Other channels reduced delay beta test error by 12.7% per patient on average in retained trials from six ds006136 patients
We tested whether other channels improve delay beta power estimates in retained trials from OpenNeuro ds006136 version 1.0.1. The test used six patients and seven recordings. Source JSON files mark these records as SEEG. Channel tables use EEG. We matched original trial IDs and fit models on earlier trials. The baseline used own fixation power, trial ID, and other-channel availability. The added model used mean delay beta power from other available channels on the same trial. The primary test used 57 target channels and 2,100 test channel trials. Mean patient test error fell by 12.7%. Five patient gains were positive. The 95% patient bootstrap interval was 2.58% to 25.03%. The exact two-sided patient test gave p = 0.0625. Evidence for a population gain is limited. Using only different named regions increased error by 9.81% on average in five eligible patients. That check had a different eligible cohort. The results concern retained processed signals, not recovery of artifact-rejected signals. The package contains exact signal windows, all trial identities, code, source hashes, models, predictions, figures, and repeat instructions. Offline extraction reproduced all 4,943 power rows byte for byte. A separate solver checked all 652 models. A fresh process reproduced all 16 model analysis files byte for byte.
working memory · beta power · intracranial EEG · SEEG · missing data · ridge regression · trial identity · held-out prediction
References
4- usesTrial ID matching gave higher mean delay beta power correlation for fixed frontal samples in 11 ds006136 patients sci:sha256:b690a0168a8f0081dff2fcdcc9a04c0272851da42eeea70d9ed46c74999e9fa2
This study used fixed source URLs and hashes, retrieval code, EDF decoder functions, and an independent beta spectrum calculation from this publication. All seven Missouri signals were downloaded anew and all features were rebuilt. It tests a new reconstruction question in the same cohort.
- usesSelection of trials present in every channel kept 315 of 719 observed trials in seven recordings marked as SEEG sci:sha256:59d73440c1bf637517d78a451ab05ab2017ab76b5c443d77a23ff259c34fca6a
The channel availability audit showed that requiring every channel kept 315 of 719 observed trials in these seven source recordings. That finding motivated a test of reconstruction from incomplete available channel sets. No features, source code, or fitted recall effects from that audit were reused.
- usesdoi:10.18112/openneuro.ds006136.v1.0.1
The analyzed source was OpenNeuro ds006136 version 1.0.1; seven Missouri recordings were retrieved and checked against fixed source hashes.
- doi:10.1016/j.neuroimage.2026.121718
This paper describes task collection and source processing methods. Its results were not used as model input.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesAdded channel features increased delay beta test error by 18.3% per patient on average in retained trials from six ds006136 patients at ridge penalty 0.1
This publication supplied exact SEEG sample windows, trial features, and split functions. They enabled a direct comparison with its channel mean model. We reproduced all 2,100 primary test predictions.