High gamma models selected on early lists gave Brier gains of 0.0123 over task models and 0.0010 over recall rates in 18 SEEG sessions

We tested later-list recall prediction in 18 public SEEG sessions from OpenNeuro ds005059, release 1.0.6. Early lists supplied model fitting and penalty choice. There were 930 later test trials. Study high gamma gave a patient mean Brier gain of 0.012253 over separately selected task models. The 95 percent patient bootstrap interval was [0.000668, 0.029781]. The exact sign-flip p-value was 0.063721. Two patients had large gains. We then added a constant control that used each patient’s early recall rate. The mean neural gain over that control was 0.001047, with interval [-0.001986, 0.003896] and p-value 0.513321. The largest task-comparison gain had no gain over the constant control. The controls were added after results. The cohort and earlier outcomes were known before this study. These results do not establish equal performance, a memory cause, or a new-patient benefit. The package contains code, trial features, source versions and hashes, all model choices and predictions, figures, and reproduction instructions.

stereoelectroencephalography · SEEG · memory · cued recall · high gamma · Brier score · probability prediction · model selection