Single-Trial Decoding of Motion Direction During Visual Attention From Local Field Potential Signals

Brain-Computer Interface (BCI) based on Local Field Potential (LFP) has recently been developed to restore communication or behavioral functions. LFP provides comprehensive information, due to its stability, robustness, and reach frequency content within the cognitive process. It has been demonstrated that spatial attention can be decoded from brain activity in the visual cortical areas. However, whether motion direction can be decoded from the LFP signal in the primate visual cortex remains uninvestigated, as well as how decoding performance may be influenced by spatial attention. In this paper, these issues were examined by recording LFP from the middle temporal area (MT) of macaque, employing machine learning algorithms. The animal was trained to report a brief direction change in a target stimulus which moved in various directions during a visual attention task. It was found that the LFP-gamma power was able to provide significant information to reliably decode motion direction, compared with other frequency bands, on a single-trial basis. Moreover, the results show that spatial attention leads to enhancements in motion direction discrimination performance. The highest decoding performance was achieved in the high-gamma frequencies (60-120Hz) when targets were presented inside the receptive field in opposite directions. Using a feature selection approach, performance was improved by optimally selecting features where the highest level of participation was observed in the gamma-band. Generally, the results suggest that in the MT area, LFP signals exhibit appreciable information about visual features like motion direction, which could thus be utilized as a control signal for cognitive BCI systems.