Supporting lung protective ventilation with predictive analytics in ICU using HPC

Rachael Hagan, Charles Gillan, Murali Shyamsundar, Ivor Spence

Research output: Contribution to conferencePosterpeer-review


We present a new methodology to generate predictive alerts indicating violation of the clinically accepted threshold for the metric tidal volume per unit of ideal body weight, a value that is used to enable lung protective ventilation of patients in an intensive care unit. We process streams of patient respiratory data recorded per minute from ventilators in an ICU and apply several state of the art time series prediction methods to forecast the behavior of the tidal volume metric per patient.

Utilizing the NI Tier 2 Kelvin supercomputer, we can train and test neural network models for prediction and further test the scalability of our work being used by multiple patients in multiple locations. Our results show that boosted regression performs better than other methods that we investigated, however we found that long short term memory neural networks can offer similar levels of accuracy when sufficient training data is available. Our work suggests that with three hours of continuous ventilator readings for a patient, the boosted regression is a viable technique for prediction of the progression of tidal volume behavior within 10% accuracy.
Original languageEnglish
Publication statusPublished - 23 Jun 2020
EventISC High Performance 2020 - Digital
Duration: 22 Jun 202025 Jun 2020


ConferenceISC High Performance 2020
Abbreviated titleISC 2020
Internet address


Dive into the research topics of 'Supporting lung protective ventilation with predictive analytics in ICU using HPC'. Together they form a unique fingerprint.

Cite this