AI/ ai · open-source · healthcare · datasets

A Free Biosignal Dataset Aims to Teach AI When Patients Are Sedated

DOSE-I packages 78.5 hours of endoscopy sedation recordings, with consciousness-transition labels, to help researchers train monitoring models.

A new open dataset called DOSE-I gives researchers annotated biosignal recordings from patients under procedural sedation during endoscopy.

Published on Zenodo, DOSE-I covers 281 endoscopic procedures across 171 records, totaling 78.5 hours of data. Individual records range from 6.7 to 70.8 minutes. The dataset includes 1,129 transitions of consciousness — a median of 6 per record — and 7,328 sedation depth labels, with a median of 39 labels per record. Beyond the raw biosignals, it ships with static subject data, recording metadata, artifact detection notes, preprocessed pEEG features, and C preprocessing code on GitHub.

Sedation monitoring is an area where AI-assisted tools have real clinical stakes: misjudging depth can mean a patient wakes mid-procedure or receives too much drug. Labeled, multimodal biosignal datasets are scarce, which limits how far researchers can push model development — DOSE-I is a direct attempt to close that gap. The pEEG component is worth watching specifically, since processed EEG is one of the more promising signals for real-time consciousness inference.

For now, this is a technical report describing a dataset, not a model with clinical results — so the cautious read is that the hard work of turning these labels into reliable monitoring software still lies ahead.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →