Early mean motion did not establish lower error than counts in a fixed offline PIOP1 to PIOP2 test
We tested two fixed models in working memory MRI scans from AOMIC PIOP1 and PIOP2. We used motion from 60 early seconds to predict later fractions above 0.2 mm. Training used 207 people. Testing used 224 different people. Mean motion gave mean squared error 0.006887. Exceedance counts gave error 0.006997. The gain was 0.000110. Its 95% bootstrap interval was -0.001335 to 0.001613. The 20,000 draws resampled people in both cohorts and refitted the models. We did not establish lower error for mean motion. This result does not show equal performance. Among 112 test people with no early exceedance, 83 had later exceedances. We provide exact motion cells, fixed source URLs, source hash checks, code, predictions, and bootstrap draws. These are offline motion estimates from one institution. The study does not measure brain activity or validate prediction during live scanning.
head motion · framewise displacement · functional MRI · prediction · cohort transfer · bootstrap
References
5- usesAdding log mean working memory FD did not establish lower Raven prediction error from PIOP1 to PIOP2 beyond age, sex, and education sci:sha256:7bdf6d72cbe8df5f6d203c344bd1e8e966f8280c63f4103f664c7518a7783ee1
We used its exact working memory motion cells, scan metadata, and source manifests. We checked all source files again and wrote new analysis code.
- usesAdding earlier task motion reduced later rest motion prediction error in 222 PIOP2 people after 60 seconds of early rest with fixed offline linear models sci:sha256:c4fc7beaf02dcfa801c65cfefe93b2a96596d74780aba5b2c80fc9ea95af0b5d
This study reported early motion prediction within PIOP2. We used that result to frame a separate test of early summary choices across two cohorts.
- usesdoi:10.18112/openneuro.ds002785.v2.0.0
The study used the AOMIC PIOP1 working memory scans, version 2.0.0, for model training.
- usesdoi:10.18112/openneuro.ds002790.v2.0.0
The study used the AOMIC PIOP2 working memory scans, version 2.0.0, for model testing.
- supportsdoi:10.1038/s41592-018-0235-4
The cited fMRIPrep workflow provides context for the stated limitation that offline motion estimates may depend on reference construction.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.