Automated COVID-19 classification using ultrasound

Start date: 01-10-2020
End date: 31-03-2021

Clinical problem

The COVID-19 pandemic causes health care challenges around the world. Its major morbidity and mortality is caused by COVID-19 pneumonia and respiratory failure. Fast diagnosis and classification of the severity of the disease is of vital importance. A physical examination and chest X-ray are not sufficient to determine the severity of the disease. Most patients with a suspicion of COVID-19 will therefore undergo a CT scan, but this results in radiation exposure and is only performed within a radiology department of a hospital which limits its accessibility. Ultrasound imaging could resolve these problems, since it is does not use ionizing radiation and can be performed at the point-of-care. Moreover, in COVID-19 pneumonia it has comparable diagnostic accuracy compared to CT. Ultrasound devices have become very portable and can be connected to laptops, tablets and even smartphones. Several point-of-care ultrasound (POCUS) devices have become available on the market in recent years. Some examples are: Butterfly iQ, Philips Lumify, Telemed MicrUs Pro and the Clarius C3. Ultrasound requires a trained sonographer that can both acquire and interpret the images, which limits the use of these devices outside the hospital environment.

Solution

In this project we will make use of a standardized acquisition protocol and develop AI algorithms that will automatically interpret the ultrasound images. This makes it possible to widely perform lung ultrasound for patients suspected of COVID-19 with minimal amount of training. This would enable fast triage also outside hospitals and would make follow-up of patients over time without any radiation burden possible.

Tasks

  • Learn about ultrasound imaging and the acquisition protocol that is used
  • Development of deep learning algorithms
  • Evaluation of the developed algorithms
  • Optimize the algorithms on computational efficiency
  • Development of a web application for automated classification of COVID-19

Innovation

When the algorithms have sufficient sensitivity and accuracy, and are able to run on a smartphone, they can be used for development of a point-of-care ultrasound solution that makes it possible for health-care personnel to perform the predefined acquisition protocol with minimal amount of training, since the developed algorithm will perform the automated interpretation. This could enable wide spread use of POCUS for fast triage of COVID-19 suspected patients. The development of a web-based application will enable widespread use of the state-of-the-art algorithm by clincians and health-care personnel.

People

Sjoerd Bos

Sjoerd Bos

Master Student

Thomas van den Heuvel

Thomas van den Heuvel

Postdoctoral Researcher

Luca Ambrogioni

Luca Ambrogioni

Assistant professor

Artificial Intelligence, Radboud University

Bram Kok

Bram Kok

Internist in training

Internal Medicine, Radboudumc

Frank Bosch

Frank Bosch

Professor

Emergency Medicine, Radboudumc

Chris de Korte

Chris de Korte

Professor

Medical UltraSound Imaging Centre