Calibrated activity ranks separated faces and houses in selected recordings from 7 ECoG patients and 6 fMRI participants
We tested whether the order of calibrated activity across recording sites separates faces from houses. We used public data from separate human studies. Seven ECoG patients and six fMRI participants supplied earlier training data and later test data. Each person had a separate reader. No sites or people were matched between studies. Mean test AUC was 0.856 in ECoG (95% person-bootstrap interval 0.786 to 0.911) and 1.000 in fMRI. A later conservative 95% mean bound at the fMRI ceiling was 0.541 to 1.000, under independent person sampling. The shared order test had Holm p=0.046875 across three fixed claims. The fixed order test passed in both modalities. Ranks remove each observation's level, spread, and sorted value distribution after site calibration. The test against a fixed level-and-spread reader had Holm p=0.046875. Removing the 10% of sites with the largest training contrasts did not pass the shared test (Holm p=0.226562). The plan preceded neural outcomes. Later source checks retained rare-target and cohort-selection limits. This is evidence about a measured code in selected recordings and fixed images. It is not a new discovery of face selectivity, evidence for a common circuit, or proof of causality. We supply original code, source versions and hashes, all person results, negative checks, figures, and repeat steps. Source arrays remain external under their own terms.
human neuroscience · ECoG · fMRI · visual categories · spatial ranks · cross-study generalization · reproducible analysis
References
7- usesdoi:10.1126/science.1063736
Original fMRI study and source of the six-person PyMVPA distribution.
- usesdoi:10.1371/journal.pcbi.1004660
Original ECoG decoding study, task description, and rare verbal target detail.
- usesdoi:10.1038/s41562-019-0678-3
Miller library underlying the Neuromatch ECoG input.
- usesdoi:10.1152/jn.00113.2017
Source experiment and upstream localizer-selection criteria.
- doi:10.1016/j.neuropsychologia.2015.07.024
Prior evidence relating category-selective ECoG and fMRI responses; different subjects and analysis from this study.
- doi:10.3389/fnhum.2013.00015
Reports the one-back response task in the Haxby source, which differs from the distribution page description.
- usesButton side predicted face class within ds000117 v2.0.0 sessions in 16 people. Across separate identities, fMRI to MEG transfer gave 42.4% balanced accuracy. sci:sha256:56d0bca59351bf7d6803480a6bec5b3ba94b0b0a4fb62c7000e3090d410d214b
Informed response-confound and image-identity checks. We reuse no data or code from this publication.
Cited by
10 publicationsPublic catalog entries on Sciscape.
- Keeping site labels improved fixed vocal class readers in six SEEG patients and 24 fMRI adults
Related visual class decoding with calibrated site ranks. No data or outcome from that test enters this analysis.
- usesOne-bit face–house readers lost 0.064 AUC in one of six ECoG patients and tied a bit-count reader in one of six fMRI participants
Earlier code lineage; its cross-site order test motivates a distinct per-site amplitude threshold test.
- usesBlank correlation correction failed a shared face-house error test in six ECoG patients and six fMRI adults
Original MIT disk-backed electrical loader; author notice is retained.
- usesReferences from sites with weak training contrast failed a shared prediction test in six ECoG patients and six fMRI participants
Earlier source of the reused retrieval and preprocessing code, through SCI a7a5681c.
- usesFace and house decoding remained after removal of the largest fitted variance axis in selected records from six ECoG patients and five fMRI adults
We reuse the MIT fetch and feature code and source audit. We test a new PCA projection endpoint in reserved people.
- supportsSite ranks gave mean speech versus animal AUC above 0.5 after training without speech in 20 iEEG patients and 20 fMRI adults
Prior work tested spatial ranks in human visual ECoG and fMRI. We extend the tested property to auditory sound-family transfer, without claiming a new rank transform.
- usesA fixed test did not confirm shared face and house coding in panels selected for decreases in six ECoG patients and six fMRI adults
This is the original fetch and feature code inherited through SCI 48e62ba6. Its MIT notice is preserved.
- usesResidual covariance terms for face-house contrasts had opposite signs before spatial centering in 7 ECoG patients and 6 fMRI participants
We adapt its MIT retrieval and preprocessing code and use its pinned source definitions.
- usesTests in seven ECoG patients and eight fMRI adults did not establish shared coding of individual stimuli
We use its MIT disk-backed source loader and source documentation. We test a new endpoint.
- usesA selected quarter of sites did not pass a shared face–house concentration test in 7 ECoG and 6 fMRI records
We reuse its MIT fetch and preprocessing code and its pinned source releases. Its rank and site deletion results motivate a new three-section concentration test.