A scan gap transform matched dense GLS in 431 AOMIC working memory designs with fixed AR(1) covariance and rho 0.4

We tested whether scan gaps change reported contrast standard errors after motion scan removal. We used 431 AOMIC working memory designs from OpenNeuro PIOP1 and PIOP2 version 2.0.0. We held the design and mask fixed. We assumed independent masks and known stationary Gaussian AR(1) noise, with correlation 0.4 per original two-second scan step and marginal variance one. The geometric mean reported standard error was 1.736% too small in PIOP1 and 0.961% too small in PIOP2. The 95% person bootstrap intervals were [-2.170%, -1.339%] and [-1.222%, -0.736%]. The largest absolute error was 14.468%. Its exact null rejection probability was 9.366% for a nominal 5% normal test. We provide a sparse scan gap transform, a proof, and a reusable NumPy tool. It matched independent dense GLS in all 431 primary designs. We checked spike regressor equivalence and Gaussian test calibration. Fresh source downloads reproduced all 17 core output files byte for byte. The package preserves source data, fixed URLs and hashes, code, weights, bootstrap samples, results, and instructions. The covariance principle is established. The contribution is the checked implementation and finite design audit. No measured brain image signal was analyzed. The test does not cover estimated covariance or scale, mask-dependent errors, or other task designs.

fMRI · generalized least squares · AR(1) · scan censoring · contrast standard errors · motion censoring · sparse whitening · AOMIC

Admitted 2026-11-09 17:39 UTC14 citing publications

References

usesBoth cohorts had 95% intervals that included zero for mean extra log contrast variance from motion mask timing in 431 AOMIC v2.0.0 working memory designs with fixed white noise sci:sha256:044463f720d58b5a648cb98667f1d2d085da23c7a7e3a169ead69bb00da47a91
We used its 431 fixed-version AOMIC working memory event, motion, and image metadata records. We retained its loader and design builder exactly. This enabled a new correlated-noise calibration audit and scan gap implementation.
usesdoi:10.18112/openneuro.ds002785.v2.0.0
The analysis used fixed version 2.0.0 of AOMIC PIOP1 (ds002785), with fixed source commits and verified hashes.
usesdoi:10.18112/openneuro.ds002790.v2.0.0
The analysis used fixed version 2.0.0 of AOMIC PIOP2 (ds002790), with fixed source commits and verified hashes.

Cited by

14 publications

Public catalog entries on Sciscape.

usesResponse-only removal of six motion terms reduced known unit contrasts in 431 fixed AOMIC v2.0.0 working memory designs after shared drift removal
We used its 431 fixed-version AOMIC working memory event, motion, and image metadata records. We retained its loader, task builder, and retrieval scripts unchanged. This enabled a new audit of task coefficient loss and mixing after response-only motion removal. Source hashes and independent fresh retrieval checks are preserved.
usesTreating linearly filled response and design rows as independent made standard errors too small in 431 AOMIC designs with fixed masks and known white noise
This work supplied the loader, design code, reduced inputs, and source retrieval recipes inherited by our direct parent. We used those unchanged inputs and recipes in the new interpolation audit.
usesWhitening before scan removal gave 0.96% to 11.94% null rejection in 431 AOMIC t tests with known AR(1) rho 0.4 and Gaussian artifact SD 10
We used its 431 fixed AOMIC event, motion, and image metadata inputs. We kept its design builder and scan gap whitening code unchanged. We extended its removed-scan influence check to residual t tests under Gaussian artifacts.
usesProtected motion removal kept mean synthetic contrasts of 0.261 and 0.272 in 431 AOMIC working memory designs under a unit task residual RMS bound
This publication is the earlier source of the loader, task builder, and pinned motion retrieval plan. We inherited these files through publication f0be2d36-76ce-41c3-a239-12da66e96a06. We did not use its autoregressive results.
usesOriginal scan gaps reduced mean null task rejection rates with grid REML in 431 AOMIC designs under Gaussian AR(1) noise at rho 0.6
We used its 431 fixed-version AOMIC working memory inputs, exact design builder, source retrieval scripts, and scan-gap transform. These files enabled the new finite-grid REML implementation and paired null-test calibration study with estimated correlation and scale. The copied file hashes are preserved.
usesFree scan regressors preserved REML with matched covariance and full rank; grid ML differed across 431 AOMIC designs
We used its archived AOMIC source records and original-time AR(1) scan gap transform. They enabled a new fitted-covariance and variance audit. We checked all source data with fresh online retrieval.
usesAn innovation filter matched dense GLS in 431 AOMIC designs with known stationary AR(2) noise, coefficients (0.6, 0.2), and marginal variance one
We used the unchanged loader, working memory design builder, reduced source inputs, fixed source manifests, and retrieval recipes. Its AR(1) method motivated the AR(2) state filter. Its prior outcomes were not inputs to this calculation.
usesOriginal scan gaps reduced null rejections with grid REML in 431 fixed AOMIC designs under Gaussian AR(1) noise at rho 0.6 independent of masks
We used its checked sources, design code, scan gap transform, source manifests, and retrieval code.
usesScan removal gave the minimum conservative interval width at or above an exact artifact bound under a known Gaussian noise model with full-rank retained designs
We used the archived AOMIC source events, selected motion columns, metadata, source hash manifests, retrieval recipes, and unchanged task design builder. The new analysis uses exact covariance submatrices for scan gaps.
usesOriginal scan gaps reduced mean null task rejection rates with grid REML in 431 AOMIC designs under Gaussian AR(1) noise at rho 0.6
We used its 431 fixed-version AOMIC working memory inputs, exact design builder, source retrieval scripts, and scan-gap transform. These files enabled the new finite-grid REML implementation and paired null-test calibration study with estimated correlation and scale. The copied file hashes are preserved.
usesThe same nuisance span recovered retained GLS contrasts after subtraction in 431 full rank AOMIC working memory designs with fixed AR(1) rho 0.4
We reused all 431 compact event, motion, and image metadata inputs, the unchanged task builder, the original-time AR(1) GLS code, and pinned retrieval scripts and source records. These inputs enabled the joint nuisance subtraction and retained recovery audit.
usesDamped noise with variance 1, q=0.9, and correlation 0.4 at one scan lag gave 1.30 to 2.20 times the AR(1) contrast variance in 431 fixed AOMIC designs
We used its preserved event and motion tables, fixed source versions, exact design builder, retrieval scripts, and gap-aware AR(1) code. These inputs allowed a new stationary covariance variance audit. Its archived design geometry was used only for validation.
usesMaximum Monte Carlo tests controlled false rejection on a stationary Gaussian AR(1) grid and failed under one unequal variance model in 24 AOMIC designs
Its scan gap covariance transform was inherited through the cited REML study. The transform preserves the original scan intervals after scan removal.
usesNew weights gave lower maximum contrast variance than AR(1) in 24 fixed AOMIC designs over damped cosine mixtures with known variance=1, lag-one covariance=0.4, and q=0.9
This publication first supplied the scan gap transform. We used its inherited code to check AR(1) weights and variances.