Pooled PCA retained more 1–80 Hz SEEG contrast variance on average than path PCA in separate records from 36 patients with training path power scales and a 20% component rule

We tested whether pooled principal component analysis (PCA) preserves more stereoelectroencephalography (SEEG) variance in a separate record than PCA fitted within each electrode path. Public inputs came from OpenNeuro ds004100 v1.1.3. The test used 36 patients and 72 interictal records. Both methods removed path means and used the same training path power scales. Each method kept the ceiling of 20% of the contrast dimensions. Path PCA used the best training rank allocation. For 1–80 Hz signals, mean retained test variance was 73.52% for pooled PCA and 68.76% for path PCA. The equal-patient gain was 4.76 percentage points. The 95% patient bootstrap interval was 3.79–5.86 points. The gain was positive in 35 patients. The interval assumes independent, exchangeable patients and holds fitted scores fixed. All planned secondary tests are reported. All 72 public raw excerpts were retrieved and hash-checked. Their rebuilt covariances matched exactly. Offline and fresh raw-data repeats each reproduced 79 analysis outputs exactly. The package preserves code, covariance inputs, fitted models, source versions, range hashes, results, and instructions. The gain concerns measured, scaled contrast variance. It does not identify neural sources or show a clinical benefit. The true time gap between files remains unverified.

principal component analysis · stereoelectroencephalography · SEEG · spatial filtering · variance retention · cross-record evaluation · signal processing