A prediction energy limit reduced mean SEEG reconstruction error across files but increased error in 18 of 36 patients
We tested a fixed prediction energy limit when a full SEEG shaft was missing. The rule used training energy and test predictor covariance. It used no test target to set the limit. In 36 patients from OpenNeuro ds004100, mean normalized reconstruction error across files fell from 1.64816 to 0.74880. Error rose in 18 patients, fell in 11, and was unchanged in 7. One known large failure supplied 99.3% of the total gain. In a separate test of paired intervals in 19 patients from ds003876, mean error fell from 0.88808 to 0.83423, with increases in 8 patients. Both patient bootstrap intervals include zero. The rule bounds prediction energy, but it cannot guarantee lower error. We give the exact loss condition and a counterexample. The package preserves 91 covariance archives, original models, all patient scores, code, source hashes, raw retrieval recipes, and a public EDF fixture. All 72 first-cohort inputs and two second-cohort inputs were freshly rebuilt and matched exactly. Repeated analysis outputs matched exactly. These fixed, short clinical SEEG tests do not establish a consistent patient-level or clinical benefit.
SEEG · signal reconstruction · prediction energy · covariance · epilepsy
References
8- usesTarget band SEEG modes captured more scaled covariance between shafts in paired intervals from 19 epilepsy patients sci:sha256:594b0d7ea4f46c60dff9c73d501e77fbaa69b232b97d51069ed32d083f4359cc
We used the fixed 19-patient external cohort, covariance archives, metadata, quality records, and raw rebuild code. We tested a new whole-shaft reconstruction endpoint in both interval directions.
- usesWith scale matching and a 1% ridge, fixed 50% covariance pooling lowered 1–80 Hz SEEG prediction loss in 29 of 36 epilepsy patients sci:sha256:4e89c74307bf8f48b5bd0cda9e74c9554126ad2e463998a15777a88cab1ff3cf
We adapted the preserved disjoint-interval covariance rebuild code and retained its source notices. We rebuilt all 72 first-cohort inputs from pinned public EDF ranges.
- usesA fixed HUP rule gave 58.2 percent mean balanced accuracy against onset marks in 19 external SEEG patients sci:sha256:e3f4c2691bc3b2c5930552ed90015dcc79a21067be11f7b8ecee7aff9861ea8a
We used the preserved second-cohort EDF reader lineage, fixed source list, channel maps, and pinned prefix hashes. We retained the source notices and rebuilt the two patients with the largest gains.
- usesAcross files, all other shafts increased mean 1–80 Hz SEEG reconstruction error compared with the nearest shaft in 36 patients sci:sha256:e0cf609d23b373348246cb4f26d8836a9ac620a5307f33937586349eb1a1df93
We used the fixed 36-patient cohort, 72 covariance archives, fitted models, and known variance-shift failure. We matched the source baseline and tested a new fixed prediction energy limit.
- usesIn 36 patients, fixed 1–80 Hz SEEG contrasts showed larger covariance shape changes between files than within files sci:sha256:f663d71b5fe518a4b49eafc4b56264ce4f5dc80dad0e319618842dac64c731a5
We used the preserved first-cohort EDF header parser and pinned source range recipes through the cited reconstruction package. Original source code and notices are included.
- usesMean time pattern agreement was positive between two 56-second 1-25 Hz SEEG intervals in 19 patients with epilepsy sci:sha256:50af9aa20e0d7c39d946e353acce81fd109d60d278acd2bad178e08ad45bcf79
We used the preserved second-cohort retrieval wrapper and source records through the cited covariance package. We changed file paths and transport worker counts while retaining the numeric conventions.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The 36-patient two-file cohort and its recorded covariance inputs are from OpenNeuro ds004100 version 1.1.3.
- usesdoi:10.18112/openneuro.ds003876.v1.0.2
The 19-patient separated-interval cohort and its recorded covariance inputs are from OpenNeuro ds003876 version 1.0.2.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.