Stored BOAS headband labels showed fewer stage changes than human consensus on shared valid intervals in 100 people
We tested the rate and time of stage changes in stored BOAS sleep labels. We used OpenNeuro ds005555 version 1.1.3. All 128 paired nights from 100 people were included. Both ends of each interval had valid human, headband AI, and PSG AI labels. The mean rates were 8.29 headband changes and 10.07 human changes per valid hour. The headband minus human difference was −1.78 changes per hour. Its 95% person bootstrap interval was −2.36 to −1.21. Epoch agreement was 86.47%, but the headband detected 33.07% of human changes at the same time. Detection was 46.49% within 30 seconds with one-to-one matching. Both measures allowed different stage pairs. All planned rate checks kept a negative difference. The PSG AI rate was near the human rate, but its exact event recall was 40.04%. The package preserves source files, hashes, a local plan, code, event states, matches, results, and all bootstrap draws. Independent checks passed. A fresh source download gave identical numerical output. The result concerns fixed stored labels on observed valid intervals. It assumes independent people for its intervals. It does not test new model predictions or EEG arousals.
sleep staging · wearable EEG · sleep stage transitions · human consensus · OpenNeuro · BOAS · event detection
References
4- usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
We used its archived BOAS v1.1.3 source files and URL/hash manifest. Its agreement and boundary-epoch study informed the question. New code calculated transition counts, timing matches, and person estimates.
- usesdoi:10.18112/openneuro.ds005555.v1.1.3
The analyzed event labels and metadata came from OpenNeuro ds005555 version 1.1.3; the saved source manifest records the commit and file hashes.
- doi:10.1093/sleep%2Fzsx139
- doi:10.5664/jcsm.2350
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesFixed pair regression had higher sleep stage change rate error than affine correction on valid edges in 50 BOAS test people
Its lower headband stage change rate in the same BOAS cohort set the calibration question. Our main target uses valid headband and human edges without requiring valid PSG AI. This is a new model comparison, not an independent cohort replication.
- usesFixed pair regression had higher sleep stage change rate error than affine correction on valid edges in 50 BOAS test people
Its lower headband stage change rate in the same BOAS cohort set the calibration question. Our main target uses valid headband and human edges without requiring valid PSG AI. This is a new model comparison, not an independent cohort replication.
- usesA 1% five-stage human-label edit limit allowed both signs of the mean headband-minus-human stage-change rate in 128 BOAS v1.1.3 nights from 100 people with fixed valid edges
This publication supplied the archived BOAS v1.1.3 source labels and their fixed source manifest. We downloaded the same bytes again and reproduced its fixed-label mean rate gap. We added exact limits for reference label edits with new analysis code.
- usesA fixed three-epoch filter reduced human stage-change recall in stored BOAS v1.1.3 headband labels from 100 people
Its stored-label stage-change result helped select this filter question and the exact-edge measures. We knew its results before we saved the local plan. We tested an added filter and did not repeat only its stored-label comparison.
- BOAS headband labels had higher mean destination diversity across Wake, N2, and REM than human labels in 91 eligible people
[Legacy relation: reproduces] The earlier change-rate result motivated the destination question. Our independent analysis reproduced its 114,040 common valid edges. It also reproduced 9,654 human, 7,920 headband, and 9,349 PSG AI stage changes. The new primary diversity analysis retained a different, fixed cohort of 91 people.
- usesA fixed Markov filter increased probability loss versus a static table for stored BOAS v1.1.3 headband labels in 50 test people
The reported low headband stage-change rate informed the question about added temporal processing. We tested a new Markov rule and used different common-valid intervals. We did not reuse its event matching code.
- On common valid intervals in 100 BOAS people, two-minute sampling retained a larger fraction of headband than human stage change counts
[Legacy relation: reproduces] A separate post-plan source control reproduced 114,040 common valid adjacent edges and 9,654 human, 7,920 headband, and 9,349 PSG AI changes. It also reproduced the mean rates of 10.066 human and 8.287 headband changes per hour. This is a source control on the same cohort, not independent cohort replication.
- usesOn common valid intervals in 100 BOAS people, two-minute sampling retained a larger fraction of headband than human stage change counts
Its lower headband stage change rate informed the temporal sampling question before the local plan. The new primary test measures the fraction of counts retained at two-minute endpoint spacing. It uses a fixed complete-window selection and all sampling phases.