IntelliLung: A human-in-the-loop AI-DSS for ICU ventilation settings

Robert Huhle^1,2, Thea Koch¹, Muhammad Hamza Yousuf^1,3, Jaume Montanya¹, Jens Lehmann¹, Martin Faltys¹, Dejan Paunović^1,4, Jakob Wittenstein^1,5, Valentina Janev^1,4, Sahar Vahdati¹, Miloš Nenadović^1,4*, Jason Li^1,3, Álvaro Ritoré Hidalgo¹ and Miguel Ángel Armengol de la Hoz¹

¹IntelliLung Consortium

²Technische Universität Dresden

³Institut für Angewandte Informatik

⁴Institute Mihajlo Pupin

⁵University Hospital Carl Gustav Carus

milos.nenadovic [at] pupin.rs

Abstract

IntelliLung is a human-in-the-loop AI decision-support system for optimizing mechanical ventilation in intensive care units (ICUs) using bedside and physiologic monitor streams. Patient data are ingested through a secure on-premises platform that normalizes, stores, and governs data transfer to ensure traceability, auditability, and regulatory compliance. Using a data aggregator that converts inputs into hourly patient-state representations, an AI engine generates hybrid recommendations that jointly propose ventilator mode selection and coordinated adjustments of key parameters. Recommendations are delivered via a clinician-facing interface embedded within existing ICU workflows, where each suggestion can be explicitly accepted or rejected. IntelliLung functions strictly as an advisory system. The demonstration replays realistic ICU patient trajectories to showcase system behavior, recommendation progression under changing physiology, and mechanisms for safe human-AI collaboration in clinical settings.

Keywords: DSS, ventilation, ICU, human-in-the-loop, AI

Acknowledgement: Our work is funded by the European Union IntelliLung project with Grant Agreement No. 101057434.