Direct information transfer rate optimisation for SSVEP-based BCI

Abstract

In this work, a classification method for SSVEP-based BCI is proposed. The classification method uses features extracted by traditional SSVEP-based BCI methods and finds optimal discrimination thresholds for each feature to classify the targets. Optimising the thresholds is formalised as a maximisation task of a performance measure of BCIs called information transfer rate (ITR). However, instead of the standard method of calculating ITR, which makes certain assumptions about the data, a more general formula is derived to avoid incorrect ITR calculation when the standard assumptions are not met. This allows the optimal discrimination thresholds to be automatically calculated and thus eliminates the need for manual parameter selection or performing computationally expensive grid searches.The proposed method shows good performance in classifying targets of a BCI, outperforming previously reported results on the same dataset by a factor of 2 in terms of ITR. The highest achieved ITR on the used dataset was 62 bit/min. The proposed method also provides a way to reduce false classifications, which is important in real-world applications.

Information Bottleneck as Optimisation Method for SSVEP-Based BCI

“…Therefore, under the assumption that the differences in ITR are normally distributed, the better performance of Classifier 2 is statistically significant. The same comparison between Classifier 2 and the approach of Ingel et al (2018) did not show statistical significance (p = 0.097). We also evaluated the classifier of Ingel et al (2018) with the added pre-processing step of subtracting Cz channel from O1 and O2 channels, and we replaced their original gradient descent algorithm with a more stable basinhopping algorithm (Wales and Doye, 1997).…”

“…Tables 4-6 show that in some cases, the proposed algorithm is outperformed by the arg max classifier and by the algorithm described by Ingel et al (2018). We assume this is because the choice of bins was not optimal.…”

“…In particular, we subtract the Cz channel signal from the O1 and O2 channels as the first step. The rest of the feature extraction is identical to Ingel et al (2018). In particular, we use channels O1 and O2 from the dataset, and we use PSDA (Cheng et al, 2002) and CCA (Lin et al, 2007) feature extraction methods.…”

“…To calculate mean detection time (MDT), we follow the approach of Ingel et al (2018). In particular, we estimate MDT as follows:…”

“…For example, this dataset has been used by Zhang et al (2018b), Zhang et al (2018a), Jiang et al (2018), , Wong et al (2020), and they report average ITRs up to 230 bits/min. Ingel et al (2018) proposed an algorithm for finding optimal rule from a fixed set of threshold-based classification rules. In particular, they use a threshold-based classification rule which has a threshold for each feature.…”