Diagnosing COVID19 with the available clinical tools has proven to be effective when patients have an evolution of the infection and therefore, a clinical presentation of patients (lung image positive image, viral RNA detection or associated symptoms). However, during the initial states of the infection, viral RNA detection is not a reliable tool. In these cases, lung imaging becomes central.

What it does

We focused on POCUS ultrasound images, because ultrasound imaging is an easy, cost-effective an non-invasive method of testing available in almost any medical facility. Our model can identify with a 93% recall and 96% precision a Covid-19 POCUS image. We also run it on X-Ray and it should be working with CT’s data as well but the data available online did not permit us to conclude.

How we built it

There have been some research to check what existed (X-ray identification is already an open source work in progress, but still we need more data). CT’s identification has been studied by Alibaba but nothing is available online. And we found nothing on POCUS. Then we looked for data! And this is the hardest part. And then we trained and tested the model on the data we collected.

Challenges we ran into

GETTING THE DATA !! We want to push for hospitals and private companies to help Covid identification by sharing their data. Help us to achieve it !!

Accomplishments that we are proud of

We are all very proud of our collaboration, mixing a pediatrician, a medical geneticist (with expertise in AI), a deep learning and drug discovery specialist, a medical biophysics and data specialist, an economist data scientist and an actuary data scientist. Everyone brought major contribution to this project and we believe we found something that could actually help covid-19 detection and therefore improve treatments when given at early stage.

What we learned

When we collaborate, we can achieve great things  !

What's next for Automatic detection of COVID-19 from POCUS ultrasound data

Step 1 :Convince data providers to share their data.

Step 2 : Set up a platform to share data and models.

Step 3 : Set up a user friendly platform for hospitals where one could upload any medical images and get a result.

More information:

** More information is available here:  **

Try It out



github, python

Devpost Software Identifier