Fig. 2. LDA classifier identified leads that carry decision information.

a Flatmaps showing, for all recorded leads across six patients whether a significant response-locked change from baseline was detected (blue dots) and/or the left/right decision was successfully decoded (red circles), or whether their activity did not differ from baseline in the interval starting 0.5 s before and ending 0.15 s after the response (white dots). Brain areas of interest are indicated by white outlines. The insets show the t-scored classification performance of four example left/right decoding leads (indicated by I–IV on the flatmaps). Dotted lines indicate the onset of temporal cluster of the significant classification. For full labeling of areas see Supplementary Fig. 1; b average spectral power per area (number of leads given in brackets), including only those leads that showed significant decoding of the left/right decision. To combine data recorded in both hemispheres, data were contrasted as contralateral (i.e., left button presses when recording in the right hemisphere) versus ipsilateral decisions (i.e., right button presses when recording in the right hemisphere); c average correlation per area between spectral power and classification performance across all left/right decoding leads. Correlations were computed using a 100 ms wide sliding window. Green indicates that power increase related to high classification performance and purple indicates that power decrease co-occurred with high classification performance. In b and c, black lines give the average response-locked classifier performance across leads in the area, normalized to the peak performance (y-scale between 0 and 1). Non-scaled performances per lead are given in Supplementary Fig. 4, stimulus-locked and response-locked classifier performance are compared in Supplementary Fig. 5, and spectral power and power-classifier correlations for all other areas of interest can be found in Supplementary Figs. 6–9.