Deep Breathe Interns Fall 21

Lung ultrasound (LUS) is a portable, non-ionizing imaging method with high accuracy for numerous respiratory conditions.1-3 These properties allow LUS – unlike other lung imaging techniques – to be delivered at scale and outside of traditional healthcare environments that may allow for broadly distributed diagnosis, triage or prognosis of respiratory conditions.4–11 As training for LUS is not widespread, it can be expected that such democratization will eventually rely on deep learning (DL), computer vision solutions to mimic real time expert oversight.
Despite the potential for DL to accelerate the scale and reach of LUS, this domain of study is complicated by a paucity of well-organized, institutional LUS datasets.
With an archive of more than 100,000 LUS clips, our group is focused on the development of computer vision models that will contribute to automated interpretation that will assist in the broadest dissemination of the tool. The current project for which we are applying, is directed at the detection of pneumothorax – a life threatening collapse of the lung – that can be readily detected or ruled out using LUS. In fact, LUS has comparable accuracy to a CT scan for this condition.

Faculty Supervisor:

Alexander Wong

Student:

Partner:

Deep Breathe

Discipline:

Life Sciences

Sector:

Professional, scientific and technical services

University:

University of Waterloo

Program: