A fixed sign rule did not confirm selective MEG and fMRI image matches in 15 adults after image and class removal
We tested whether detailed image relations favor early MEG with early visual cortex (EVC) fMRI and late MEG with inferior temporal cortex (IT) fMRI. Development used 78 photographs from one human study. A local fixed plan then tested 92 cutout images and 15 adults in another study. The same people supplied both modalities within each study; anonymous rows prevented cross-modal person matching. We removed image terms and six-class pair means. Four matched-minus-unmatched correlations used five-person references and bounds over every possible ten-person test group. All four smallest mean preferences were positive: 0.05085, 0.02335, 0.02609, and 0.04447. Their outer 95% person t limits were positive. The fixed rule also required exact person-sign and whole-image label support in all four directions. Its conjunction p was 0.0546875, so it failed. One omitted reference person could reverse the smallest late-MEG mean. Four visual controls made early mean limits include zero. A one-pattern model gave a positive overall interaction without selective matching. These results do not confirm the planned shared selectivity. The package gives code, source hashes and terms, all person scores, negative checks, figures, and repeat steps. Independent arithmetic and a full replay passed. Fixed images, reused cohorts, source-selected windows, unknown row links, and unmeasured noise limit scope. No common circuit or cause is established.
MEG · fMRI · representational similarity · visual cortex · image-pair selectivity · subject dependence · negative result
References
6- usesIn 15 adults, MEG and fMRI maps shared pair relations for 92 images after image and class terms were removed sci:sha256:8c80fe1a9be5de808741be43744593fe4ed549e89b8571e17deb256a7a5163f4
Its positive matched residual correlations motivate our stronger four-preference endpoint. We adapt its input, image-feature, and fetch code under MIT, reuse its exact source records, and build on its method for anonymous person links.
- usesdoi:10.1038/nn.3635
Source human study for the fixed 92-image MEG and fMRI targets.
- usesdoi:10.1093/cercor/bhw135
Source fusion study and required citation for the public training targets.
- usesdoi:10.3390/vision3010008
Source human study for the 78-image development targets.
- usesdoi:10.48550/arXiv.1905.05675
Algonauts 2019 public release and required citation.
- doi:10.7554/eLife.82566
Inference on image-pair maps must distinguish variation over people and over images. Our person intervals condition on fixed images and references.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- A fixed test did not confirm shared prediction of session changes in eight EEG people and three fMRI people
Its image and person holdout discipline informed this new endpoint. No neural arrays or outcome scores were reused.
- usesA fixed determinant test did not confirm visual map transfer between MEG and fMRI for 118 images
Its selective-preference test and person bounds informed our question. We knew its results before we planned our test.
- usesA fixed test did not confirm shared variation directions in late MEG and IT fMRI maps for 118 images
We build on its source checks and conservative treatment of anonymous person links. The present contrast-direction score and code are new.
- usesIn 15 adults, 24 reference images supported within-class matching of 68 other images between MEG and fMRI
We reuse pinned source records and credited input and visual-feature helpers, and adapt bounds for unknown person links. Our image identification endpoint differs from its selective time-region matching test.