optimizes class separation within a pre-defined window (Blankertz et al, 2008; Onaran et al, 2011; Wu et al, 2013; Kapeller et al, 2014, 2018; Lotte et al, 2015; Gruenwald et al, 2017a).…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Time-Variant Linear Discriminant Analysis Improves Hand Gesture and Finger Movement Decoding for Invasive Brain-Computer Interfaces

et al. 2019

View full text Add to dashboard Cite

Invasive brain-computer interfaces yield remarkable performance in a multitude of applications. For classification experiments, high-gamma bandpower features and linear discriminant analysis (LDA) are commonly used due to simplicity and robustness. However, LDA is inherently static and not suited to account for transient information that is typically present in high-gamma features. To resolve this issue, we here present an extension of LDA to the time-variant feature space. We call this method time-variant linear discriminant analysis (TVLDA). It intrinsically provides a feature reduction stage, which makes external approaches thereto obsolete, such as feature selection techniques or common spatial patterns (CSPs). As well, we propose a time-domain whitening stage which equalizes the pronounced 1/f-shape of the typical brain-wave spectrum. We evaluated our proposed architecture based on recordings from 15 epilepsy patients with temporarily implanted subdural grids, who participated in additional research experiments besides clinical treatment. The experiments featured two different motor tasks involving three high-level gestures and individual finger movement. We used log-transformed bandpower features from the high-gamma band (50–300 Hz, excluding power-line harmonics) for classification. On average, whitening improved the classification performance by about 11%. On whitened data, TVLDA outperformed LDA with feature selection by 11.8%, LDA with CSPs by 13.9%, and regularized LDA with vectorized features by 16.4%. At the same time, TVLDA only required one or two internal features to achieve this. TVLDA provides stable results even if very few trials are available. It is easy to implement, fully automatic and deterministic. Due to its low complexity, TVLDA is suited for real-time brain-computer interfaces. Training is done in less than a second. TVLDA performed particularly well in experiments with data from high-density electrode arrays. For example, the three high-level gestures were correctly identified at a rate of 99% over all subjects. Similarly, the decoding accuracy of individual fingers was 96% on average over all subjects. To our knowledge, these mean accuracies are the highest ever reported for three-class and five-class motor-control BCIs.

show abstract