Fig. 1: Overview of the RMS decision support system for respiratory state management, and its extension for ICU-level ventilator resource planning.
From: RMS: a ML-based system for ICU respiratory monitoring and resource planning
a Flow diagram for the development of RMS predictors at the individual patient level. Time series were extracted from the HiRID-II dataset and gridded to a 5-min resolution, and features were computed. Respiratory failure/ventilation/ready-to-extubate periods were annotated and machine learning labels created. b The respiratory monitoring system consists of four scores which are active during different time periods of the ICU stay, according to the current respiratory and ventilation state of the patient. c Flow diagram for the development of a resource monitoring system at the ICU level. For all current patients in the ICU, the four scores were integrated to predict the probability that a patient will require mechanical ventilation within a future time horizon. d Example time period of 3 months, displaying the actual number of ventilated patients and the predicted number as estimated by RMS in the next 8–16 h. e Overview of prediction tasks solved by RMS for individual patients (RMS-RF/RMS-EF/RMS-MVStart/RMS-MVEnd) as well as at the ICU level. For RF, MVStart and MVEnd we provide the event prevalences in the test set at times when the patient is stable, not ventilated, or ventilated, respectively.
