In 40 RAM FR1 SEEG patients, a 5-band global model trained in session 0 did not predict session-1 word recall (AUC 0.508) or outperform the pair model
We tested whether mean spectral power across matched SEEG pairs predicts word recall across two RAM FR1 sessions. For 40 patients, a ridge model with five global band-power features trained on session 0 and tested on session 1 had mean AUC 0.508 (95% CI 0.492–0.522; 22/40 above 0.5; Holm-adjusted p = 0.68). Its AUC minus that of a pair-level model was −0.005 (95% CI −0.024 to 0.013; Holm-adjusted p = 0.68); the pair model mean AUC was 0.513. Exploratory within-session global-model AUCs were 0.511 and 0.499. The person-level global subsequent-memory effect was not stable across sessions, while recall rate was stable (r = 0.81, p < 0.0001). In this sample and with equal weighting of pairs, a 0–1.6 s window, and fixed ridge regularization, global spectral features did not predict recall across or within sessions. The sample was 40 patients; effects smaller than AUC 0.53 are not excluded. Regional weighting and other time-frequency features were not tested.
intracranial EEG · subsequent memory effect · word recall · spectral power · cross-session transfer · ridge regression · memory encoding
References
3- supportsSEEG Pair-Level Subsequent Memory Maps Did Not Recur Across Sessions in 40 Patients sci:sha256:d645a896595d4d59648dc061ae2134791e89cbd7dca4009d4e598435d419b8ed
Its pair-model cross-session AUC (0.513) is reproduced. A global all-pair model also does not transfer across sessions.
- usesSEEG Pair-Level Subsequent Memory Maps Did Not Recur Across Sessions in 40 Patients sci:sha256:d645a896595d4d59648dc061ae2134791e89cbd7dca4009d4e598435d419b8ed
We use its session-1 FR1 features, its subject set and its pair-level ridge model as the comparator.
- usesIn 83 pure-SEEG patients, mean HFA and theta slopes were not significant, and HFA did not explain primacy sci:sha256:f677a6fd27b2fd81b0e22487cc5ecd4440063dedc4bacf3999b8bdd88c1974c0
We use its session-0 FR1 band-power features (features/*.npz) and its artifact rule.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- supportsIn 84 same-implant RAM FR1 patients, 60-second SEEG spectral changes did not track free-recall changes across sessions
Also finds that the spectral state does not predict FR1 recall across sessions; here we test the within-patient change.
- challengesAcross 48 SEEG patients, a model trained on other patients predicted word recall above chance (mean AUC 0.549)
737f9be8 found that a global 5-band model trained inside one patient did not predict recall (AUC about 0.51). Here a global 5-band model trained on 47 other patients predicts recall above chance, also without serial position (AUC 0.535). The features are different, so the challenge is limited.
- challengesSample/fs timing reveals cross-session SEEG memory-effect map repeatability in 15 patients
With sample/fs timing, a session-0-trained ridge model predicts session-1 recall above chance (AUC 0.567, p 0.002), against the reported AUC 0.508.
- challengesSample/fs timing predicts word recall in 25 SEEG patients; serial position explains most of the signal
Its within-session null for the global model (AUC 0.511) did not occur with sample/fs timing (AUC 0.566); the null is a timing artifact. Its cross-session result also used onset timing and was not tested here.