Single-trial classification of antagonistic oxyhemoglobin responses during mental arithmetic

Near-infrared spectroscopy (NIRS) is a non-invasive optical technique that can be used for brain-computer interfaces (BCIs) systems. A common challenge for BCIs is a stable and reliable classification of single-trial data, especially for cognitive (mental) tasks. With antagonistic activation pattern, recently found for mental arithmetic (MA) tasks, an improved online classification for optical BCIs using MA should become possible. For this investigation, we used the data of a previous study where we found antagonistic activation patterns (focal bilateral increase of [oxy-Hb] in the dorsolateral prefrontal cortex in parallel with a [oxy-Hb] decrease in the medial area of the anterior prefrontal cortex) in eight subjects. We used the [oxy-Hb] responses to search for the best antagonistic feature combination and compared it to individual features from the same regions. In addition, we investigated the use of antagonistic [deoxy-Hb], total hemoglobin [Hbtot] and pairs of [oxy-Hb] and [deoxy-Hb] features as well as the existence of a group-related feature set. Our results indicate that the use of the antagonistic [oxy-Hb] features significantly increases the classification accuracy from 63.3 to 79.7%. These results support the hypothesis that antagonistic hemodynamic response patterns are a suitable control strategy for optical BCI, and that only two prefrontal NIRS channels are needed for good performance.