Spread sampling reduced median-pair spectral-shape error by 0.58% in 36 patients, but not mean-pair error

We tested two ways to select 20 seconds of public SEEG signal. We used paired five minute interictal files from 36 patients. One rule used a continuous part. The other used ten two second parts spread through the file. Each rule had 15 alternatives and used every window once. The target was the other file's full spectrum. The score measured centered log10 spectral shape at 30 to 80 Hz, with 55 to 65 Hz removed. The analysis used 3,686 adjacent named contact pairs. Each patient used the median pair loss. Each patient had equal group weight. Mean loss was 0.02250021 for continuous samples and 0.02237007 for spread samples. The relative reduction was 0.58 percent. Its 95 percent patient bootstrap interval was 0.22 to 0.94 percent. Gains were positive in 24 of 36 patients. The fixed mean pair check gave a 0.40 percent increase in error. Its reduction interval was -1.01 to 0.16 percent. Patient intervals assume independent, exchangeable patients. The target is a finite released file, not a known true spectrum. The spread rule requires a longer observation period. This result does not establish a neural cause or clinical benefit. The package preserves code, compact spectra, a native signal example, source hashes, retrieval recipes, all scores, and checks.

stereoelectroencephalography · SEEG · spectral estimation · sampling · power spectral density · OpenNeuro