Development and evaluation of a reinforcement learning-based decision support system for mechanical ventilation in acute respiratory failure
Contributing person | Jakob Wittenstein | |
Contributing person | Raphael Theilen | |
Contributing person | Franziska Fischer | |
Contributing person | Muhammut Hamza Yousuf | |
Contributing person | Jason Li | |
Contributing person | Sahar Vahdati | |
Contributing person | Robert Huhle | |
Countries to which the data refer | GERMANY | |
Countries to which the data refer | UNITED STATES OF AMERICA | |
References to related material | https://github.com/NIMI-research/rl-acute-respiratory-failure-dss | |
Description of the data | further information may be found in dataset_description.md | |
Type of the data | Dataset | |
Total size of the dataset | 44793886 | |
Author | Huhle, Robert | |
Upload date | 2026-09-12T14:52:35Z | |
Publication date | 2026-09-12T14:52:35Z | |
Data of data creation | 2024-10-14 | |
Publication date | 2026-09-12 | |
Abstract of the dataset | This data publication contains development (Department of Anesthesiology and Intensive Care Medicine, Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Local Ethical Committee Approval: BO-EK-423082021) and external validation data set (extracted and preprocessed from MIMIC-IV) for the developed first version of the IntelliLung-AI algorithm based on Reinforcement Learning. Data holds the input/output of the training, testing (development), as well as external validation pipeline as published in the GitLab repository (https://github.com/NIMI-research/rl-acute-respiratory-failure-dss). | |
Public reference to this page | https://opara.zih.tu-dresden.de/handle/123456789/1906 | |
Public reference to this page | https://doi.org/10.25532/OPARA-1045 | |
Publisher | Technische Universität Dresden | |
Licence | Attribution-NonCommercial-NoDerivatives 4.0 International | en |
URI of the licence text | http://creativecommons.org/licenses/by-nc-nd/4.0/ | |
Specification of the discipline(s) | 2::22::205::205-11 | |
Title of the dataset | Development and evaluation of a reinforcement learning-based decision support system for mechanical ventilation in acute respiratory failure | |
Project abstract | Rationale: It is unknown whether artificial intelligence might be useful to guide mechanical ventilation in acute respiratory failure. Objectives: To develop and compare the performance of an artificial intelligence-based decision support system (IntelliLung) with clinical practice and current recommendations. Methods, Measurements and Main Results: Data from 751 adult patients with acute respiratory failure and treated with invasive mechanical ventilation at the University Hospital Dresden between 2010 and 2020 were used to develop IntelliLung. An offline reinforcement learning algorithm (batch-constrained deep Q-learning) was used to identify settings of positive end-expiratory pressure (PEEP), inspiratory fraction of oxygen (FIO2), and respiratory rate that were able to optimize a reward function consisting of short-term physiological variables and ICU mortality. IntelliLung recommendations were compared with ventilation settings chosen by clinicians in another 191 patients treated in the same period. Additionally, using a cross-over design, IntelliLung was compared to ARDSnet protocol recommendations in four pigs undergoing sequential lung injury with lung saline washout (LAV), ventilator-induced lung injury (VILI), and lipopolysaccharide infusion (LPS). In patients, IntelliLung more frequently suggested lower PEEP (<10cmH2O), lower FIO2 (≤40%) and a respiratory rate <18min-1 and 25-29min-1 compared to clinicians. In animals, IntelliLung proposed higher FIO2 during LAV, (43±22% vs 37±21%, p=0.038), higher PEEP during VILI (7.4±1.3cmH2O vs 5.2±0.7cmH2O, p<0.001), and similar settings during LPS compared to the ARDSnet protocol. Mechanical power and hemodynamics did not differ between ventilation strategies. Conclusions: The recommendations of an artificial intelligence-based algorithm for mechanical ventilation in acute respiratory failure differed from clinical practice and current recommendations. | |
Funding Acknowledgement | Funded by EKFZ for Digital Health, TU Dresden and the European Union Horizon 2021 (Grant agreement ID: 101057434 DOI:[10.3030/101057434](https://doi.org/10.3030/101057434)) | |
Public project website(s) | https://intellilung-project.eu/ | |
Project title | Development and evaluation of a reinforcement learning-based decision support system for mechanical ventilation in acute respiratory failure |
Files
License bundle
- Name:
- license.txt
- Size:
- 3.86 KB
- Format:
- Item-specific license agreed to upon submission
- Description:
Collections
