Clipping at three robust scale units reduced mean calibrated prediction error for recorded 1–80 Hz SEEG correlation maps in 36 patients
We tested whether fixed amplitude clipping improves prediction of a recorded SEEG correlation map in the other file from the same patient. We used 72 public interictal files from 36 epilepsy patients in OpenNeuro ds004100 version 1.1.3. Each file supplied 60 seconds of 1–80 Hz bipolar SEEG. Both methods used the same affine calibration from two halves of the training file. Clipping at three robust scale units reduced mean prediction error from 0.0041740 to 0.0041333. The mean gain was 0.0000407 squared correlation units. The patient bootstrap 95 percent interval was [0.0000135, 0.0000708]. Gain was positive in 30 of 36 patients. The interval excludes zero. The target was the measured unclipped Pearson map. The result does not establish true neural connections, artifact removal, or clinical benefit. The package contains all maps, patient results, source code, verified source retrieval records, and an offline raw-data example.
stereoelectroencephalography · SEEG · Pearson correlation · amplitude clipping · robust scale · prediction error · signal processing · epilepsy
References
4- usesOne second beta power averaging retained shared contact correlation in independent stationary proper complex Gaussian models fitted to 36 SEEG patients sci:sha256:588f4ae1d7299099a824e145e72910deeef2a5162a2f91e6e764e00284846b98
This work supplied the pinned public source recipe, eligibility rules, metadata, raw-data example, and reviewed retrieval and EDF decoding code. We downloaded all 72 source excerpts anew and calculated new clipping and Pearson maps. Its beta power and simulation results did not enter our endpoint.
- usesIn 36 epilepsy patients, mean squared cross-shaft correlation change in 1–80 Hz bipolar SEEG was larger between files than between adjacent 30-second blocks sci:sha256:38d19c44f4536f57be9f1fefd05f78c1bce318501627bff7a6447439e051326f
The prior result showed larger changes between files than between adjacent blocks in this cohort. It motivated our other-file prediction target. Its numerical maps and scores did not enter our analysis.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis used 72 interictal files from OpenNeuro ds004100 version 1.1.3; this exact version and its CC0 license are recorded in the source metadata.
- doi:10.1371/journal.pone.0077916
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesSpearman gave higher agreement than Pearson between 1–80 Hz bipolar SEEG maps in 18 patients and lower agreement in 18 patients
This work supplied the fixed clipping threshold of three robust scale units for a comparator. The present study uses the same cohort. It tests agreement between files with channels that share no contacts. It does not repeat the prior calibrated prediction endpoint or provide an independent cohort test.
- usesRemoving six high-energy seconds gave mixed gains against random removal for calibrated 1–80 Hz SEEG maps in 36 patients
The pinned source recipe, metadata, decoder, and calibration context enabled fresh retrieval and the new time-removal comparison. Prior prediction maps were not used.