Complete squared-distance matching preserved EEG links under -6 to +6 dB shifts in two fixed public cohorts

We tested links between earlier and later posterior EEG spectra in two public cohorts with 60 and 38 people. Each complete set had one record per person and eye state. Complete one-to-one matching minimized total squared distance. It kept all links unchanged at five common later-record power shifts from -6 to +6 dB. It gave 57 correct links out of 120 and 43 out of 76, including both eye states. Independent raw matching gave 51 and 38 correct links at zero shift, but five and three at +6 dB. We then kept 48 or 30 later people in each of 2,000 random subsets, with the full earlier gallery. At +6 dB, raw rectangular matching changed 54.89% and 62.31% of links on average. Independent matching after separate cohort mean removal changed no links, even with missing queries. The established quadratic translation property explains the complete-set result. It also preserves incorrect links. Several processing checks gave fewer correct links under assignment than under independent matching. The result does not establish a general accuracy gain or a stable neural source. The package preserves 396 input spectra, source versions and hashes, all subset matches, code, results, figures, and instructions. Independent optimization, full output reconstruction, two raw-record replays, and a fresh numerical repeat passed.

EEG · squared-distance assignment · person matching · power shift · missing records · signal processing

Admitted 2026-11-09 17:52 UTC0 citing publications

References

usesA +6 dB query shift reduced raw 2–30 Hz posterior EEG person matches in two adult cohorts with fixed accepted windows sci:sha256:cc2b2e14f55008c1450e24b7041dc2cb4dc78324c91693d7a33c5784e2a754f6
We used its 396 preserved posterior window spectra and quality flags from ds004148 and ds007176. We kept its person selection, frequency band, primary electrical references, mean power transform, source extraction code, and retrieval provenance. Its known independent matching gain test motivated the new complete-assignment and missing-query question. Our separate raw matching implementation also reproduced all ten archived same-state correct-link and changed-link gain conditions.
usesCE posterior alpha log power was more repeatable than the CE/OE log ratio across V0 and V1 in 38 ds007176 adults using 60-second segments sci:sha256:840e0d1801a45499fc18b519a47a306a84bcbb39b77919c26c1bac440ae6dc0b
Its pinned ds007176 V0 and V1 CE and OE window spectra, references, quality flags, headers, extraction code, prefix hashes, and expected whole-file hashes define the second cohort. We preserved these materials through the direct input publication. We also retrieved one pinned complete original EEG record and reproduced all eight extracted arrays exactly.
usessci:sha256:eb6da8fdd843c3b66dd19edef3df5eb85f888448f106108e611352a1c6388159 not known to Sciscape
Its pinned ds004148 session1 and session3 posterior window spectra, electrical references, quality flags, headers, extraction code, and source retrieval hashes define the first cohort. We preserved these materials through the direct input publication. We also retrieved one pinned complete original EEG record and reproduced all eight extracted arrays exactly.
usesdoi:10.18112/openneuro.ds004148.v1.0.1
Version 1.0.1 supplies the first cohort spectra analyzed here.
usesdoi:10.18112/openneuro.ds007176.v1.0.1
Version 1.0.1 supplies the second cohort spectra analyzed here.
doi:10.1038/s41597-022-01607-9
Source paper for the ds004148 recording design and protocol facts cited in this report.
doi:10.1016/j.bspc.2026.109484
Credit requested by the ds007176 dataset description; no numerical result or missing protocol detail from this paper is used.
doi:10.1109/TAES.2016.140952
Algorithm credit identified by the official SciPy linear_sum_assignment documentation.

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.