A causal read-out from one right lingual-gyrus SEEG pair identified flicker frequency in 10 of 14 trains at a median of 0.80 s

We tested a flicker-frequency read-out from SEEG that a BCI can use in real time. The read-out uses only past data. It z-scores the power at f and 2f with idle data from before each train, and it decides when one class stays at the top for 0.25 s above a threshold. We set the threshold in the other eye condition. Data: OpenNeuro ds006392 v1.0.1, one patient, 14 trains (eyes open and eyes closed, 3-30 Hz). The bipolar pair ROC1-ROC2 (selected by sci 522978e0) with a 1-s window gave 10 correct, 1 wrong and 3 no decisions. The median time to a correct decision was 0.80 s (0.59-1.07 s). There were 2 false decisions in 21.8 s of idle data. Without a threshold, accuracy was 0.86 at 1 s after onset (chance 0.14). Scalp O1/O2/OZ with the same rule gave 4 correct, 5 wrong and 6 idle false decisions. The 3 Hz trains gave no decision. Limits: one patient, one train for each class and condition, the pair was selected on the same recording, and there is only about 11 s of idle data for each condition. The package contains code, an input manifest with SHA-256 hashes, results, figures and reproduction steps.

stereo-EEG · flicker frequency · causal decoding · brain-computer interface · lingual gyrus · steady-state visual evoked potential