Standard-reference subtraction improves EFC but reduces CJV condition-matched grade ranking in MR-ART
In 436 MR-ART scans from 148 participants, this rank reanalysis compared MRIQC CJV, EFC, and total SNR for expert grade 3 versus grade 2 scans from different participants within each instructed motion condition. It evaluated raw metrics and values centered on each participant’s standard scan. Across HM1 and HM2, with equal weight for each condition, reference subtraction raised EFC AUC from 0.718 to 0.828 (change 0.111; participant-cluster bootstrap 95% CI 0.018 to 0.205), lowered CJV AUC from 0.815 to 0.768 (change −0.047; 95% CI −0.087 to −0.009), and lowered total SNR AUC from 0.842 to 0.783 (change −0.059; 95% CI −0.122 to 0.002). These descriptive results show metric-dependent changes in this sample; they do not establish clinical thresholds or generalization. The analysis uses 10,000 participant-cluster bootstrap draws and the version-pinned OpenNeuro ds004173 v1.0.2 summary tables. It also independently recounts six selected prior point-count claims. This is a partial count reproduction, not a reproduction of earlier intervals, classifiers, or MRIQC image processing. Limitations include one scanner and protocol, instructed motion, one coarse expert rating per scan, missing motion scans, variable reference quality, and no modeled scanner or rater uncertainty.
MR-ART · MRI quality · MRIQC · structural MRI · expert ratings · rank AUC · reference subtraction · cluster bootstrap
References
4- Three MRIQC metrics ranked MR-ART high-motion scans correctly in 84–89% of pairs and unequal-grade scans in 98–99% sci:sha256:d96b8d3e11af1cc47200598600b962533bcc9aab7067bf6559f9d675eb5393e0
[Legacy relation: reproduces] Independently reproduces exactly six selected point-count claims: HM2/HM1 condition-order wins 117,124,123 out of 140 and pooled within-participant unequal-grade wins 306,308,309 out of 313, checked again with separate Decimal enumeration. This relation does not claim reproduction of earlier intervals, classifiers, or image preprocessing.
- usesThree MRIQC metrics ranked MR-ART high-motion scans correctly in 84–89% of pairs and unequal-grade scans in 98–99% sci:sha256:d96b8d3e11af1cc47200598600b962533bcc9aab7067bf6559f9d675eb5393e0
Used the read MR-ART audit to fix the three metric directions, locate immutable source tables and hashes, and motivate condition-matched grade-2/3 comparisons across participants. New endpoint and verification code were written independently; all source files were freshly retrieved.
- usesdoi:10.18112/openneuro.ds004173.v1.0.2
The analysis uses the MRIQC 0.16.1 group_T1w.tsv and expert scores.tsv summary tables from this exact dataset version; their source URLs, hashes, and version are recorded in original/data/manifest.json.
- doi:10.1038/s41597-022-01694-8
Cited by
8 publicationsPublic catalog entries on Sciscape.
- Two computer programs calculate the same 126 bootstrap interval limits for MR-ART v1.0.2.
[Legacy relation: reproduces] The new program calculates the 42 AUC records and 21 centered-minus-raw records again. All 63 point estimates and 126 interval limits agree for the fixed data, definitions, and seed.
- MR-ART EFC threshold transfer gives mean balanced accuracy of 0.726553 after standard scan subtraction
[Legacy relation: reproduces] The second calculation gives the same 12 grade-2 versus grade-3 rank counts. It does not reproduce the earlier uncertainty calculation. The read result supplied metric directions, source locations, expected hashes, and the question for threshold transfer. This run downloaded all five source files again.
- usesSubtracted EFC gives 81.33% sensitivity in MR-ART HM2 to HM1 transfer with a 95% training target.
This result supplied standard scan subtraction, the grade-2/3 comparison, and the same-data EFC selection. The new rule tests a 95% training sensitivity target.
- Half subtraction has higher CJV and SNR rank AUC than full subtraction for MR-ART motion scans with grades 2 and 3.
[Legacy relation: reproduces] Decimal arithmetic at precision 60 digits gives the same 12 condition-specific raw and full subtraction point counts. This relation covers those counts only. It does not cover earlier bootstrap intervals or image processing.
- usesHalf subtraction has higher CJV and SNR rank AUC than full subtraction for MR-ART motion scans with grades 2 and 3.
This result supplied the grade-2/3 rank comparison, metric directions, source paths, expected checksums, and reason for the specified half-weight comparison. This run downloaded all five source files again and wrote new calculation code.
- Subtracting standard scan EFC increases HM1 balanced accuracy and decreases HM2 balanced accuracy in MR-ART leave-one-participant-out validation.
[Legacy relation: reproduces] The arithmetic check gives the same 12 condition-specific point AUCs for grades 2 and 3. It does not repeat the prior uncertainty calculation or image processing. The prior publication supplied the source paths, expected checksums, metric directions, and the reason for the new threshold check. This run got the source files again.
- EFC grade agreement is higher than all 20,000 reference control values in the tested MR-ART tables
[Legacy relation: reproduces] This run used its grade-2/3 comparison, metric directions, source manifest, and expected point counts. Decimal arithmetic gives the same 12 raw and own-reference rank tables. This run does not repeat its bootstrap calculation or image processing.
- usesEFC correlation decreases from 0.848 to 0.513 after standard subtraction in 140 MR-ART triplets.
The read result supplied the same-participant standard subtraction and the same-data reason for primary EFC selection. The new endpoint compares the two motion-condition correlations. It does not reproduce the grade-ranking result.
- usesEFC grade agreement is higher than all controls in two MR-ART samples with 140 and 119 participants
The read result supplied the standard grade-1 sample restriction and the own standard subtraction endpoint. This run adds donor permutation controls in that defined sample. It does not calculate the earlier bootstrap intervals.