Fixed models did not confirm shared gains for heat ratings in 70 EEG adults and 23 fMRI adults
We tested neural estimates of mean heat ratings after stimulus strength was known. Public laser EEG and thermal BOLD fMRI came from separate studies. Development used 25 EEG adults and 10 fMRI adults. Fixed tests used 70 other EEG adults and 23 other fMRI adults. Adding neural features changed mean squared error by -0.16% and -14.58%. A positive value means less error. Adding one personal level mean report gave point gains of 33.79% and 18.87%. Mean normalized gains were 0.006804 and 0.004256. Their 95% person Student intervals were [0.002175,0.011432] and [-0.003785,0.012296]. Gains occurred in 39/70 and 12/23 people. A third claim compared dose plus a mean report with the neural model. This claim was selected in development. All three shared mean and prevalence rules failed. Holm p values were 1, 1 and 0.6073. The MRI model had mean within-person correlation 0.975. This did not establish added information beyond dose. An exact finite-condition identity separates offset error from residual shape variation. It explains why a perfect correlation need not make calibration useful. All 190 EEG source hashes passed. An independent SVD fit checked all 418 reserved neural predictions. The package gives a benchmark, code, compact EEG inputs, exact MRI retrieval, source terms, all outcomes and figures. The MRI target combines warmth and pain codes. A report is a condition mean, not one trial. Different cohorts, source processing, lost EEG hand labels and omitted gamma activity limit scope. The MRI training set was small and lacked new nuisance traces. No shared positive neural effect or clinical or causal claim is established.
human neuroscience · EEG · fMRI · pain · thermal sensation · prediction · calibration · subject holdout · negative results
References
9- usessci:sha256:e18d147dec0dcf3e381fa07c4ba7962547c06b0b871862111dcde62168d9ad49 not known to Sciscape
We build on its practice of person-weighted pain prediction tests, source pins and preservation of uncertain outcomes. No data or code are reused.
- usesdoi:10.18112/openneuro.ds005293.v1.0.0
Exact public EEG source version. All 95 rerefer SET/FDT pairs were checked against the tag hashes.
- usesdoi:10.1038/s41597-025-05900-1
EEG dataset description, collection and preprocessing.
- usesdoi:10.1056/NEJMoa1204471
BMRK3 source study and warmth/pain rating code. We use the public author-linked six-level file, not request-only NPS weights.
- usesdoi:10.1371/journal.pbio.1002036
One of the three required source references recorded in the public BMRK3 file.
- usesdoi:10.1016/j.neuroimage.2015.10.074
Prior personal pain prediction methods and a required BMRK3 source reference. We do not claim the first calibration method.
- doi:10.1073/pnas.1812499116
Prior distinction between within-person and between-person EEG pain markers. Our filtered waveform features cannot test gamma markers.
- doi:10.1038/s41467-022-31039-3
Prior evidence and debate about individual differences in reported pain and fMRI activity.
- usesdoi:10.1038/s41592-019-0686-2
Scientific computing library used for source reading, numerical fits and statistical tests.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesFixed cubic heat tests passed in 70 EEG and 23 fMRI adults; another EEG test and some MRI template tests failed
We reuse its CC0-derived compact EEG features and MIT source preparation code. Our new endpoints test polynomial response contrasts and population mean rank, not its report-prediction claims.
- usesFixed tests did not confirm shared links between neural and report heat curvature in 70 EEG and 23 fMRI adults
Verified compact CC0 EEG input ancestry and MIT source preparation; the new target is between-person curvature association, not report prediction.
- usesStudent confidence limits supported an extra mean heat response pattern in 70 EEG adults but not 23 fMRI adults
We reuse its verified CC0-derived EEG condition table, pinned source provenance and MIT preparation code. The response-rank endpoint is new.
- usesWith fixed models, past reports reduced pain report error in 115 EEG adults and 20 fMRI adults; a shared neural gain was not confirmed
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.