Screening half of matched SEEG link weight for low block disagreement reduced mean squared correlation error between files at 1 to 80 Hz in 36 patients

This study tested whether agreement between two blocks in one SEEG file can select correlations with lower error against another file. It used all 36 patients with two interictal files in OpenNeuro ds004100 version 1.1.3. The fixed rule used donor correlation values only. It retained half of the link weight within each named shaft pair and coarse donor correlation range. At 1 to 80 Hz, mean squared error fell from 0.00458459 to 0.00438368, a 4.38 percent reduction. Error fell in 29 of 36 patients. The descriptive 95 percent patient bootstrap interval was 1.91 to 7.48 percent under independent, exchangeable patient sampling. A later check with shorter blocks and a 10-second gap gave a 4.44 percent reduction. The low-band and absolute-group intervals included zero. The result measures recorded correlation agreement at the cost of half the link weight. It does not measure true anatomical links or clinical accuracy. All 72 public source reconstructions matched the imported covariance inputs exactly. The package preserves code, source versions and hashes, covariance inputs, link decisions, results, figures, and repeat instructions.

stereoelectroencephalography · SEEG · intracranial EEG · correlation networks · stability screening · patient bootstrap · OpenNeuro