A foundational transformer leveraging full night, multichannel sleep study data accurately classifies sleep stages
Bibliographic record
Abstract
STUDY OBJECTIVES: To evaluate whether a foundational transformer using 8-hour, multichannel polysomnogram (PSG) data can effectively encode signals and classify sleep stages with state-of-the-art performance. METHODS: The Sleep Heart Health Study, Wisconsin Sleep Cohort, and Osteoporotic Fractures in Men (MrOS) Study visit 1 were used for training, and the Multi-Ethnic Study of Atherosclerosis (MESA), Apnea Positive Pressure Long-term Efficacy Study (APPLES), and MrOS visit 2 served as independent test sets. We developed PFTSleep, a self-supervised foundational transformer that encodes full-night sleep studies with brain, movement, cardiac, oxygen, and respiratory channels. These representations were used to train another model to classify sleep stages. We compared our results to existing methods, examined differences in performance by varying channel input data and training dataset size, and investigated an AI explainability tool to analyze decision processes. RESULTS: PFTSleep was trained with 13 888 sleep studies and tested on 4169 independent studies. Cohen's Kappa scores were 0.81 for our held-out set, 0.59 for APPLES, 0.60 for MESA, and 0.75 for MrOS Visit 2. Performance increases to 0.76 on a held-out MESA set when MESA is included in the training of the classifier head but not the transformer. Compared to other state-of-the-art AI models, our model shows high performance across diverse datasets while only using task-agnostic PSG representations from a foundational transformer as input for sleep stage classification. CONCLUSIONS: Full night, multichannel PSG representations from a foundational transformer enable accurate sleep stage classification comparable to state-of-the-art AI methods across diverse datasets.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".