A patient offset from distant SEEG pairs reduced normative spectral error by 8–34% in 249 RAM patients

In 249 patients from the RAM free-recall datasets, a patient-specific offset estimated from the same patient’s covered bipolar SEEG pairs at least 40 mm from a target reduced leave-one-patient-out normative mean squared error by 8% to 34% across five spectral features. The analysis used 27,974 pairs from 256 patients, a map built from the 30 nearest pairs of other patients within 15 mm, and a prespecified plan. Reductions were 16% for relative theta power, 8% for relative alpha, 25% for relative beta, 25% for the 2–40 Hz aperiodic exponent, and 34% for log broadband power; Holm-adjusted p values ranged from 0.0002 to 0.0011. About half of patients improved for theta and alpha. An offset from another patient increased error by 48% to 76%. An offset from one distant pair increased error for every feature, but with two pairs broadband error decreased by 7%; therefore the original claim that fewer than five pairs always increased error was corrected. Results apply to one 60 s word-encoding segment, bipolar recordings, and RAM patients. There were no seizure-onset-zone or pathology labels, so the study did not test abnormality detection. The observed feature includes possible pathology, and the cause of the patient offset is unknown.

stereo-EEG · SEEG · normative maps · patient offset · spectral features · free recall · RAM