Abstract
Previous electrophysiological studies have demonstrated that single neurons in area 8a of macaques are involved in the planning and execution of saccadic eye movements. However, most of these studies have recorded from one neuron at a time and pooled neuronal responses over multiple non-simultaneous single trial presentations of the same condition. This procedure assumes that responses of single neurons are independent (non-correlated), an assumption that has proven to be erroneous. Here, we implanted 96-channels multielectrode arrays in the area 8a of two macaques and recorded single and multiunit activity while they performed a visually guided saccade task to one out of four simultaneously presented targets. Simultaneous neuronal spiking activity aligned to the initiation of the saccade was inputted into a support vector machine (SVM) algorithm to optimize a model predicting to which quadrant of the screen the monkey would saccade. Including all neurons, the model achieved 67% accuracy in predicting saccade end point. Including only neurons with significant spatial tuning increased the model accuracy to 94%. Interestingly, the analysis period that yielded the best accuracy spanned from saccade initiation to 80 msec before, longer or shorter time intervals slightly diminishing model performance. In order to assess the effect of the simultaneity of the neuronal activity on model performance, we destroyed simultaneity by shuffling trials’ identity within experimental conditions, which concomitantly abolishes noise correlations. Surprisingly, noise-correlations-free activity yielded significantly better prediction accuracy both when all neurons were included (+9%) and when only spatially tuned neurons were selected (+4%). These preliminary results suggest that neuronal population activity in the macaque area 8a can be decoded to predict stereotyped saccade end point with high accuracy, and that noise correlations have a slight detrimental effect on population decoding using a SVM classifier.
Meeting abstract presented at VSS 2013