Paperdoi:10.5281/zenodo.23638031
Predicting hypotension from a pulse waveform: masked-reconstruction pretrained transformer on VitalDB PPG (Zenodo report 2026)
Zenodo report (record 23638032; concept DOI 10.5281/zenodo.23638031), 2026-10-25; Digital Twin Biomedical
Licence: CC BY 4.0
TL;DR
An 85.1 M-parameter pre-norm transformer (width 768, depth 12, 12 heads) is pretrained by masked reconstruction on VitalDB intraoperative waveforms (2× NVIDIA B200, 38.2 min). It is then fine-tuned to predict, from 30 s of PPG, whether MAP falls below 65 mmHg within 5 min. Patients are split by hash of subject ID (80/10/10, no patient overlap).
- Form factor: clinical_finger (intraoperative)
Abstract
An 85-million-parameter transformer was pretrained by masked reconstruction on intraoperative waveforms from VitalDB, then fine-tuned to predict hypotension. Pretraining used two NVIDIA B200 GPUs for 38.2 minutes. On patients held out of training, 30 seconds of photoplethysmography predicted whether mean arterial pressure would fall below 65 mmHg within 5 minutes (test AUROC 0.693), compared with 0.676 for gradient-boosted trees on hand-crafted pulse features and 0.641 for the same transformer trained from scratch.
Verbatim from DataCite (opens in a new tab) · Licence: https://creativecommons.org/licenses/by/4.0/legalcode
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Tasks
Signals
Keywords
Datasets
How it works
Cite this paper
APA
Smith, I. (2026). Predicting hypotension from a pulse waveform: masked-reconstruction pretrained transformer on VitalDB PPG (Zenodo report 2026). Zenodo report , 2026-10-25. https://doi.org/10.5281/zenodo.23638031
BibTeX
@article{smith2026predicting,
title = {{Predicting hypotension from a pulse waveform: masked-reconstruction pretrained transformer on VitalDB PPG (Zenodo report 2026)}},
author = {Smith, Isak},
year = {2026},
journal = {Zenodo report , 2026-10-25},
doi = {10.5281/zenodo.23638031}
}Built from the hub record. Check it against the publisher before submitting.
Added to the ChatPPG research hub
