DOSE-I: A Multimodal Biosignal Dataset of Procedural Sedation for Endoscopy

DOSE-I is a multimodal clinical biosignal dataset recorded during routine gastrointestinal and bronchoscopic endoscopy under propofol sedation. It contains 78.5 hours of data from 171 recordings, with a mean length of 27.5 minutes. Recordings include two-channel fronto-temporal EEG sampled at 125 Hz, ECG, plethysmography, ECG-derived respiration, and blood pressure. Propofol administration was annotated in 10 mg increments, alongside endoscopy and other clinical event markers.

To the best of our knowledge, DOSE-I is the first public large-scale EEG-based multimodal dataset of continuous sedation with temporally dense annotation of propofol sedation depth.

Sedation depth (MOAA/S) and state of consciousness (SoC) were assessed repeatedly at the bedside by trained study personnel. The dataset contains 7,328 sedation-depth assessments using a five-level Modified Observer's Assessment of Alertness/Sedation (MOAA/S) scale and 1,129 annotated transitions of consciousness. Recordings are accompanied by subject information (static data), e.g. age, sex, BMI, ASA physical status, relevant medication classes, preexisting conditions, handedness and endoscopy type. For quality assessment, DOSE-I additionally provides metadata, overview visualization plots for each recording, EEG artifact annotations and artifact summary plots.

For machine learning use, the dataset provides 40 precomputed processed EEG parameters (pEEG) at 1 Hz. These include, among others, absolute and relative band power, phase synchronisation, entropy, spectral and median frequency measures, and bispectral features. Their suitability for classification modelling is demonstrated in the AIME 2026 conference paper introducing the dataset, "Towards Predicting Sedation Depth in Endoscopy with Large Clinically Annotated EEG Data of Continuous Propofol Sedation".

Technical details relevant to dataset use are provided in the accompanying dataset documentation "DOSE-I: A Multimodal Biosignal Dataset of Procedural Sedation for Endoscopy -- Technical Report" and on the dataset website.

Suggested Citation

Garbe, J., Nguyen, Q. V., Kantelhardt, J. W., Dünninghaus, F., Erffmeier, K., Seeliger, K., & Schmid, T. (2027). Towards Predicting Sedation Depth in Endoscopy with Large Clinically Annotated EEG Data of Continuous Propofol Sedation. In: Andreev, P., Van Woensel, W., Holmes, J., Sauré, A. (eds) Artificial Intelligence in Medicine. AIME 2026. Lecture Notes in Computer Science, vol 16749. Springer, Cham. https://doi.org/10.1007/978-3-032-30813-9_58

Garbe, J., Nguyen, Q. V., Kantelhardt, J. W., Dünninghaus, F., Erffmeier, K., Seeliger, K., & Schmid, T. (2026). DOSE-I: A Multimodal Biosignal Dataset of Procedural Sedation for Endoscopy [Data set]. 24th International Conference on Artificial Intelligence in Medicine (AIME 2026), Ottawa, Canada. Zenodo. https://doi.org/10.5281/zenodo.18483292