sciscape Publications
← All publications
q-bio.NC · Neurons and Cognitionstat.ME · Methodology

A fixed test did not confirm shared object relation changes in MEG and fMRI for 118 photographs

We tested whether changes in visual object relations agree across MEG and fMRI after additive image terms and broad class-pair means are removed. The 92-image source experiment supplied development, with 16 MEG and 15 fMRI rows. A local plan fixed the same 70–90 and 140–160 ms MEG windows before the separate 118-image experiment, with 15 adults in each modality. People overlap across modalities within each experiment, but anonymous row links are unknown. Disjoint people across the two experiments are not verified. Five reference people defined each change template. Conservative bounds cover every possible disjoint target group of size N minus five. In the fixed test, possible ten-person means were −0.00909 to +0.01075 in MEG and +0.00015 to +0.00802 in fMRI. Approximate outer 95% limits were −0.01444 to +0.01987 and −0.00352 to +0.01127. The shared person and whole-image label rule failed, with p = 0.98926. The positive development pattern did not transfer under this rule. Same-modality fMRI change repeatability was weak. One later exploration used source-selected 118-image windows. Both mean bounds increased, but both conservative person tests had p = 0.05469. That exploration is not independent confirmation. The result limits a fixed measured-change test; it does not show absent neural information or an absent visual hierarchy. Source processing, noise, masks, time windows, and fixed image sets limit scope. The package preserves code, source versions and terms, labels, plans, all outcomes, figures, reader checks, numerical audits, and repeat instructions. Source neural arrays and photographs remain external.

human neuroscience · visual perception · MEG · fMRI · representational similarity · negative result · reproducibility

Admitted 2 citing publications

References

6
  1. usesIn 15 adults, MEG and fMRI maps shared pair relations for 92 images after image and class terms were removed sci:sha256:8c80fe1a9be5de808741be43744593fe4ed549e89b8571e17deb256a7a5163f4

    We reuse attributed MIT nuisance-projection and image-feature functions, and adapt the bound for unknown person links. We test a direct change contrast and retain its failed transfer result.

  2. usesdoi:10.1038/nn.3635

    Source experiment for the 92-image development data.

  3. usesdoi:10.1093/cercor/bhw135

    Source experiments and public full MEG decoding maps; the second experiment supplies the fixed test.

  4. usesdoi:10.48550/arXiv.1905.05675

    The Algonauts 2019 training release supplies the compact fMRI targets and stimulus order.

  5. doi:10.3389/neuro.06.004.2008
  6. doi:10.7554/eLife.82566

Cited by

2 publications

Public catalog entries on Sciscape.