Python in Medicine: A toolkit for exploring neonatal ventilator data

David Chong
Gusztav Belteki

Modern ventilators generate large amounts of pressure and flow data that clinicians cannot realistically monitor over long periods of time. A toolkit for segmentation and extracting breath-wise features would provide a means for summarising this information and investigating the relationship of different ventilator and patient characteristics to individual breaths.

Mechanical ventilation is an essential therapy during intensive care for patients who are unable to breathe alone. Modern mechanical ventilators (breathing machines) generate large amount of data about the airway pressure and flow used during lung inflations and the volume of ventilator inflations (breaths). These data are displayed on the ventilator screen in real time but traditionally they are not downloaded or stored in the longer term. The newest models of ventilators allow for downloading these raw data with a high sampling rate (100 Hz or more). These data can be used to analyse how the ventilator has been performing and how it has been interacting with the patient. However, analysis of these large datasets requires computational tools.

Using Python, we have developed a toolkit to analyse these high throughput data obtained from neonatal ventilators in more detail. The toolkit allows for segmenting the ventilator data streaming continuously from the medical device into individual ventilator inflations (breaths). It also extracts important features of these inflations such as duration of different parts (lung inflation time, deflation time etc) and the presence or absence of ventilator-patient interactions. We would like to use these tools to develop quantitative indicators of how comfortable the baby has been during the mechanical ventilation. Such indicators would be of great interest to clinicians.