MR-ART EFC threshold transfer gives mean balanced accuracy of 0.726553 after standard scan subtraction
This analysis tests whether entropy focus criterion (EFC) thresholds selected in one MR-ART head-motion condition transfer to the other. It uses OpenNeuro ds004173 v1.0.2 and 278 motion scans with grade 2 or 3 from 147 participants. For each target scan, all source scans from that participant are excluded while selecting the threshold. The equal-weight mean balanced accuracy changes from 0.700563 for raw EFC to 0.726553 after subtracting each participant’s standard-scan EFC. The paired change is +0.025990 (pointwise 95% participant-bootstrap interval −0.081390 to +0.219474), which includes zero. Condition-specific changes are mixed: +0.055706 for HM1 and −0.003726 for HM2. In HM2, sensitivity rises while the correct grade-2 count falls from 20 of 22 to 11 of 22. The HM2 change direction is sensitive to restricting the sample to complete scan triplets. These results do not show a population benefit, biological causality, or future clinical performance. They come from one scanner and protocol, requested head motion, a three-point image grade, and one researcher’s checks; HM2 includes only 22 eligible grade-2 scans.
magnetic resonance imaging · image quality · entropy focus criterion · threshold transfer · balanced accuracy · participant bootstrap · MR-ART
References
4- Standard-reference subtraction improves EFC but reduces CJV condition-matched grade ranking in MR-ART sci:sha256:5246205abcf30a204e7b86cba2a1fa9d96a7077d08902ee630dba181fa75a49d
[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.
- Subtracting standard scan EFC increases HM1 balanced accuracy and decreases HM2 balanced accuracy in MR-ART leave-one-participant-out validation. sci:sha256:b4ae455173df30614d7232a12837b8276f10e618238bcbe6f1f814f4b3ce99e6
[Legacy relation: reproduces] The second calculation gives the same 12 primary local threshold confusion tables. It does not reproduce the earlier uncertainty calculation. The read result supplied the training rule and vector method design. The new analysis adds threshold transfer between HM1 and HM2 with the test participant removed from training.
- usesdoi:10.18112/openneuro.ds004173.v1.0.2
This is the exact OpenNeuro ds004173 version 1.0.2 analyzed; the package manifest pins the source files to Git commit fb45b349f21faa4b65ad23089788bbea30d07d74.
- doi:10.1038/s41597-022-01694-8
Cited by
3 publicationsPublic catalog entries on Sciscape.
- Subtracted EFC gives 81.33% sensitivity in MR-ART HM2 to HM1 transfer with a 95% training target.
[Legacy relation: reproduces] The new code gives the same four EFC transfer point count tables for the maximum-balanced-accuracy rule. Its uncertainty and other metrics were not reproduced.
- Exclusion of one more source participant changes at most 9 of 134 subtracted EFC decisions for MR-ART HM1 scans.
[Legacy relation: reproduces] This run used the read source locations, hashes, and threshold rule. Fresh downloads and two arithmetic code paths give the same four EFC transfer confusion tables. The relation covers these point counts only. It does not cover bootstrap intervals, other quality measures, or image processing. The new check excludes one more eligible source participant while all target scans stay fixed.
- EFC increases from HM1 to HM2 in 15 of 16 MR-ART participants with two grade-2 scans.
[Legacy relation: reproduces] The new code gives the same 1,112 EFC point decisions and eight point confusion tables for the main grade-2/3 analysis. It uses the same source tables and threshold rule. It does not reproduce bootstrap intervals or other metrics.