Permutation entropy adds no SOZ information beyond the power spectrum in 34 HUP SEEG patients

We tested a pre-registered question: does permutation entropy (PE) of interictal SEEG give information about the seizure onset zone (SOZ) that the power spectrum does not give? We used the first 60 s of interictal run 01 for 34 SEEG subjects of OpenNeuro ds004100 (HUP). We calculated one AUC for each subject (SOZ vs non-SOZ bipolar pairs) and used sign-flip tests across subjects. Raw PE did not separate the SOZ (mean AUC 0.48, Holm p = 0.94). PE minus the PE of spectrum-preserving phase-randomized surrogates did not separate the SOZ (AUC 0.52, Holm p = 0.94). PE residualized on log power and line length gave AUC 0.68 (Holm p = 0.0001), which is opposite to the expected direction. A post-hoc control showed that the same residual model applied to the surrogate PE gives the same effect (AUC 0.66; difference p = 0.38). Within each subject, PE and surrogate PE had a median Spearman rho of 0.985. Power (AUC 0.69), line length (0.66) and spike rate (0.69) separated the SOZ better than PE did. Limits: one center, one short clip for each subject, clinical SOZ labels, one PE order and delay, and no replication in a second dataset. The package contains the plan with its hash, the code, the per-pair and per-subject results, and the input hashes.

permutation entropy · stereo-EEG · seizure onset zone · epilepsy · power spectrum · phase-randomized surrogates · signal processing