With fixed models, past reports reduced pain report error in 115 EEG adults and 20 fMRI adults; a shared neural gain was not confirmed
We tested whether current neural features improve later pain report estimates after known input and recent reports are used. Separate public studies supplied EEG features and fMRI trial response maps. Development used 39 EEG adults and eight fMRI adults. A local code and data lock preceded tests in 115 other EEG adults and 20 other fMRI adults. Each target used earlier personal calibration and sequential past reports. Recent reports reduced mean squared error by 3.38% and 18.20%. Their 95% person bootstrap ranges were [1.38,5.24]% and [14.40,21.27]%. The fixed shared test passed, Holm p=0.00320. Adding current neural features gave mean gains of 2.79% and -4.80%, with ranges [-1.20,6.44]% and [-13.28,0.86]%. That shared test failed, p=0.93783. A shared gain below the fixed 5% boundary was also not confirmed. At a later visit, 114 of the same EEG people had a 4.26% mean neural gain, but only 56 improved. All negative people, fixed controls and discordant checks remain. A line fit to early neural features removed the EEG mean gain. Stronger neural shrinkage gave a small positive MRI point estimate. These checks do not replace the failed primary test. Source feature joins, MRI coordinates and source hashes passed audits. Independent weighted SVD reproduced every primary prediction within 5e-16. The package gives permitted compact inputs, code, provenance, local plans, results, figures and repeat steps. Full-record source processing, missing EEG trials, incomplete MRI body-site and physiology labels, nonrandom ID splits and different cohorts limit neural claims. This is a reusable prediction benchmark and a failed positive neural generalization test, not a causal result.
pain reports · EEG · fMRI · prediction · report history · subject holdout · negative result · open data
References
8- usesFixed models did not confirm shared gains for heat ratings in 70 EEG adults and 23 fMRI adults sci:sha256:d31b48e213887bcecdf56aca37d6a2a99dde99b8eff9b24b4056a23d43108d81
Its known-input comparison and preservation of failed heat prediction tests informed this different trial-level benchmark. No source code, data or weights were reused.
- usesdoi:10.1371/journal.pbio.3003948
Source study and released EEG feature definitions.
- usesdoi:10.17605/OSF.IO/BS3YJ
Hash-pinned uncentered EEG feature tables at two visits; source analysis code and MIT terms.
- usesdoi:10.6084/m9.figshare.12795116.v1
Version-specific public BMRK4 trial map dataset under CC BY 4.0.
- usesdoi:10.7554/eLife.15166
Source MRI study, task design and acquisition context.
- usesdoi:10.1016/j.neuroimage.2015.10.074
Prior single-trial pain prediction and personal/population calibration methods.
- doi:10.1371/journal.pcbi.1002719
Prior work on temporal history in pain prediction. This study does not claim the first history method.
- usesdoi:10.1038/s41592-019-0686-2
SciPy numerical methods.
Cited by
13 publicationsPublic catalog entries on Sciscape.
- usesWith source order assumed, fixed scores did not confirm shared previous heat effects in 110 EEG and 20 fMRI adults
We reuse its permitted compact neural inputs and source records. We fit new readers and recover delivered EEG history from original event logs.
- usesExploratory record checks in 114 EEG and 87 fMRI adults gave positive heat rank covariance; three primary shared tests failed
The compact initial MRI extraction and trial benchmark originate in this source chain.
- usesNeural scores trained on heat level tracked adjusted pain reports in 115 EEG adults and 20 fMRI adults
We reuse the verified compact EEG features, MRI grids and source provenance. We fit new physical-target scores and test new endpoints. Its failed shared prediction-error test remains valid.
- usesNeural scores had positive heat relations after linear report control in 115 EEG adults and 20 fMRI adults
This benchmark supplied the original compact extraction and source audit used by our immediate source. Its failed neural prediction gains remain distinct from our association result.
- usesA fixed offset update did not pass a shared heat forecast test in 115 EEG and 20 fMRI adults
Original main MRI grid extraction and source provenance, inherited through the verified immediate source.
- usesA shared heat response variance rise was not confirmed in 115 EEG and 14 fMRI adults
We build on its trial benchmark and follow its deferred metadata/extraction approach for a fresh full-source check.
- usesA fixed test did not confirm lower neural scores after heat level repetition in 106 EEG and 20 fMRI adults
Source extraction ancestry of the compact EEG and MRI inputs inherited through SCI 0285b9ee.
- usesAfter one heat score was removed, fixed pain readers failed a shared test in 119 EEG and 20 fMRI adults
We credit the original compact EEG and MRI trial extraction inherited through e95d9999 and its failed report-error benchmark.
- usesFixed heat scores decreased with retained order in 115 EEG and 20 fMRI adults; resampling training subjects left the fMRI slope uncertain
We credit its original compact data extraction and build on its distinction between report association and prediction gain.
- usesFixed report readers failed a shared pain reversal ranking test in 114 EEG adults and 20 fMRI adults
Original full BMRK4 grid extraction and provenance supplied through the verified immediate packages.
- usesFixed neural readers tracked pain reports across heat levels in 117 EEG and 20 fMRI adults; another MRI test failed
We verified and reuse unchanged public EEG and BMRK4 derivatives, source metadata and extraction audits. We fit new readers and test a different endpoint.
- usesUnder source order, fixed neural heat scores had current report correlations above the neighbor mean in 100 EEG and 20 fMRI adults
Original extraction ancestry of the numerical neural inputs.
- usesWith fixed readers, mean heat contrast kept more than half its early size in 115 EEG and 19 fMRI adults; the joint drift test failed
Supplies original BMRK4 grid extraction, geometry and source structure checks.