In a four-patient RAM PAL1 pilot, study-phase SEEG band power did not predict cued recall consistently

We analyzed study-phase stereo-electroencephalography (SEEG) band power as a predictor of later cued recall in OpenNeuro ds005059, the RAM PAL1 word-pair task. We screened 13 session-0 participants using pre-specified rules; four had at least 10 SEEG pairs and at least 15 recalled and 15 not-recalled pairs. Across leave-one-list-out tests, an L2 logistic decoder with fixed penalty C = 7.2e-4 had mean AUC 0.534 (3 of 4 participants above 0.5; exact two-sided sign-flip p = 0.625). A fixed mean z(70–150 Hz) minus z(4–8 Hz) index had mean AUC 0.495 (2 of 4 above 0.5; p = 0.875). With n = 4, the smallest possible two-sided sign-flip p is 0.125, so this test cannot show a group effect at alpha = 0.05. Neither method predicted cued recall consistently in this pilot; these results do not show that PAL1 recall cannot be decoded. The planned 200-permutation label control was not run because of the work limit. Only session 0 was used, and the sample is small. Screening suggests that about one in three of the 13 sessions had both enough SEEG contacts and balanced recall. The two dense-SEEG participants showed a descriptive power pattern opposite to the usual free-recall effect; this observation involves only two participants and was not tested.

stereo-electroencephalography · cued recall · paired-associate learning · memory decoding · band power · RAM PAL1 · OpenNeuro · leave-one-list-out cross-validation