ez-vent: AI solution to help critical care specialists optimise mechanical ventilation settings

Project Reference :


Institution :

National University of Singapore (NUS), Singapore General Hospital (SGH)

Principal Investigator :

Prof Feng Mengling

Technology Readiness :

3 (Experimental proof of concept)

Technology Categories :

AI-Reinforcement Learning

Background/Problem Statement

In critical care medicine, mechanical ventilation is pivotal for ICU patients, especially amid increasing demand driven by factors like the COVID-19 pandemic and aging demographics. Despite extensive research, choosing the best ventilator strategy remains challenging due to patient variability. Current clinical guidelines lack personalization, neglecting essential factors such as personalized settings for ventilation like PEEP, FiO2, and ideal body weight-adjusted tidal volume.

The AI in critical care market is projected to reach a valuation of US$ 50 Billion by 2030 and the mechanical ventilator market is projected to reach a valuation of US$ 10.5 Billion by 2033.


The reinforced learning AI model (EZ-Vent), developed in close collaboration with SGH, uses Batch Constrained deep Q-learning (BCQ) algorithm to:

  1. learn and create an optimal ventilation policy for critically ill patients reliant on mechanical ventilation.
  2. recommend the optimal ventilation settings for levels of PEEP, FiO2, and ideal body weight-adjusted tidal volume by taking into account the individual patient’s conditions including their demographic features, physiological status, and multiple comorbidities.

Link to publication:

Reinforcement learning to help intensivists optimize mechanical ventilation settings (EZ-Vent): Derivation and validation using large databases


  1. Trained on extensive ICU data, EZ-Vent excels in learning ventilation policies, outperforming observed physician policies based on rigorous evaluation metrics. It demonstrates potential to surpass current clinical practices, providing valuable support to physicians in enhancing treatment decisions for critically ill patients requiring respiratory support.
  2. Compared to the existing guidelines, EZ-Vent can adjust ventilation treatment recommendations based on changes in a patient’s condition.
  3. EZ-Vent employs batch learning, drawing insights from fixed datasets without direct interaction with real patients. Unlike state-of-the-art reinforcement learning algorithms that struggle in a batch setting, leading to overestimation and diminished performance with new data, EZ-Vent utilizes the BCQ algorithm to overcome these challenges. The BCQ algorithm ensures that the learned policy closely aligns with physicians’ decisions. Consequently, EZ-Vent holds promise as a clinical decision support (CDS) tool, assisting physicians in making enhanced decisions for critically ill patients requiring mechanical ventilation.

Potential Application(s)

EZ-Vent can be applied as a clinical decision support tool that helps physicians make better treatment decisions and to improve the survival and prognosis of critically ill patients requiring invasive respiratory support.

We welcome interest from the industry for collaboration/ co-development / customisation of the technology into a new product or service. If you have any enquiries or are keen to collaborate, please contact us.