On common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages
We compared stored headband and clinical polysomnography (PSG) AI sleep labels with the human reference. We used OpenNeuro ds005555 version 1.1.3. The study included 128 nights from 100 people. It used 114,578 epochs with valid labels from all three sources. Each person received equal weight. AI agreement was 89.06%. Headband agreement with humans was 86.47%. The paired difference was 2.58 percentage points. Its 95% participant bootstrap interval was 1.72 to 3.47 points. When the AI labels agreed, their human-reference error rate was 8.14%. In the human N1 group, that error rate was 70.49%. All planned weighting and missing-label checks kept the overall difference positive. Thus, AI agreement did not equal headband accuracy against the human reference in these stored labels. The intervals assume independent people and fixed model outputs. The study does not test new model predictions. The package preserves source labels, hashes, the local plan, code, an epoch ledger, results, and reproduction checks.
sleep staging · polysomnography · headband · AI agreement · human reference · paired comparison · participant bootstrap · BOAS
References
1- usesdoi:10.18112/openneuro.ds005555.v1.1.3
The publication analyzes stored headband AI, PSG AI, and human sleep-stage labels from OpenNeuro ds005555 version 1.1.3.
Cited by
66 publicationsPublic catalog entries on Sciscape.
- usesWith fixed BOAS labels and known masks, adjacent pair audits had 83.6% less stage change sampling variance than epoch audits at equal 10% label caps
We reused its 773 BOAS source files and fixed commit manifest. A fresh public retrieval matched all file sizes, SHA-256 hashes, and Git blob hashes. We wrote a new analysis for adjacent pair audits of stage changes.
- usesWith fixed BOAS labels and known masks, 5 minute block audits had 5.14 times epoch audit variance at equal budgets
We reused its unchanged BOAS source tables, label definitions, person grouping, and pinned retrieval manifest. These supplied the fixed census for a new audit selection study.
- usesStored headband labels had a lower mean wake-sleep change rate than human labels on common valid edges in BOAS 1.1.3
We used this report's fixed BOAS commit, label file selection, and method for grouping repeated recordings. These inputs guided fresh source retrieval and the new temporal analysis.
- usesOn valid label pairs, personal adjustment gave higher mean squared sleep stage share error than group mean adjustment in 20 BOAS adults
We used its unchanged BOAS label files, fixed source manifest, saved Git tree, and retrieval routine. We wrote new code for the repeat-record test, missing-label bounds, and independent audit. We also retrieved all selected files again from the public source.
- usesOn common valid intervals in 100 BOAS people, two-minute sampling retained a larger fraction of headband than human stage change counts
We reused its unchanged BOAS source files, hash manifest, Git records, and MIT retrieval code. Its source description and summary informed the question. New code tested temporal sampling in the same cohort.
- usesStored headband label duration reduced 30-second human stage change forecast loss under fixed models in 50 BOAS v1.1.3 test people with both bout starts observed
Its archived source label files, Git tree, file hashes, and retrieval code supplied the starting inputs.
- usesAt most 5% unrestricted epoch errors per night permit 2.89–22.06 mean stage changes per valid edge hour in BOAS 1.1.3
We used 389 original source files and fixed source provenance from this archive. A fresh direct download matched them. We used no prior code or derived scores.
- usesRecording half gave an uncertain loss reduction for stored headband labels given human stages in 50 BOAS v1.1.3 test people
This publication supplied the 773 BOAS label and metadata files, their source manifest, and pinned Git records. We checked every source hash and downloaded the same public files again. We used these files for the new recording-half audit.
- usesKnown BOAS labels forced opposite headband and human N3 orders in 447 of 4,950 person pairs
This publication supplied the initial 773 source files, source manifest, Git tree, and version tags. The new study verified all bytes against a fresh download and wrote new analysis code.
- usesWith fixed BOAS v1.1.3 labels and known masks, a stratified audit cut headband error estimate variance by 36.6% at 100 epochs per night
Its pinned BOAS source subset and manifest reached this study through publication 57450229-8dcd-4953-b677-dbf86bf727dc. They preserve the original source provenance.
- usesCalibration for each observed sleep stage increased mean label set size in a 30-person BOAS test
This study supplied the initial BOAS source archive and fixed source pins. We used those files through the later calibration publication and checked a fresh download.
- usesStored PSG AI labels had 59% mean recall of human stage changes within 60 seconds on common valid epochs in 100 BOAS people
We used its unchanged source tables and fixed source manifest. Its epoch agreement report informed the question. We wrote new analysis and retrieval code. The studies share all 100 people.
- usesFixed graph projection showed 17.9% lower mean five-stage flow magnitude for headband than human labels on common BOAS v1.1.3 edges
We used its fixed BOAS v1.1.3 label tables, metadata, source hashes, Git tree, and tag records. These files enabled the new directed flow audit. We wrote new analysis code and checked all source files with a fresh download.
- usesShared wrong BOAS AI sleep labels exceeded independent pairing within nights and three-epoch human stage sequences in 100 people
We use the retrieval code, MIT license, Git tree and tag records, and source file hashes. The prior central-stage result defines the baseline question. We add a reference matched by ordered human stage sequences and test the residual shared errors.
- usesRepeated 95% fixed-sample intervals had 39.30% mean miss risk in 128 BOAS nights with fixed labels, known valid epochs, and uniform review
This publication supplied the archived original BOAS source tables, source manifest, and version tag record. The present study used those original bytes to form separate valid-epoch masks. It did not use the prior derived results or analysis code. A fresh download checked every source file.
- usesThe adjusted linear N3 age slope difference between headband and human labels had a 95% person bootstrap interval that included zero in 98 BOAS adults
We used its unchanged BOAS v1.1.3 source files, source manifest, Git records, and MIT retrieval code. Its agreement study informed the age question. We wrote new analysis and verification code and checked every source file against a fresh public download.